2026-02-04 14:31:01 -07:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2009-05-21 20:33:58 +01:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-04-12 15:50:21 +02:00
|
|
|
|
2020-06-08 21:33:54 -07:00
|
|
|
|
2021-01-28 19:19:45 +01:00
|
|
|
|
2024-03-26 17:10:31 +00:00
|
|
|
|
2022-09-14 15:18:07 -07:00
|
|
|
|
2023-02-27 09:36:14 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2009-09-16 11:50:15 +02:00
|
|
|
|
2021-04-12 15:50:21 +02:00
|
|
|
|
2011-06-27 16:18:09 -07:00
|
|
|
|
2009-09-16 11:50:15 +02:00
|
|
|
|
hugetlbfs: use i_mmap_rwsem for more pmd sharing synchronization
Patch series "hugetlbfs: use i_mmap_rwsem for more synchronization", v2.
While discussing the issue with huge_pte_offset [1], I remembered that
there were more outstanding hugetlb races. These issues are:
1) For shared pmds, huge PTE pointers returned by huge_pte_alloc can become
invalid via a call to huge_pmd_unshare by another thread.
2) hugetlbfs page faults can race with truncation causing invalid global
reserve counts and state.
A previous attempt was made to use i_mmap_rwsem in this manner as
described at [2]. However, those patches were reverted starting with [3]
due to locking issues.
To effectively use i_mmap_rwsem to address the above issues it needs to be
held (in read mode) during page fault processing. However, during fault
processing we need to lock the page we will be adding. Lock ordering
requires we take page lock before i_mmap_rwsem. Waiting until after
taking the page lock is too late in the fault process for the
synchronization we want to do.
To address this lock ordering issue, the following patches change the lock
ordering for hugetlb pages. This is not too invasive as hugetlbfs
processing is done separate from core mm in many places. However, I don't
really like this idea. Much ugliness is contained in the new routine
hugetlb_page_mapping_lock_write() of patch 1.
The only other way I can think of to address these issues is by catching
all the races. After catching a race, cleanup, backout, retry ... etc,
as needed. This can get really ugly, especially for huge page
reservations. At one time, I started writing some of the reservation
backout code for page faults and it got so ugly and complicated I went
down the path of adding synchronization to avoid the races. Any other
suggestions would be welcome.
[1] https://lore.kernel.org/linux-mm/1582342427-230392-1-git-send-email-longpeng2@huawei.com/
[2] https://lore.kernel.org/linux-mm/20181222223013.22193-1-mike.kravetz@oracle.com/
[3] https://lore.kernel.org/linux-mm/20190103235452.29335-1-mike.kravetz@oracle.com
[4] https://lore.kernel.org/linux-mm/1584028670.7365.182.camel@lca.pw/
[5] https://lore.kernel.org/lkml/20200312183142.108df9ac@canb.auug.org.au/
This patch (of 2):
While looking at BUGs associated with invalid huge page map counts, it was
discovered and observed that a huge pte pointer could become 'invalid' and
point to another task's page table. Consider the following:
A task takes a page fault on a shared hugetlbfs file and calls
huge_pte_alloc to get a ptep. Suppose the returned ptep points to a
shared pmd.
Now, another task truncates the hugetlbfs file. As part of truncation, it
unmaps everyone who has the file mapped. If the range being truncated is
covered by a shared pmd, huge_pmd_unshare will be called. For all but the
last user of the shared pmd, huge_pmd_unshare will clear the pud pointing
to the pmd. If the task in the middle of the page fault is not the last
user, the ptep returned by huge_pte_alloc now points to another task's
page table or worse. This leads to bad things such as incorrect page
map/reference counts or invalid memory references.
To fix, expand the use of i_mmap_rwsem as follows:
- i_mmap_rwsem is held in read mode whenever huge_pmd_share is called.
huge_pmd_share is only called via huge_pte_alloc, so callers of
huge_pte_alloc take i_mmap_rwsem before calling. In addition, callers
of huge_pte_alloc continue to hold the semaphore until finished with
the ptep.
- i_mmap_rwsem is held in write mode whenever huge_pmd_unshare is called.
One problem with this scheme is that it requires taking i_mmap_rwsem
before taking the page lock during page faults. This is not the order
specified in the rest of mm code. Handling of hugetlbfs pages is mostly
isolated today. Therefore, we use this alternative locking order for
PageHuge() pages.
mapping->i_mmap_rwsem
hugetlb_fault_mutex (hugetlbfs specific page fault mutex)
page->flags PG_locked (lock_page)
To help with lock ordering issues, hugetlb_page_mapping_lock_write() is
introduced to write lock the i_mmap_rwsem associated with a page.
In most cases it is easy to get address_space via vma->vm_file->f_mapping.
However, in the case of migration or memory errors for anon pages we do
not have an associated vma. A new routine _get_hugetlb_page_mapping()
will use anon_vma to get address_space in these cases.
Signed-off-by: Mike Kravetz <mike.kravetz@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Hugh Dickins <hughd@google.com>
Cc: Naoya Horiguchi <n-horiguchi@ah.jp.nec.com>
Cc: "Aneesh Kumar K . V" <aneesh.kumar@linux.vnet.ibm.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: "Kirill A . Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Davidlohr Bueso <dave@stgolabs.net>
Cc: Prakash Sangappa <prakash.sangappa@oracle.com>
Link: http://lkml.kernel.org/r/20200316205756.146666-2-mike.kravetz@oracle.com
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2020-04-01 21:11:05 -07:00
|
|
|
|
2022-09-14 15:18:07 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-03-26 17:10:31 +00:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-02-08 18:51:29 +01:00
|
|
|
|
2017-02-08 18:51:36 +01:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
2025-11-10 22:21:28 +00:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
ksm: let shared pages be swappable
Initial implementation for swapping out KSM's shared pages: add
page_referenced_ksm() and try_to_unmap_ksm(), which rmap.c calls when
faced with a PageKsm page.
Most of what's needed can be got from the rmap_items listed from the
stable_node of the ksm page, without discovering the actual vma: so in
this patch just fake up a struct vma for page_referenced_one() or
try_to_unmap_one(), then refine that in the next patch.
Add VM_NONLINEAR to ksm_madvise()'s list of exclusions: it has always been
implicit there (being only set with VM_SHARED, already excluded), but
let's make it explicit, to help justify the lack of nonlinear unmap.
Rely on the page lock to protect against concurrent modifications to that
page's node of the stable tree.
The awkward part is not swapout but swapin: do_swap_page() and
page_add_anon_rmap() now have to allow for new possibilities - perhaps a
ksm page still in swapcache, perhaps a swapcache page associated with one
location in one anon_vma now needed for another location or anon_vma.
(And the vma might even be no longer VM_MERGEABLE when that happens.)
ksm_might_need_to_copy() checks for that case, and supplies a duplicate
page when necessary, simply leaving it to a subsequent pass of ksmd to
rediscover the identity and merge them back into one ksm page.
Disappointingly primitive: but the alternative would have to accumulate
unswappable info about the swapped out ksm pages, limiting swappability.
Remove page_add_ksm_rmap(): page_add_anon_rmap() now has to allow for the
particular case it was handling, so just use it instead.
Signed-off-by: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Cc: Izik Eidus <ieidus@redhat.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Chris Wright <chrisw@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-12-14 17:59:24 -08:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
2011-10-16 02:01:52 -04:00
|
|
|
|
2008-02-07 00:13:53 -08:00
|
|
|
|
mmu-notifiers: core
With KVM/GFP/XPMEM there isn't just the primary CPU MMU pointing to pages.
There are secondary MMUs (with secondary sptes and secondary tlbs) too.
sptes in the kvm case are shadow pagetables, but when I say spte in
mmu-notifier context, I mean "secondary pte". In GRU case there's no
actual secondary pte and there's only a secondary tlb because the GRU
secondary MMU has no knowledge about sptes and every secondary tlb miss
event in the MMU always generates a page fault that has to be resolved by
the CPU (this is not the case of KVM where the a secondary tlb miss will
walk sptes in hardware and it will refill the secondary tlb transparently
to software if the corresponding spte is present). The same way
zap_page_range has to invalidate the pte before freeing the page, the spte
(and secondary tlb) must also be invalidated before any page is freed and
reused.
Currently we take a page_count pin on every page mapped by sptes, but that
means the pages can't be swapped whenever they're mapped by any spte
because they're part of the guest working set. Furthermore a spte unmap
event can immediately lead to a page to be freed when the pin is released
(so requiring the same complex and relatively slow tlb_gather smp safe
logic we have in zap_page_range and that can be avoided completely if the
spte unmap event doesn't require an unpin of the page previously mapped in
the secondary MMU).
The mmu notifiers allow kvm/GRU/XPMEM to attach to the tsk->mm and know
when the VM is swapping or freeing or doing anything on the primary MMU so
that the secondary MMU code can drop sptes before the pages are freed,
avoiding all page pinning and allowing 100% reliable swapping of guest
physical address space. Furthermore it avoids the code that teardown the
mappings of the secondary MMU, to implement a logic like tlb_gather in
zap_page_range that would require many IPI to flush other cpu tlbs, for
each fixed number of spte unmapped.
To make an example: if what happens on the primary MMU is a protection
downgrade (from writeable to wrprotect) the secondary MMU mappings will be
invalidated, and the next secondary-mmu-page-fault will call
get_user_pages and trigger a do_wp_page through get_user_pages if it
called get_user_pages with write=1, and it'll re-establishing an updated
spte or secondary-tlb-mapping on the copied page. Or it will setup a
readonly spte or readonly tlb mapping if it's a guest-read, if it calls
get_user_pages with write=0. This is just an example.
This allows to map any page pointed by any pte (and in turn visible in the
primary CPU MMU), into a secondary MMU (be it a pure tlb like GRU, or an
full MMU with both sptes and secondary-tlb like the shadow-pagetable layer
with kvm), or a remote DMA in software like XPMEM (hence needing of
schedule in XPMEM code to send the invalidate to the remote node, while no
need to schedule in kvm/gru as it's an immediate event like invalidating
primary-mmu pte).
At least for KVM without this patch it's impossible to swap guests
reliably. And having this feature and removing the page pin allows
several other optimizations that simplify life considerably.
Dependencies:
1) mm_take_all_locks() to register the mmu notifier when the whole VM
isn't doing anything with "mm". This allows mmu notifier users to keep
track if the VM is in the middle of the invalidate_range_begin/end
critical section with an atomic counter incraese in range_begin and
decreased in range_end. No secondary MMU page fault is allowed to map
any spte or secondary tlb reference, while the VM is in the middle of
range_begin/end as any page returned by get_user_pages in that critical
section could later immediately be freed without any further
->invalidate_page notification (invalidate_range_begin/end works on
ranges and ->invalidate_page isn't called immediately before freeing
the page). To stop all page freeing and pagetable overwrites the
mmap_sem must be taken in write mode and all other anon_vma/i_mmap
locks must be taken too.
2) It'd be a waste to add branches in the VM if nobody could possibly
run KVM/GRU/XPMEM on the kernel, so mmu notifiers will only enabled if
CONFIG_KVM=m/y. In the current kernel kvm won't yet take advantage of
mmu notifiers, but this already allows to compile a KVM external module
against a kernel with mmu notifiers enabled and from the next pull from
kvm.git we'll start using them. And GRU/XPMEM will also be able to
continue the development by enabling KVM=m in their config, until they
submit all GRU/XPMEM GPLv2 code to the mainline kernel. Then they can
also enable MMU_NOTIFIERS in the same way KVM does it (even if KVM=n).
This guarantees nobody selects MMU_NOTIFIER=y if KVM and GRU and XPMEM
are all =n.
The mmu_notifier_register call can fail because mm_take_all_locks may be
interrupted by a signal and return -EINTR. Because mmu_notifier_reigster
is used when a driver startup, a failure can be gracefully handled. Here
an example of the change applied to kvm to register the mmu notifiers.
Usually when a driver startups other allocations are required anyway and
-ENOMEM failure paths exists already.
struct kvm *kvm_arch_create_vm(void)
{
struct kvm *kvm = kzalloc(sizeof(struct kvm), GFP_KERNEL);
+ int err;
if (!kvm)
return ERR_PTR(-ENOMEM);
INIT_LIST_HEAD(&kvm->arch.active_mmu_pages);
+ kvm->arch.mmu_notifier.ops = &kvm_mmu_notifier_ops;
+ err = mmu_notifier_register(&kvm->arch.mmu_notifier, current->mm);
+ if (err) {
+ kfree(kvm);
+ return ERR_PTR(err);
+ }
+
return kvm;
}
mmu_notifier_unregister returns void and it's reliable.
The patch also adds a few needed but missing includes that would prevent
kernel to compile after these changes on non-x86 archs (x86 didn't need
them by luck).
[akpm@linux-foundation.org: coding-style fixes]
[akpm@linux-foundation.org: fix mm/filemap_xip.c build]
[akpm@linux-foundation.org: fix mm/mmu_notifier.c build]
Signed-off-by: Andrea Arcangeli <andrea@qumranet.com>
Signed-off-by: Nick Piggin <npiggin@suse.de>
Signed-off-by: Christoph Lameter <cl@linux-foundation.org>
Cc: Jack Steiner <steiner@sgi.com>
Cc: Robin Holt <holt@sgi.com>
Cc: Nick Piggin <npiggin@suse.de>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Kanoj Sarcar <kanojsarcar@yahoo.com>
Cc: Roland Dreier <rdreier@cisco.com>
Cc: Steve Wise <swise@opengridcomputing.com>
Cc: Avi Kivity <avi@qumranet.com>
Cc: Hugh Dickins <hugh@veritas.com>
Cc: Rusty Russell <rusty@rustcorp.com.au>
Cc: Anthony Liguori <aliguori@us.ibm.com>
Cc: Chris Wright <chrisw@redhat.com>
Cc: Marcelo Tosatti <marcelo@kvack.org>
Cc: Eric Dumazet <dada1@cosmosbay.com>
Cc: "Paul E. McKenney" <paulmck@us.ibm.com>
Cc: Izik Eidus <izike@qumranet.com>
Cc: Anthony Liguori <aliguori@us.ibm.com>
Cc: Rik van Riel <riel@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-07-28 15:46:29 -07:00
|
|
|
|
2009-01-06 14:39:16 -08:00
|
|
|
|
2010-05-28 09:29:16 +09:00
|
|
|
|
2019-10-18 20:20:17 -07:00
|
|
|
|
2012-10-25 13:37:31 -07:00
|
|
|
|
mm: introduce idle page tracking
Knowing the portion of memory that is not used by a certain application or
memory cgroup (idle memory) can be useful for partitioning the system
efficiently, e.g. by setting memory cgroup limits appropriately.
Currently, the only means to estimate the amount of idle memory provided
by the kernel is /proc/PID/{clear_refs,smaps}: the user can clear the
access bit for all pages mapped to a particular process by writing 1 to
clear_refs, wait for some time, and then count smaps:Referenced. However,
this method has two serious shortcomings:
- it does not count unmapped file pages
- it affects the reclaimer logic
To overcome these drawbacks, this patch introduces two new page flags,
Idle and Young, and a new sysfs file, /sys/kernel/mm/page_idle/bitmap.
A page's Idle flag can only be set from userspace by setting bit in
/sys/kernel/mm/page_idle/bitmap at the offset corresponding to the page,
and it is cleared whenever the page is accessed either through page tables
(it is cleared in page_referenced() in this case) or using the read(2)
system call (mark_page_accessed()). Thus by setting the Idle flag for
pages of a particular workload, which can be found e.g. by reading
/proc/PID/pagemap, waiting for some time to let the workload access its
working set, and then reading the bitmap file, one can estimate the amount
of pages that are not used by the workload.
The Young page flag is used to avoid interference with the memory
reclaimer. A page's Young flag is set whenever the Access bit of a page
table entry pointing to the page is cleared by writing to the bitmap file.
If page_referenced() is called on a Young page, it will add 1 to its
return value, therefore concealing the fact that the Access bit was
cleared.
Note, since there is no room for extra page flags on 32 bit, this feature
uses extended page flags when compiled on 32 bit.
[akpm@linux-foundation.org: fix build]
[akpm@linux-foundation.org: kpageidle requires an MMU]
[akpm@linux-foundation.org: decouple from page-flags rework]
Signed-off-by: Vladimir Davydov <vdavydov@parallels.com>
Reviewed-by: Andres Lagar-Cavilla <andreslc@google.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Raghavendra K T <raghavendra.kt@linux.vnet.ibm.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michal Hocko <mhocko@suse.cz>
Cc: Greg Thelen <gthelen@google.com>
Cc: Michel Lespinasse <walken@google.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Pavel Emelyanov <xemul@parallels.com>
Cc: Cyrill Gorcunov <gorcunov@openvz.org>
Cc: Jonathan Corbet <corbet@lwn.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2015-09-09 15:35:45 -07:00
|
|
|
|
2017-09-08 16:12:17 -07:00
|
|
|
|
2018-07-13 16:58:52 -07:00
|
|
|
|
2022-05-12 20:22:53 -07:00
|
|
|
|
2024-07-10 16:36:41 +08:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
mm/hugetlb: fix excessive IPI broadcasts when unsharing PMD tables using mmu_gather
As reported, ever since commit 1013af4f585f ("mm/hugetlb: fix
huge_pmd_unshare() vs GUP-fast race") we can end up in some situations
where we perform so many IPI broadcasts when unsharing hugetlb PMD page
tables that it severely regresses some workloads.
In particular, when we fork()+exit(), or when we munmap() a large
area backed by many shared PMD tables, we perform one IPI broadcast per
unshared PMD table.
There are two optimizations to be had:
(1) When we process (unshare) multiple such PMD tables, such as during
exit(), it is sufficient to send a single IPI broadcast (as long as
we respect locking rules) instead of one per PMD table.
Locking prevents that any of these PMD tables could get reused before
we drop the lock.
(2) When we are not the last sharer (> 2 users including us), there is
no need to send the IPI broadcast. The shared PMD tables cannot
become exclusive (fully unshared) before an IPI will be broadcasted
by the last sharer.
Concurrent GUP-fast could walk into a PMD table just before we
unshared it. It could then succeed in grabbing a page from the
shared page table even after munmap() etc succeeded (and supressed
an IPI). But there is not difference compared to GUP-fast just
sleeping for a while after grabbing the page and re-enabling IRQs.
Most importantly, GUP-fast will never walk into page tables that are
no-longer shared, because the last sharer will issue an IPI
broadcast.
(if ever required, checking whether the PUD changed in GUP-fast
after grabbing the page like we do in the PTE case could handle
this)
So let's rework PMD sharing TLB flushing + IPI sync to use the mmu_gather
infrastructure so we can implement these optimizations and demystify the
code at least a bit. Extend the mmu_gather infrastructure to be able to
deal with our special hugetlb PMD table sharing implementation.
To make initialization of the mmu_gather easier when working on a single
VMA (in particular, when dealing with hugetlb), provide
tlb_gather_mmu_vma().
We'll consolidate the handling for (full) unsharing of PMD tables in
tlb_unshare_pmd_ptdesc() and tlb_flush_unshared_tables(), and track
in "struct mmu_gather" whether we had (full) unsharing of PMD tables.
Because locking is very special (concurrent unsharing+reuse must be
prevented), we disallow deferring flushing to tlb_finish_mmu() and instead
require an explicit earlier call to tlb_flush_unshared_tables().
From hugetlb code, we call huge_pmd_unshare_flush() where we make sure
that the expected lock protecting us from concurrent unsharing+reuse is
still held.
Check with a VM_WARN_ON_ONCE() in tlb_finish_mmu() that
tlb_flush_unshared_tables() was properly called earlier.
Document it all properly.
Notes about tlb_remove_table_sync_one() interaction with unsharing:
There are two fairly tricky things:
(1) tlb_remove_table_sync_one() is a NOP on architectures without
CONFIG_MMU_GATHER_RCU_TABLE_FREE.
Here, the assumption is that the previous TLB flush would send an
IPI to all relevant CPUs. Careful: some architectures like x86 only
send IPIs to all relevant CPUs when tlb->freed_tables is set.
The relevant architectures should be selecting
MMU_GATHER_RCU_TABLE_FREE, but x86 might not do that in stable
kernels and it might have been problematic before this patch.
Also, the arch flushing behavior (independent of IPIs) is different
when tlb->freed_tables is set. Do we have to enlighten them to also
take care of tlb->unshared_tables? So far we didn't care, so
hopefully we are fine. Of course, we could be setting
tlb->freed_tables as well, but that might then unnecessarily flush
too much, because the semantics of tlb->freed_tables are a bit
fuzzy.
This patch changes nothing in this regard.
(2) tlb_remove_table_sync_one() is not a NOP on architectures with
CONFIG_MMU_GATHER_RCU_TABLE_FREE that actually don't need a sync.
Take x86 as an example: in the common case (!pv, !X86_FEATURE_INVLPGB)
we still issue IPIs during TLB flushes and don't actually need the
second tlb_remove_table_sync_one().
This optimized can be implemented on top of this, by checking e.g., in
tlb_remove_table_sync_one() whether we really need IPIs. But as
described in (1), it really must honor tlb->freed_tables then to
send IPIs to all relevant CPUs.
Notes on TLB flushing changes:
(1) Flushing for non-shared PMD tables
We're converting from flush_hugetlb_tlb_range() to
tlb_remove_huge_tlb_entry(). Given that we properly initialize the
MMU gather in tlb_gather_mmu_vma() to be hugetlb aware, similar to
__unmap_hugepage_range(), that should be fine.
(2) Flushing for shared PMD tables
We're converting from various things (flush_hugetlb_tlb_range(),
tlb_flush_pmd_range(), flush_tlb_range()) to tlb_flush_pmd_range().
tlb_flush_pmd_range() achieves the same that
tlb_remove_huge_tlb_entry() would achieve in these scenarios.
Note that tlb_remove_huge_tlb_entry() also calls
__tlb_remove_tlb_entry(), however that is only implemented on
powerpc, which does not support PMD table sharing.
Similar to (1), tlb_gather_mmu_vma() should make sure that TLB
flushing keeps on working as expected.
Further, note that the ptdesc_pmd_pts_dec() in huge_pmd_share() is not a
concern, as we are holding the i_mmap_lock the whole time, preventing
concurrent unsharing. That ptdesc_pmd_pts_dec() usage will be removed
separately as a cleanup later.
There are plenty more cleanups to be had, but they have to wait until
this is fixed.
[david@kernel.org: fix kerneldoc]
Link: https://lkml.kernel.org/r/f223dd74-331c-412d-93fc-69e360a5006c@kernel.org
Link: https://lkml.kernel.org/r/20251223214037.580860-5-david@kernel.org
Fixes: 1013af4f585f ("mm/hugetlb: fix huge_pmd_unshare() vs GUP-fast race")
Signed-off-by: David Hildenbrand (Red Hat) <david@kernel.org>
Reported-by: Uschakow, Stanislav" <suschako@amazon.de>
Closes: https://lore.kernel.org/all/4d3878531c76479d9f8ca9789dc6485d@amazon.de/
Tested-by: Laurence Oberman <loberman@redhat.com>
Acked-by: Harry Yoo <harry.yoo@oracle.com>
Reviewed-by: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: Lance Yang <lance.yang@linux.dev>
Cc: Liu Shixin <liushixin2@huawei.com>
Cc: Oscar Salvador <osalvador@suse.de>
Cc: Rik van Riel <riel@surriel.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2025-12-23 22:40:37 +01:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2022-03-24 18:10:01 -07:00
|
|
|
|
|
|
|
|
|
2015-09-04 15:47:32 -07:00
|
|
|
|
mlock: mlocked pages are unevictable
Make sure that mlocked pages also live on the unevictable LRU, so kswapd
will not scan them over and over again.
This is achieved through various strategies:
1) add yet another page flag--PG_mlocked--to indicate that
the page is locked for efficient testing in vmscan and,
optionally, fault path. This allows early culling of
unevictable pages, preventing them from getting to
page_referenced()/try_to_unmap(). Also allows separate
accounting of mlock'd pages, as Nick's original patch
did.
Note: Nick's original mlock patch used a PG_mlocked
flag. I had removed this in favor of the PG_unevictable
flag + an mlock_count [new page struct member]. I
restored the PG_mlocked flag to eliminate the new
count field.
2) add the mlock/unevictable infrastructure to mm/mlock.c,
with internal APIs in mm/internal.h. This is a rework
of Nick's original patch to these files, taking into
account that mlocked pages are now kept on unevictable
LRU list.
3) update vmscan.c:page_evictable() to check PageMlocked()
and, if vma passed in, the vm_flags. Note that the vma
will only be passed in for new pages in the fault path;
and then only if the "cull unevictable pages in fault
path" patch is included.
4) add try_to_unlock() to rmap.c to walk a page's rmap and
ClearPageMlocked() if no other vmas have it mlocked.
Reuses as much of try_to_unmap() as possible. This
effectively replaces the use of one of the lru list links
as an mlock count. If this mechanism let's pages in mlocked
vmas leak through w/o PG_mlocked set [I don't know that it
does], we should catch them later in try_to_unmap(). One
hopes this will be rare, as it will be relatively expensive.
Original mm/internal.h, mm/rmap.c and mm/mlock.c changes:
Signed-off-by: Nick Piggin <npiggin@suse.de>
splitlru: introduce __get_user_pages():
New munlock processing need to GUP_FLAGS_IGNORE_VMA_PERMISSIONS.
because current get_user_pages() can't grab PROT_NONE pages theresore it
cause PROT_NONE pages can't munlock.
[akpm@linux-foundation.org: fix this for pagemap-pass-mm-into-pagewalkers.patch]
[akpm@linux-foundation.org: untangle patch interdependencies]
[akpm@linux-foundation.org: fix things after out-of-order merging]
[hugh@veritas.com: fix page-flags mess]
[lee.schermerhorn@hp.com: fix munlock page table walk - now requires 'mm']
[kosaki.motohiro@jp.fujitsu.com: build fix]
[kosaki.motohiro@jp.fujitsu.com: fix truncate race and sevaral comments]
[kosaki.motohiro@jp.fujitsu.com: splitlru: introduce __get_user_pages()]
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Rik van Riel <riel@redhat.com>
Signed-off-by: Lee Schermerhorn <lee.schermerhorn@hp.com>
Cc: Nick Piggin <npiggin@suse.de>
Cc: Dave Hansen <dave@linux.vnet.ibm.com>
Cc: Matt Mackall <mpm@selenic.com>
Signed-off-by: Hugh Dickins <hugh@veritas.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-10-18 20:26:44 -07:00
|
|
|
|
mm, swap: cleanup swap entry management workflow
The current swap entry allocation/freeing workflow has never had a clear
definition. This makes it hard to debug or add new optimizations.
This commit introduces a proper definition of how swap entries would be
allocated and freed. Now, most operations are folio based, so they will
never exceed one swap cluster, and we now have a cleaner border between
swap and the rest of mm, making it much easier to follow and debug,
especially with new added sanity checks. Also making more optimization
possible.
Swap entry will be mostly freed and free with a folio bound. The folio
lock will be useful for resolving many swap related races.
Now swap allocation (except hibernation) always starts with a folio in the
swap cache, and gets duped/freed protected by the folio lock:
- folio_alloc_swap() - The only allocation entry point now.
Context: The folio must be locked.
This allocates one or a set of continuous swap slots for a folio and
binds them to the folio by adding the folio to the swap cache. The
swap slots' swap count start with zero value.
- folio_dup_swap() - Increase the swap count of one or more entries.
Context: The folio must be locked and in the swap cache. For now, the
caller still has to lock the new swap entry owner (e.g., PTL).
This increases the ref count of swap entries allocated to a folio.
Newly allocated swap slots' count has to be increased by this helper
as the folio got unmapped (and swap entries got installed).
- folio_put_swap() - Decrease the swap count of one or more entries.
Context: The folio must be locked and in the swap cache. For now, the
caller still has to lock the new swap entry owner (e.g., PTL).
This decreases the ref count of swap entries allocated to a folio.
Typically, swapin will decrease the swap count as the folio got
installed back and the swap entry got uninstalled
This won't remove the folio from the swap cache and free the
slot. Lazy freeing of swap cache is helpful for reducing IO.
There is already a folio_free_swap() for immediate cache reclaim.
This part could be further optimized later.
The above locking constraints could be further relaxed when the swap table
is fully implemented. Currently dup still needs the caller to lock the
swap entry container (e.g. PTL), or a concurrent zap may underflow the
swap count.
Some swap users need to interact with swap count without involving folio
(e.g. forking/zapping the page table or mapping truncate without swapin).
In such cases, the caller has to ensure there is no race condition on
whatever owns the swap count and use the below helpers:
- swap_put_entries_direct() - Decrease the swap count directly.
Context: The caller must lock whatever is referencing the slots to
avoid a race.
Typically the page table zapping or shmem mapping truncate will need
to free swap slots directly. If a slot is cached (has a folio bound),
this will also try to release the swap cache.
- swap_dup_entry_direct() - Increase the swap count directly.
Context: The caller must lock whatever is referencing the entries to
avoid race, and the entries must already have a swap count > 1.
Typically, forking will need to copy the page table and hence needs to
increase the swap count of the entries in the table. The page table is
locked while referencing the swap entries, so the entries all have a
swap count > 1 and can't be freed.
Hibernation subsystem is a bit different, so two special wrappers are here:
- swap_alloc_hibernation_slot() - Allocate one entry from one device.
- swap_free_hibernation_slot() - Free one entry allocated by the above
helper.
All hibernation entries are exclusive to the hibernation subsystem and
should not interact with ordinary swap routines.
By separating the workflows, it will be possible to bind folio more
tightly with swap cache and get rid of the SWAP_HAS_CACHE as a temporary
pin.
This commit should not introduce any behavior change
[kasong@tencent.com: fix leak, per Chris Mason. Remove WARN_ON, per Lai Yi]
Link: https://lkml.kernel.org/r/CAMgjq7AUz10uETVm8ozDWcB3XohkOqf0i33KGrAquvEVvfp5cg@mail.gmail.com
[ryncsn@gmail.com: fix KSM copy pages for swapoff, per Chris]
Link: https://lkml.kernel.org/r/aXxkANcET3l2Xu6J@KASONG-MC4
Link: https://lkml.kernel.org/r/20251220-swap-table-p2-v5-14-8862a265a033@tencent.com
Signed-off-by: Kairui Song <kasong@tencent.com>
Signed-off-by: Kairui Song <ryncsn@gmail.com>
Acked-by: Rafael J. Wysocki (Intel) <rafael@kernel.org>
Reviewed-by: Baoquan He <bhe@redhat.com>
Cc: Baolin Wang <baolin.wang@linux.alibaba.com>
Cc: Barry Song <baohua@kernel.org>
Cc: Chris Li <chrisl@kernel.org>
Cc: Nhat Pham <nphamcs@gmail.com>
Cc: Yosry Ahmed <yosry.ahmed@linux.dev>
Cc: Deepanshu Kartikey <kartikey406@gmail.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Kairui Song <ryncsn@gmail.com>
Cc: Chris Mason <clm@fb.com>
Cc: Chris Mason <clm@meta.com>
Cc: Lai Yi <yi1.lai@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2025-12-20 03:43:43 +08:00
|
|
|
|
mlock: mlocked pages are unevictable
Make sure that mlocked pages also live on the unevictable LRU, so kswapd
will not scan them over and over again.
This is achieved through various strategies:
1) add yet another page flag--PG_mlocked--to indicate that
the page is locked for efficient testing in vmscan and,
optionally, fault path. This allows early culling of
unevictable pages, preventing them from getting to
page_referenced()/try_to_unmap(). Also allows separate
accounting of mlock'd pages, as Nick's original patch
did.
Note: Nick's original mlock patch used a PG_mlocked
flag. I had removed this in favor of the PG_unevictable
flag + an mlock_count [new page struct member]. I
restored the PG_mlocked flag to eliminate the new
count field.
2) add the mlock/unevictable infrastructure to mm/mlock.c,
with internal APIs in mm/internal.h. This is a rework
of Nick's original patch to these files, taking into
account that mlocked pages are now kept on unevictable
LRU list.
3) update vmscan.c:page_evictable() to check PageMlocked()
and, if vma passed in, the vm_flags. Note that the vma
will only be passed in for new pages in the fault path;
and then only if the "cull unevictable pages in fault
path" patch is included.
4) add try_to_unlock() to rmap.c to walk a page's rmap and
ClearPageMlocked() if no other vmas have it mlocked.
Reuses as much of try_to_unmap() as possible. This
effectively replaces the use of one of the lru list links
as an mlock count. If this mechanism let's pages in mlocked
vmas leak through w/o PG_mlocked set [I don't know that it
does], we should catch them later in try_to_unmap(). One
hopes this will be rare, as it will be relatively expensive.
Original mm/internal.h, mm/rmap.c and mm/mlock.c changes:
Signed-off-by: Nick Piggin <npiggin@suse.de>
splitlru: introduce __get_user_pages():
New munlock processing need to GUP_FLAGS_IGNORE_VMA_PERMISSIONS.
because current get_user_pages() can't grab PROT_NONE pages theresore it
cause PROT_NONE pages can't munlock.
[akpm@linux-foundation.org: fix this for pagemap-pass-mm-into-pagewalkers.patch]
[akpm@linux-foundation.org: untangle patch interdependencies]
[akpm@linux-foundation.org: fix things after out-of-order merging]
[hugh@veritas.com: fix page-flags mess]
[lee.schermerhorn@hp.com: fix munlock page table walk - now requires 'mm']
[kosaki.motohiro@jp.fujitsu.com: build fix]
[kosaki.motohiro@jp.fujitsu.com: fix truncate race and sevaral comments]
[kosaki.motohiro@jp.fujitsu.com: splitlru: introduce __get_user_pages()]
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Rik van Riel <riel@redhat.com>
Signed-off-by: Lee Schermerhorn <lee.schermerhorn@hp.com>
Cc: Nick Piggin <npiggin@suse.de>
Cc: Dave Hansen <dave@linux.vnet.ibm.com>
Cc: Matt Mackall <mpm@selenic.com>
Signed-off-by: Hugh Dickins <hugh@veritas.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-10-18 20:26:44 -07:00
|
|
|
|
2008-10-18 20:28:38 -07:00
|
|
|
|
mm: change anon_vma linking to fix multi-process server scalability issue
The old anon_vma code can lead to scalability issues with heavily forking
workloads. Specifically, each anon_vma will be shared between the parent
process and all its child processes.
In a workload with 1000 child processes and a VMA with 1000 anonymous
pages per process that get COWed, this leads to a system with a million
anonymous pages in the same anon_vma, each of which is mapped in just one
of the 1000 processes. However, the current rmap code needs to walk them
all, leading to O(N) scanning complexity for each page.
This can result in systems where one CPU is walking the page tables of
1000 processes in page_referenced_one, while all other CPUs are stuck on
the anon_vma lock. This leads to catastrophic failure for a benchmark
like AIM7, where the total number of processes can reach in the tens of
thousands. Real workloads are still a factor 10 less process intensive
than AIM7, but they are catching up.
This patch changes the way anon_vmas and VMAs are linked, which allows us
to associate multiple anon_vmas with a VMA. At fork time, each child
process gets its own anon_vmas, in which its COWed pages will be
instantiated. The parents' anon_vma is also linked to the VMA, because
non-COWed pages could be present in any of the children.
This reduces rmap scanning complexity to O(1) for the pages of the 1000
child processes, with O(N) complexity for at most 1/N pages in the system.
This reduces the average scanning cost in heavily forking workloads from
O(N) to 2.
The only real complexity in this patch stems from the fact that linking a
VMA to anon_vmas now involves memory allocations. This means vma_adjust
can fail, if it needs to attach a VMA to anon_vma structures. This in
turn means error handling needs to be added to the calling functions.
A second source of complexity is that, because there can be multiple
anon_vmas, the anon_vma linking in vma_adjust can no longer be done under
"the" anon_vma lock. To prevent the rmap code from walking up an
incomplete VMA, this patch introduces the VM_LOCK_RMAP VMA flag. This bit
flag uses the same slot as the NOMMU VM_MAPPED_COPY, with an ifdef in mm.h
to make sure it is impossible to compile a kernel that needs both symbolic
values for the same bitflag.
Some test results:
Without the anon_vma changes, when AIM7 hits around 9.7k users (on a test
box with 16GB RAM and not quite enough IO), the system ends up running
>99% in system time, with every CPU on the same anon_vma lock in the
pageout code.
With these changes, AIM7 hits the cross-over point around 29.7k users.
This happens with ~99% IO wait time, there never seems to be any spike in
system time. The anon_vma lock contention appears to be resolved.
[akpm@linux-foundation.org: cleanups]
Signed-off-by: Rik van Riel <riel@redhat.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Larry Woodman <lwoodman@redhat.com>
Cc: Lee Schermerhorn <Lee.Schermerhorn@hp.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-03-05 13:42:07 -08:00
|
|
|
|
2008-10-18 20:28:38 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2011-03-22 16:32:49 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-08-31 19:06:00 +02:00
|
|
|
|
|
|
|
|
|
2015-01-08 14:32:15 -08:00
|
|
|
|
2011-03-22 16:32:49 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2008-10-18 20:28:38 -07:00
|
|
|
|
|
|
|
|
|
2011-03-22 16:32:49 -07:00
|
|
|
|
2008-10-18 20:28:38 -07:00
|
|
|
|
2011-03-22 16:32:49 -07:00
|
|
|
|
2011-05-24 17:12:13 -07:00
|
|
|
|
|
|
|
|
|
2022-01-29 16:06:53 -05:00
|
|
|
|
2011-05-24 17:12:13 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-01-29 16:06:53 -05:00
|
|
|
|
2011-05-24 17:12:13 -07:00
|
|
|
|
2022-01-29 16:06:53 -05:00
|
|
|
|
2012-12-02 19:56:50 +00:00
|
|
|
|
2011-05-24 17:12:13 -07:00
|
|
|
|
2012-12-02 19:56:50 +00:00
|
|
|
|
2011-05-24 17:12:13 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2014-06-04 16:05:33 -07:00
|
|
|
|
2012-12-02 19:56:46 +00:00
|
|
|
|
2012-12-02 19:56:50 +00:00
|
|
|
|
2013-02-22 16:34:40 -08:00
|
|
|
|
2011-05-24 17:12:13 -07:00
|
|
|
|
|
|
|
|
|
2008-10-18 20:28:38 -07:00
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2011-06-17 19:05:36 -07:00
|
|
|
|
mm: change anon_vma linking to fix multi-process server scalability issue
The old anon_vma code can lead to scalability issues with heavily forking
workloads. Specifically, each anon_vma will be shared between the parent
process and all its child processes.
In a workload with 1000 child processes and a VMA with 1000 anonymous
pages per process that get COWed, this leads to a system with a million
anonymous pages in the same anon_vma, each of which is mapped in just one
of the 1000 processes. However, the current rmap code needs to walk them
all, leading to O(N) scanning complexity for each page.
This can result in systems where one CPU is walking the page tables of
1000 processes in page_referenced_one, while all other CPUs are stuck on
the anon_vma lock. This leads to catastrophic failure for a benchmark
like AIM7, where the total number of processes can reach in the tens of
thousands. Real workloads are still a factor 10 less process intensive
than AIM7, but they are catching up.
This patch changes the way anon_vmas and VMAs are linked, which allows us
to associate multiple anon_vmas with a VMA. At fork time, each child
process gets its own anon_vmas, in which its COWed pages will be
instantiated. The parents' anon_vma is also linked to the VMA, because
non-COWed pages could be present in any of the children.
This reduces rmap scanning complexity to O(1) for the pages of the 1000
child processes, with O(N) complexity for at most 1/N pages in the system.
This reduces the average scanning cost in heavily forking workloads from
O(N) to 2.
The only real complexity in this patch stems from the fact that linking a
VMA to anon_vmas now involves memory allocations. This means vma_adjust
can fail, if it needs to attach a VMA to anon_vma structures. This in
turn means error handling needs to be added to the calling functions.
A second source of complexity is that, because there can be multiple
anon_vmas, the anon_vma linking in vma_adjust can no longer be done under
"the" anon_vma lock. To prevent the rmap code from walking up an
incomplete VMA, this patch introduces the VM_LOCK_RMAP VMA flag. This bit
flag uses the same slot as the NOMMU VM_MAPPED_COPY, with an ifdef in mm.h
to make sure it is impossible to compile a kernel that needs both symbolic
values for the same bitflag.
Some test results:
Without the anon_vma changes, when AIM7 hits around 9.7k users (on a test
box with 16GB RAM and not quite enough IO), the system ends up running
>99% in system time, with every CPU on the same anon_vma lock in the
pageout code.
With these changes, AIM7 hits the cross-over point around 29.7k users.
This happens with ~99% IO wait time, there never seems to be any spike in
system time. The anon_vma lock contention appears to be resolved.
[akpm@linux-foundation.org: cleanups]
Signed-off-by: Rik van Riel <riel@redhat.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Larry Woodman <lwoodman@redhat.com>
Cc: Lee Schermerhorn <Lee.Schermerhorn@hp.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-03-05 13:42:07 -08:00
|
|
|
|
2011-06-17 19:05:36 -07:00
|
|
|
|
mm: change anon_vma linking to fix multi-process server scalability issue
The old anon_vma code can lead to scalability issues with heavily forking
workloads. Specifically, each anon_vma will be shared between the parent
process and all its child processes.
In a workload with 1000 child processes and a VMA with 1000 anonymous
pages per process that get COWed, this leads to a system with a million
anonymous pages in the same anon_vma, each of which is mapped in just one
of the 1000 processes. However, the current rmap code needs to walk them
all, leading to O(N) scanning complexity for each page.
This can result in systems where one CPU is walking the page tables of
1000 processes in page_referenced_one, while all other CPUs are stuck on
the anon_vma lock. This leads to catastrophic failure for a benchmark
like AIM7, where the total number of processes can reach in the tens of
thousands. Real workloads are still a factor 10 less process intensive
than AIM7, but they are catching up.
This patch changes the way anon_vmas and VMAs are linked, which allows us
to associate multiple anon_vmas with a VMA. At fork time, each child
process gets its own anon_vmas, in which its COWed pages will be
instantiated. The parents' anon_vma is also linked to the VMA, because
non-COWed pages could be present in any of the children.
This reduces rmap scanning complexity to O(1) for the pages of the 1000
child processes, with O(N) complexity for at most 1/N pages in the system.
This reduces the average scanning cost in heavily forking workloads from
O(N) to 2.
The only real complexity in this patch stems from the fact that linking a
VMA to anon_vmas now involves memory allocations. This means vma_adjust
can fail, if it needs to attach a VMA to anon_vma structures. This in
turn means error handling needs to be added to the calling functions.
A second source of complexity is that, because there can be multiple
anon_vmas, the anon_vma linking in vma_adjust can no longer be done under
"the" anon_vma lock. To prevent the rmap code from walking up an
incomplete VMA, this patch introduces the VM_LOCK_RMAP VMA flag. This bit
flag uses the same slot as the NOMMU VM_MAPPED_COPY, with an ifdef in mm.h
to make sure it is impossible to compile a kernel that needs both symbolic
values for the same bitflag.
Some test results:
Without the anon_vma changes, when AIM7 hits around 9.7k users (on a test
box with 16GB RAM and not quite enough IO), the system ends up running
>99% in system time, with every CPU on the same anon_vma lock in the
pageout code.
With these changes, AIM7 hits the cross-over point around 29.7k users.
This happens with ~99% IO wait time, there never seems to be any spike in
system time. The anon_vma lock contention appears to be resolved.
[akpm@linux-foundation.org: cleanups]
Signed-off-by: Rik van Riel <riel@redhat.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Larry Woodman <lwoodman@redhat.com>
Cc: Lee Schermerhorn <Lee.Schermerhorn@hp.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-03-05 13:42:07 -08:00
|
|
|
|
|
|
|
|
|
2010-10-26 14:22:02 -07:00
|
|
|
|
mm: change anon_vma linking to fix multi-process server scalability issue
The old anon_vma code can lead to scalability issues with heavily forking
workloads. Specifically, each anon_vma will be shared between the parent
process and all its child processes.
In a workload with 1000 child processes and a VMA with 1000 anonymous
pages per process that get COWed, this leads to a system with a million
anonymous pages in the same anon_vma, each of which is mapped in just one
of the 1000 processes. However, the current rmap code needs to walk them
all, leading to O(N) scanning complexity for each page.
This can result in systems where one CPU is walking the page tables of
1000 processes in page_referenced_one, while all other CPUs are stuck on
the anon_vma lock. This leads to catastrophic failure for a benchmark
like AIM7, where the total number of processes can reach in the tens of
thousands. Real workloads are still a factor 10 less process intensive
than AIM7, but they are catching up.
This patch changes the way anon_vmas and VMAs are linked, which allows us
to associate multiple anon_vmas with a VMA. At fork time, each child
process gets its own anon_vmas, in which its COWed pages will be
instantiated. The parents' anon_vma is also linked to the VMA, because
non-COWed pages could be present in any of the children.
This reduces rmap scanning complexity to O(1) for the pages of the 1000
child processes, with O(N) complexity for at most 1/N pages in the system.
This reduces the average scanning cost in heavily forking workloads from
O(N) to 2.
The only real complexity in this patch stems from the fact that linking a
VMA to anon_vmas now involves memory allocations. This means vma_adjust
can fail, if it needs to attach a VMA to anon_vma structures. This in
turn means error handling needs to be added to the calling functions.
A second source of complexity is that, because there can be multiple
anon_vmas, the anon_vma linking in vma_adjust can no longer be done under
"the" anon_vma lock. To prevent the rmap code from walking up an
incomplete VMA, this patch introduces the VM_LOCK_RMAP VMA flag. This bit
flag uses the same slot as the NOMMU VM_MAPPED_COPY, with an ifdef in mm.h
to make sure it is impossible to compile a kernel that needs both symbolic
values for the same bitflag.
Some test results:
Without the anon_vma changes, when AIM7 hits around 9.7k users (on a test
box with 16GB RAM and not quite enough IO), the system ends up running
>99% in system time, with every CPU on the same anon_vma lock in the
pageout code.
With these changes, AIM7 hits the cross-over point around 29.7k users.
This happens with ~99% IO wait time, there never seems to be any spike in
system time. The anon_vma lock contention appears to be resolved.
[akpm@linux-foundation.org: cleanups]
Signed-off-by: Rik van Riel <riel@redhat.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Larry Woodman <lwoodman@redhat.com>
Cc: Lee Schermerhorn <Lee.Schermerhorn@hp.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-03-05 13:42:07 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-18 14:50:44 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2012-03-21 16:34:01 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2008-10-19 10:32:20 -07:00
|
|
|
|
2016-12-12 16:44:38 -08:00
|
|
|
|
2008-10-19 10:32:20 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2016-12-12 16:44:38 -08:00
|
|
|
|
|
|
|
|
|
2010-12-27 15:14:06 +01:00
|
|
|
|
2008-10-19 10:32:20 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-01-29 16:06:53 -05:00
|
|
|
|
2021-02-25 17:17:53 -08:00
|
|
|
|
2008-10-19 10:32:20 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2016-12-12 16:44:38 -08:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2016-12-12 16:44:38 -08:00
|
|
|
|
|
|
|
|
|
mm: change anon_vma linking to fix multi-process server scalability issue
The old anon_vma code can lead to scalability issues with heavily forking
workloads. Specifically, each anon_vma will be shared between the parent
process and all its child processes.
In a workload with 1000 child processes and a VMA with 1000 anonymous
pages per process that get COWed, this leads to a system with a million
anonymous pages in the same anon_vma, each of which is mapped in just one
of the 1000 processes. However, the current rmap code needs to walk them
all, leading to O(N) scanning complexity for each page.
This can result in systems where one CPU is walking the page tables of
1000 processes in page_referenced_one, while all other CPUs are stuck on
the anon_vma lock. This leads to catastrophic failure for a benchmark
like AIM7, where the total number of processes can reach in the tens of
thousands. Real workloads are still a factor 10 less process intensive
than AIM7, but they are catching up.
This patch changes the way anon_vmas and VMAs are linked, which allows us
to associate multiple anon_vmas with a VMA. At fork time, each child
process gets its own anon_vmas, in which its COWed pages will be
instantiated. The parents' anon_vma is also linked to the VMA, because
non-COWed pages could be present in any of the children.
This reduces rmap scanning complexity to O(1) for the pages of the 1000
child processes, with O(N) complexity for at most 1/N pages in the system.
This reduces the average scanning cost in heavily forking workloads from
O(N) to 2.
The only real complexity in this patch stems from the fact that linking a
VMA to anon_vmas now involves memory allocations. This means vma_adjust
can fail, if it needs to attach a VMA to anon_vma structures. This in
turn means error handling needs to be added to the calling functions.
A second source of complexity is that, because there can be multiple
anon_vmas, the anon_vma linking in vma_adjust can no longer be done under
"the" anon_vma lock. To prevent the rmap code from walking up an
incomplete VMA, this patch introduces the VM_LOCK_RMAP VMA flag. This bit
flag uses the same slot as the NOMMU VM_MAPPED_COPY, with an ifdef in mm.h
to make sure it is impossible to compile a kernel that needs both symbolic
values for the same bitflag.
Some test results:
Without the anon_vma changes, when AIM7 hits around 9.7k users (on a test
box with 16GB RAM and not quite enough IO), the system ends up running
>99% in system time, with every CPU on the same anon_vma lock in the
pageout code.
With these changes, AIM7 hits the cross-over point around 29.7k users.
This happens with ~99% IO wait time, there never seems to be any spike in
system time. The anon_vma lock contention appears to be resolved.
[akpm@linux-foundation.org: cleanups]
Signed-off-by: Rik van Riel <riel@redhat.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Larry Woodman <lwoodman@redhat.com>
Cc: Lee Schermerhorn <Lee.Schermerhorn@hp.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-03-05 13:42:07 -08:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2024-04-26 15:45:00 +01:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
2016-12-12 16:44:38 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
mm: change anon_vma linking to fix multi-process server scalability issue
The old anon_vma code can lead to scalability issues with heavily forking
workloads. Specifically, each anon_vma will be shared between the parent
process and all its child processes.
In a workload with 1000 child processes and a VMA with 1000 anonymous
pages per process that get COWed, this leads to a system with a million
anonymous pages in the same anon_vma, each of which is mapped in just one
of the 1000 processes. However, the current rmap code needs to walk them
all, leading to O(N) scanning complexity for each page.
This can result in systems where one CPU is walking the page tables of
1000 processes in page_referenced_one, while all other CPUs are stuck on
the anon_vma lock. This leads to catastrophic failure for a benchmark
like AIM7, where the total number of processes can reach in the tens of
thousands. Real workloads are still a factor 10 less process intensive
than AIM7, but they are catching up.
This patch changes the way anon_vmas and VMAs are linked, which allows us
to associate multiple anon_vmas with a VMA. At fork time, each child
process gets its own anon_vmas, in which its COWed pages will be
instantiated. The parents' anon_vma is also linked to the VMA, because
non-COWed pages could be present in any of the children.
This reduces rmap scanning complexity to O(1) for the pages of the 1000
child processes, with O(N) complexity for at most 1/N pages in the system.
This reduces the average scanning cost in heavily forking workloads from
O(N) to 2.
The only real complexity in this patch stems from the fact that linking a
VMA to anon_vmas now involves memory allocations. This means vma_adjust
can fail, if it needs to attach a VMA to anon_vma structures. This in
turn means error handling needs to be added to the calling functions.
A second source of complexity is that, because there can be multiple
anon_vmas, the anon_vma linking in vma_adjust can no longer be done under
"the" anon_vma lock. To prevent the rmap code from walking up an
incomplete VMA, this patch introduces the VM_LOCK_RMAP VMA flag. This bit
flag uses the same slot as the NOMMU VM_MAPPED_COPY, with an ifdef in mm.h
to make sure it is impossible to compile a kernel that needs both symbolic
values for the same bitflag.
Some test results:
Without the anon_vma changes, when AIM7 hits around 9.7k users (on a test
box with 16GB RAM and not quite enough IO), the system ends up running
>99% in system time, with every CPU on the same anon_vma lock in the
pageout code.
With these changes, AIM7 hits the cross-over point around 29.7k users.
This happens with ~99% IO wait time, there never seems to be any spike in
system time. The anon_vma lock contention appears to be resolved.
[akpm@linux-foundation.org: cleanups]
Signed-off-by: Rik van Riel <riel@redhat.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Larry Woodman <lwoodman@redhat.com>
Cc: Lee Schermerhorn <Lee.Schermerhorn@hp.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-03-05 13:42:07 -08:00
|
|
|
|
2016-12-12 16:44:38 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-08-31 19:06:00 +02:00
|
|
|
|
2016-12-12 16:44:38 -08:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2016-12-12 16:44:38 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-18 14:50:44 +00:00
|
|
|
|
|
|
|
|
|
2022-08-31 19:06:00 +02:00
|
|
|
|
2016-12-12 16:44:38 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
mm: change anon_vma linking to fix multi-process server scalability issue
The old anon_vma code can lead to scalability issues with heavily forking
workloads. Specifically, each anon_vma will be shared between the parent
process and all its child processes.
In a workload with 1000 child processes and a VMA with 1000 anonymous
pages per process that get COWed, this leads to a system with a million
anonymous pages in the same anon_vma, each of which is mapped in just one
of the 1000 processes. However, the current rmap code needs to walk them
all, leading to O(N) scanning complexity for each page.
This can result in systems where one CPU is walking the page tables of
1000 processes in page_referenced_one, while all other CPUs are stuck on
the anon_vma lock. This leads to catastrophic failure for a benchmark
like AIM7, where the total number of processes can reach in the tens of
thousands. Real workloads are still a factor 10 less process intensive
than AIM7, but they are catching up.
This patch changes the way anon_vmas and VMAs are linked, which allows us
to associate multiple anon_vmas with a VMA. At fork time, each child
process gets its own anon_vmas, in which its COWed pages will be
instantiated. The parents' anon_vma is also linked to the VMA, because
non-COWed pages could be present in any of the children.
This reduces rmap scanning complexity to O(1) for the pages of the 1000
child processes, with O(N) complexity for at most 1/N pages in the system.
This reduces the average scanning cost in heavily forking workloads from
O(N) to 2.
The only real complexity in this patch stems from the fact that linking a
VMA to anon_vmas now involves memory allocations. This means vma_adjust
can fail, if it needs to attach a VMA to anon_vma structures. This in
turn means error handling needs to be added to the calling functions.
A second source of complexity is that, because there can be multiple
anon_vmas, the anon_vma linking in vma_adjust can no longer be done under
"the" anon_vma lock. To prevent the rmap code from walking up an
incomplete VMA, this patch introduces the VM_LOCK_RMAP VMA flag. This bit
flag uses the same slot as the NOMMU VM_MAPPED_COPY, with an ifdef in mm.h
to make sure it is impossible to compile a kernel that needs both symbolic
values for the same bitflag.
Some test results:
Without the anon_vma changes, when AIM7 hits around 9.7k users (on a test
box with 16GB RAM and not quite enough IO), the system ends up running
>99% in system time, with every CPU on the same anon_vma lock in the
pageout code.
With these changes, AIM7 hits the cross-over point around 29.7k users.
This happens with ~99% IO wait time, there never seems to be any spike in
system time. The anon_vma lock contention appears to be resolved.
[akpm@linux-foundation.org: cleanups]
Signed-off-by: Rik van Riel <riel@redhat.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Larry Woodman <lwoodman@redhat.com>
Cc: Lee Schermerhorn <Lee.Schermerhorn@hp.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-03-05 13:42:07 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
mm/rmap: improve anon_vma_clone(), unlink_anon_vmas() comments, add asserts
Patch series "mm: clean up anon_vma implementation", v3.
The anon_vma logic is hugely confusing and, much like a bundle of wires
entangled with one another, pulling on one thread seems only to lead to
more entanglement elsewhere.
There is a mish-mash of the core implementation, how that implementation
is invoked, how helper functions are invoked and concepts such as adjacent
anon_vma merge and anon_vma object reuse.
This series tries to improve the situation somewhat.
It starts by establishing some invariants in the core anon_vma_clone() and
unlink_anon_vmas() functions, largely expressed via VM_WARN_ON_ONCE()
asserts.
These act as some form of self-documentation as to the conditions we find
ourselves in when invoking these functions.
We also add kdoc comments for anon_vma_clone() and unlink_anon_vmas().
We then update anon_vma_fork() to avoid passing a partially set up (and
thus invalid) VMA to unlink_anon_vmas() - functions which are used both
for partially set up and valid data types has historically been the source
of a lot of confusion and bugs.
We then makes use of the established known conditions to directly skip
unfaulted VMAs (rather than implicitly via an empty vma->anon_vma_chain
list).
We remove the confusing anon_vma_merge() function (we already have a
concept of anon_vma merge in that we merge anon_vma's that would otherwise
be compatible except for attributes that mprotect() could change - which
anon_vma_merge() has nothing to do with).
We make the anon_vma functions internal to mm as they do not need to be
used by any other part of the kernel, which allows for future abstraction
without concern about this.
We then reduce the time over which we hold the anon rmap lock in
anon_vma_clone(), as it turns out we can allocate anon_vma_chain objects
without holding this lock, since the objects are not yet accessible from
the rmap.
This should reduce anon_vma lock contention.
This additionally allows us to remove a confusing GFP_NOWAIT, GFP_KERNEL
allocation fallback strategy.
Finally, we explicitly indicate which operation is being performed upon
anon_vma_clone(), and separate out fork-only logic to make it very clear
that anon_vma reuse only occurs on fork.
This patch (of 9):
Add kdoc comments and describe exactly what these functions are used for
in detail, pointing out importantly that the anon_vma_clone()
!dst->anon_vma && src->anon_vma dance is ONLY for fork.
Both are confusing functions that will be refactored in a subsequent patch
but the first stage is establishing documentation and some invariants.
Add some basic CONFIG_DEBUG_VM asserts that help document expected state,
specifically:
anon_vma_clone()
- mmap write lock held.
- We do nothing if src VMA is not faulted.
- The destination VMA has no anon_vma_chain yet.
- We are always operating on the same active VMA (i.e. vma->anon_vma).
- If not forking, must operate on the same mm_struct.
unlink_anon_vmas()
- mmap lock held (write lock except when freeing page tables).
- That unfaulted VMAs are no-ops.
We are presented with a special case when anon_vma_clone() fails to
allocate memory, where we have a VMA with partially set up anon_vma state.
Since we hold the exclusive mmap write lock, and since we are cloning
from a source VMA which consequently can't also have its anon_vma state
modified, we know no anon_vma referenced can be empty.
This allows us to significantly simplify this case and just remove
anon_vma_chain objects associated with the VMA, so we add a specific
partial cleanup path for this scenario.
This also allows us to drop the hack of setting vma->anon_vma to NULL
before unlinking anon_vma state in this scenario.
Link: https://lkml.kernel.org/r/cover.1768746221.git.lorenzo.stoakes@oracle.com
Link: https://lkml.kernel.org/r/8644e89369be0cc89d7ac57443dff9e822803c91.1768746221.git.lorenzo.stoakes@oracle.com
Signed-off-by: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Reviewed-by: Liam R. Howlett <Liam.Howlett@oracle.com>
Cc: Barry Song <v-songbaohua@oppo.com>
Cc: Chris Li <chriscli@google.com>
Cc: David Hildenbrand <david@kernel.org>
Cc: Harry Yoo <harry.yoo@oracle.com>
Cc: Jann Horn <jannh@google.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Mike Rapoport <rppt@kernel.org>
Cc: Pedro Falcato <pfalcato@suse.de>
Cc: Rik van Riel <riel@surriel.com>
Cc: Shakeel Butt <shakeel.butt@linux.dev>
Cc: Suren Baghdasaryan <surenb@google.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2026-01-18 14:50:37 +00:00
|
|
|
|
2026-01-18 14:50:45 +00:00
|
|
|
|
|
|
|
|
|
mm/rmap: improve anon_vma_clone(), unlink_anon_vmas() comments, add asserts
Patch series "mm: clean up anon_vma implementation", v3.
The anon_vma logic is hugely confusing and, much like a bundle of wires
entangled with one another, pulling on one thread seems only to lead to
more entanglement elsewhere.
There is a mish-mash of the core implementation, how that implementation
is invoked, how helper functions are invoked and concepts such as adjacent
anon_vma merge and anon_vma object reuse.
This series tries to improve the situation somewhat.
It starts by establishing some invariants in the core anon_vma_clone() and
unlink_anon_vmas() functions, largely expressed via VM_WARN_ON_ONCE()
asserts.
These act as some form of self-documentation as to the conditions we find
ourselves in when invoking these functions.
We also add kdoc comments for anon_vma_clone() and unlink_anon_vmas().
We then update anon_vma_fork() to avoid passing a partially set up (and
thus invalid) VMA to unlink_anon_vmas() - functions which are used both
for partially set up and valid data types has historically been the source
of a lot of confusion and bugs.
We then makes use of the established known conditions to directly skip
unfaulted VMAs (rather than implicitly via an empty vma->anon_vma_chain
list).
We remove the confusing anon_vma_merge() function (we already have a
concept of anon_vma merge in that we merge anon_vma's that would otherwise
be compatible except for attributes that mprotect() could change - which
anon_vma_merge() has nothing to do with).
We make the anon_vma functions internal to mm as they do not need to be
used by any other part of the kernel, which allows for future abstraction
without concern about this.
We then reduce the time over which we hold the anon rmap lock in
anon_vma_clone(), as it turns out we can allocate anon_vma_chain objects
without holding this lock, since the objects are not yet accessible from
the rmap.
This should reduce anon_vma lock contention.
This additionally allows us to remove a confusing GFP_NOWAIT, GFP_KERNEL
allocation fallback strategy.
Finally, we explicitly indicate which operation is being performed upon
anon_vma_clone(), and separate out fork-only logic to make it very clear
that anon_vma reuse only occurs on fork.
This patch (of 9):
Add kdoc comments and describe exactly what these functions are used for
in detail, pointing out importantly that the anon_vma_clone()
!dst->anon_vma && src->anon_vma dance is ONLY for fork.
Both are confusing functions that will be refactored in a subsequent patch
but the first stage is establishing documentation and some invariants.
Add some basic CONFIG_DEBUG_VM asserts that help document expected state,
specifically:
anon_vma_clone()
- mmap write lock held.
- We do nothing if src VMA is not faulted.
- The destination VMA has no anon_vma_chain yet.
- We are always operating on the same active VMA (i.e. vma->anon_vma).
- If not forking, must operate on the same mm_struct.
unlink_anon_vmas()
- mmap lock held (write lock except when freeing page tables).
- That unfaulted VMAs are no-ops.
We are presented with a special case when anon_vma_clone() fails to
allocate memory, where we have a VMA with partially set up anon_vma state.
Since we hold the exclusive mmap write lock, and since we are cloning
from a source VMA which consequently can't also have its anon_vma state
modified, we know no anon_vma referenced can be empty.
This allows us to significantly simplify this case and just remove
anon_vma_chain objects associated with the VMA, so we add a specific
partial cleanup path for this scenario.
This also allows us to drop the hack of setting vma->anon_vma to NULL
before unlinking anon_vma state in this scenario.
Link: https://lkml.kernel.org/r/cover.1768746221.git.lorenzo.stoakes@oracle.com
Link: https://lkml.kernel.org/r/8644e89369be0cc89d7ac57443dff9e822803c91.1768746221.git.lorenzo.stoakes@oracle.com
Signed-off-by: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Reviewed-by: Liam R. Howlett <Liam.Howlett@oracle.com>
Cc: Barry Song <v-songbaohua@oppo.com>
Cc: Chris Li <chriscli@google.com>
Cc: David Hildenbrand <david@kernel.org>
Cc: Harry Yoo <harry.yoo@oracle.com>
Cc: Jann Horn <jannh@google.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Mike Rapoport <rppt@kernel.org>
Cc: Pedro Falcato <pfalcato@suse.de>
Cc: Rik van Riel <riel@surriel.com>
Cc: Shakeel Butt <shakeel.butt@linux.dev>
Cc: Suren Baghdasaryan <surenb@google.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2026-01-18 14:50:37 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-18 14:50:45 +00:00
|
|
|
|
|
|
|
|
|
mm/rmap: improve anon_vma_clone(), unlink_anon_vmas() comments, add asserts
Patch series "mm: clean up anon_vma implementation", v3.
The anon_vma logic is hugely confusing and, much like a bundle of wires
entangled with one another, pulling on one thread seems only to lead to
more entanglement elsewhere.
There is a mish-mash of the core implementation, how that implementation
is invoked, how helper functions are invoked and concepts such as adjacent
anon_vma merge and anon_vma object reuse.
This series tries to improve the situation somewhat.
It starts by establishing some invariants in the core anon_vma_clone() and
unlink_anon_vmas() functions, largely expressed via VM_WARN_ON_ONCE()
asserts.
These act as some form of self-documentation as to the conditions we find
ourselves in when invoking these functions.
We also add kdoc comments for anon_vma_clone() and unlink_anon_vmas().
We then update anon_vma_fork() to avoid passing a partially set up (and
thus invalid) VMA to unlink_anon_vmas() - functions which are used both
for partially set up and valid data types has historically been the source
of a lot of confusion and bugs.
We then makes use of the established known conditions to directly skip
unfaulted VMAs (rather than implicitly via an empty vma->anon_vma_chain
list).
We remove the confusing anon_vma_merge() function (we already have a
concept of anon_vma merge in that we merge anon_vma's that would otherwise
be compatible except for attributes that mprotect() could change - which
anon_vma_merge() has nothing to do with).
We make the anon_vma functions internal to mm as they do not need to be
used by any other part of the kernel, which allows for future abstraction
without concern about this.
We then reduce the time over which we hold the anon rmap lock in
anon_vma_clone(), as it turns out we can allocate anon_vma_chain objects
without holding this lock, since the objects are not yet accessible from
the rmap.
This should reduce anon_vma lock contention.
This additionally allows us to remove a confusing GFP_NOWAIT, GFP_KERNEL
allocation fallback strategy.
Finally, we explicitly indicate which operation is being performed upon
anon_vma_clone(), and separate out fork-only logic to make it very clear
that anon_vma reuse only occurs on fork.
This patch (of 9):
Add kdoc comments and describe exactly what these functions are used for
in detail, pointing out importantly that the anon_vma_clone()
!dst->anon_vma && src->anon_vma dance is ONLY for fork.
Both are confusing functions that will be refactored in a subsequent patch
but the first stage is establishing documentation and some invariants.
Add some basic CONFIG_DEBUG_VM asserts that help document expected state,
specifically:
anon_vma_clone()
- mmap write lock held.
- We do nothing if src VMA is not faulted.
- The destination VMA has no anon_vma_chain yet.
- We are always operating on the same active VMA (i.e. vma->anon_vma).
- If not forking, must operate on the same mm_struct.
unlink_anon_vmas()
- mmap lock held (write lock except when freeing page tables).
- That unfaulted VMAs are no-ops.
We are presented with a special case when anon_vma_clone() fails to
allocate memory, where we have a VMA with partially set up anon_vma state.
Since we hold the exclusive mmap write lock, and since we are cloning
from a source VMA which consequently can't also have its anon_vma state
modified, we know no anon_vma referenced can be empty.
This allows us to significantly simplify this case and just remove
anon_vma_chain objects associated with the VMA, so we add a specific
partial cleanup path for this scenario.
This also allows us to drop the hack of setting vma->anon_vma to NULL
before unlinking anon_vma state in this scenario.
Link: https://lkml.kernel.org/r/cover.1768746221.git.lorenzo.stoakes@oracle.com
Link: https://lkml.kernel.org/r/8644e89369be0cc89d7ac57443dff9e822803c91.1768746221.git.lorenzo.stoakes@oracle.com
Signed-off-by: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Reviewed-by: Liam R. Howlett <Liam.Howlett@oracle.com>
Cc: Barry Song <v-songbaohua@oppo.com>
Cc: Chris Li <chriscli@google.com>
Cc: David Hildenbrand <david@kernel.org>
Cc: Harry Yoo <harry.yoo@oracle.com>
Cc: Jann Horn <jannh@google.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Mike Rapoport <rppt@kernel.org>
Cc: Pedro Falcato <pfalcato@suse.de>
Cc: Rik van Riel <riel@surriel.com>
Cc: Shakeel Butt <shakeel.butt@linux.dev>
Cc: Suren Baghdasaryan <surenb@google.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2026-01-18 14:50:37 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-18 14:50:45 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
mm/rmap: improve anon_vma_clone(), unlink_anon_vmas() comments, add asserts
Patch series "mm: clean up anon_vma implementation", v3.
The anon_vma logic is hugely confusing and, much like a bundle of wires
entangled with one another, pulling on one thread seems only to lead to
more entanglement elsewhere.
There is a mish-mash of the core implementation, how that implementation
is invoked, how helper functions are invoked and concepts such as adjacent
anon_vma merge and anon_vma object reuse.
This series tries to improve the situation somewhat.
It starts by establishing some invariants in the core anon_vma_clone() and
unlink_anon_vmas() functions, largely expressed via VM_WARN_ON_ONCE()
asserts.
These act as some form of self-documentation as to the conditions we find
ourselves in when invoking these functions.
We also add kdoc comments for anon_vma_clone() and unlink_anon_vmas().
We then update anon_vma_fork() to avoid passing a partially set up (and
thus invalid) VMA to unlink_anon_vmas() - functions which are used both
for partially set up and valid data types has historically been the source
of a lot of confusion and bugs.
We then makes use of the established known conditions to directly skip
unfaulted VMAs (rather than implicitly via an empty vma->anon_vma_chain
list).
We remove the confusing anon_vma_merge() function (we already have a
concept of anon_vma merge in that we merge anon_vma's that would otherwise
be compatible except for attributes that mprotect() could change - which
anon_vma_merge() has nothing to do with).
We make the anon_vma functions internal to mm as they do not need to be
used by any other part of the kernel, which allows for future abstraction
without concern about this.
We then reduce the time over which we hold the anon rmap lock in
anon_vma_clone(), as it turns out we can allocate anon_vma_chain objects
without holding this lock, since the objects are not yet accessible from
the rmap.
This should reduce anon_vma lock contention.
This additionally allows us to remove a confusing GFP_NOWAIT, GFP_KERNEL
allocation fallback strategy.
Finally, we explicitly indicate which operation is being performed upon
anon_vma_clone(), and separate out fork-only logic to make it very clear
that anon_vma reuse only occurs on fork.
This patch (of 9):
Add kdoc comments and describe exactly what these functions are used for
in detail, pointing out importantly that the anon_vma_clone()
!dst->anon_vma && src->anon_vma dance is ONLY for fork.
Both are confusing functions that will be refactored in a subsequent patch
but the first stage is establishing documentation and some invariants.
Add some basic CONFIG_DEBUG_VM asserts that help document expected state,
specifically:
anon_vma_clone()
- mmap write lock held.
- We do nothing if src VMA is not faulted.
- The destination VMA has no anon_vma_chain yet.
- We are always operating on the same active VMA (i.e. vma->anon_vma).
- If not forking, must operate on the same mm_struct.
unlink_anon_vmas()
- mmap lock held (write lock except when freeing page tables).
- That unfaulted VMAs are no-ops.
We are presented with a special case when anon_vma_clone() fails to
allocate memory, where we have a VMA with partially set up anon_vma state.
Since we hold the exclusive mmap write lock, and since we are cloning
from a source VMA which consequently can't also have its anon_vma state
modified, we know no anon_vma referenced can be empty.
This allows us to significantly simplify this case and just remove
anon_vma_chain objects associated with the VMA, so we add a specific
partial cleanup path for this scenario.
This also allows us to drop the hack of setting vma->anon_vma to NULL
before unlinking anon_vma state in this scenario.
Link: https://lkml.kernel.org/r/cover.1768746221.git.lorenzo.stoakes@oracle.com
Link: https://lkml.kernel.org/r/8644e89369be0cc89d7ac57443dff9e822803c91.1768746221.git.lorenzo.stoakes@oracle.com
Signed-off-by: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Reviewed-by: Liam R. Howlett <Liam.Howlett@oracle.com>
Cc: Barry Song <v-songbaohua@oppo.com>
Cc: Chris Li <chriscli@google.com>
Cc: David Hildenbrand <david@kernel.org>
Cc: Harry Yoo <harry.yoo@oracle.com>
Cc: Jann Horn <jannh@google.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Mike Rapoport <rppt@kernel.org>
Cc: Pedro Falcato <pfalcato@suse.de>
Cc: Rik van Riel <riel@surriel.com>
Cc: Shakeel Butt <shakeel.butt@linux.dev>
Cc: Suren Baghdasaryan <surenb@google.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2026-01-18 14:50:37 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-18 14:50:45 +00:00
|
|
|
|
2015-01-08 14:32:15 -08:00
|
|
|
|
mm/rmap: improve anon_vma_clone(), unlink_anon_vmas() comments, add asserts
Patch series "mm: clean up anon_vma implementation", v3.
The anon_vma logic is hugely confusing and, much like a bundle of wires
entangled with one another, pulling on one thread seems only to lead to
more entanglement elsewhere.
There is a mish-mash of the core implementation, how that implementation
is invoked, how helper functions are invoked and concepts such as adjacent
anon_vma merge and anon_vma object reuse.
This series tries to improve the situation somewhat.
It starts by establishing some invariants in the core anon_vma_clone() and
unlink_anon_vmas() functions, largely expressed via VM_WARN_ON_ONCE()
asserts.
These act as some form of self-documentation as to the conditions we find
ourselves in when invoking these functions.
We also add kdoc comments for anon_vma_clone() and unlink_anon_vmas().
We then update anon_vma_fork() to avoid passing a partially set up (and
thus invalid) VMA to unlink_anon_vmas() - functions which are used both
for partially set up and valid data types has historically been the source
of a lot of confusion and bugs.
We then makes use of the established known conditions to directly skip
unfaulted VMAs (rather than implicitly via an empty vma->anon_vma_chain
list).
We remove the confusing anon_vma_merge() function (we already have a
concept of anon_vma merge in that we merge anon_vma's that would otherwise
be compatible except for attributes that mprotect() could change - which
anon_vma_merge() has nothing to do with).
We make the anon_vma functions internal to mm as they do not need to be
used by any other part of the kernel, which allows for future abstraction
without concern about this.
We then reduce the time over which we hold the anon rmap lock in
anon_vma_clone(), as it turns out we can allocate anon_vma_chain objects
without holding this lock, since the objects are not yet accessible from
the rmap.
This should reduce anon_vma lock contention.
This additionally allows us to remove a confusing GFP_NOWAIT, GFP_KERNEL
allocation fallback strategy.
Finally, we explicitly indicate which operation is being performed upon
anon_vma_clone(), and separate out fork-only logic to make it very clear
that anon_vma reuse only occurs on fork.
This patch (of 9):
Add kdoc comments and describe exactly what these functions are used for
in detail, pointing out importantly that the anon_vma_clone()
!dst->anon_vma && src->anon_vma dance is ONLY for fork.
Both are confusing functions that will be refactored in a subsequent patch
but the first stage is establishing documentation and some invariants.
Add some basic CONFIG_DEBUG_VM asserts that help document expected state,
specifically:
anon_vma_clone()
- mmap write lock held.
- We do nothing if src VMA is not faulted.
- The destination VMA has no anon_vma_chain yet.
- We are always operating on the same active VMA (i.e. vma->anon_vma).
- If not forking, must operate on the same mm_struct.
unlink_anon_vmas()
- mmap lock held (write lock except when freeing page tables).
- That unfaulted VMAs are no-ops.
We are presented with a special case when anon_vma_clone() fails to
allocate memory, where we have a VMA with partially set up anon_vma state.
Since we hold the exclusive mmap write lock, and since we are cloning
from a source VMA which consequently can't also have its anon_vma state
modified, we know no anon_vma referenced can be empty.
This allows us to significantly simplify this case and just remove
anon_vma_chain objects associated with the VMA, so we add a specific
partial cleanup path for this scenario.
This also allows us to drop the hack of setting vma->anon_vma to NULL
before unlinking anon_vma state in this scenario.
Link: https://lkml.kernel.org/r/cover.1768746221.git.lorenzo.stoakes@oracle.com
Link: https://lkml.kernel.org/r/8644e89369be0cc89d7ac57443dff9e822803c91.1768746221.git.lorenzo.stoakes@oracle.com
Signed-off-by: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Reviewed-by: Liam R. Howlett <Liam.Howlett@oracle.com>
Cc: Barry Song <v-songbaohua@oppo.com>
Cc: Chris Li <chriscli@google.com>
Cc: David Hildenbrand <david@kernel.org>
Cc: Harry Yoo <harry.yoo@oracle.com>
Cc: Jann Horn <jannh@google.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Mike Rapoport <rppt@kernel.org>
Cc: Pedro Falcato <pfalcato@suse.de>
Cc: Rik van Riel <riel@surriel.com>
Cc: Shakeel Butt <shakeel.butt@linux.dev>
Cc: Suren Baghdasaryan <surenb@google.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2026-01-18 14:50:37 +00:00
|
|
|
|
|
|
|
|
|
2019-11-30 17:50:56 -08:00
|
|
|
|
mm/rmap: improve anon_vma_clone(), unlink_anon_vmas() comments, add asserts
Patch series "mm: clean up anon_vma implementation", v3.
The anon_vma logic is hugely confusing and, much like a bundle of wires
entangled with one another, pulling on one thread seems only to lead to
more entanglement elsewhere.
There is a mish-mash of the core implementation, how that implementation
is invoked, how helper functions are invoked and concepts such as adjacent
anon_vma merge and anon_vma object reuse.
This series tries to improve the situation somewhat.
It starts by establishing some invariants in the core anon_vma_clone() and
unlink_anon_vmas() functions, largely expressed via VM_WARN_ON_ONCE()
asserts.
These act as some form of self-documentation as to the conditions we find
ourselves in when invoking these functions.
We also add kdoc comments for anon_vma_clone() and unlink_anon_vmas().
We then update anon_vma_fork() to avoid passing a partially set up (and
thus invalid) VMA to unlink_anon_vmas() - functions which are used both
for partially set up and valid data types has historically been the source
of a lot of confusion and bugs.
We then makes use of the established known conditions to directly skip
unfaulted VMAs (rather than implicitly via an empty vma->anon_vma_chain
list).
We remove the confusing anon_vma_merge() function (we already have a
concept of anon_vma merge in that we merge anon_vma's that would otherwise
be compatible except for attributes that mprotect() could change - which
anon_vma_merge() has nothing to do with).
We make the anon_vma functions internal to mm as they do not need to be
used by any other part of the kernel, which allows for future abstraction
without concern about this.
We then reduce the time over which we hold the anon rmap lock in
anon_vma_clone(), as it turns out we can allocate anon_vma_chain objects
without holding this lock, since the objects are not yet accessible from
the rmap.
This should reduce anon_vma lock contention.
This additionally allows us to remove a confusing GFP_NOWAIT, GFP_KERNEL
allocation fallback strategy.
Finally, we explicitly indicate which operation is being performed upon
anon_vma_clone(), and separate out fork-only logic to make it very clear
that anon_vma reuse only occurs on fork.
This patch (of 9):
Add kdoc comments and describe exactly what these functions are used for
in detail, pointing out importantly that the anon_vma_clone()
!dst->anon_vma && src->anon_vma dance is ONLY for fork.
Both are confusing functions that will be refactored in a subsequent patch
but the first stage is establishing documentation and some invariants.
Add some basic CONFIG_DEBUG_VM asserts that help document expected state,
specifically:
anon_vma_clone()
- mmap write lock held.
- We do nothing if src VMA is not faulted.
- The destination VMA has no anon_vma_chain yet.
- We are always operating on the same active VMA (i.e. vma->anon_vma).
- If not forking, must operate on the same mm_struct.
unlink_anon_vmas()
- mmap lock held (write lock except when freeing page tables).
- That unfaulted VMAs are no-ops.
We are presented with a special case when anon_vma_clone() fails to
allocate memory, where we have a VMA with partially set up anon_vma state.
Since we hold the exclusive mmap write lock, and since we are cloning
from a source VMA which consequently can't also have its anon_vma state
modified, we know no anon_vma referenced can be empty.
This allows us to significantly simplify this case and just remove
anon_vma_chain objects associated with the VMA, so we add a specific
partial cleanup path for this scenario.
This also allows us to drop the hack of setting vma->anon_vma to NULL
before unlinking anon_vma state in this scenario.
Link: https://lkml.kernel.org/r/cover.1768746221.git.lorenzo.stoakes@oracle.com
Link: https://lkml.kernel.org/r/8644e89369be0cc89d7ac57443dff9e822803c91.1768746221.git.lorenzo.stoakes@oracle.com
Signed-off-by: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Reviewed-by: Liam R. Howlett <Liam.Howlett@oracle.com>
Cc: Barry Song <v-songbaohua@oppo.com>
Cc: Chris Li <chriscli@google.com>
Cc: David Hildenbrand <david@kernel.org>
Cc: Harry Yoo <harry.yoo@oracle.com>
Cc: Jann Horn <jannh@google.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Mike Rapoport <rppt@kernel.org>
Cc: Pedro Falcato <pfalcato@suse.de>
Cc: Rik van Riel <riel@surriel.com>
Cc: Shakeel Butt <shakeel.butt@linux.dev>
Cc: Suren Baghdasaryan <surenb@google.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2026-01-18 14:50:37 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
mm: change anon_vma linking to fix multi-process server scalability issue
The old anon_vma code can lead to scalability issues with heavily forking
workloads. Specifically, each anon_vma will be shared between the parent
process and all its child processes.
In a workload with 1000 child processes and a VMA with 1000 anonymous
pages per process that get COWed, this leads to a system with a million
anonymous pages in the same anon_vma, each of which is mapped in just one
of the 1000 processes. However, the current rmap code needs to walk them
all, leading to O(N) scanning complexity for each page.
This can result in systems where one CPU is walking the page tables of
1000 processes in page_referenced_one, while all other CPUs are stuck on
the anon_vma lock. This leads to catastrophic failure for a benchmark
like AIM7, where the total number of processes can reach in the tens of
thousands. Real workloads are still a factor 10 less process intensive
than AIM7, but they are catching up.
This patch changes the way anon_vmas and VMAs are linked, which allows us
to associate multiple anon_vmas with a VMA. At fork time, each child
process gets its own anon_vmas, in which its COWed pages will be
instantiated. The parents' anon_vma is also linked to the VMA, because
non-COWed pages could be present in any of the children.
This reduces rmap scanning complexity to O(1) for the pages of the 1000
child processes, with O(N) complexity for at most 1/N pages in the system.
This reduces the average scanning cost in heavily forking workloads from
O(N) to 2.
The only real complexity in this patch stems from the fact that linking a
VMA to anon_vmas now involves memory allocations. This means vma_adjust
can fail, if it needs to attach a VMA to anon_vma structures. This in
turn means error handling needs to be added to the calling functions.
A second source of complexity is that, because there can be multiple
anon_vmas, the anon_vma linking in vma_adjust can no longer be done under
"the" anon_vma lock. To prevent the rmap code from walking up an
incomplete VMA, this patch introduces the VM_LOCK_RMAP VMA flag. This bit
flag uses the same slot as the NOMMU VM_MAPPED_COPY, with an ifdef in mm.h
to make sure it is impossible to compile a kernel that needs both symbolic
values for the same bitflag.
Some test results:
Without the anon_vma changes, when AIM7 hits around 9.7k users (on a test
box with 16GB RAM and not quite enough IO), the system ends up running
>99% in system time, with every CPU on the same anon_vma lock in the
pageout code.
With these changes, AIM7 hits the cross-over point around 29.7k users.
This happens with ~99% IO wait time, there never seems to be any spike in
system time. The anon_vma lock contention appears to be resolved.
[akpm@linux-foundation.org: cleanups]
Signed-off-by: Rik van Riel <riel@redhat.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Larry Woodman <lwoodman@redhat.com>
Cc: Lee Schermerhorn <Lee.Schermerhorn@hp.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-03-05 13:42:07 -08:00
|
|
|
|
2026-01-18 14:50:45 +00:00
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
mm: change anon_vma linking to fix multi-process server scalability issue
The old anon_vma code can lead to scalability issues with heavily forking
workloads. Specifically, each anon_vma will be shared between the parent
process and all its child processes.
In a workload with 1000 child processes and a VMA with 1000 anonymous
pages per process that get COWed, this leads to a system with a million
anonymous pages in the same anon_vma, each of which is mapped in just one
of the 1000 processes. However, the current rmap code needs to walk them
all, leading to O(N) scanning complexity for each page.
This can result in systems where one CPU is walking the page tables of
1000 processes in page_referenced_one, while all other CPUs are stuck on
the anon_vma lock. This leads to catastrophic failure for a benchmark
like AIM7, where the total number of processes can reach in the tens of
thousands. Real workloads are still a factor 10 less process intensive
than AIM7, but they are catching up.
This patch changes the way anon_vmas and VMAs are linked, which allows us
to associate multiple anon_vmas with a VMA. At fork time, each child
process gets its own anon_vmas, in which its COWed pages will be
instantiated. The parents' anon_vma is also linked to the VMA, because
non-COWed pages could be present in any of the children.
This reduces rmap scanning complexity to O(1) for the pages of the 1000
child processes, with O(N) complexity for at most 1/N pages in the system.
This reduces the average scanning cost in heavily forking workloads from
O(N) to 2.
The only real complexity in this patch stems from the fact that linking a
VMA to anon_vmas now involves memory allocations. This means vma_adjust
can fail, if it needs to attach a VMA to anon_vma structures. This in
turn means error handling needs to be added to the calling functions.
A second source of complexity is that, because there can be multiple
anon_vmas, the anon_vma linking in vma_adjust can no longer be done under
"the" anon_vma lock. To prevent the rmap code from walking up an
incomplete VMA, this patch introduces the VM_LOCK_RMAP VMA flag. This bit
flag uses the same slot as the NOMMU VM_MAPPED_COPY, with an ifdef in mm.h
to make sure it is impossible to compile a kernel that needs both symbolic
values for the same bitflag.
Some test results:
Without the anon_vma changes, when AIM7 hits around 9.7k users (on a test
box with 16GB RAM and not quite enough IO), the system ends up running
>99% in system time, with every CPU on the same anon_vma lock in the
pageout code.
With these changes, AIM7 hits the cross-over point around 29.7k users.
This happens with ~99% IO wait time, there never seems to be any spike in
system time. The anon_vma lock contention appears to be resolved.
[akpm@linux-foundation.org: cleanups]
Signed-off-by: Rik van Riel <riel@redhat.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Larry Woodman <lwoodman@redhat.com>
Cc: Lee Schermerhorn <Lee.Schermerhorn@hp.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-03-05 13:42:07 -08:00
|
|
|
|
2026-01-18 14:50:45 +00:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2026-01-18 14:50:45 +00:00
|
|
|
|
mm/rmap: improve anon_vma_clone(), unlink_anon_vmas() comments, add asserts
Patch series "mm: clean up anon_vma implementation", v3.
The anon_vma logic is hugely confusing and, much like a bundle of wires
entangled with one another, pulling on one thread seems only to lead to
more entanglement elsewhere.
There is a mish-mash of the core implementation, how that implementation
is invoked, how helper functions are invoked and concepts such as adjacent
anon_vma merge and anon_vma object reuse.
This series tries to improve the situation somewhat.
It starts by establishing some invariants in the core anon_vma_clone() and
unlink_anon_vmas() functions, largely expressed via VM_WARN_ON_ONCE()
asserts.
These act as some form of self-documentation as to the conditions we find
ourselves in when invoking these functions.
We also add kdoc comments for anon_vma_clone() and unlink_anon_vmas().
We then update anon_vma_fork() to avoid passing a partially set up (and
thus invalid) VMA to unlink_anon_vmas() - functions which are used both
for partially set up and valid data types has historically been the source
of a lot of confusion and bugs.
We then makes use of the established known conditions to directly skip
unfaulted VMAs (rather than implicitly via an empty vma->anon_vma_chain
list).
We remove the confusing anon_vma_merge() function (we already have a
concept of anon_vma merge in that we merge anon_vma's that would otherwise
be compatible except for attributes that mprotect() could change - which
anon_vma_merge() has nothing to do with).
We make the anon_vma functions internal to mm as they do not need to be
used by any other part of the kernel, which allows for future abstraction
without concern about this.
We then reduce the time over which we hold the anon rmap lock in
anon_vma_clone(), as it turns out we can allocate anon_vma_chain objects
without holding this lock, since the objects are not yet accessible from
the rmap.
This should reduce anon_vma lock contention.
This additionally allows us to remove a confusing GFP_NOWAIT, GFP_KERNEL
allocation fallback strategy.
Finally, we explicitly indicate which operation is being performed upon
anon_vma_clone(), and separate out fork-only logic to make it very clear
that anon_vma reuse only occurs on fork.
This patch (of 9):
Add kdoc comments and describe exactly what these functions are used for
in detail, pointing out importantly that the anon_vma_clone()
!dst->anon_vma && src->anon_vma dance is ONLY for fork.
Both are confusing functions that will be refactored in a subsequent patch
but the first stage is establishing documentation and some invariants.
Add some basic CONFIG_DEBUG_VM asserts that help document expected state,
specifically:
anon_vma_clone()
- mmap write lock held.
- We do nothing if src VMA is not faulted.
- The destination VMA has no anon_vma_chain yet.
- We are always operating on the same active VMA (i.e. vma->anon_vma).
- If not forking, must operate on the same mm_struct.
unlink_anon_vmas()
- mmap lock held (write lock except when freeing page tables).
- That unfaulted VMAs are no-ops.
We are presented with a special case when anon_vma_clone() fails to
allocate memory, where we have a VMA with partially set up anon_vma state.
Since we hold the exclusive mmap write lock, and since we are cloning
from a source VMA which consequently can't also have its anon_vma state
modified, we know no anon_vma referenced can be empty.
This allows us to significantly simplify this case and just remove
anon_vma_chain objects associated with the VMA, so we add a specific
partial cleanup path for this scenario.
This also allows us to drop the hack of setting vma->anon_vma to NULL
before unlinking anon_vma state in this scenario.
Link: https://lkml.kernel.org/r/cover.1768746221.git.lorenzo.stoakes@oracle.com
Link: https://lkml.kernel.org/r/8644e89369be0cc89d7ac57443dff9e822803c91.1768746221.git.lorenzo.stoakes@oracle.com
Signed-off-by: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Reviewed-by: Liam R. Howlett <Liam.Howlett@oracle.com>
Cc: Barry Song <v-songbaohua@oppo.com>
Cc: Chris Li <chriscli@google.com>
Cc: David Hildenbrand <david@kernel.org>
Cc: Harry Yoo <harry.yoo@oracle.com>
Cc: Jann Horn <jannh@google.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Mike Rapoport <rppt@kernel.org>
Cc: Pedro Falcato <pfalcato@suse.de>
Cc: Rik van Riel <riel@surriel.com>
Cc: Shakeel Butt <shakeel.butt@linux.dev>
Cc: Suren Baghdasaryan <surenb@google.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2026-01-18 14:50:37 +00:00
|
|
|
|
2026-01-18 14:50:45 +00:00
|
|
|
|
2026-01-18 14:50:39 +00:00
|
|
|
|
|
|
|
|
|
2026-01-18 14:50:44 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2011-06-16 20:44:51 -07:00
|
|
|
|
2026-01-18 14:50:44 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-18 14:50:45 +00:00
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2026-01-18 14:50:45 +00:00
|
|
|
|
|
|
|
|
|
2022-08-31 19:06:00 +02:00
|
|
|
|
2026-01-18 14:50:45 +00:00
|
|
|
|
|
|
|
|
|
mm: change anon_vma linking to fix multi-process server scalability issue
The old anon_vma code can lead to scalability issues with heavily forking
workloads. Specifically, each anon_vma will be shared between the parent
process and all its child processes.
In a workload with 1000 child processes and a VMA with 1000 anonymous
pages per process that get COWed, this leads to a system with a million
anonymous pages in the same anon_vma, each of which is mapped in just one
of the 1000 processes. However, the current rmap code needs to walk them
all, leading to O(N) scanning complexity for each page.
This can result in systems where one CPU is walking the page tables of
1000 processes in page_referenced_one, while all other CPUs are stuck on
the anon_vma lock. This leads to catastrophic failure for a benchmark
like AIM7, where the total number of processes can reach in the tens of
thousands. Real workloads are still a factor 10 less process intensive
than AIM7, but they are catching up.
This patch changes the way anon_vmas and VMAs are linked, which allows us
to associate multiple anon_vmas with a VMA. At fork time, each child
process gets its own anon_vmas, in which its COWed pages will be
instantiated. The parents' anon_vma is also linked to the VMA, because
non-COWed pages could be present in any of the children.
This reduces rmap scanning complexity to O(1) for the pages of the 1000
child processes, with O(N) complexity for at most 1/N pages in the system.
This reduces the average scanning cost in heavily forking workloads from
O(N) to 2.
The only real complexity in this patch stems from the fact that linking a
VMA to anon_vmas now involves memory allocations. This means vma_adjust
can fail, if it needs to attach a VMA to anon_vma structures. This in
turn means error handling needs to be added to the calling functions.
A second source of complexity is that, because there can be multiple
anon_vmas, the anon_vma linking in vma_adjust can no longer be done under
"the" anon_vma lock. To prevent the rmap code from walking up an
incomplete VMA, this patch introduces the VM_LOCK_RMAP VMA flag. This bit
flag uses the same slot as the NOMMU VM_MAPPED_COPY, with an ifdef in mm.h
to make sure it is impossible to compile a kernel that needs both symbolic
values for the same bitflag.
Some test results:
Without the anon_vma changes, when AIM7 hits around 9.7k users (on a test
box with 16GB RAM and not quite enough IO), the system ends up running
>99% in system time, with every CPU on the same anon_vma lock in the
pageout code.
With these changes, AIM7 hits the cross-over point around 29.7k users.
This happens with ~99% IO wait time, there never seems to be any spike in
system time. The anon_vma lock contention appears to be resolved.
[akpm@linux-foundation.org: cleanups]
Signed-off-by: Rik van Riel <riel@redhat.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Larry Woodman <lwoodman@redhat.com>
Cc: Lee Schermerhorn <Lee.Schermerhorn@hp.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-03-05 13:42:07 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
mm/rmap: improve anon_vma_clone(), unlink_anon_vmas() comments, add asserts
Patch series "mm: clean up anon_vma implementation", v3.
The anon_vma logic is hugely confusing and, much like a bundle of wires
entangled with one another, pulling on one thread seems only to lead to
more entanglement elsewhere.
There is a mish-mash of the core implementation, how that implementation
is invoked, how helper functions are invoked and concepts such as adjacent
anon_vma merge and anon_vma object reuse.
This series tries to improve the situation somewhat.
It starts by establishing some invariants in the core anon_vma_clone() and
unlink_anon_vmas() functions, largely expressed via VM_WARN_ON_ONCE()
asserts.
These act as some form of self-documentation as to the conditions we find
ourselves in when invoking these functions.
We also add kdoc comments for anon_vma_clone() and unlink_anon_vmas().
We then update anon_vma_fork() to avoid passing a partially set up (and
thus invalid) VMA to unlink_anon_vmas() - functions which are used both
for partially set up and valid data types has historically been the source
of a lot of confusion and bugs.
We then makes use of the established known conditions to directly skip
unfaulted VMAs (rather than implicitly via an empty vma->anon_vma_chain
list).
We remove the confusing anon_vma_merge() function (we already have a
concept of anon_vma merge in that we merge anon_vma's that would otherwise
be compatible except for attributes that mprotect() could change - which
anon_vma_merge() has nothing to do with).
We make the anon_vma functions internal to mm as they do not need to be
used by any other part of the kernel, which allows for future abstraction
without concern about this.
We then reduce the time over which we hold the anon rmap lock in
anon_vma_clone(), as it turns out we can allocate anon_vma_chain objects
without holding this lock, since the objects are not yet accessible from
the rmap.
This should reduce anon_vma lock contention.
This additionally allows us to remove a confusing GFP_NOWAIT, GFP_KERNEL
allocation fallback strategy.
Finally, we explicitly indicate which operation is being performed upon
anon_vma_clone(), and separate out fork-only logic to make it very clear
that anon_vma reuse only occurs on fork.
This patch (of 9):
Add kdoc comments and describe exactly what these functions are used for
in detail, pointing out importantly that the anon_vma_clone()
!dst->anon_vma && src->anon_vma dance is ONLY for fork.
Both are confusing functions that will be refactored in a subsequent patch
but the first stage is establishing documentation and some invariants.
Add some basic CONFIG_DEBUG_VM asserts that help document expected state,
specifically:
anon_vma_clone()
- mmap write lock held.
- We do nothing if src VMA is not faulted.
- The destination VMA has no anon_vma_chain yet.
- We are always operating on the same active VMA (i.e. vma->anon_vma).
- If not forking, must operate on the same mm_struct.
unlink_anon_vmas()
- mmap lock held (write lock except when freeing page tables).
- That unfaulted VMAs are no-ops.
We are presented with a special case when anon_vma_clone() fails to
allocate memory, where we have a VMA with partially set up anon_vma state.
Since we hold the exclusive mmap write lock, and since we are cloning
from a source VMA which consequently can't also have its anon_vma state
modified, we know no anon_vma referenced can be empty.
This allows us to significantly simplify this case and just remove
anon_vma_chain objects associated with the VMA, so we add a specific
partial cleanup path for this scenario.
This also allows us to drop the hack of setting vma->anon_vma to NULL
before unlinking anon_vma state in this scenario.
Link: https://lkml.kernel.org/r/cover.1768746221.git.lorenzo.stoakes@oracle.com
Link: https://lkml.kernel.org/r/8644e89369be0cc89d7ac57443dff9e822803c91.1768746221.git.lorenzo.stoakes@oracle.com
Signed-off-by: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Reviewed-by: Liam R. Howlett <Liam.Howlett@oracle.com>
Cc: Barry Song <v-songbaohua@oppo.com>
Cc: Chris Li <chriscli@google.com>
Cc: David Hildenbrand <david@kernel.org>
Cc: Harry Yoo <harry.yoo@oracle.com>
Cc: Jann Horn <jannh@google.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Mike Rapoport <rppt@kernel.org>
Cc: Pedro Falcato <pfalcato@suse.de>
Cc: Rik van Riel <riel@surriel.com>
Cc: Shakeel Butt <shakeel.butt@linux.dev>
Cc: Suren Baghdasaryan <surenb@google.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2026-01-18 14:50:37 +00:00
|
|
|
|
mm: change anon_vma linking to fix multi-process server scalability issue
The old anon_vma code can lead to scalability issues with heavily forking
workloads. Specifically, each anon_vma will be shared between the parent
process and all its child processes.
In a workload with 1000 child processes and a VMA with 1000 anonymous
pages per process that get COWed, this leads to a system with a million
anonymous pages in the same anon_vma, each of which is mapped in just one
of the 1000 processes. However, the current rmap code needs to walk them
all, leading to O(N) scanning complexity for each page.
This can result in systems where one CPU is walking the page tables of
1000 processes in page_referenced_one, while all other CPUs are stuck on
the anon_vma lock. This leads to catastrophic failure for a benchmark
like AIM7, where the total number of processes can reach in the tens of
thousands. Real workloads are still a factor 10 less process intensive
than AIM7, but they are catching up.
This patch changes the way anon_vmas and VMAs are linked, which allows us
to associate multiple anon_vmas with a VMA. At fork time, each child
process gets its own anon_vmas, in which its COWed pages will be
instantiated. The parents' anon_vma is also linked to the VMA, because
non-COWed pages could be present in any of the children.
This reduces rmap scanning complexity to O(1) for the pages of the 1000
child processes, with O(N) complexity for at most 1/N pages in the system.
This reduces the average scanning cost in heavily forking workloads from
O(N) to 2.
The only real complexity in this patch stems from the fact that linking a
VMA to anon_vmas now involves memory allocations. This means vma_adjust
can fail, if it needs to attach a VMA to anon_vma structures. This in
turn means error handling needs to be added to the calling functions.
A second source of complexity is that, because there can be multiple
anon_vmas, the anon_vma linking in vma_adjust can no longer be done under
"the" anon_vma lock. To prevent the rmap code from walking up an
incomplete VMA, this patch introduces the VM_LOCK_RMAP VMA flag. This bit
flag uses the same slot as the NOMMU VM_MAPPED_COPY, with an ifdef in mm.h
to make sure it is impossible to compile a kernel that needs both symbolic
values for the same bitflag.
Some test results:
Without the anon_vma changes, when AIM7 hits around 9.7k users (on a test
box with 16GB RAM and not quite enough IO), the system ends up running
>99% in system time, with every CPU on the same anon_vma lock in the
pageout code.
With these changes, AIM7 hits the cross-over point around 29.7k users.
This happens with ~99% IO wait time, there never seems to be any spike in
system time. The anon_vma lock contention appears to be resolved.
[akpm@linux-foundation.org: cleanups]
Signed-off-by: Rik van Riel <riel@redhat.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Larry Woodman <lwoodman@redhat.com>
Cc: Lee Schermerhorn <Lee.Schermerhorn@hp.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-03-05 13:42:07 -08:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
mm: change anon_vma linking to fix multi-process server scalability issue
The old anon_vma code can lead to scalability issues with heavily forking
workloads. Specifically, each anon_vma will be shared between the parent
process and all its child processes.
In a workload with 1000 child processes and a VMA with 1000 anonymous
pages per process that get COWed, this leads to a system with a million
anonymous pages in the same anon_vma, each of which is mapped in just one
of the 1000 processes. However, the current rmap code needs to walk them
all, leading to O(N) scanning complexity for each page.
This can result in systems where one CPU is walking the page tables of
1000 processes in page_referenced_one, while all other CPUs are stuck on
the anon_vma lock. This leads to catastrophic failure for a benchmark
like AIM7, where the total number of processes can reach in the tens of
thousands. Real workloads are still a factor 10 less process intensive
than AIM7, but they are catching up.
This patch changes the way anon_vmas and VMAs are linked, which allows us
to associate multiple anon_vmas with a VMA. At fork time, each child
process gets its own anon_vmas, in which its COWed pages will be
instantiated. The parents' anon_vma is also linked to the VMA, because
non-COWed pages could be present in any of the children.
This reduces rmap scanning complexity to O(1) for the pages of the 1000
child processes, with O(N) complexity for at most 1/N pages in the system.
This reduces the average scanning cost in heavily forking workloads from
O(N) to 2.
The only real complexity in this patch stems from the fact that linking a
VMA to anon_vmas now involves memory allocations. This means vma_adjust
can fail, if it needs to attach a VMA to anon_vma structures. This in
turn means error handling needs to be added to the calling functions.
A second source of complexity is that, because there can be multiple
anon_vmas, the anon_vma linking in vma_adjust can no longer be done under
"the" anon_vma lock. To prevent the rmap code from walking up an
incomplete VMA, this patch introduces the VM_LOCK_RMAP VMA flag. This bit
flag uses the same slot as the NOMMU VM_MAPPED_COPY, with an ifdef in mm.h
to make sure it is impossible to compile a kernel that needs both symbolic
values for the same bitflag.
Some test results:
Without the anon_vma changes, when AIM7 hits around 9.7k users (on a test
box with 16GB RAM and not quite enough IO), the system ends up running
>99% in system time, with every CPU on the same anon_vma lock in the
pageout code.
With these changes, AIM7 hits the cross-over point around 29.7k users.
This happens with ~99% IO wait time, there never seems to be any spike in
system time. The anon_vma lock contention appears to be resolved.
[akpm@linux-foundation.org: cleanups]
Signed-off-by: Rik van Riel <riel@redhat.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Larry Woodman <lwoodman@redhat.com>
Cc: Lee Schermerhorn <Lee.Schermerhorn@hp.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-03-05 13:42:07 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
mm: change anon_vma linking to fix multi-process server scalability issue
The old anon_vma code can lead to scalability issues with heavily forking
workloads. Specifically, each anon_vma will be shared between the parent
process and all its child processes.
In a workload with 1000 child processes and a VMA with 1000 anonymous
pages per process that get COWed, this leads to a system with a million
anonymous pages in the same anon_vma, each of which is mapped in just one
of the 1000 processes. However, the current rmap code needs to walk them
all, leading to O(N) scanning complexity for each page.
This can result in systems where one CPU is walking the page tables of
1000 processes in page_referenced_one, while all other CPUs are stuck on
the anon_vma lock. This leads to catastrophic failure for a benchmark
like AIM7, where the total number of processes can reach in the tens of
thousands. Real workloads are still a factor 10 less process intensive
than AIM7, but they are catching up.
This patch changes the way anon_vmas and VMAs are linked, which allows us
to associate multiple anon_vmas with a VMA. At fork time, each child
process gets its own anon_vmas, in which its COWed pages will be
instantiated. The parents' anon_vma is also linked to the VMA, because
non-COWed pages could be present in any of the children.
This reduces rmap scanning complexity to O(1) for the pages of the 1000
child processes, with O(N) complexity for at most 1/N pages in the system.
This reduces the average scanning cost in heavily forking workloads from
O(N) to 2.
The only real complexity in this patch stems from the fact that linking a
VMA to anon_vmas now involves memory allocations. This means vma_adjust
can fail, if it needs to attach a VMA to anon_vma structures. This in
turn means error handling needs to be added to the calling functions.
A second source of complexity is that, because there can be multiple
anon_vmas, the anon_vma linking in vma_adjust can no longer be done under
"the" anon_vma lock. To prevent the rmap code from walking up an
incomplete VMA, this patch introduces the VM_LOCK_RMAP VMA flag. This bit
flag uses the same slot as the NOMMU VM_MAPPED_COPY, with an ifdef in mm.h
to make sure it is impossible to compile a kernel that needs both symbolic
values for the same bitflag.
Some test results:
Without the anon_vma changes, when AIM7 hits around 9.7k users (on a test
box with 16GB RAM and not quite enough IO), the system ends up running
>99% in system time, with every CPU on the same anon_vma lock in the
pageout code.
With these changes, AIM7 hits the cross-over point around 29.7k users.
This happens with ~99% IO wait time, there never seems to be any spike in
system time. The anon_vma lock contention appears to be resolved.
[akpm@linux-foundation.org: cleanups]
Signed-off-by: Rik van Riel <riel@redhat.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Larry Woodman <lwoodman@redhat.com>
Cc: Lee Schermerhorn <Lee.Schermerhorn@hp.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-03-05 13:42:07 -08:00
|
|
|
|
|
|
|
|
|
mm/rmap: eliminate partial anon_vma tear-down in anon_vma_fork()
We have spun a web of unnecessary headaches for ourselves in
anon_vma_fork() with the introduction of the anon_vma reuse logic, as
introduced by commit 7a3ef208e662 ("mm: prevent endless growth of anon_vma
hierarchy").
When we clone anon_vma's linked to a VMA via vma->anon_vma_chain, we check
each anon_vma for specific conditions, and if met we set vma->anon_vma to
this anon_vma to indicate we will reuse it rather than allocating a new
one.
It triggers upon the first ancestor anon_vma found that possesses at most
1 child, and no active VMAs.
This was implemented such that if you continually fork and free VMAs, you
would achieve anon_vma reuse rather than continually allocating
unnecessary new anon_vma's.
This however brings an unfortunate situation should a memory allocation
fail during this process. anon_vma_fork():
1. Clones the anon_vma.
2. If no reuse (i.e. !vma->anon_vma), tries to allocate anon_vma, AVC.
3. If 2 fails, we are forced to unwind step 1 by invoking
unlink_anon_vmas(vma).
This means that we send a partially set up (i.e. invalid) VMA to
unlink_anon_vmas().
Doing this is dangerous and confusing - it is reasonable for kernel
developers to assume unlink_anon_vmas() is called on a correctly
established vma, and thus confusion arises if logic is implemented there
to account for invalid VMAs, and further development might introduce
subtle bugs.
It is especially problematic in the anon rmap implementation which is
essentially a broken abstraction.
The patch solves the issue by simply trying to allocate the anon_vma and
AVC ahead of time - i.e. optimising for the usual case - and freeing them
should reuse occur or an error arise in anon_vma_clone().
This is not egregious performance-wise, as this function is called on the
fork path which already performs a great number of allocations, and thus
it is already a slow-path in this respect.
It is additionally not egregious in terms of memory usage - the
allocations are too-small-to-fail anyway unless, for instance, a fatal
signal may have arisen, and any OOM for such tiny allocations that may
arise would indicate the system is under so much memory pressure that the
associated process is not long for this world anyway.
We also update anon_vma->num_active_vmas to 1 directly rather than
incrementing the newly allocated anon_vma's active VMA count - this makes
it clear that this detached anon_vma can have only 1 num_active_vma at
this point.
Finally we eliminate the out_error and out_error_free_anon_vma labels
which makes the logic much easier to follow. We also correct a small
comment typo.
Link: https://lkml.kernel.org/r/9923da5f8b095dd1e8d677692dcaf95859de0ef5.1768746221.git.lorenzo.stoakes@oracle.com
Signed-off-by: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: Barry Song <v-songbaohua@oppo.com>
Cc: Chris Li <chriscli@google.com>
Cc: David Hildenbrand <david@kernel.org>
Cc: Harry Yoo <harry.yoo@oracle.com>
Cc: Jann Horn <jannh@google.com>
Cc: Liam R. Howlett <Liam.Howlett@oracle.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Mike Rapoport <rppt@kernel.org>
Cc: Pedro Falcato <pfalcato@suse.de>
Cc: Rik van Riel <riel@surriel.com>
Cc: Shakeel Butt <shakeel.butt@linux.dev>
Cc: Suren Baghdasaryan <surenb@google.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2026-01-18 14:50:38 +00:00
|
|
|
|
mm: change anon_vma linking to fix multi-process server scalability issue
The old anon_vma code can lead to scalability issues with heavily forking
workloads. Specifically, each anon_vma will be shared between the parent
process and all its child processes.
In a workload with 1000 child processes and a VMA with 1000 anonymous
pages per process that get COWed, this leads to a system with a million
anonymous pages in the same anon_vma, each of which is mapped in just one
of the 1000 processes. However, the current rmap code needs to walk them
all, leading to O(N) scanning complexity for each page.
This can result in systems where one CPU is walking the page tables of
1000 processes in page_referenced_one, while all other CPUs are stuck on
the anon_vma lock. This leads to catastrophic failure for a benchmark
like AIM7, where the total number of processes can reach in the tens of
thousands. Real workloads are still a factor 10 less process intensive
than AIM7, but they are catching up.
This patch changes the way anon_vmas and VMAs are linked, which allows us
to associate multiple anon_vmas with a VMA. At fork time, each child
process gets its own anon_vmas, in which its COWed pages will be
instantiated. The parents' anon_vma is also linked to the VMA, because
non-COWed pages could be present in any of the children.
This reduces rmap scanning complexity to O(1) for the pages of the 1000
child processes, with O(N) complexity for at most 1/N pages in the system.
This reduces the average scanning cost in heavily forking workloads from
O(N) to 2.
The only real complexity in this patch stems from the fact that linking a
VMA to anon_vmas now involves memory allocations. This means vma_adjust
can fail, if it needs to attach a VMA to anon_vma structures. This in
turn means error handling needs to be added to the calling functions.
A second source of complexity is that, because there can be multiple
anon_vmas, the anon_vma linking in vma_adjust can no longer be done under
"the" anon_vma lock. To prevent the rmap code from walking up an
incomplete VMA, this patch introduces the VM_LOCK_RMAP VMA flag. This bit
flag uses the same slot as the NOMMU VM_MAPPED_COPY, with an ifdef in mm.h
to make sure it is impossible to compile a kernel that needs both symbolic
values for the same bitflag.
Some test results:
Without the anon_vma changes, when AIM7 hits around 9.7k users (on a test
box with 16GB RAM and not quite enough IO), the system ends up running
>99% in system time, with every CPU on the same anon_vma lock in the
pageout code.
With these changes, AIM7 hits the cross-over point around 29.7k users.
This happens with ~99% IO wait time, there never seems to be any spike in
system time. The anon_vma lock contention appears to be resolved.
[akpm@linux-foundation.org: cleanups]
Signed-off-by: Rik van Riel <riel@redhat.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Larry Woodman <lwoodman@redhat.com>
Cc: Lee Schermerhorn <Lee.Schermerhorn@hp.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-03-05 13:42:07 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-01-08 14:32:15 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
mm/rmap: eliminate partial anon_vma tear-down in anon_vma_fork()
We have spun a web of unnecessary headaches for ourselves in
anon_vma_fork() with the introduction of the anon_vma reuse logic, as
introduced by commit 7a3ef208e662 ("mm: prevent endless growth of anon_vma
hierarchy").
When we clone anon_vma's linked to a VMA via vma->anon_vma_chain, we check
each anon_vma for specific conditions, and if met we set vma->anon_vma to
this anon_vma to indicate we will reuse it rather than allocating a new
one.
It triggers upon the first ancestor anon_vma found that possesses at most
1 child, and no active VMAs.
This was implemented such that if you continually fork and free VMAs, you
would achieve anon_vma reuse rather than continually allocating
unnecessary new anon_vma's.
This however brings an unfortunate situation should a memory allocation
fail during this process. anon_vma_fork():
1. Clones the anon_vma.
2. If no reuse (i.e. !vma->anon_vma), tries to allocate anon_vma, AVC.
3. If 2 fails, we are forced to unwind step 1 by invoking
unlink_anon_vmas(vma).
This means that we send a partially set up (i.e. invalid) VMA to
unlink_anon_vmas().
Doing this is dangerous and confusing - it is reasonable for kernel
developers to assume unlink_anon_vmas() is called on a correctly
established vma, and thus confusion arises if logic is implemented there
to account for invalid VMAs, and further development might introduce
subtle bugs.
It is especially problematic in the anon rmap implementation which is
essentially a broken abstraction.
The patch solves the issue by simply trying to allocate the anon_vma and
AVC ahead of time - i.e. optimising for the usual case - and freeing them
should reuse occur or an error arise in anon_vma_clone().
This is not egregious performance-wise, as this function is called on the
fork path which already performs a great number of allocations, and thus
it is already a slow-path in this respect.
It is additionally not egregious in terms of memory usage - the
allocations are too-small-to-fail anyway unless, for instance, a fatal
signal may have arisen, and any OOM for such tiny allocations that may
arise would indicate the system is under so much memory pressure that the
associated process is not long for this world anyway.
We also update anon_vma->num_active_vmas to 1 directly rather than
incrementing the newly allocated anon_vma's active VMA count - this makes
it clear that this detached anon_vma can have only 1 num_active_vma at
this point.
Finally we eliminate the out_error and out_error_free_anon_vma labels
which makes the logic much easier to follow. We also correct a small
comment typo.
Link: https://lkml.kernel.org/r/9923da5f8b095dd1e8d677692dcaf95859de0ef5.1768746221.git.lorenzo.stoakes@oracle.com
Signed-off-by: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: Barry Song <v-songbaohua@oppo.com>
Cc: Chris Li <chriscli@google.com>
Cc: David Hildenbrand <david@kernel.org>
Cc: Harry Yoo <harry.yoo@oracle.com>
Cc: Jann Horn <jannh@google.com>
Cc: Liam R. Howlett <Liam.Howlett@oracle.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Mike Rapoport <rppt@kernel.org>
Cc: Pedro Falcato <pfalcato@suse.de>
Cc: Rik van Riel <riel@surriel.com>
Cc: Shakeel Butt <shakeel.butt@linux.dev>
Cc: Suren Baghdasaryan <surenb@google.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2026-01-18 14:50:38 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
mm: change anon_vma linking to fix multi-process server scalability issue
The old anon_vma code can lead to scalability issues with heavily forking
workloads. Specifically, each anon_vma will be shared between the parent
process and all its child processes.
In a workload with 1000 child processes and a VMA with 1000 anonymous
pages per process that get COWed, this leads to a system with a million
anonymous pages in the same anon_vma, each of which is mapped in just one
of the 1000 processes. However, the current rmap code needs to walk them
all, leading to O(N) scanning complexity for each page.
This can result in systems where one CPU is walking the page tables of
1000 processes in page_referenced_one, while all other CPUs are stuck on
the anon_vma lock. This leads to catastrophic failure for a benchmark
like AIM7, where the total number of processes can reach in the tens of
thousands. Real workloads are still a factor 10 less process intensive
than AIM7, but they are catching up.
This patch changes the way anon_vmas and VMAs are linked, which allows us
to associate multiple anon_vmas with a VMA. At fork time, each child
process gets its own anon_vmas, in which its COWed pages will be
instantiated. The parents' anon_vma is also linked to the VMA, because
non-COWed pages could be present in any of the children.
This reduces rmap scanning complexity to O(1) for the pages of the 1000
child processes, with O(N) complexity for at most 1/N pages in the system.
This reduces the average scanning cost in heavily forking workloads from
O(N) to 2.
The only real complexity in this patch stems from the fact that linking a
VMA to anon_vmas now involves memory allocations. This means vma_adjust
can fail, if it needs to attach a VMA to anon_vma structures. This in
turn means error handling needs to be added to the calling functions.
A second source of complexity is that, because there can be multiple
anon_vmas, the anon_vma linking in vma_adjust can no longer be done under
"the" anon_vma lock. To prevent the rmap code from walking up an
incomplete VMA, this patch introduces the VM_LOCK_RMAP VMA flag. This bit
flag uses the same slot as the NOMMU VM_MAPPED_COPY, with an ifdef in mm.h
to make sure it is impossible to compile a kernel that needs both symbolic
values for the same bitflag.
Some test results:
Without the anon_vma changes, when AIM7 hits around 9.7k users (on a test
box with 16GB RAM and not quite enough IO), the system ends up running
>99% in system time, with every CPU on the same anon_vma lock in the
pageout code.
With these changes, AIM7 hits the cross-over point around 29.7k users.
This happens with ~99% IO wait time, there never seems to be any spike in
system time. The anon_vma lock contention appears to be resolved.
[akpm@linux-foundation.org: cleanups]
Signed-off-by: Rik van Riel <riel@redhat.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Larry Woodman <lwoodman@redhat.com>
Cc: Lee Schermerhorn <Lee.Schermerhorn@hp.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-03-05 13:42:07 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-18 14:50:45 +00:00
|
|
|
|
mm/rmap: eliminate partial anon_vma tear-down in anon_vma_fork()
We have spun a web of unnecessary headaches for ourselves in
anon_vma_fork() with the introduction of the anon_vma reuse logic, as
introduced by commit 7a3ef208e662 ("mm: prevent endless growth of anon_vma
hierarchy").
When we clone anon_vma's linked to a VMA via vma->anon_vma_chain, we check
each anon_vma for specific conditions, and if met we set vma->anon_vma to
this anon_vma to indicate we will reuse it rather than allocating a new
one.
It triggers upon the first ancestor anon_vma found that possesses at most
1 child, and no active VMAs.
This was implemented such that if you continually fork and free VMAs, you
would achieve anon_vma reuse rather than continually allocating
unnecessary new anon_vma's.
This however brings an unfortunate situation should a memory allocation
fail during this process. anon_vma_fork():
1. Clones the anon_vma.
2. If no reuse (i.e. !vma->anon_vma), tries to allocate anon_vma, AVC.
3. If 2 fails, we are forced to unwind step 1 by invoking
unlink_anon_vmas(vma).
This means that we send a partially set up (i.e. invalid) VMA to
unlink_anon_vmas().
Doing this is dangerous and confusing - it is reasonable for kernel
developers to assume unlink_anon_vmas() is called on a correctly
established vma, and thus confusion arises if logic is implemented there
to account for invalid VMAs, and further development might introduce
subtle bugs.
It is especially problematic in the anon rmap implementation which is
essentially a broken abstraction.
The patch solves the issue by simply trying to allocate the anon_vma and
AVC ahead of time - i.e. optimising for the usual case - and freeing them
should reuse occur or an error arise in anon_vma_clone().
This is not egregious performance-wise, as this function is called on the
fork path which already performs a great number of allocations, and thus
it is already a slow-path in this respect.
It is additionally not egregious in terms of memory usage - the
allocations are too-small-to-fail anyway unless, for instance, a fatal
signal may have arisen, and any OOM for such tiny allocations that may
arise would indicate the system is under so much memory pressure that the
associated process is not long for this world anyway.
We also update anon_vma->num_active_vmas to 1 directly rather than
incrementing the newly allocated anon_vma's active VMA count - this makes
it clear that this detached anon_vma can have only 1 num_active_vma at
this point.
Finally we eliminate the out_error and out_error_free_anon_vma labels
which makes the logic much easier to follow. We also correct a small
comment typo.
Link: https://lkml.kernel.org/r/9923da5f8b095dd1e8d677692dcaf95859de0ef5.1768746221.git.lorenzo.stoakes@oracle.com
Signed-off-by: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: Barry Song <v-songbaohua@oppo.com>
Cc: Chris Li <chriscli@google.com>
Cc: David Hildenbrand <david@kernel.org>
Cc: Harry Yoo <harry.yoo@oracle.com>
Cc: Jann Horn <jannh@google.com>
Cc: Liam R. Howlett <Liam.Howlett@oracle.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Mike Rapoport <rppt@kernel.org>
Cc: Pedro Falcato <pfalcato@suse.de>
Cc: Rik van Riel <riel@surriel.com>
Cc: Shakeel Butt <shakeel.butt@linux.dev>
Cc: Suren Baghdasaryan <surenb@google.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2026-01-18 14:50:38 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
mm: change anon_vma linking to fix multi-process server scalability issue
The old anon_vma code can lead to scalability issues with heavily forking
workloads. Specifically, each anon_vma will be shared between the parent
process and all its child processes.
In a workload with 1000 child processes and a VMA with 1000 anonymous
pages per process that get COWed, this leads to a system with a million
anonymous pages in the same anon_vma, each of which is mapped in just one
of the 1000 processes. However, the current rmap code needs to walk them
all, leading to O(N) scanning complexity for each page.
This can result in systems where one CPU is walking the page tables of
1000 processes in page_referenced_one, while all other CPUs are stuck on
the anon_vma lock. This leads to catastrophic failure for a benchmark
like AIM7, where the total number of processes can reach in the tens of
thousands. Real workloads are still a factor 10 less process intensive
than AIM7, but they are catching up.
This patch changes the way anon_vmas and VMAs are linked, which allows us
to associate multiple anon_vmas with a VMA. At fork time, each child
process gets its own anon_vmas, in which its COWed pages will be
instantiated. The parents' anon_vma is also linked to the VMA, because
non-COWed pages could be present in any of the children.
This reduces rmap scanning complexity to O(1) for the pages of the 1000
child processes, with O(N) complexity for at most 1/N pages in the system.
This reduces the average scanning cost in heavily forking workloads from
O(N) to 2.
The only real complexity in this patch stems from the fact that linking a
VMA to anon_vmas now involves memory allocations. This means vma_adjust
can fail, if it needs to attach a VMA to anon_vma structures. This in
turn means error handling needs to be added to the calling functions.
A second source of complexity is that, because there can be multiple
anon_vmas, the anon_vma linking in vma_adjust can no longer be done under
"the" anon_vma lock. To prevent the rmap code from walking up an
incomplete VMA, this patch introduces the VM_LOCK_RMAP VMA flag. This bit
flag uses the same slot as the NOMMU VM_MAPPED_COPY, with an ifdef in mm.h
to make sure it is impossible to compile a kernel that needs both symbolic
values for the same bitflag.
Some test results:
Without the anon_vma changes, when AIM7 hits around 9.7k users (on a test
box with 16GB RAM and not quite enough IO), the system ends up running
>99% in system time, with every CPU on the same anon_vma lock in the
pageout code.
With these changes, AIM7 hits the cross-over point around 29.7k users.
This happens with ~99% IO wait time, there never seems to be any spike in
system time. The anon_vma lock contention appears to be resolved.
[akpm@linux-foundation.org: cleanups]
Signed-off-by: Rik van Riel <riel@redhat.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Larry Woodman <lwoodman@redhat.com>
Cc: Lee Schermerhorn <Lee.Schermerhorn@hp.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-03-05 13:42:07 -08:00
|
|
|
|
mm/rmap: eliminate partial anon_vma tear-down in anon_vma_fork()
We have spun a web of unnecessary headaches for ourselves in
anon_vma_fork() with the introduction of the anon_vma reuse logic, as
introduced by commit 7a3ef208e662 ("mm: prevent endless growth of anon_vma
hierarchy").
When we clone anon_vma's linked to a VMA via vma->anon_vma_chain, we check
each anon_vma for specific conditions, and if met we set vma->anon_vma to
this anon_vma to indicate we will reuse it rather than allocating a new
one.
It triggers upon the first ancestor anon_vma found that possesses at most
1 child, and no active VMAs.
This was implemented such that if you continually fork and free VMAs, you
would achieve anon_vma reuse rather than continually allocating
unnecessary new anon_vma's.
This however brings an unfortunate situation should a memory allocation
fail during this process. anon_vma_fork():
1. Clones the anon_vma.
2. If no reuse (i.e. !vma->anon_vma), tries to allocate anon_vma, AVC.
3. If 2 fails, we are forced to unwind step 1 by invoking
unlink_anon_vmas(vma).
This means that we send a partially set up (i.e. invalid) VMA to
unlink_anon_vmas().
Doing this is dangerous and confusing - it is reasonable for kernel
developers to assume unlink_anon_vmas() is called on a correctly
established vma, and thus confusion arises if logic is implemented there
to account for invalid VMAs, and further development might introduce
subtle bugs.
It is especially problematic in the anon rmap implementation which is
essentially a broken abstraction.
The patch solves the issue by simply trying to allocate the anon_vma and
AVC ahead of time - i.e. optimising for the usual case - and freeing them
should reuse occur or an error arise in anon_vma_clone().
This is not egregious performance-wise, as this function is called on the
fork path which already performs a great number of allocations, and thus
it is already a slow-path in this respect.
It is additionally not egregious in terms of memory usage - the
allocations are too-small-to-fail anyway unless, for instance, a fatal
signal may have arisen, and any OOM for such tiny allocations that may
arise would indicate the system is under so much memory pressure that the
associated process is not long for this world anyway.
We also update anon_vma->num_active_vmas to 1 directly rather than
incrementing the newly allocated anon_vma's active VMA count - this makes
it clear that this detached anon_vma can have only 1 num_active_vma at
this point.
Finally we eliminate the out_error and out_error_free_anon_vma labels
which makes the logic much easier to follow. We also correct a small
comment typo.
Link: https://lkml.kernel.org/r/9923da5f8b095dd1e8d677692dcaf95859de0ef5.1768746221.git.lorenzo.stoakes@oracle.com
Signed-off-by: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: Barry Song <v-songbaohua@oppo.com>
Cc: Chris Li <chriscli@google.com>
Cc: David Hildenbrand <david@kernel.org>
Cc: Harry Yoo <harry.yoo@oracle.com>
Cc: Jann Horn <jannh@google.com>
Cc: Liam R. Howlett <Liam.Howlett@oracle.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Mike Rapoport <rppt@kernel.org>
Cc: Pedro Falcato <pfalcato@suse.de>
Cc: Rik van Riel <riel@surriel.com>
Cc: Shakeel Butt <shakeel.butt@linux.dev>
Cc: Suren Baghdasaryan <surenb@google.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2026-01-18 14:50:38 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2010-08-09 17:18:39 -07:00
|
|
|
|
mm/rmap: eliminate partial anon_vma tear-down in anon_vma_fork()
We have spun a web of unnecessary headaches for ourselves in
anon_vma_fork() with the introduction of the anon_vma reuse logic, as
introduced by commit 7a3ef208e662 ("mm: prevent endless growth of anon_vma
hierarchy").
When we clone anon_vma's linked to a VMA via vma->anon_vma_chain, we check
each anon_vma for specific conditions, and if met we set vma->anon_vma to
this anon_vma to indicate we will reuse it rather than allocating a new
one.
It triggers upon the first ancestor anon_vma found that possesses at most
1 child, and no active VMAs.
This was implemented such that if you continually fork and free VMAs, you
would achieve anon_vma reuse rather than continually allocating
unnecessary new anon_vma's.
This however brings an unfortunate situation should a memory allocation
fail during this process. anon_vma_fork():
1. Clones the anon_vma.
2. If no reuse (i.e. !vma->anon_vma), tries to allocate anon_vma, AVC.
3. If 2 fails, we are forced to unwind step 1 by invoking
unlink_anon_vmas(vma).
This means that we send a partially set up (i.e. invalid) VMA to
unlink_anon_vmas().
Doing this is dangerous and confusing - it is reasonable for kernel
developers to assume unlink_anon_vmas() is called on a correctly
established vma, and thus confusion arises if logic is implemented there
to account for invalid VMAs, and further development might introduce
subtle bugs.
It is especially problematic in the anon rmap implementation which is
essentially a broken abstraction.
The patch solves the issue by simply trying to allocate the anon_vma and
AVC ahead of time - i.e. optimising for the usual case - and freeing them
should reuse occur or an error arise in anon_vma_clone().
This is not egregious performance-wise, as this function is called on the
fork path which already performs a great number of allocations, and thus
it is already a slow-path in this respect.
It is additionally not egregious in terms of memory usage - the
allocations are too-small-to-fail anyway unless, for instance, a fatal
signal may have arisen, and any OOM for such tiny allocations that may
arise would indicate the system is under so much memory pressure that the
associated process is not long for this world anyway.
We also update anon_vma->num_active_vmas to 1 directly rather than
incrementing the newly allocated anon_vma's active VMA count - this makes
it clear that this detached anon_vma can have only 1 num_active_vma at
this point.
Finally we eliminate the out_error and out_error_free_anon_vma labels
which makes the logic much easier to follow. We also correct a small
comment typo.
Link: https://lkml.kernel.org/r/9923da5f8b095dd1e8d677692dcaf95859de0ef5.1768746221.git.lorenzo.stoakes@oracle.com
Signed-off-by: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: Barry Song <v-songbaohua@oppo.com>
Cc: Chris Li <chriscli@google.com>
Cc: David Hildenbrand <david@kernel.org>
Cc: Harry Yoo <harry.yoo@oracle.com>
Cc: Jann Horn <jannh@google.com>
Cc: Liam R. Howlett <Liam.Howlett@oracle.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Mike Rapoport <rppt@kernel.org>
Cc: Pedro Falcato <pfalcato@suse.de>
Cc: Rik van Riel <riel@surriel.com>
Cc: Shakeel Butt <shakeel.butt@linux.dev>
Cc: Suren Baghdasaryan <surenb@google.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2026-01-18 14:50:38 +00:00
|
|
|
|
2010-08-09 17:18:39 -07:00
|
|
|
|
2021-02-25 17:17:53 -08:00
|
|
|
|
2010-08-09 17:18:39 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-01-08 14:32:15 -08:00
|
|
|
|
2010-08-09 17:18:41 -07:00
|
|
|
|
2011-03-22 16:32:49 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2010-08-09 17:18:41 -07:00
|
|
|
|
|
|
|
|
|
mm: change anon_vma linking to fix multi-process server scalability issue
The old anon_vma code can lead to scalability issues with heavily forking
workloads. Specifically, each anon_vma will be shared between the parent
process and all its child processes.
In a workload with 1000 child processes and a VMA with 1000 anonymous
pages per process that get COWed, this leads to a system with a million
anonymous pages in the same anon_vma, each of which is mapped in just one
of the 1000 processes. However, the current rmap code needs to walk them
all, leading to O(N) scanning complexity for each page.
This can result in systems where one CPU is walking the page tables of
1000 processes in page_referenced_one, while all other CPUs are stuck on
the anon_vma lock. This leads to catastrophic failure for a benchmark
like AIM7, where the total number of processes can reach in the tens of
thousands. Real workloads are still a factor 10 less process intensive
than AIM7, but they are catching up.
This patch changes the way anon_vmas and VMAs are linked, which allows us
to associate multiple anon_vmas with a VMA. At fork time, each child
process gets its own anon_vmas, in which its COWed pages will be
instantiated. The parents' anon_vma is also linked to the VMA, because
non-COWed pages could be present in any of the children.
This reduces rmap scanning complexity to O(1) for the pages of the 1000
child processes, with O(N) complexity for at most 1/N pages in the system.
This reduces the average scanning cost in heavily forking workloads from
O(N) to 2.
The only real complexity in this patch stems from the fact that linking a
VMA to anon_vmas now involves memory allocations. This means vma_adjust
can fail, if it needs to attach a VMA to anon_vma structures. This in
turn means error handling needs to be added to the calling functions.
A second source of complexity is that, because there can be multiple
anon_vmas, the anon_vma linking in vma_adjust can no longer be done under
"the" anon_vma lock. To prevent the rmap code from walking up an
incomplete VMA, this patch introduces the VM_LOCK_RMAP VMA flag. This bit
flag uses the same slot as the NOMMU VM_MAPPED_COPY, with an ifdef in mm.h
to make sure it is impossible to compile a kernel that needs both symbolic
values for the same bitflag.
Some test results:
Without the anon_vma changes, when AIM7 hits around 9.7k users (on a test
box with 16GB RAM and not quite enough IO), the system ends up running
>99% in system time, with every CPU on the same anon_vma lock in the
pageout code.
With these changes, AIM7 hits the cross-over point around 29.7k users.
This happens with ~99% IO wait time, there never seems to be any spike in
system time. The anon_vma lock contention appears to be resolved.
[akpm@linux-foundation.org: cleanups]
Signed-off-by: Rik van Riel <riel@redhat.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Larry Woodman <lwoodman@redhat.com>
Cc: Lee Schermerhorn <Lee.Schermerhorn@hp.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-03-05 13:42:07 -08:00
|
|
|
|
|
|
|
|
|
2026-01-18 14:50:44 +00:00
|
|
|
|
|
|
|
|
|
2012-12-02 19:56:50 +00:00
|
|
|
|
2026-01-18 14:50:44 +00:00
|
|
|
|
2022-08-31 19:06:00 +02:00
|
|
|
|
2013-02-22 16:34:40 -08:00
|
|
|
|
mm: change anon_vma linking to fix multi-process server scalability issue
The old anon_vma code can lead to scalability issues with heavily forking
workloads. Specifically, each anon_vma will be shared between the parent
process and all its child processes.
In a workload with 1000 child processes and a VMA with 1000 anonymous
pages per process that get COWed, this leads to a system with a million
anonymous pages in the same anon_vma, each of which is mapped in just one
of the 1000 processes. However, the current rmap code needs to walk them
all, leading to O(N) scanning complexity for each page.
This can result in systems where one CPU is walking the page tables of
1000 processes in page_referenced_one, while all other CPUs are stuck on
the anon_vma lock. This leads to catastrophic failure for a benchmark
like AIM7, where the total number of processes can reach in the tens of
thousands. Real workloads are still a factor 10 less process intensive
than AIM7, but they are catching up.
This patch changes the way anon_vmas and VMAs are linked, which allows us
to associate multiple anon_vmas with a VMA. At fork time, each child
process gets its own anon_vmas, in which its COWed pages will be
instantiated. The parents' anon_vma is also linked to the VMA, because
non-COWed pages could be present in any of the children.
This reduces rmap scanning complexity to O(1) for the pages of the 1000
child processes, with O(N) complexity for at most 1/N pages in the system.
This reduces the average scanning cost in heavily forking workloads from
O(N) to 2.
The only real complexity in this patch stems from the fact that linking a
VMA to anon_vmas now involves memory allocations. This means vma_adjust
can fail, if it needs to attach a VMA to anon_vma structures. This in
turn means error handling needs to be added to the calling functions.
A second source of complexity is that, because there can be multiple
anon_vmas, the anon_vma linking in vma_adjust can no longer be done under
"the" anon_vma lock. To prevent the rmap code from walking up an
incomplete VMA, this patch introduces the VM_LOCK_RMAP VMA flag. This bit
flag uses the same slot as the NOMMU VM_MAPPED_COPY, with an ifdef in mm.h
to make sure it is impossible to compile a kernel that needs both symbolic
values for the same bitflag.
Some test results:
Without the anon_vma changes, when AIM7 hits around 9.7k users (on a test
box with 16GB RAM and not quite enough IO), the system ends up running
>99% in system time, with every CPU on the same anon_vma lock in the
pageout code.
With these changes, AIM7 hits the cross-over point around 29.7k users.
This happens with ~99% IO wait time, there never seems to be any spike in
system time. The anon_vma lock contention appears to be resolved.
[akpm@linux-foundation.org: cleanups]
Signed-off-by: Rik van Riel <riel@redhat.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Larry Woodman <lwoodman@redhat.com>
Cc: Lee Schermerhorn <Lee.Schermerhorn@hp.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-03-05 13:42:07 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
mm/rmap: improve anon_vma_clone(), unlink_anon_vmas() comments, add asserts
Patch series "mm: clean up anon_vma implementation", v3.
The anon_vma logic is hugely confusing and, much like a bundle of wires
entangled with one another, pulling on one thread seems only to lead to
more entanglement elsewhere.
There is a mish-mash of the core implementation, how that implementation
is invoked, how helper functions are invoked and concepts such as adjacent
anon_vma merge and anon_vma object reuse.
This series tries to improve the situation somewhat.
It starts by establishing some invariants in the core anon_vma_clone() and
unlink_anon_vmas() functions, largely expressed via VM_WARN_ON_ONCE()
asserts.
These act as some form of self-documentation as to the conditions we find
ourselves in when invoking these functions.
We also add kdoc comments for anon_vma_clone() and unlink_anon_vmas().
We then update anon_vma_fork() to avoid passing a partially set up (and
thus invalid) VMA to unlink_anon_vmas() - functions which are used both
for partially set up and valid data types has historically been the source
of a lot of confusion and bugs.
We then makes use of the established known conditions to directly skip
unfaulted VMAs (rather than implicitly via an empty vma->anon_vma_chain
list).
We remove the confusing anon_vma_merge() function (we already have a
concept of anon_vma merge in that we merge anon_vma's that would otherwise
be compatible except for attributes that mprotect() could change - which
anon_vma_merge() has nothing to do with).
We make the anon_vma functions internal to mm as they do not need to be
used by any other part of the kernel, which allows for future abstraction
without concern about this.
We then reduce the time over which we hold the anon rmap lock in
anon_vma_clone(), as it turns out we can allocate anon_vma_chain objects
without holding this lock, since the objects are not yet accessible from
the rmap.
This should reduce anon_vma lock contention.
This additionally allows us to remove a confusing GFP_NOWAIT, GFP_KERNEL
allocation fallback strategy.
Finally, we explicitly indicate which operation is being performed upon
anon_vma_clone(), and separate out fork-only logic to make it very clear
that anon_vma reuse only occurs on fork.
This patch (of 9):
Add kdoc comments and describe exactly what these functions are used for
in detail, pointing out importantly that the anon_vma_clone()
!dst->anon_vma && src->anon_vma dance is ONLY for fork.
Both are confusing functions that will be refactored in a subsequent patch
but the first stage is establishing documentation and some invariants.
Add some basic CONFIG_DEBUG_VM asserts that help document expected state,
specifically:
anon_vma_clone()
- mmap write lock held.
- We do nothing if src VMA is not faulted.
- The destination VMA has no anon_vma_chain yet.
- We are always operating on the same active VMA (i.e. vma->anon_vma).
- If not forking, must operate on the same mm_struct.
unlink_anon_vmas()
- mmap lock held (write lock except when freeing page tables).
- That unfaulted VMAs are no-ops.
We are presented with a special case when anon_vma_clone() fails to
allocate memory, where we have a VMA with partially set up anon_vma state.
Since we hold the exclusive mmap write lock, and since we are cloning
from a source VMA which consequently can't also have its anon_vma state
modified, we know no anon_vma referenced can be empty.
This allows us to significantly simplify this case and just remove
anon_vma_chain objects associated with the VMA, so we add a specific
partial cleanup path for this scenario.
This also allows us to drop the hack of setting vma->anon_vma to NULL
before unlinking anon_vma state in this scenario.
Link: https://lkml.kernel.org/r/cover.1768746221.git.lorenzo.stoakes@oracle.com
Link: https://lkml.kernel.org/r/8644e89369be0cc89d7ac57443dff9e822803c91.1768746221.git.lorenzo.stoakes@oracle.com
Signed-off-by: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Reviewed-by: Liam R. Howlett <Liam.Howlett@oracle.com>
Cc: Barry Song <v-songbaohua@oppo.com>
Cc: Chris Li <chriscli@google.com>
Cc: David Hildenbrand <david@kernel.org>
Cc: Harry Yoo <harry.yoo@oracle.com>
Cc: Jann Horn <jannh@google.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Mike Rapoport <rppt@kernel.org>
Cc: Pedro Falcato <pfalcato@suse.de>
Cc: Rik van Riel <riel@surriel.com>
Cc: Shakeel Butt <shakeel.butt@linux.dev>
Cc: Suren Baghdasaryan <surenb@google.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2026-01-18 14:50:37 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-18 14:50:44 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
mm/rmap: improve anon_vma_clone(), unlink_anon_vmas() comments, add asserts
Patch series "mm: clean up anon_vma implementation", v3.
The anon_vma logic is hugely confusing and, much like a bundle of wires
entangled with one another, pulling on one thread seems only to lead to
more entanglement elsewhere.
There is a mish-mash of the core implementation, how that implementation
is invoked, how helper functions are invoked and concepts such as adjacent
anon_vma merge and anon_vma object reuse.
This series tries to improve the situation somewhat.
It starts by establishing some invariants in the core anon_vma_clone() and
unlink_anon_vmas() functions, largely expressed via VM_WARN_ON_ONCE()
asserts.
These act as some form of self-documentation as to the conditions we find
ourselves in when invoking these functions.
We also add kdoc comments for anon_vma_clone() and unlink_anon_vmas().
We then update anon_vma_fork() to avoid passing a partially set up (and
thus invalid) VMA to unlink_anon_vmas() - functions which are used both
for partially set up and valid data types has historically been the source
of a lot of confusion and bugs.
We then makes use of the established known conditions to directly skip
unfaulted VMAs (rather than implicitly via an empty vma->anon_vma_chain
list).
We remove the confusing anon_vma_merge() function (we already have a
concept of anon_vma merge in that we merge anon_vma's that would otherwise
be compatible except for attributes that mprotect() could change - which
anon_vma_merge() has nothing to do with).
We make the anon_vma functions internal to mm as they do not need to be
used by any other part of the kernel, which allows for future abstraction
without concern about this.
We then reduce the time over which we hold the anon rmap lock in
anon_vma_clone(), as it turns out we can allocate anon_vma_chain objects
without holding this lock, since the objects are not yet accessible from
the rmap.
This should reduce anon_vma lock contention.
This additionally allows us to remove a confusing GFP_NOWAIT, GFP_KERNEL
allocation fallback strategy.
Finally, we explicitly indicate which operation is being performed upon
anon_vma_clone(), and separate out fork-only logic to make it very clear
that anon_vma reuse only occurs on fork.
This patch (of 9):
Add kdoc comments and describe exactly what these functions are used for
in detail, pointing out importantly that the anon_vma_clone()
!dst->anon_vma && src->anon_vma dance is ONLY for fork.
Both are confusing functions that will be refactored in a subsequent patch
but the first stage is establishing documentation and some invariants.
Add some basic CONFIG_DEBUG_VM asserts that help document expected state,
specifically:
anon_vma_clone()
- mmap write lock held.
- We do nothing if src VMA is not faulted.
- The destination VMA has no anon_vma_chain yet.
- We are always operating on the same active VMA (i.e. vma->anon_vma).
- If not forking, must operate on the same mm_struct.
unlink_anon_vmas()
- mmap lock held (write lock except when freeing page tables).
- That unfaulted VMAs are no-ops.
We are presented with a special case when anon_vma_clone() fails to
allocate memory, where we have a VMA with partially set up anon_vma state.
Since we hold the exclusive mmap write lock, and since we are cloning
from a source VMA which consequently can't also have its anon_vma state
modified, we know no anon_vma referenced can be empty.
This allows us to significantly simplify this case and just remove
anon_vma_chain objects associated with the VMA, so we add a specific
partial cleanup path for this scenario.
This also allows us to drop the hack of setting vma->anon_vma to NULL
before unlinking anon_vma state in this scenario.
Link: https://lkml.kernel.org/r/cover.1768746221.git.lorenzo.stoakes@oracle.com
Link: https://lkml.kernel.org/r/8644e89369be0cc89d7ac57443dff9e822803c91.1768746221.git.lorenzo.stoakes@oracle.com
Signed-off-by: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Reviewed-by: Liam R. Howlett <Liam.Howlett@oracle.com>
Cc: Barry Song <v-songbaohua@oppo.com>
Cc: Chris Li <chriscli@google.com>
Cc: David Hildenbrand <david@kernel.org>
Cc: Harry Yoo <harry.yoo@oracle.com>
Cc: Jann Horn <jannh@google.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Mike Rapoport <rppt@kernel.org>
Cc: Pedro Falcato <pfalcato@suse.de>
Cc: Rik van Riel <riel@surriel.com>
Cc: Shakeel Butt <shakeel.butt@linux.dev>
Cc: Suren Baghdasaryan <surenb@google.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2026-01-18 14:50:37 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
mm/rmap: clear vma->anon_vma on error
Commit 542eda1a8329 ("mm/rmap: improve anon_vma_clone(),
unlink_anon_vmas() comments, add asserts") alters the way errors are
handled, but overlooked one important aspect of clean up.
When a VMA encounters an error state in anon_vma_clone() (that is, on
attempted allocation of anon_vma_chain objects), it cleans up partially
established state in cleanup_partial_anon_vmas(), before returning an
error.
However, this occurs prior to anon_vma->num_active_vmas being incremented,
and it also fails to clear the VMA's vma->anon_vma field, which remains in
place.
This is immediately an inconsistent state, because
anon_vma->num_active_vmas is supposed to track the number of VMAs whose
vma->anon_vma field references that anon_vma, and now that count is
off-by-negative-1 for each VMA for which this error state has occurred.
When VMAs are unlinked from this anon_vma, unlink_anon_vmas() will
eventually underflow anon_vma->num_active_vmas, which will trigger a
warning.
This will always eventually happen, as we unlink anon_vma's at process
teardown.
It could also cause maybe_reuse_anon_vma() to incorrectly permit the reuse
of an anon_vma which has active VMAs attached, which will lead to a
persistently invalid state.
The solution is to clear the VMA's anon_vma field when we clean up partial
state, as the fact we are doing so indicates clearly that the VMA is not
correctly integrated into the anon_vma tree and thus this field is
invalid.
Link: https://lkml.kernel.org/r/20260318122632.63404-1-ljs@kernel.org
Fixes: 542eda1a8329 ("mm/rmap: improve anon_vma_clone(), unlink_anon_vmas() comments, add asserts")
Signed-off-by: Lorenzo Stoakes (Oracle) <ljs@kernel.org>
Reported-by: Sasha Levin <sashal@kernel.org>
Closes: https://lore.kernel.org/linux-mm/20260302151547.2389070-1-sashal@kernel.org/
Reported-by: Jiakai Xu <jiakaipeanut@gmail.com>
Closes: https://lore.kernel.org/linux-mm/CAFb8wJvRhatRD-9DVmr5v5pixTMPEr3UKjYBJjCd09OfH55CKg@mail.gmail.com/
Acked-by: David Hildenbrand (Arm) <david@kernel.org>
Acked-by: Vlastimil Babka (SUSE) <vbabka@kernel.org>
Tested-by: Jiakai Xu <jiakaipeanut@gmail.com>
Acked-by: Harry Yoo <harry.yoo@oracle.com>
Cc: Jann Horn <jannh@google.com>
Cc: Liam Howlett <liam.howlett@oracle.com>
Cc: Rik van Riel <riel@surriel.com>
Cc: Sasha Levin (Microsoft) <sashal@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2026-03-18 12:26:32 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
mm/rmap: improve anon_vma_clone(), unlink_anon_vmas() comments, add asserts
Patch series "mm: clean up anon_vma implementation", v3.
The anon_vma logic is hugely confusing and, much like a bundle of wires
entangled with one another, pulling on one thread seems only to lead to
more entanglement elsewhere.
There is a mish-mash of the core implementation, how that implementation
is invoked, how helper functions are invoked and concepts such as adjacent
anon_vma merge and anon_vma object reuse.
This series tries to improve the situation somewhat.
It starts by establishing some invariants in the core anon_vma_clone() and
unlink_anon_vmas() functions, largely expressed via VM_WARN_ON_ONCE()
asserts.
These act as some form of self-documentation as to the conditions we find
ourselves in when invoking these functions.
We also add kdoc comments for anon_vma_clone() and unlink_anon_vmas().
We then update anon_vma_fork() to avoid passing a partially set up (and
thus invalid) VMA to unlink_anon_vmas() - functions which are used both
for partially set up and valid data types has historically been the source
of a lot of confusion and bugs.
We then makes use of the established known conditions to directly skip
unfaulted VMAs (rather than implicitly via an empty vma->anon_vma_chain
list).
We remove the confusing anon_vma_merge() function (we already have a
concept of anon_vma merge in that we merge anon_vma's that would otherwise
be compatible except for attributes that mprotect() could change - which
anon_vma_merge() has nothing to do with).
We make the anon_vma functions internal to mm as they do not need to be
used by any other part of the kernel, which allows for future abstraction
without concern about this.
We then reduce the time over which we hold the anon rmap lock in
anon_vma_clone(), as it turns out we can allocate anon_vma_chain objects
without holding this lock, since the objects are not yet accessible from
the rmap.
This should reduce anon_vma lock contention.
This additionally allows us to remove a confusing GFP_NOWAIT, GFP_KERNEL
allocation fallback strategy.
Finally, we explicitly indicate which operation is being performed upon
anon_vma_clone(), and separate out fork-only logic to make it very clear
that anon_vma reuse only occurs on fork.
This patch (of 9):
Add kdoc comments and describe exactly what these functions are used for
in detail, pointing out importantly that the anon_vma_clone()
!dst->anon_vma && src->anon_vma dance is ONLY for fork.
Both are confusing functions that will be refactored in a subsequent patch
but the first stage is establishing documentation and some invariants.
Add some basic CONFIG_DEBUG_VM asserts that help document expected state,
specifically:
anon_vma_clone()
- mmap write lock held.
- We do nothing if src VMA is not faulted.
- The destination VMA has no anon_vma_chain yet.
- We are always operating on the same active VMA (i.e. vma->anon_vma).
- If not forking, must operate on the same mm_struct.
unlink_anon_vmas()
- mmap lock held (write lock except when freeing page tables).
- That unfaulted VMAs are no-ops.
We are presented with a special case when anon_vma_clone() fails to
allocate memory, where we have a VMA with partially set up anon_vma state.
Since we hold the exclusive mmap write lock, and since we are cloning
from a source VMA which consequently can't also have its anon_vma state
modified, we know no anon_vma referenced can be empty.
This allows us to significantly simplify this case and just remove
anon_vma_chain objects associated with the VMA, so we add a specific
partial cleanup path for this scenario.
This also allows us to drop the hack of setting vma->anon_vma to NULL
before unlinking anon_vma state in this scenario.
Link: https://lkml.kernel.org/r/cover.1768746221.git.lorenzo.stoakes@oracle.com
Link: https://lkml.kernel.org/r/8644e89369be0cc89d7ac57443dff9e822803c91.1768746221.git.lorenzo.stoakes@oracle.com
Signed-off-by: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Reviewed-by: Liam R. Howlett <Liam.Howlett@oracle.com>
Cc: Barry Song <v-songbaohua@oppo.com>
Cc: Chris Li <chriscli@google.com>
Cc: David Hildenbrand <david@kernel.org>
Cc: Harry Yoo <harry.yoo@oracle.com>
Cc: Jann Horn <jannh@google.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Mike Rapoport <rppt@kernel.org>
Cc: Pedro Falcato <pfalcato@suse.de>
Cc: Rik van Riel <riel@surriel.com>
Cc: Shakeel Butt <shakeel.butt@linux.dev>
Cc: Suren Baghdasaryan <surenb@google.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2026-01-18 14:50:37 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
mm: change anon_vma linking to fix multi-process server scalability issue
The old anon_vma code can lead to scalability issues with heavily forking
workloads. Specifically, each anon_vma will be shared between the parent
process and all its child processes.
In a workload with 1000 child processes and a VMA with 1000 anonymous
pages per process that get COWed, this leads to a system with a million
anonymous pages in the same anon_vma, each of which is mapped in just one
of the 1000 processes. However, the current rmap code needs to walk them
all, leading to O(N) scanning complexity for each page.
This can result in systems where one CPU is walking the page tables of
1000 processes in page_referenced_one, while all other CPUs are stuck on
the anon_vma lock. This leads to catastrophic failure for a benchmark
like AIM7, where the total number of processes can reach in the tens of
thousands. Real workloads are still a factor 10 less process intensive
than AIM7, but they are catching up.
This patch changes the way anon_vmas and VMAs are linked, which allows us
to associate multiple anon_vmas with a VMA. At fork time, each child
process gets its own anon_vmas, in which its COWed pages will be
instantiated. The parents' anon_vma is also linked to the VMA, because
non-COWed pages could be present in any of the children.
This reduces rmap scanning complexity to O(1) for the pages of the 1000
child processes, with O(N) complexity for at most 1/N pages in the system.
This reduces the average scanning cost in heavily forking workloads from
O(N) to 2.
The only real complexity in this patch stems from the fact that linking a
VMA to anon_vmas now involves memory allocations. This means vma_adjust
can fail, if it needs to attach a VMA to anon_vma structures. This in
turn means error handling needs to be added to the calling functions.
A second source of complexity is that, because there can be multiple
anon_vmas, the anon_vma linking in vma_adjust can no longer be done under
"the" anon_vma lock. To prevent the rmap code from walking up an
incomplete VMA, this patch introduces the VM_LOCK_RMAP VMA flag. This bit
flag uses the same slot as the NOMMU VM_MAPPED_COPY, with an ifdef in mm.h
to make sure it is impossible to compile a kernel that needs both symbolic
values for the same bitflag.
Some test results:
Without the anon_vma changes, when AIM7 hits around 9.7k users (on a test
box with 16GB RAM and not quite enough IO), the system ends up running
>99% in system time, with every CPU on the same anon_vma lock in the
pageout code.
With these changes, AIM7 hits the cross-over point around 29.7k users.
This happens with ~99% IO wait time, there never seems to be any spike in
system time. The anon_vma lock contention appears to be resolved.
[akpm@linux-foundation.org: cleanups]
Signed-off-by: Rik van Riel <riel@redhat.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Larry Woodman <lwoodman@redhat.com>
Cc: Lee Schermerhorn <Lee.Schermerhorn@hp.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-03-05 13:42:07 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-18 14:50:40 +00:00
|
|
|
|
mm: change anon_vma linking to fix multi-process server scalability issue
The old anon_vma code can lead to scalability issues with heavily forking
workloads. Specifically, each anon_vma will be shared between the parent
process and all its child processes.
In a workload with 1000 child processes and a VMA with 1000 anonymous
pages per process that get COWed, this leads to a system with a million
anonymous pages in the same anon_vma, each of which is mapped in just one
of the 1000 processes. However, the current rmap code needs to walk them
all, leading to O(N) scanning complexity for each page.
This can result in systems where one CPU is walking the page tables of
1000 processes in page_referenced_one, while all other CPUs are stuck on
the anon_vma lock. This leads to catastrophic failure for a benchmark
like AIM7, where the total number of processes can reach in the tens of
thousands. Real workloads are still a factor 10 less process intensive
than AIM7, but they are catching up.
This patch changes the way anon_vmas and VMAs are linked, which allows us
to associate multiple anon_vmas with a VMA. At fork time, each child
process gets its own anon_vmas, in which its COWed pages will be
instantiated. The parents' anon_vma is also linked to the VMA, because
non-COWed pages could be present in any of the children.
This reduces rmap scanning complexity to O(1) for the pages of the 1000
child processes, with O(N) complexity for at most 1/N pages in the system.
This reduces the average scanning cost in heavily forking workloads from
O(N) to 2.
The only real complexity in this patch stems from the fact that linking a
VMA to anon_vmas now involves memory allocations. This means vma_adjust
can fail, if it needs to attach a VMA to anon_vma structures. This in
turn means error handling needs to be added to the calling functions.
A second source of complexity is that, because there can be multiple
anon_vmas, the anon_vma linking in vma_adjust can no longer be done under
"the" anon_vma lock. To prevent the rmap code from walking up an
incomplete VMA, this patch introduces the VM_LOCK_RMAP VMA flag. This bit
flag uses the same slot as the NOMMU VM_MAPPED_COPY, with an ifdef in mm.h
to make sure it is impossible to compile a kernel that needs both symbolic
values for the same bitflag.
Some test results:
Without the anon_vma changes, when AIM7 hits around 9.7k users (on a test
box with 16GB RAM and not quite enough IO), the system ends up running
>99% in system time, with every CPU on the same anon_vma lock in the
pageout code.
With these changes, AIM7 hits the cross-over point around 29.7k users.
This happens with ~99% IO wait time, there never seems to be any spike in
system time. The anon_vma lock contention appears to be resolved.
[akpm@linux-foundation.org: cleanups]
Signed-off-by: Rik van Riel <riel@redhat.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Larry Woodman <lwoodman@redhat.com>
Cc: Lee Schermerhorn <Lee.Schermerhorn@hp.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-03-05 13:42:07 -08:00
|
|
|
|
mm/rmap: improve anon_vma_clone(), unlink_anon_vmas() comments, add asserts
Patch series "mm: clean up anon_vma implementation", v3.
The anon_vma logic is hugely confusing and, much like a bundle of wires
entangled with one another, pulling on one thread seems only to lead to
more entanglement elsewhere.
There is a mish-mash of the core implementation, how that implementation
is invoked, how helper functions are invoked and concepts such as adjacent
anon_vma merge and anon_vma object reuse.
This series tries to improve the situation somewhat.
It starts by establishing some invariants in the core anon_vma_clone() and
unlink_anon_vmas() functions, largely expressed via VM_WARN_ON_ONCE()
asserts.
These act as some form of self-documentation as to the conditions we find
ourselves in when invoking these functions.
We also add kdoc comments for anon_vma_clone() and unlink_anon_vmas().
We then update anon_vma_fork() to avoid passing a partially set up (and
thus invalid) VMA to unlink_anon_vmas() - functions which are used both
for partially set up and valid data types has historically been the source
of a lot of confusion and bugs.
We then makes use of the established known conditions to directly skip
unfaulted VMAs (rather than implicitly via an empty vma->anon_vma_chain
list).
We remove the confusing anon_vma_merge() function (we already have a
concept of anon_vma merge in that we merge anon_vma's that would otherwise
be compatible except for attributes that mprotect() could change - which
anon_vma_merge() has nothing to do with).
We make the anon_vma functions internal to mm as they do not need to be
used by any other part of the kernel, which allows for future abstraction
without concern about this.
We then reduce the time over which we hold the anon rmap lock in
anon_vma_clone(), as it turns out we can allocate anon_vma_chain objects
without holding this lock, since the objects are not yet accessible from
the rmap.
This should reduce anon_vma lock contention.
This additionally allows us to remove a confusing GFP_NOWAIT, GFP_KERNEL
allocation fallback strategy.
Finally, we explicitly indicate which operation is being performed upon
anon_vma_clone(), and separate out fork-only logic to make it very clear
that anon_vma reuse only occurs on fork.
This patch (of 9):
Add kdoc comments and describe exactly what these functions are used for
in detail, pointing out importantly that the anon_vma_clone()
!dst->anon_vma && src->anon_vma dance is ONLY for fork.
Both are confusing functions that will be refactored in a subsequent patch
but the first stage is establishing documentation and some invariants.
Add some basic CONFIG_DEBUG_VM asserts that help document expected state,
specifically:
anon_vma_clone()
- mmap write lock held.
- We do nothing if src VMA is not faulted.
- The destination VMA has no anon_vma_chain yet.
- We are always operating on the same active VMA (i.e. vma->anon_vma).
- If not forking, must operate on the same mm_struct.
unlink_anon_vmas()
- mmap lock held (write lock except when freeing page tables).
- That unfaulted VMAs are no-ops.
We are presented with a special case when anon_vma_clone() fails to
allocate memory, where we have a VMA with partially set up anon_vma state.
Since we hold the exclusive mmap write lock, and since we are cloning
from a source VMA which consequently can't also have its anon_vma state
modified, we know no anon_vma referenced can be empty.
This allows us to significantly simplify this case and just remove
anon_vma_chain objects associated with the VMA, so we add a specific
partial cleanup path for this scenario.
This also allows us to drop the hack of setting vma->anon_vma to NULL
before unlinking anon_vma state in this scenario.
Link: https://lkml.kernel.org/r/cover.1768746221.git.lorenzo.stoakes@oracle.com
Link: https://lkml.kernel.org/r/8644e89369be0cc89d7ac57443dff9e822803c91.1768746221.git.lorenzo.stoakes@oracle.com
Signed-off-by: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Reviewed-by: Liam R. Howlett <Liam.Howlett@oracle.com>
Cc: Barry Song <v-songbaohua@oppo.com>
Cc: Chris Li <chriscli@google.com>
Cc: David Hildenbrand <david@kernel.org>
Cc: Harry Yoo <harry.yoo@oracle.com>
Cc: Jann Horn <jannh@google.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Mike Rapoport <rppt@kernel.org>
Cc: Pedro Falcato <pfalcato@suse.de>
Cc: Rik van Riel <riel@surriel.com>
Cc: Shakeel Butt <shakeel.butt@linux.dev>
Cc: Suren Baghdasaryan <surenb@google.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2026-01-18 14:50:37 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-18 14:50:40 +00:00
|
|
|
|
2026-01-18 14:50:39 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
mm/rmap: improve anon_vma_clone(), unlink_anon_vmas() comments, add asserts
Patch series "mm: clean up anon_vma implementation", v3.
The anon_vma logic is hugely confusing and, much like a bundle of wires
entangled with one another, pulling on one thread seems only to lead to
more entanglement elsewhere.
There is a mish-mash of the core implementation, how that implementation
is invoked, how helper functions are invoked and concepts such as adjacent
anon_vma merge and anon_vma object reuse.
This series tries to improve the situation somewhat.
It starts by establishing some invariants in the core anon_vma_clone() and
unlink_anon_vmas() functions, largely expressed via VM_WARN_ON_ONCE()
asserts.
These act as some form of self-documentation as to the conditions we find
ourselves in when invoking these functions.
We also add kdoc comments for anon_vma_clone() and unlink_anon_vmas().
We then update anon_vma_fork() to avoid passing a partially set up (and
thus invalid) VMA to unlink_anon_vmas() - functions which are used both
for partially set up and valid data types has historically been the source
of a lot of confusion and bugs.
We then makes use of the established known conditions to directly skip
unfaulted VMAs (rather than implicitly via an empty vma->anon_vma_chain
list).
We remove the confusing anon_vma_merge() function (we already have a
concept of anon_vma merge in that we merge anon_vma's that would otherwise
be compatible except for attributes that mprotect() could change - which
anon_vma_merge() has nothing to do with).
We make the anon_vma functions internal to mm as they do not need to be
used by any other part of the kernel, which allows for future abstraction
without concern about this.
We then reduce the time over which we hold the anon rmap lock in
anon_vma_clone(), as it turns out we can allocate anon_vma_chain objects
without holding this lock, since the objects are not yet accessible from
the rmap.
This should reduce anon_vma lock contention.
This additionally allows us to remove a confusing GFP_NOWAIT, GFP_KERNEL
allocation fallback strategy.
Finally, we explicitly indicate which operation is being performed upon
anon_vma_clone(), and separate out fork-only logic to make it very clear
that anon_vma reuse only occurs on fork.
This patch (of 9):
Add kdoc comments and describe exactly what these functions are used for
in detail, pointing out importantly that the anon_vma_clone()
!dst->anon_vma && src->anon_vma dance is ONLY for fork.
Both are confusing functions that will be refactored in a subsequent patch
but the first stage is establishing documentation and some invariants.
Add some basic CONFIG_DEBUG_VM asserts that help document expected state,
specifically:
anon_vma_clone()
- mmap write lock held.
- We do nothing if src VMA is not faulted.
- The destination VMA has no anon_vma_chain yet.
- We are always operating on the same active VMA (i.e. vma->anon_vma).
- If not forking, must operate on the same mm_struct.
unlink_anon_vmas()
- mmap lock held (write lock except when freeing page tables).
- That unfaulted VMAs are no-ops.
We are presented with a special case when anon_vma_clone() fails to
allocate memory, where we have a VMA with partially set up anon_vma state.
Since we hold the exclusive mmap write lock, and since we are cloning
from a source VMA which consequently can't also have its anon_vma state
modified, we know no anon_vma referenced can be empty.
This allows us to significantly simplify this case and just remove
anon_vma_chain objects associated with the VMA, so we add a specific
partial cleanup path for this scenario.
This also allows us to drop the hack of setting vma->anon_vma to NULL
before unlinking anon_vma state in this scenario.
Link: https://lkml.kernel.org/r/cover.1768746221.git.lorenzo.stoakes@oracle.com
Link: https://lkml.kernel.org/r/8644e89369be0cc89d7ac57443dff9e822803c91.1768746221.git.lorenzo.stoakes@oracle.com
Signed-off-by: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Reviewed-by: Liam R. Howlett <Liam.Howlett@oracle.com>
Cc: Barry Song <v-songbaohua@oppo.com>
Cc: Chris Li <chriscli@google.com>
Cc: David Hildenbrand <david@kernel.org>
Cc: Harry Yoo <harry.yoo@oracle.com>
Cc: Jann Horn <jannh@google.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Mike Rapoport <rppt@kernel.org>
Cc: Pedro Falcato <pfalcato@suse.de>
Cc: Rik van Riel <riel@surriel.com>
Cc: Shakeel Butt <shakeel.butt@linux.dev>
Cc: Suren Baghdasaryan <surenb@google.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2026-01-18 14:50:37 +00:00
|
|
|
|
2026-01-18 14:50:40 +00:00
|
|
|
|
|
|
|
|
|
2010-08-09 17:18:39 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
mm: change anon_vma linking to fix multi-process server scalability issue
The old anon_vma code can lead to scalability issues with heavily forking
workloads. Specifically, each anon_vma will be shared between the parent
process and all its child processes.
In a workload with 1000 child processes and a VMA with 1000 anonymous
pages per process that get COWed, this leads to a system with a million
anonymous pages in the same anon_vma, each of which is mapped in just one
of the 1000 processes. However, the current rmap code needs to walk them
all, leading to O(N) scanning complexity for each page.
This can result in systems where one CPU is walking the page tables of
1000 processes in page_referenced_one, while all other CPUs are stuck on
the anon_vma lock. This leads to catastrophic failure for a benchmark
like AIM7, where the total number of processes can reach in the tens of
thousands. Real workloads are still a factor 10 less process intensive
than AIM7, but they are catching up.
This patch changes the way anon_vmas and VMAs are linked, which allows us
to associate multiple anon_vmas with a VMA. At fork time, each child
process gets its own anon_vmas, in which its COWed pages will be
instantiated. The parents' anon_vma is also linked to the VMA, because
non-COWed pages could be present in any of the children.
This reduces rmap scanning complexity to O(1) for the pages of the 1000
child processes, with O(N) complexity for at most 1/N pages in the system.
This reduces the average scanning cost in heavily forking workloads from
O(N) to 2.
The only real complexity in this patch stems from the fact that linking a
VMA to anon_vmas now involves memory allocations. This means vma_adjust
can fail, if it needs to attach a VMA to anon_vma structures. This in
turn means error handling needs to be added to the calling functions.
A second source of complexity is that, because there can be multiple
anon_vmas, the anon_vma linking in vma_adjust can no longer be done under
"the" anon_vma lock. To prevent the rmap code from walking up an
incomplete VMA, this patch introduces the VM_LOCK_RMAP VMA flag. This bit
flag uses the same slot as the NOMMU VM_MAPPED_COPY, with an ifdef in mm.h
to make sure it is impossible to compile a kernel that needs both symbolic
values for the same bitflag.
Some test results:
Without the anon_vma changes, when AIM7 hits around 9.7k users (on a test
box with 16GB RAM and not quite enough IO), the system ends up running
>99% in system time, with every CPU on the same anon_vma lock in the
pageout code.
With these changes, AIM7 hits the cross-over point around 29.7k users.
This happens with ~99% IO wait time, there never seems to be any spike in
system time. The anon_vma lock contention appears to be resolved.
[akpm@linux-foundation.org: cleanups]
Signed-off-by: Rik van Riel <riel@redhat.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Larry Woodman <lwoodman@redhat.com>
Cc: Lee Schermerhorn <Lee.Schermerhorn@hp.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-03-05 13:42:07 -08:00
|
|
|
|
2011-06-17 13:54:23 +02:00
|
|
|
|
|
|
|
|
|
mm anon rmap: replace same_anon_vma linked list with an interval tree.
When a large VMA (anon or private file mapping) is first touched, which
will populate its anon_vma field, and then split into many regions through
the use of mprotect(), the original anon_vma ends up linking all of the
vmas on a linked list. This can cause rmap to become inefficient, as we
have to walk potentially thousands of irrelevent vmas before finding the
one a given anon page might fall into.
By replacing the same_anon_vma linked list with an interval tree (where
each avc's interval is determined by its vma's start and last pgoffs), we
can make rmap efficient for this use case again.
While the change is large, all of its pieces are fairly simple.
Most places that were walking the same_anon_vma list were looking for a
known pgoff, so they can just use the anon_vma_interval_tree_foreach()
interval tree iterator instead. The exception here is ksm, where the
page's index is not known. It would probably be possible to rework ksm so
that the index would be known, but for now I have decided to keep things
simple and just walk the entirety of the interval tree there.
When updating vma's that already have an anon_vma assigned, we must take
care to re-index the corresponding avc's on their interval tree. This is
done through the use of anon_vma_interval_tree_pre_update_vma() and
anon_vma_interval_tree_post_update_vma(), which remove the avc's from
their interval tree before the update and re-insert them after the update.
The anon_vma stays locked during the update, so there is no chance that
rmap would miss the vmas that are being updated.
Signed-off-by: Michel Lespinasse <walken@google.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Rik van Riel <riel@redhat.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Daniel Santos <daniel.santos@pobox.com>
Cc: Hugh Dickins <hughd@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2012-10-08 16:31:39 -07:00
|
|
|
|
2011-06-17 13:54:23 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-09-08 16:15:08 -07:00
|
|
|
|
2022-08-31 19:06:00 +02:00
|
|
|
|
2011-06-17 13:54:23 +02:00
|
|
|
|
2015-01-08 14:32:15 -08:00
|
|
|
|
2011-06-17 13:54:23 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-02-24 12:04:49 -08:00
|
|
|
|
2026-01-18 14:50:40 +00:00
|
|
|
|
2026-01-18 14:50:39 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-01-18 14:50:40 +00:00
|
|
|
|
|
|
|
|
|
2011-06-17 13:54:23 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2012-12-02 19:56:46 +00:00
|
|
|
|
2011-06-17 13:54:23 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-08-31 19:06:00 +02:00
|
|
|
|
|
|
|
|
|
2011-06-17 13:54:23 +02:00
|
|
|
|
|
|
|
|
|
mm: change anon_vma linking to fix multi-process server scalability issue
The old anon_vma code can lead to scalability issues with heavily forking
workloads. Specifically, each anon_vma will be shared between the parent
process and all its child processes.
In a workload with 1000 child processes and a VMA with 1000 anonymous
pages per process that get COWed, this leads to a system with a million
anonymous pages in the same anon_vma, each of which is mapped in just one
of the 1000 processes. However, the current rmap code needs to walk them
all, leading to O(N) scanning complexity for each page.
This can result in systems where one CPU is walking the page tables of
1000 processes in page_referenced_one, while all other CPUs are stuck on
the anon_vma lock. This leads to catastrophic failure for a benchmark
like AIM7, where the total number of processes can reach in the tens of
thousands. Real workloads are still a factor 10 less process intensive
than AIM7, but they are catching up.
This patch changes the way anon_vmas and VMAs are linked, which allows us
to associate multiple anon_vmas with a VMA. At fork time, each child
process gets its own anon_vmas, in which its COWed pages will be
instantiated. The parents' anon_vma is also linked to the VMA, because
non-COWed pages could be present in any of the children.
This reduces rmap scanning complexity to O(1) for the pages of the 1000
child processes, with O(N) complexity for at most 1/N pages in the system.
This reduces the average scanning cost in heavily forking workloads from
O(N) to 2.
The only real complexity in this patch stems from the fact that linking a
VMA to anon_vmas now involves memory allocations. This means vma_adjust
can fail, if it needs to attach a VMA to anon_vma structures. This in
turn means error handling needs to be added to the calling functions.
A second source of complexity is that, because there can be multiple
anon_vmas, the anon_vma linking in vma_adjust can no longer be done under
"the" anon_vma lock. To prevent the rmap code from walking up an
incomplete VMA, this patch introduces the VM_LOCK_RMAP VMA flag. This bit
flag uses the same slot as the NOMMU VM_MAPPED_COPY, with an ifdef in mm.h
to make sure it is impossible to compile a kernel that needs both symbolic
values for the same bitflag.
Some test results:
Without the anon_vma changes, when AIM7 hits around 9.7k users (on a test
box with 16GB RAM and not quite enough IO), the system ends up running
>99% in system time, with every CPU on the same anon_vma lock in the
pageout code.
With these changes, AIM7 hits the cross-over point around 29.7k users.
This happens with ~99% IO wait time, there never seems to be any spike in
system time. The anon_vma lock contention appears to be resolved.
[akpm@linux-foundation.org: cleanups]
Signed-off-by: Rik van Riel <riel@redhat.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Larry Woodman <lwoodman@redhat.com>
Cc: Lee Schermerhorn <Lee.Schermerhorn@hp.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-03-05 13:42:07 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2008-07-25 19:45:34 -07:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2007-05-16 22:10:57 -07:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2012-12-02 19:56:46 +00:00
|
|
|
|
2011-03-22 16:32:48 -07:00
|
|
|
|
2017-09-08 16:15:08 -07:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-01-18 02:53:44 -08:00
|
|
|
|
2016-01-14 15:18:21 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2011-05-24 17:12:08 -07:00
|
|
|
|
|
|
|
|
|
2023-12-20 23:44:56 +01:00
|
|
|
|
2021-02-25 17:18:06 -08:00
|
|
|
|
|
|
|
|
|
2011-05-24 17:12:08 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2011-05-29 10:33:44 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-04-27 13:43:16 +02:00
|
|
|
|
2011-05-29 10:33:44 +02:00
|
|
|
|
2011-05-24 17:12:08 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2019-11-30 17:51:23 -08:00
|
|
|
|
2023-12-20 23:44:56 +01:00
|
|
|
|
2019-11-30 17:51:23 -08:00
|
|
|
|
|
|
|
|
|
userfaultfd: UFFDIO_MOVE uABI
Implement the uABI of UFFDIO_MOVE ioctl.
UFFDIO_COPY performs ~20% better than UFFDIO_MOVE when the application
needs pages to be allocated [1]. However, with UFFDIO_MOVE, if pages are
available (in userspace) for recycling, as is usually the case in heap
compaction algorithms, then we can avoid the page allocation and memcpy
(done by UFFDIO_COPY). Also, since the pages are recycled in the
userspace, we avoid the need to release (via madvise) the pages back to
the kernel [2].
We see over 40% reduction (on a Google pixel 6 device) in the compacting
thread's completion time by using UFFDIO_MOVE vs. UFFDIO_COPY. This was
measured using a benchmark that emulates a heap compaction implementation
using userfaultfd (to allow concurrent accesses by application threads).
More details of the usecase are explained in [2]. Furthermore,
UFFDIO_MOVE enables moving swapped-out pages without touching them within
the same vma. Today, it can only be done by mremap, however it forces
splitting the vma.
[1] https://lore.kernel.org/all/1425575884-2574-1-git-send-email-aarcange@redhat.com/
[2] https://lore.kernel.org/linux-mm/CA+EESO4uO84SSnBhArH4HvLNhaUQ5nZKNKXqxRCyjniNVjp0Aw@mail.gmail.com/
Update for the ioctl_userfaultfd(2) manpage:
UFFDIO_MOVE
(Since Linux xxx) Move a continuous memory chunk into the
userfault registered range and optionally wake up the blocked
thread. The source and destination addresses and the number of
bytes to move are specified by the src, dst, and len fields of
the uffdio_move structure pointed to by argp:
struct uffdio_move {
__u64 dst; /* Destination of move */
__u64 src; /* Source of move */
__u64 len; /* Number of bytes to move */
__u64 mode; /* Flags controlling behavior of move */
__s64 move; /* Number of bytes moved, or negated error */
};
The following value may be bitwise ORed in mode to change the
behavior of the UFFDIO_MOVE operation:
UFFDIO_MOVE_MODE_DONTWAKE
Do not wake up the thread that waits for page-fault
resolution
UFFDIO_MOVE_MODE_ALLOW_SRC_HOLES
Allow holes in the source virtual range that is being moved.
When not specified, the holes will result in ENOENT error.
When specified, the holes will be accounted as successfully
moved memory. This is mostly useful to move hugepage aligned
virtual regions without knowing if there are transparent
hugepages in the regions or not, but preventing the risk of
having to split the hugepage during the operation.
The move field is used by the kernel to return the number of
bytes that was actually moved, or an error (a negated errno-
style value). If the value returned in move doesn't match the
value that was specified in len, the operation fails with the
error EAGAIN. The move field is output-only; it is not read by
the UFFDIO_MOVE operation.
The operation may fail for various reasons. Usually, remapping of
pages that are not exclusive to the given process fail; once KSM
might deduplicate pages or fork() COW-shares pages during fork()
with child processes, they are no longer exclusive. Further, the
kernel might only perform lightweight checks for detecting whether
the pages are exclusive, and return -EBUSY in case that check fails.
To make the operation more likely to succeed, KSM should be
disabled, fork() should be avoided or MADV_DONTFORK should be
configured for the source VMA before fork().
This ioctl(2) operation returns 0 on success. In this case, the
entire area was moved. On error, -1 is returned and errno is
set to indicate the error. Possible errors include:
EAGAIN The number of bytes moved (i.e., the value returned in
the move field) does not equal the value that was
specified in the len field.
EINVAL Either dst or len was not a multiple of the system page
size, or the range specified by src and len or dst and len
was invalid.
EINVAL An invalid bit was specified in the mode field.
ENOENT
The source virtual memory range has unmapped holes and
UFFDIO_MOVE_MODE_ALLOW_SRC_HOLES is not set.
EEXIST
The destination virtual memory range is fully or partially
mapped.
EBUSY
The pages in the source virtual memory range are either
pinned or not exclusive to the process. The kernel might
only perform lightweight checks for detecting whether the
pages are exclusive. To make the operation more likely to
succeed, KSM should be disabled, fork() should be avoided
or MADV_DONTFORK should be configured for the source virtual
memory area before fork().
ENOMEM Allocating memory needed for the operation failed.
ESRCH
The target process has exited at the time of a UFFDIO_MOVE
operation.
Link: https://lkml.kernel.org/r/20231206103702.3873743-3-surenb@google.com
Signed-off-by: Andrea Arcangeli <aarcange@redhat.com>
Signed-off-by: Suren Baghdasaryan <surenb@google.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Axel Rasmussen <axelrasmussen@google.com>
Cc: Brian Geffon <bgeffon@google.com>
Cc: Christian Brauner <brauner@kernel.org>
Cc: David Hildenbrand <david@redhat.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Jann Horn <jannh@google.com>
Cc: Kalesh Singh <kaleshsingh@google.com>
Cc: Liam R. Howlett <Liam.Howlett@oracle.com>
Cc: Lokesh Gidra <lokeshgidra@google.com>
Cc: Matthew Wilcox (Oracle) <willy@infradead.org>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Mike Rapoport (IBM) <rppt@kernel.org>
Cc: Nicolas Geoffray <ngeoffray@google.com>
Cc: Peter Xu <peterx@redhat.com>
Cc: Ryan Roberts <ryan.roberts@arm.com>
Cc: Shuah Khan <shuah@kernel.org>
Cc: ZhangPeng <zhangpeng362@huawei.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2023-12-06 02:36:56 -08:00
|
|
|
|
mm: always call rmap_walk() on locked folios
Patch series "Improve UFFDIO_MOVE scalability by removing anon_vma lock", v2.
Userfaultfd has a scalability issue in its UFFDIO_MOVE ioctl, which is
heavily used in Android as its java garbage collector uses it for
concurrent heap compaction.
The issue arises because UFFDIO_MOVE updates folio->mapping to an anon_vma
with a different root, in order to move the folio from a src VMA to dst
VMA. It performs the operation with the folio locked, but this is
insufficient, because rmap_walk() can be performed on non-KSM anonymous
folios without folio lock.
This means that UFFDIO_MOVE has to acquire the anon_vma write lock of the
root anon_vma belonging to the folio it wishes to move.
This causes scalability bottleneck when multiple threads perform
UFFDIO_MOVE simultanously on distinct pages of the same src VMA. In field
traces of arm64 android devices, we have observed janky user interactions
due to long (sometimes over ~50ms) uninterruptible sleeps on main UI
thread caused by anon_vma lock contention in UFFDIO_MOVE. This is
particularly severe during the beginning of GC's compaction phase when it
is likely to have multiple threads involved.
This patch resolves the issue by removing the exception in rmap_walk() for
non-KSM anon folios by ensuring that all folios are locked during rmap
walk. This is less problematic than it might seem, as the only major
caller which utilises this mode is shrink_active_list(), which is covered
in detail in the first patch of this series.
As a result of changing our approach to locking, we can remove all the
code that took steps to acquire an anon_vma write lock instead of a folio
lock. This results in a significant simplification and scalability
improvement of the code (currently only in UFFDIO_MOVE). Furthermore, as
a side-effect, folio_lock_anon_vma_read() gets simpler as we don't need to
worry that folio->mapping may have changed under us.
This patch (of 2):
Guarantee that rmap_walk() is called on locked folios so that threads
changing folio->mapping and folio->index for non-KSM anon folios can
serialize on fine-grained folio lock rather than anon_vma lock. Other
folio types are already always locked before rmap_walk(). With this, we
are going from 'not necessarily' locking the non-KSM anon folio to
'definitely' locking it during rmap walks.
This patch is in preparation for removing anon_vma write-lock from
UFFDIO_MOVE.
With this patch, three functions are now expected to be called with a
locked folio. To be careful of not missing any case, here is the
exhaustive list of all their callers.
1) rmap_walk() is called from:
a) folio_referenced()
b) damon_folio_mkold()
c) damon_folio_young()
d) page_idle_clear_pte_refs()
e) try_to_unmap()
f) try_to_migrate()
g) folio_mkclean()
h) remove_migration_ptes()
In the above list, first 4 are changed in this patch to try-lock non-KSM
anon folios, similar to other types of folios. The remaining functions in
the list already hold folio lock when calling rmap_walk().
2) folio_lock_anon_vma_read() is called from following functions:
a) collect_procs_anon()
b) page_idle_clear_pte_refs()
c) damon_folio_mkold()
d) damon_folio_young()
e) folio_referenced()
f) try_to_unmap()
g) try_to_migrate()
All the functions in above list, except collect_procs_anon(), are covered
by the rmap_walk() list above. For collect_procs_anon(), with
kill_procs_now() changed to take folio lock in this patch ensures that all
callers of folio_lock_anon_vma_read() now hold the lock.
3) folio_get_anon_vma() is called from following functions, all of which
already hold the folio lock:
a) move_pages_huge_pmd()
b) __folio_split()
c) move_pages_ptes()
d) migrate_folio_unmap()
e) unmap_and_move_huge_page()
Functionally, this patch doesn't break the logic because rmap walkers
generally do some other check to see if what is expected to mapped did
happen so it's fine, or otherwise treat things as best-effort.
Among the 4 functions changed in this patch, folio_referenced() is the
only core-mm function, and is also frequently accessed. To assess the
impact of locking non-KSM anon folios in
shrink_active_list()->folio_referenced() path, we performed an app cycle
test on an arm64 android device. During the whole duration of the test
there were over 140k invocations of shrink_active_list(), out of which
over 29k had at least one non-KSM anon folio on which folio_referenced()
was called. In none of these invocations folio_trylock() failed.
Of course, we now take a lock where we wouldn't previously have. In the
past it would have had a major impact in causing a CoW write fault to copy
a page in do_wp_page(), as commit 09854ba94c6a ("mm: do_wp_page()
simplification") caused a failure to obtain folio lock to result in a page
copy even if one wasn't necessary.
However, since commit 6c287605fd56 ("mm: remember exclusively mapped
anonymous pages with PG_anon_exclusive"), and the introduction of the
folio anon exclusive flag, this issue is significantly mitigated.
The only case remaining that we might worry about from this perspective is
that of read-only folios immediately after fork where the anon exclusive
bit will not have been set yet.
We note however in the case of read-only just-forked folios that
wp_can_reuse_anon_folio() will notice the raised reference count
established by shrink_active_list() via isolate_lru_folios() and refuse to
reuse in any case, so this will in fact have no impact - the folio lock is
ultimately immaterial here.
All-in-all it appears that there is little opportunity for meaningful
negative impact from this change.
Link: https://lkml.kernel.org/r/20250923071019.775806-1-lokeshgidra@google.com
Link: https://lkml.kernel.org/r/20250923071019.775806-2-lokeshgidra@google.com
Signed-off-by: Lokesh Gidra <lokeshgidra@google.com>
Acked-by: David Hildenbrand <david@redhat.com>
Acked-by: Peter Xu <peterx@redhat.com>
Reviewed-by: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: Harry Yoo <harry.yoo@oracle.com>
Cc: Suren Baghdasaryan <surenb@google.com>
Cc: Barry Song <baohua@kernel.org>
Cc: SeongJae Park <sj@kernel.org>
Cc: Jann Horn <jannh@google.com>
Cc: Kalesh Singh <kaleshsingh@google.com>
Cc: Nicolas Geoffray <ngeoffray@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2025-09-23 00:10:18 -07:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2024-10-05 21:01:15 +01:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2011-05-24 17:12:10 -07:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
mm: always call rmap_walk() on locked folios
Patch series "Improve UFFDIO_MOVE scalability by removing anon_vma lock", v2.
Userfaultfd has a scalability issue in its UFFDIO_MOVE ioctl, which is
heavily used in Android as its java garbage collector uses it for
concurrent heap compaction.
The issue arises because UFFDIO_MOVE updates folio->mapping to an anon_vma
with a different root, in order to move the folio from a src VMA to dst
VMA. It performs the operation with the folio locked, but this is
insufficient, because rmap_walk() can be performed on non-KSM anonymous
folios without folio lock.
This means that UFFDIO_MOVE has to acquire the anon_vma write lock of the
root anon_vma belonging to the folio it wishes to move.
This causes scalability bottleneck when multiple threads perform
UFFDIO_MOVE simultanously on distinct pages of the same src VMA. In field
traces of arm64 android devices, we have observed janky user interactions
due to long (sometimes over ~50ms) uninterruptible sleeps on main UI
thread caused by anon_vma lock contention in UFFDIO_MOVE. This is
particularly severe during the beginning of GC's compaction phase when it
is likely to have multiple threads involved.
This patch resolves the issue by removing the exception in rmap_walk() for
non-KSM anon folios by ensuring that all folios are locked during rmap
walk. This is less problematic than it might seem, as the only major
caller which utilises this mode is shrink_active_list(), which is covered
in detail in the first patch of this series.
As a result of changing our approach to locking, we can remove all the
code that took steps to acquire an anon_vma write lock instead of a folio
lock. This results in a significant simplification and scalability
improvement of the code (currently only in UFFDIO_MOVE). Furthermore, as
a side-effect, folio_lock_anon_vma_read() gets simpler as we don't need to
worry that folio->mapping may have changed under us.
This patch (of 2):
Guarantee that rmap_walk() is called on locked folios so that threads
changing folio->mapping and folio->index for non-KSM anon folios can
serialize on fine-grained folio lock rather than anon_vma lock. Other
folio types are already always locked before rmap_walk(). With this, we
are going from 'not necessarily' locking the non-KSM anon folio to
'definitely' locking it during rmap walks.
This patch is in preparation for removing anon_vma write-lock from
UFFDIO_MOVE.
With this patch, three functions are now expected to be called with a
locked folio. To be careful of not missing any case, here is the
exhaustive list of all their callers.
1) rmap_walk() is called from:
a) folio_referenced()
b) damon_folio_mkold()
c) damon_folio_young()
d) page_idle_clear_pte_refs()
e) try_to_unmap()
f) try_to_migrate()
g) folio_mkclean()
h) remove_migration_ptes()
In the above list, first 4 are changed in this patch to try-lock non-KSM
anon folios, similar to other types of folios. The remaining functions in
the list already hold folio lock when calling rmap_walk().
2) folio_lock_anon_vma_read() is called from following functions:
a) collect_procs_anon()
b) page_idle_clear_pte_refs()
c) damon_folio_mkold()
d) damon_folio_young()
e) folio_referenced()
f) try_to_unmap()
g) try_to_migrate()
All the functions in above list, except collect_procs_anon(), are covered
by the rmap_walk() list above. For collect_procs_anon(), with
kill_procs_now() changed to take folio lock in this patch ensures that all
callers of folio_lock_anon_vma_read() now hold the lock.
3) folio_get_anon_vma() is called from following functions, all of which
already hold the folio lock:
a) move_pages_huge_pmd()
b) __folio_split()
c) move_pages_ptes()
d) migrate_folio_unmap()
e) unmap_and_move_huge_page()
Functionally, this patch doesn't break the logic because rmap walkers
generally do some other check to see if what is expected to mapped did
happen so it's fine, or otherwise treat things as best-effort.
Among the 4 functions changed in this patch, folio_referenced() is the
only core-mm function, and is also frequently accessed. To assess the
impact of locking non-KSM anon folios in
shrink_active_list()->folio_referenced() path, we performed an app cycle
test on an arm64 android device. During the whole duration of the test
there were over 140k invocations of shrink_active_list(), out of which
over 29k had at least one non-KSM anon folio on which folio_referenced()
was called. In none of these invocations folio_trylock() failed.
Of course, we now take a lock where we wouldn't previously have. In the
past it would have had a major impact in causing a CoW write fault to copy
a page in do_wp_page(), as commit 09854ba94c6a ("mm: do_wp_page()
simplification") caused a failure to obtain folio lock to result in a page
copy even if one wasn't necessary.
However, since commit 6c287605fd56 ("mm: remember exclusively mapped
anonymous pages with PG_anon_exclusive"), and the introduction of the
folio anon exclusive flag, this issue is significantly mitigated.
The only case remaining that we might worry about from this perspective is
that of read-only folios immediately after fork where the anon exclusive
bit will not have been set yet.
We note however in the case of read-only just-forked folios that
wp_can_reuse_anon_folio() will notice the raised reference count
established by shrink_active_list() via isolate_lru_folios() and refuse to
reuse in any case, so this will in fact have no impact - the folio lock is
ultimately immaterial here.
All-in-all it appears that there is little opportunity for meaningful
negative impact from this change.
Link: https://lkml.kernel.org/r/20250923071019.775806-1-lokeshgidra@google.com
Link: https://lkml.kernel.org/r/20250923071019.775806-2-lokeshgidra@google.com
Signed-off-by: Lokesh Gidra <lokeshgidra@google.com>
Acked-by: David Hildenbrand <david@redhat.com>
Acked-by: Peter Xu <peterx@redhat.com>
Reviewed-by: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: Harry Yoo <harry.yoo@oracle.com>
Cc: Suren Baghdasaryan <surenb@google.com>
Cc: Barry Song <baohua@kernel.org>
Cc: SeongJae Park <sj@kernel.org>
Cc: Jann Horn <jannh@google.com>
Cc: Kalesh Singh <kaleshsingh@google.com>
Cc: Nicolas Geoffray <ngeoffray@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2025-09-23 00:10:18 -07:00
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2022-09-02 20:46:50 +01:00
|
|
|
|
2025-07-04 12:25:20 +02:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2022-09-02 20:46:50 +01:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
2025-07-04 12:25:20 +02:00
|
|
|
|
2011-05-24 17:12:10 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2010-08-25 23:12:54 -07:00
|
|
|
|
|
|
|
|
|
2022-09-02 20:46:50 +01:00
|
|
|
|
2011-05-24 17:12:10 -07:00
|
|
|
|
|
|
|
|
|
2017-01-18 02:53:44 -08:00
|
|
|
|
2011-05-24 17:12:10 -07:00
|
|
|
|
2010-08-25 23:12:54 -07:00
|
|
|
|
2022-09-02 20:46:50 +01:00
|
|
|
|
2014-06-04 16:05:33 -07:00
|
|
|
|
2011-05-24 17:12:10 -07:00
|
|
|
|
2014-06-04 16:05:33 -07:00
|
|
|
|
2011-05-24 17:12:10 -07:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
2011-05-24 17:12:10 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2011-05-24 17:12:13 -07:00
|
|
|
|
2022-09-02 20:46:50 +01:00
|
|
|
|
2011-05-24 17:12:13 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-09-02 20:46:50 +01:00
|
|
|
|
mm: don't be stuck to rmap lock on reclaim path
The rmap locks(i_mmap_rwsem and anon_vma->root->rwsem) could be contended
under memory pressure if processes keep working on their vmas(e.g., fork,
mmap, munmap). It makes reclaim path stuck. In our real workload traces,
we see kswapd is waiting the lock for 300ms+(worst case, a sec) and it
makes other processes entering direct reclaim, which were also stuck on
the lock.
This patch makes lru aging path try_lock mode like shink_page_list so the
reclaim context will keep working with next lru pages without being stuck.
if it found the rmap lock contended, it rotates the page back to head of
lru in both active/inactive lrus to make them consistent behavior, which
is basic starting point rather than adding more heristic.
Since this patch introduces a new "contended" field as out-param along
with try_lock in-param in rmap_walk_control, it's not immutable any longer
if the try_lock is set so remove const keywords on rmap related functions.
Since rmap walking is already expensive operation, I doubt the const
would help sizable benefit( And we didn't have it until 5.17).
In a heavy app workload in Android, trace shows following statistics. It
almost removes rmap lock contention from reclaim path.
Martin Liu reported:
Before:
max_dur(ms) min_dur(ms) max-min(dur)ms avg_dur(ms) sum_dur(ms) count blocked_function
1632 0 1631 151.542173 31672 209 page_lock_anon_vma_read
601 0 601 145.544681 28817 198 rmap_walk_file
After:
max_dur(ms) min_dur(ms) max-min(dur)ms avg_dur(ms) sum_dur(ms) count blocked_function
NaN NaN NaN NaN NaN 0.0 NaN
0 0 0 0.127645 1 12 rmap_walk_file
[minchan@kernel.org: add comment, per Matthew]
Link: https://lkml.kernel.org/r/YnNqeB5tUf6LZ57b@google.com
Link: https://lkml.kernel.org/r/20220510215423.164547-1-minchan@kernel.org
Signed-off-by: Minchan Kim <minchan@kernel.org>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: Suren Baghdasaryan <surenb@google.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: John Dias <joaodias@google.com>
Cc: Tim Murray <timmurray@google.com>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Vladimir Davydov <vdavydov.dev@gmail.com>
Cc: Martin Liu <liumartin@google.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Matthew Wilcox <willy@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2022-05-19 14:08:54 -07:00
|
|
|
|
2011-05-24 17:12:13 -07:00
|
|
|
|
2024-10-05 21:01:15 +01:00
|
|
|
|
mm: don't be stuck to rmap lock on reclaim path
The rmap locks(i_mmap_rwsem and anon_vma->root->rwsem) could be contended
under memory pressure if processes keep working on their vmas(e.g., fork,
mmap, munmap). It makes reclaim path stuck. In our real workload traces,
we see kswapd is waiting the lock for 300ms+(worst case, a sec) and it
makes other processes entering direct reclaim, which were also stuck on
the lock.
This patch makes lru aging path try_lock mode like shink_page_list so the
reclaim context will keep working with next lru pages without being stuck.
if it found the rmap lock contended, it rotates the page back to head of
lru in both active/inactive lrus to make them consistent behavior, which
is basic starting point rather than adding more heristic.
Since this patch introduces a new "contended" field as out-param along
with try_lock in-param in rmap_walk_control, it's not immutable any longer
if the try_lock is set so remove const keywords on rmap related functions.
Since rmap walking is already expensive operation, I doubt the const
would help sizable benefit( And we didn't have it until 5.17).
In a heavy app workload in Android, trace shows following statistics. It
almost removes rmap lock contention from reclaim path.
Martin Liu reported:
Before:
max_dur(ms) min_dur(ms) max-min(dur)ms avg_dur(ms) sum_dur(ms) count blocked_function
1632 0 1631 151.542173 31672 209 page_lock_anon_vma_read
601 0 601 145.544681 28817 198 rmap_walk_file
After:
max_dur(ms) min_dur(ms) max-min(dur)ms avg_dur(ms) sum_dur(ms) count blocked_function
NaN NaN NaN NaN NaN 0.0 NaN
0 0 0 0.127645 1 12 rmap_walk_file
[minchan@kernel.org: add comment, per Matthew]
Link: https://lkml.kernel.org/r/YnNqeB5tUf6LZ57b@google.com
Link: https://lkml.kernel.org/r/20220510215423.164547-1-minchan@kernel.org
Signed-off-by: Minchan Kim <minchan@kernel.org>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: Suren Baghdasaryan <surenb@google.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: John Dias <joaodias@google.com>
Cc: Tim Murray <timmurray@google.com>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Vladimir Davydov <vdavydov.dev@gmail.com>
Cc: Martin Liu <liumartin@google.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Matthew Wilcox <willy@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2022-05-19 14:08:54 -07:00
|
|
|
|
2011-05-24 17:12:10 -07:00
|
|
|
|
2011-05-24 17:12:13 -07:00
|
|
|
|
2011-05-28 13:20:21 -07:00
|
|
|
|
2011-05-24 17:12:13 -07:00
|
|
|
|
2011-05-24 17:12:10 -07:00
|
|
|
|
mm: always call rmap_walk() on locked folios
Patch series "Improve UFFDIO_MOVE scalability by removing anon_vma lock", v2.
Userfaultfd has a scalability issue in its UFFDIO_MOVE ioctl, which is
heavily used in Android as its java garbage collector uses it for
concurrent heap compaction.
The issue arises because UFFDIO_MOVE updates folio->mapping to an anon_vma
with a different root, in order to move the folio from a src VMA to dst
VMA. It performs the operation with the folio locked, but this is
insufficient, because rmap_walk() can be performed on non-KSM anonymous
folios without folio lock.
This means that UFFDIO_MOVE has to acquire the anon_vma write lock of the
root anon_vma belonging to the folio it wishes to move.
This causes scalability bottleneck when multiple threads perform
UFFDIO_MOVE simultanously on distinct pages of the same src VMA. In field
traces of arm64 android devices, we have observed janky user interactions
due to long (sometimes over ~50ms) uninterruptible sleeps on main UI
thread caused by anon_vma lock contention in UFFDIO_MOVE. This is
particularly severe during the beginning of GC's compaction phase when it
is likely to have multiple threads involved.
This patch resolves the issue by removing the exception in rmap_walk() for
non-KSM anon folios by ensuring that all folios are locked during rmap
walk. This is less problematic than it might seem, as the only major
caller which utilises this mode is shrink_active_list(), which is covered
in detail in the first patch of this series.
As a result of changing our approach to locking, we can remove all the
code that took steps to acquire an anon_vma write lock instead of a folio
lock. This results in a significant simplification and scalability
improvement of the code (currently only in UFFDIO_MOVE). Furthermore, as
a side-effect, folio_lock_anon_vma_read() gets simpler as we don't need to
worry that folio->mapping may have changed under us.
This patch (of 2):
Guarantee that rmap_walk() is called on locked folios so that threads
changing folio->mapping and folio->index for non-KSM anon folios can
serialize on fine-grained folio lock rather than anon_vma lock. Other
folio types are already always locked before rmap_walk(). With this, we
are going from 'not necessarily' locking the non-KSM anon folio to
'definitely' locking it during rmap walks.
This patch is in preparation for removing anon_vma write-lock from
UFFDIO_MOVE.
With this patch, three functions are now expected to be called with a
locked folio. To be careful of not missing any case, here is the
exhaustive list of all their callers.
1) rmap_walk() is called from:
a) folio_referenced()
b) damon_folio_mkold()
c) damon_folio_young()
d) page_idle_clear_pte_refs()
e) try_to_unmap()
f) try_to_migrate()
g) folio_mkclean()
h) remove_migration_ptes()
In the above list, first 4 are changed in this patch to try-lock non-KSM
anon folios, similar to other types of folios. The remaining functions in
the list already hold folio lock when calling rmap_walk().
2) folio_lock_anon_vma_read() is called from following functions:
a) collect_procs_anon()
b) page_idle_clear_pte_refs()
c) damon_folio_mkold()
d) damon_folio_young()
e) folio_referenced()
f) try_to_unmap()
g) try_to_migrate()
All the functions in above list, except collect_procs_anon(), are covered
by the rmap_walk() list above. For collect_procs_anon(), with
kill_procs_now() changed to take folio lock in this patch ensures that all
callers of folio_lock_anon_vma_read() now hold the lock.
3) folio_get_anon_vma() is called from following functions, all of which
already hold the folio lock:
a) move_pages_huge_pmd()
b) __folio_split()
c) move_pages_ptes()
d) migrate_folio_unmap()
e) unmap_and_move_huge_page()
Functionally, this patch doesn't break the logic because rmap walkers
generally do some other check to see if what is expected to mapped did
happen so it's fine, or otherwise treat things as best-effort.
Among the 4 functions changed in this patch, folio_referenced() is the
only core-mm function, and is also frequently accessed. To assess the
impact of locking non-KSM anon folios in
shrink_active_list()->folio_referenced() path, we performed an app cycle
test on an arm64 android device. During the whole duration of the test
there were over 140k invocations of shrink_active_list(), out of which
over 29k had at least one non-KSM anon folio on which folio_referenced()
was called. In none of these invocations folio_trylock() failed.
Of course, we now take a lock where we wouldn't previously have. In the
past it would have had a major impact in causing a CoW write fault to copy
a page in do_wp_page(), as commit 09854ba94c6a ("mm: do_wp_page()
simplification") caused a failure to obtain folio lock to result in a page
copy even if one wasn't necessary.
However, since commit 6c287605fd56 ("mm: remember exclusively mapped
anonymous pages with PG_anon_exclusive"), and the introduction of the
folio anon exclusive flag, this issue is significantly mitigated.
The only case remaining that we might worry about from this perspective is
that of read-only folios immediately after fork where the anon exclusive
bit will not have been set yet.
We note however in the case of read-only just-forked folios that
wp_can_reuse_anon_folio() will notice the raised reference count
established by shrink_active_list() via isolate_lru_folios() and refuse to
reuse in any case, so this will in fact have no impact - the folio lock is
ultimately immaterial here.
All-in-all it appears that there is little opportunity for meaningful
negative impact from this change.
Link: https://lkml.kernel.org/r/20250923071019.775806-1-lokeshgidra@google.com
Link: https://lkml.kernel.org/r/20250923071019.775806-2-lokeshgidra@google.com
Signed-off-by: Lokesh Gidra <lokeshgidra@google.com>
Acked-by: David Hildenbrand <david@redhat.com>
Acked-by: Peter Xu <peterx@redhat.com>
Reviewed-by: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: Harry Yoo <harry.yoo@oracle.com>
Cc: Suren Baghdasaryan <surenb@google.com>
Cc: Barry Song <baohua@kernel.org>
Cc: SeongJae Park <sj@kernel.org>
Cc: Jann Horn <jannh@google.com>
Cc: Kalesh Singh <kaleshsingh@google.com>
Cc: Nicolas Geoffray <ngeoffray@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2025-09-23 00:10:18 -07:00
|
|
|
|
|
|
|
|
|
2011-05-24 17:12:13 -07:00
|
|
|
|
2022-02-01 23:33:08 -05:00
|
|
|
|
2025-07-04 12:25:20 +02:00
|
|
|
|
2011-05-24 17:12:13 -07:00
|
|
|
|
2022-02-01 23:33:08 -05:00
|
|
|
|
2011-05-24 17:12:13 -07:00
|
|
|
|
2011-05-24 17:12:10 -07:00
|
|
|
|
2025-07-04 12:25:20 +02:00
|
|
|
|
2015-04-15 16:14:08 -07:00
|
|
|
|
2012-12-02 19:56:50 +00:00
|
|
|
|
2011-05-24 17:12:13 -07:00
|
|
|
|
2022-02-01 23:33:08 -05:00
|
|
|
|
2011-05-28 13:20:21 -07:00
|
|
|
|
2011-05-29 10:33:44 +02:00
|
|
|
|
2011-05-24 17:12:13 -07:00
|
|
|
|
2022-02-01 23:33:08 -05:00
|
|
|
|
2012-12-02 19:56:50 +00:00
|
|
|
|
2011-05-24 17:12:13 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
mm: don't be stuck to rmap lock on reclaim path
The rmap locks(i_mmap_rwsem and anon_vma->root->rwsem) could be contended
under memory pressure if processes keep working on their vmas(e.g., fork,
mmap, munmap). It makes reclaim path stuck. In our real workload traces,
we see kswapd is waiting the lock for 300ms+(worst case, a sec) and it
makes other processes entering direct reclaim, which were also stuck on
the lock.
This patch makes lru aging path try_lock mode like shink_page_list so the
reclaim context will keep working with next lru pages without being stuck.
if it found the rmap lock contended, it rotates the page back to head of
lru in both active/inactive lrus to make them consistent behavior, which
is basic starting point rather than adding more heristic.
Since this patch introduces a new "contended" field as out-param along
with try_lock in-param in rmap_walk_control, it's not immutable any longer
if the try_lock is set so remove const keywords on rmap related functions.
Since rmap walking is already expensive operation, I doubt the const
would help sizable benefit( And we didn't have it until 5.17).
In a heavy app workload in Android, trace shows following statistics. It
almost removes rmap lock contention from reclaim path.
Martin Liu reported:
Before:
max_dur(ms) min_dur(ms) max-min(dur)ms avg_dur(ms) sum_dur(ms) count blocked_function
1632 0 1631 151.542173 31672 209 page_lock_anon_vma_read
601 0 601 145.544681 28817 198 rmap_walk_file
After:
max_dur(ms) min_dur(ms) max-min(dur)ms avg_dur(ms) sum_dur(ms) count blocked_function
NaN NaN NaN NaN NaN 0.0 NaN
0 0 0 0.127645 1 12 rmap_walk_file
[minchan@kernel.org: add comment, per Matthew]
Link: https://lkml.kernel.org/r/YnNqeB5tUf6LZ57b@google.com
Link: https://lkml.kernel.org/r/20220510215423.164547-1-minchan@kernel.org
Signed-off-by: Minchan Kim <minchan@kernel.org>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: Suren Baghdasaryan <surenb@google.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: John Dias <joaodias@google.com>
Cc: Tim Murray <timmurray@google.com>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Vladimir Davydov <vdavydov.dev@gmail.com>
Cc: Martin Liu <liumartin@google.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Matthew Wilcox <willy@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2022-05-19 14:08:54 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2011-05-24 17:12:13 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-02-01 23:33:08 -05:00
|
|
|
|
2014-06-04 16:05:33 -07:00
|
|
|
|
2011-05-24 17:12:13 -07:00
|
|
|
|
2014-06-04 16:05:33 -07:00
|
|
|
|
2011-05-24 17:12:13 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2012-12-02 19:56:50 +00:00
|
|
|
|
2011-05-24 17:12:13 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2012-12-02 19:56:50 +00:00
|
|
|
|
2011-05-24 17:12:13 -07:00
|
|
|
|
2012-12-02 19:56:50 +00:00
|
|
|
|
2011-05-24 17:12:13 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2011-05-24 17:12:10 -07:00
|
|
|
|
2007-02-28 20:13:49 -08:00
|
|
|
|
|
|
|
|
|
2015-09-04 15:47:32 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-05-22 15:30:03 -07:00
|
|
|
|
2015-09-04 15:47:32 -07:00
|
|
|
|
2015-09-04 15:47:35 -07:00
|
|
|
|
2015-09-04 15:47:32 -07:00
|
|
|
|
|
|
|
|
|
2015-09-04 15:47:35 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-01-14 14:09:16 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-07-17 21:10:02 +08:00
|
|
|
|
2025-02-14 22:30:13 +13:00
|
|
|
|
2015-09-04 15:47:32 -07:00
|
|
|
|
|
|
|
|
|
2023-02-27 22:42:28 +01:00
|
|
|
|
2023-04-24 14:54:08 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-09-04 15:47:32 -07:00
|
|
|
|
2025-02-14 22:30:13 +13:00
|
|
|
|
2015-09-04 15:47:32 -07:00
|
|
|
|
2015-09-04 15:47:35 -07:00
|
|
|
|
2017-08-02 13:31:52 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-01-14 14:09:16 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-02-27 22:42:28 +01:00
|
|
|
|
2022-01-14 14:09:16 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-08-02 13:31:52 -07:00
|
|
|
|
2015-09-04 15:47:35 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2015-09-04 15:47:32 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-07-17 21:10:01 +08:00
|
|
|
|
2015-09-04 15:47:32 -07:00
|
|
|
|
2017-08-02 13:31:52 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-01-14 14:09:16 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-08-02 13:31:52 -07:00
|
|
|
|
2022-01-14 14:09:16 -08:00
|
|
|
|
2025-06-09 11:31:30 +01:00
|
|
|
|
2017-08-02 13:31:52 -07:00
|
|
|
|
2022-01-14 14:09:16 -08:00
|
|
|
|
|
|
|
|
|
2017-08-02 13:31:52 -07:00
|
|
|
|
2022-01-14 14:09:16 -08:00
|
|
|
|
|
|
|
|
|
2017-08-02 13:31:52 -07:00
|
|
|
|
|
|
|
|
|
2015-09-04 15:47:32 -07:00
|
|
|
|
2023-07-17 21:10:02 +08:00
|
|
|
|
2025-02-14 22:30:13 +13:00
|
|
|
|
2015-09-04 15:47:32 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-10-05 21:01:14 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-04-21 16:57:29 +08:00
|
|
|
|
2024-10-05 21:01:14 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2024-10-05 21:01:14 +01:00
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2022-01-29 11:52:52 -05:00
|
|
|
|
2025-04-21 16:57:28 +08:00
|
|
|
|
2010-10-02 17:46:06 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-04-21 16:57:28 +08:00
|
|
|
|
|
|
|
|
|
2010-08-09 17:19:10 -07:00
|
|
|
|
2021-06-15 18:24:00 -07:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2022-01-29 11:52:52 -05:00
|
|
|
|
2021-06-15 18:24:00 -07:00
|
|
|
|
|
|
|
|
|
mm/thp: fix vma_address() if virtual address below file offset
Running certain tests with a DEBUG_VM kernel would crash within hours,
on the total_mapcount BUG() in split_huge_page_to_list(), while trying
to free up some memory by punching a hole in a shmem huge page: split's
try_to_unmap() was unable to find all the mappings of the page (which,
on a !DEBUG_VM kernel, would then keep the huge page pinned in memory).
When that BUG() was changed to a WARN(), it would later crash on the
VM_BUG_ON_VMA(end < vma->vm_start || start >= vma->vm_end, vma) in
mm/internal.h:vma_address(), used by rmap_walk_file() for
try_to_unmap().
vma_address() is usually correct, but there's a wraparound case when the
vm_start address is unusually low, but vm_pgoff not so low:
vma_address() chooses max(start, vma->vm_start), but that decides on the
wrong address, because start has become almost ULONG_MAX.
Rewrite vma_address() to be more careful about vm_pgoff; move the
VM_BUG_ON_VMA() out of it, returning -EFAULT for errors, so that it can
be safely used from page_mapped_in_vma() and page_address_in_vma() too.
Add vma_address_end() to apply similar care to end address calculation,
in page_vma_mapped_walk() and page_mkclean_one() and try_to_unmap_one();
though it raises a question of whether callers would do better to supply
pvmw->end to page_vma_mapped_walk() - I chose not, for a smaller patch.
An irritation is that their apparent generality breaks down on KSM
pages, which cannot be located by the page->index that page_to_pgoff()
uses: as commit 4b0ece6fa016 ("mm: migrate: fix remove_migration_pte()
for ksm pages") once discovered. I dithered over the best thing to do
about that, and have ended up with a VM_BUG_ON_PAGE(PageKsm) in both
vma_address() and vma_address_end(); though the only place in danger of
using it on them was try_to_unmap_one().
Sidenote: vma_address() and vma_address_end() now use compound_nr() on a
head page, instead of thp_size(): to make the right calculation on a
hugetlbfs page, whether or not THPs are configured. try_to_unmap() is
used on hugetlbfs pages, but perhaps the wrong calculation never
mattered.
Link: https://lkml.kernel.org/r/caf1c1a3-7cfb-7f8f-1beb-ba816e932825@google.com
Fixes: a8fa41ad2f6f ("mm, rmap: check all VMAs that PTE-mapped THP can be part of")
Signed-off-by: Hugh Dickins <hughd@google.com>
Acked-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Alistair Popple <apopple@nvidia.com>
Cc: Jan Kara <jack@suse.cz>
Cc: Jue Wang <juew@google.com>
Cc: "Matthew Wilcox (Oracle)" <willy@infradead.org>
Cc: Miaohe Lin <linmiaohe@huawei.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Naoya Horiguchi <naoya.horiguchi@nec.com>
Cc: Oscar Salvador <osalvador@suse.de>
Cc: Peter Xu <peterx@redhat.com>
Cc: Ralph Campbell <rcampbell@nvidia.com>
Cc: Shakeel Butt <shakeelb@google.com>
Cc: Wang Yugui <wangyugui@e16-tech.com>
Cc: Yang Shi <shy828301@gmail.com>
Cc: Zi Yan <ziy@nvidia.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2021-06-15 18:23:56 -07:00
|
|
|
|
2025-04-21 16:57:28 +08:00
|
|
|
|
2024-10-05 21:01:13 +01:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
2022-07-06 16:59:26 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2012-12-11 16:00:37 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-03-09 17:24:07 +03:00
|
|
|
|
2012-12-11 16:00:37 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-03-09 17:24:07 +03:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2012-12-11 16:00:37 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-01-21 11:27:31 -05:00
|
|
|
|
2014-01-21 15:49:53 -08:00
|
|
|
|
|
|
|
|
|
2025-06-18 20:42:53 +01:00
|
|
|
|
2014-01-21 15:49:53 -08:00
|
|
|
|
|
|
|
|
|
2023-09-18 15:33:17 +08:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2022-01-21 11:27:31 -05:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2022-01-29 16:06:53 -05:00
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2022-01-21 11:27:31 -05:00
|
|
|
|
|
|
|
|
|
2025-09-23 12:07:07 +01:00
|
|
|
|
mm: rmap: support batched checks of the references for large folios
Patch series "support batch checking of references and unmapping for large
folios", v6.
Currently, folio_referenced_one() always checks the young flag for each
PTE sequentially, which is inefficient for large folios. This
inefficiency is especially noticeable when reclaiming clean file-backed
large folios, where folio_referenced() is observed as a significant
performance hotspot.
Moreover, on Arm architecture, which supports contiguous PTEs, there is
already an optimization to clear the young flags for PTEs within a
contiguous range. However, this is not sufficient. We can extend this to
perform batched operations for the entire large folio (which might exceed
the contiguous range: CONT_PTE_SIZE).
Similar to folio_referenced_one(), we can also apply batched unmapping for
large file folios to optimize the performance of file folio reclamation.
By supporting batched checking of the young flags, flushing TLB entries,
and unmapping, I can observed a significant performance improvements in my
performance tests for file folios reclamation. Please check the
performance data in the commit message of each patch.
This patch (of 5):
Currently, folio_referenced_one() always checks the young flag for each
PTE sequentially, which is inefficient for large folios. This
inefficiency is especially noticeable when reclaiming clean file-backed
large folios, where folio_referenced() is observed as a significant
performance hotspot.
Moreover, on Arm64 architecture, which supports contiguous PTEs, there is
already an optimization to clear the young flags for PTEs within a
contiguous range. However, this is not sufficient. We can extend this to
perform batched operations for the entire large folio (which might exceed
the contiguous range: CONT_PTE_SIZE).
Introduce a new API: clear_flush_young_ptes() to facilitate batched
checking of the young flags and flushing TLB entries, thereby improving
performance during large folio reclamation. And it will be overridden by
the architecture that implements a more efficient batch operation in the
following patches.
While we are at it, rename ptep_clear_flush_young_notify() to
clear_flush_young_ptes_notify() to indicate that this is a batch
operation.
Link: https://lkml.kernel.org/r/cover.1770645603.git.baolin.wang@linux.alibaba.com
Link: https://lkml.kernel.org/r/12132694536834262062d1fb304f8f8a064b6750.1770645603.git.baolin.wang@linux.alibaba.com
Signed-off-by: Baolin Wang <baolin.wang@linux.alibaba.com>
Reviewed-by: Harry Yoo <harry.yoo@oracle.com>
Reviewed-by: Ryan Roberts <ryan.roberts@arm.com>
Acked-by: David Hildenbrand (Arm) <david@kernel.org>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Jann Horn <jannh@google.com>
Cc: Liam Howlett <liam.howlett@oracle.com>
Cc: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: Matthew Wilcox (Oracle) <willy@infradead.org>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Mike Rapoport <rppt@kernel.org>
Cc: Rik van Riel <riel@surriel.com>
Cc: Suren Baghdasaryan <surenb@google.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Will Deacon <will@kernel.org>
Cc: Barry Song <baohua@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2026-02-09 22:07:24 +08:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2017-02-24 14:57:48 -08:00
|
|
|
|
|
|
|
|
|
mm: rmap: support batched checks of the references for large folios
Patch series "support batch checking of references and unmapping for large
folios", v6.
Currently, folio_referenced_one() always checks the young flag for each
PTE sequentially, which is inefficient for large folios. This
inefficiency is especially noticeable when reclaiming clean file-backed
large folios, where folio_referenced() is observed as a significant
performance hotspot.
Moreover, on Arm architecture, which supports contiguous PTEs, there is
already an optimization to clear the young flags for PTEs within a
contiguous range. However, this is not sufficient. We can extend this to
perform batched operations for the entire large folio (which might exceed
the contiguous range: CONT_PTE_SIZE).
Similar to folio_referenced_one(), we can also apply batched unmapping for
large file folios to optimize the performance of file folio reclamation.
By supporting batched checking of the young flags, flushing TLB entries,
and unmapping, I can observed a significant performance improvements in my
performance tests for file folios reclamation. Please check the
performance data in the commit message of each patch.
This patch (of 5):
Currently, folio_referenced_one() always checks the young flag for each
PTE sequentially, which is inefficient for large folios. This
inefficiency is especially noticeable when reclaiming clean file-backed
large folios, where folio_referenced() is observed as a significant
performance hotspot.
Moreover, on Arm64 architecture, which supports contiguous PTEs, there is
already an optimization to clear the young flags for PTEs within a
contiguous range. However, this is not sufficient. We can extend this to
perform batched operations for the entire large folio (which might exceed
the contiguous range: CONT_PTE_SIZE).
Introduce a new API: clear_flush_young_ptes() to facilitate batched
checking of the young flags and flushing TLB entries, thereby improving
performance during large folio reclamation. And it will be overridden by
the architecture that implements a more efficient batch operation in the
following patches.
While we are at it, rename ptep_clear_flush_young_notify() to
clear_flush_young_ptes_notify() to indicate that this is a batch
operation.
Link: https://lkml.kernel.org/r/cover.1770645603.git.baolin.wang@linux.alibaba.com
Link: https://lkml.kernel.org/r/12132694536834262062d1fb304f8f8a064b6750.1770645603.git.baolin.wang@linux.alibaba.com
Signed-off-by: Baolin Wang <baolin.wang@linux.alibaba.com>
Reviewed-by: Harry Yoo <harry.yoo@oracle.com>
Reviewed-by: Ryan Roberts <ryan.roberts@arm.com>
Acked-by: David Hildenbrand (Arm) <david@kernel.org>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Jann Horn <jannh@google.com>
Cc: Liam Howlett <liam.howlett@oracle.com>
Cc: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: Matthew Wilcox (Oracle) <willy@infradead.org>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Mike Rapoport <rppt@kernel.org>
Cc: Rik van Riel <riel@surriel.com>
Cc: Suren Baghdasaryan <surenb@google.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Will Deacon <will@kernel.org>
Cc: Barry Song <baohua@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2026-02-09 22:07:24 +08:00
|
|
|
|
2016-01-15 16:54:37 -08:00
|
|
|
|
2023-09-18 15:33:17 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-09-23 12:07:07 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2016-01-15 16:54:45 -08:00
|
|
|
|
2017-02-24 14:57:48 -08:00
|
|
|
|
2024-07-10 16:36:41 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
mm: convert folio_likely_mapped_shared() to folio_maybe_mapped_shared()
Let's reuse our new MM ownership tracking infrastructure for large folios
to make folio_likely_mapped_shared() never return false negatives -- never
indicating "not mapped shared" although the folio *is* mapped shared.
With that, we can rename it to folio_maybe_mapped_shared() and get rid of
the dependency on the mapcount of the first folio page.
The semantics are now arguably clearer: no mixture of "false negatives"
and "false positives", only the remaining possibility for "false
positives".
Thoroughly document the new semantics. We might now detect that a large
folio is "maybe mapped shared" although it *no longer* is -- but once was.
Now, if more than two MMs mapped a folio at the same time, and the MM
mapping the folio exclusively at the end is not one tracked in the two
folio MM slots, we will detect the folio as "maybe mapped shared".
For anonymous folios, usually (except weird corner cases) all PTEs that
target a "maybe mapped shared" folio are R/O. As soon as a child process
would write to them (iow, actively use them), we would CoW and effectively
replace these PTEs. Most cases (below) are not expected to really matter
with large anonymous folios for this reason.
Most importantly, there will be no change at all for:
* small folios
* hugetlb folios
* PMD-mapped PMD-sized THPs (single mapping)
This change has the potential to affect existing callers of
folio_likely_mapped_shared() -> folio_maybe_mapped_shared():
(1) fs/proc/task_mmu.c: no change (hugetlb)
(2) khugepaged counts PTEs that target shared folios towards
max_ptes_shared (default: HPAGE_PMD_NR / 2), meaning we could skip a
collapse where we would have previously collapsed. This only applies
to anonymous folios and is not expected to matter in practice.
Worth noting that this change sorts out case (A) documented in
commit 1bafe96e89f0 ("mm/khugepaged: replace page_mapcount() check by
folio_likely_mapped_shared()") by removing the possibility for "false
negatives".
(3) MADV_COLD / MADV_PAGEOUT / MADV_FREE will not try splitting
PTE-mapped THPs that are considered shared but not fully covered by
the requested range, consequently not processing them.
PMD-mapped PMD-sized THP are not affected, or when all PTEs are
covered. These functions are usually only called on anon/file folios
that are exclusively mapped most of the time (no other file mappings
or no fork()), so the "false negatives" are not expected to matter in
practice.
(4) mbind() / migrate_pages() / move_pages() will refuse to migrate
shared folios unless MPOL_MF_MOVE_ALL is effective (requires
CAP_SYS_NICE). We will now reject some folios that could be migrated.
Similar to (3), especially with MPOL_MF_MOVE_ALL, so this is not
expected to matter in practice.
Note that cpuset_migrate_mm_workfn() calls do_migrate_pages() with
MPOL_MF_MOVE_ALL.
(5) NUMA hinting
mm/migrate.c:migrate_misplaced_folio_prepare() will skip file
folios that are probably shared libraries (-> "mapped shared" and
executable). This check would have detected it as a shared library at
some point (at least 3 MMs mapping it), so detecting it afterwards
does not sound wrong (still a shared library). Not expected to
matter.
mm/memory.c:numa_migrate_check() will indicate TNF_SHARED in
MAP_SHARED file mappings when encountering a shared folio. Similar
reasoning, not expected to matter.
mm/mprotect.c:change_pte_range() will skip folios detected as
shared in CoW mappings. Similarly, this is not expected to matter in
practice, but if it would ever be a problem we could relax that check
a bit (e.g., basing it on the average page-mapcount in a folio),
because it was only an optimization when many (e.g., 288) processes
were mapping the same folios -- see commit 859d4adc3415 ("mm: numa: do
not trap faults on shared data section pages.")
(6) mm/rmap.c:folio_referenced_one() will skip exclusive swapbacked
folios in dying processes. Applies to anonymous folios only. Without
"false negatives", we'll now skip all actually shared ones. Skipping
ones that are actually exclusive won't really matter, it's a pure
optimization, and is not expected to matter in practice.
In theory, one can detect the problematic scenario: folio_mapcount() > 0
and no folio MM slot is occupied ("state unknown"). One could reset the
MM slots while doing an rmap walk, which migration / folio split already
do when setting everything up. Further, when batching PTEs we might
naturally learn about a owner (e.g., folio_mapcount() == nr_ptes) and
could update the owner. However, we'll defer that until the scenarios
where it would really matter are clear.
Link: https://lkml.kernel.org/r/20250303163014.1128035-15-david@redhat.com
Signed-off-by: David Hildenbrand <david@redhat.com>
Cc: Andy Lutomirks^H^Hski <luto@kernel.org>
Cc: Borislav Betkov <bp@alien8.de>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Jann Horn <jannh@google.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Lance Yang <ioworker0@gmail.com>
Cc: Liam Howlett <liam.howlett@oracle.com>
Cc: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: Matthew Wilcow (Oracle) <willy@infradead.org>
Cc: Michal Koutn <mkoutny@suse.com>
Cc: Muchun Song <muchun.song@linux.dev>
Cc: tejun heo <tj@kernel.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Zefan Li <lizefan.x@bytedance.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2025-03-03 17:30:07 +01:00
|
|
|
|
2024-07-10 16:36:41 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
mm: support batched checking of the young flag for MGLRU
Use the batched helper test_and_clear_young_ptes_notify() to check and
clear the young flag to improve the performance during large folio
reclamation when MGLRU is enabled.
Meanwhile, we can also support batched checking the young and dirty flag
when MGLRU walks the mm's pagetable to update the folios' generation
counter. Since MGLRU also checks the PTE dirty bit, use
folio_pte_batch_flags() with FPB_MERGE_YOUNG_DIRTY set to detect batches
of PTEs for a large folio.
Then we can remove the ptep_test_and_clear_young_notify() since it has no
users now.
Note that we also update the 'young' counter and 'mm_stats[MM_LEAF_YOUNG]'
counter with the batched count in the lru_gen_look_around() and
walk_pte_range(). However, the batched operations may inflate these two
counters, because in a large folio not all PTEs may have been accessed.
(Additionally, tracking how many PTEs have been accessed within a large
folio is not very meaningful, since the mm core actually tracks
access/dirty on a per-folio basis, not per page). The impact analysis is
as follows:
1. The 'mm_stats[MM_LEAF_YOUNG]' counter has no functional impact and
is mainly for debugging.
2. The 'young' counter is used to decide whether to place the current
PMD entry into the bloom filters by suitable_to_scan() (so that next
time we can check whether it has been accessed again), which may set
the hash bit in the bloom filters for a PMD entry that hasn't seen much
access. However, bloom filters inherently allow some error, so this
effect appears negligible.
Link: https://lkml.kernel.org/r/378f4acf7d07410aa7c2e4b49d56bb165918eb34.1772778858.git.baolin.wang@linux.alibaba.com
Signed-off-by: Baolin Wang <baolin.wang@linux.alibaba.com>
Reviewed-by: Rik van Riel <riel@surriel.com>
Acked-by: David Hildenbrand (Arm) <david@kernel.org>
Cc: Alistair Popple <apopple@nvidia.com>
Cc: Axel Rasmussen <axelrasmussen@google.com>
Cc: Barry Song <baohua@kernel.org>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Dev Jain <dev.jain@arm.com>
Cc: Jann Horn <jannh@google.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Liam Howlett <liam.howlett@oracle.com>
Cc: Lorenzo Stoakes (Oracle) <ljs@kernel.org>
Cc: Matthew Wilcox (Oracle) <willy@infradead.org>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Mike Rapoport <rppt@kernel.org>
Cc: Qi Zheng <zhengqi.arch@bytedance.com>
Cc: Ryan Roberts <ryan.roberts@arm.com>
Cc: Shakeel Butt <shakeel.butt@linux.dev>
Cc: Suren Baghdasaryan <surenb@google.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Wei Xu <weixugc@google.com>
Cc: Will Deacon <will@kernel.org>
Cc: Yuanchu Xie <yuanchu@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2026-03-06 14:43:41 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
mm/mglru: fix cgroup OOM during MGLRU state switching
When the Multi-Gen LRU (MGLRU) state is toggled dynamically, a race
condition exists between the state switching and the memory reclaim path.
This can lead to unexpected cgroup OOM kills, even when plenty of
reclaimable memory is available.
Problem Description
==================
The issue arises from a "reclaim vacuum" during the transition.
1. When disabling MGLRU, lru_gen_change_state() sets lrugen->enabled to
false before the pages are drained from MGLRU lists back to traditional
LRU lists.
2. Concurrent reclaimers in shrink_lruvec() see lrugen->enabled as false
and skip the MGLRU path.
3. However, these pages might not have reached the traditional LRU lists
yet, or the changes are not yet visible to all CPUs due to a lack
of synchronization.
4. get_scan_count() subsequently finds traditional LRU lists empty,
concludes there is no reclaimable memory, and triggers an OOM kill.
A similar race can occur during enablement, where the reclaimer sees the
new state but the MGLRU lists haven't been populated via fill_evictable()
yet.
Solution
========
Introduce a 'switching' state (`lru_switch`) to bridge the transition.
When transitioning, the system enters this intermediate state where
the reclaimer is forced to attempt both MGLRU and traditional reclaim
paths sequentially. This ensures that folios remain visible to at least
one reclaim mechanism until the transition is fully materialized across
all CPUs.
Race & Mitigation
================
A race window exists between checking the 'draining' state and performing
the actual list operations. For instance, a reclaimer might observe the
draining state as false just before it changes, leading to a suboptimal
reclaim path decision.
However, this impact is effectively mitigated by the kernel's reclaim
retry mechanism (e.g., in do_try_to_free_pages). If a reclaimer pass fails
to find eligible folios due to a state transition race, subsequent retries
in the loop will observe the updated state and correctly direct the scan
to the appropriate LRU lists. This ensures the transient inconsistency
does not escalate into a terminal OOM kill.
This effectively reduce the race window that previously triggered OOMs
under high memory pressure.
This fix has been verified on v7.0.0-rc1; dynamic toggling of MGLRU
functions correctly without triggering unexpected OOM kills.
Link: https://lkml.kernel.org/r/20260319-b4-switch-mglru-v2-v5-1-8898491e5f17@gmail.com
Signed-off-by: Leno Hou <lenohou@gmail.com>
Acked-by: Yafang Shao <laoar.shao@gmail.com>
Reviewed-by: Barry Song <baohua@kernel.org>
Reviewed-by: Axel Rasmussen <axelrasmussen@google.com>
Cc: Yuanchu Xie <yuanchu@google.com>
Cc: Wei Xu <weixugc@google.com>
Cc: Jialing Wang <wjl.linux@gmail.com>
Cc: Yu Zhao <yuzhao@google.com>
Cc: Kairui Song <ryncsn@gmail.com>
Cc: Bingfang Guo <bfguo@icloud.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2026-03-19 00:30:49 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
mm: support batched checking of the young flag for MGLRU
Use the batched helper test_and_clear_young_ptes_notify() to check and
clear the young flag to improve the performance during large folio
reclamation when MGLRU is enabled.
Meanwhile, we can also support batched checking the young and dirty flag
when MGLRU walks the mm's pagetable to update the folios' generation
counter. Since MGLRU also checks the PTE dirty bit, use
folio_pte_batch_flags() with FPB_MERGE_YOUNG_DIRTY set to detect batches
of PTEs for a large folio.
Then we can remove the ptep_test_and_clear_young_notify() since it has no
users now.
Note that we also update the 'young' counter and 'mm_stats[MM_LEAF_YOUNG]'
counter with the batched count in the lru_gen_look_around() and
walk_pte_range(). However, the batched operations may inflate these two
counters, because in a large folio not all PTEs may have been accessed.
(Additionally, tracking how many PTEs have been accessed within a large
folio is not very meaningful, since the mm core actually tracks
access/dirty on a per-folio basis, not per page). The impact analysis is
as follows:
1. The 'mm_stats[MM_LEAF_YOUNG]' counter has no functional impact and
is mainly for debugging.
2. The 'young' counter is used to decide whether to place the current
PMD entry into the bloom filters by suitable_to_scan() (so that next
time we can check whether it has been accessed again), which may set
the hash bit in the bloom filters for a PMD entry that hasn't seen much
access. However, bloom filters inherently allow some error, so this
effect appears negligible.
Link: https://lkml.kernel.org/r/378f4acf7d07410aa7c2e4b49d56bb165918eb34.1772778858.git.baolin.wang@linux.alibaba.com
Signed-off-by: Baolin Wang <baolin.wang@linux.alibaba.com>
Reviewed-by: Rik van Riel <riel@surriel.com>
Acked-by: David Hildenbrand (Arm) <david@kernel.org>
Cc: Alistair Popple <apopple@nvidia.com>
Cc: Axel Rasmussen <axelrasmussen@google.com>
Cc: Barry Song <baohua@kernel.org>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Dev Jain <dev.jain@arm.com>
Cc: Jann Horn <jannh@google.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Liam Howlett <liam.howlett@oracle.com>
Cc: Lorenzo Stoakes (Oracle) <ljs@kernel.org>
Cc: Matthew Wilcox (Oracle) <willy@infradead.org>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Mike Rapoport <rppt@kernel.org>
Cc: Qi Zheng <zhengqi.arch@bytedance.com>
Cc: Ryan Roberts <ryan.roberts@arm.com>
Cc: Shakeel Butt <shakeel.butt@linux.dev>
Cc: Suren Baghdasaryan <surenb@google.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Wei Xu <weixugc@google.com>
Cc: Will Deacon <will@kernel.org>
Cc: Yuanchu Xie <yuanchu@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2026-03-06 14:43:41 +08:00
|
|
|
|
2022-09-18 02:00:04 -06:00
|
|
|
|
2024-10-19 01:29:39 +00:00
|
|
|
|
mm: rmap: support batched checks of the references for large folios
Patch series "support batch checking of references and unmapping for large
folios", v6.
Currently, folio_referenced_one() always checks the young flag for each
PTE sequentially, which is inefficient for large folios. This
inefficiency is especially noticeable when reclaiming clean file-backed
large folios, where folio_referenced() is observed as a significant
performance hotspot.
Moreover, on Arm architecture, which supports contiguous PTEs, there is
already an optimization to clear the young flags for PTEs within a
contiguous range. However, this is not sufficient. We can extend this to
perform batched operations for the entire large folio (which might exceed
the contiguous range: CONT_PTE_SIZE).
Similar to folio_referenced_one(), we can also apply batched unmapping for
large file folios to optimize the performance of file folio reclamation.
By supporting batched checking of the young flags, flushing TLB entries,
and unmapping, I can observed a significant performance improvements in my
performance tests for file folios reclamation. Please check the
performance data in the commit message of each patch.
This patch (of 5):
Currently, folio_referenced_one() always checks the young flag for each
PTE sequentially, which is inefficient for large folios. This
inefficiency is especially noticeable when reclaiming clean file-backed
large folios, where folio_referenced() is observed as a significant
performance hotspot.
Moreover, on Arm64 architecture, which supports contiguous PTEs, there is
already an optimization to clear the young flags for PTEs within a
contiguous range. However, this is not sufficient. We can extend this to
perform batched operations for the entire large folio (which might exceed
the contiguous range: CONT_PTE_SIZE).
Introduce a new API: clear_flush_young_ptes() to facilitate batched
checking of the young flags and flushing TLB entries, thereby improving
performance during large folio reclamation. And it will be overridden by
the architecture that implements a more efficient batch operation in the
following patches.
While we are at it, rename ptep_clear_flush_young_notify() to
clear_flush_young_ptes_notify() to indicate that this is a batch
operation.
Link: https://lkml.kernel.org/r/cover.1770645603.git.baolin.wang@linux.alibaba.com
Link: https://lkml.kernel.org/r/12132694536834262062d1fb304f8f8a064b6750.1770645603.git.baolin.wang@linux.alibaba.com
Signed-off-by: Baolin Wang <baolin.wang@linux.alibaba.com>
Reviewed-by: Harry Yoo <harry.yoo@oracle.com>
Reviewed-by: Ryan Roberts <ryan.roberts@arm.com>
Acked-by: David Hildenbrand (Arm) <david@kernel.org>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Jann Horn <jannh@google.com>
Cc: Liam Howlett <liam.howlett@oracle.com>
Cc: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: Matthew Wilcox (Oracle) <willy@infradead.org>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Mike Rapoport <rppt@kernel.org>
Cc: Rik van Riel <riel@surriel.com>
Cc: Suren Baghdasaryan <surenb@google.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Will Deacon <will@kernel.org>
Cc: Barry Song <baohua@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2026-02-09 22:07:24 +08:00
|
|
|
|
2022-12-30 14:52:51 -07:00
|
|
|
|
2017-02-24 14:57:48 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-01-21 11:27:31 -05:00
|
|
|
|
2017-02-24 14:57:48 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
mm: support batched checking of the young flag for MGLRU
Use the batched helper test_and_clear_young_ptes_notify() to check and
clear the young flag to improve the performance during large folio
reclamation when MGLRU is enabled.
Meanwhile, we can also support batched checking the young and dirty flag
when MGLRU walks the mm's pagetable to update the folios' generation
counter. Since MGLRU also checks the PTE dirty bit, use
folio_pte_batch_flags() with FPB_MERGE_YOUNG_DIRTY set to detect batches
of PTEs for a large folio.
Then we can remove the ptep_test_and_clear_young_notify() since it has no
users now.
Note that we also update the 'young' counter and 'mm_stats[MM_LEAF_YOUNG]'
counter with the batched count in the lru_gen_look_around() and
walk_pte_range(). However, the batched operations may inflate these two
counters, because in a large folio not all PTEs may have been accessed.
(Additionally, tracking how many PTEs have been accessed within a large
folio is not very meaningful, since the mm core actually tracks
access/dirty on a per-folio basis, not per page). The impact analysis is
as follows:
1. The 'mm_stats[MM_LEAF_YOUNG]' counter has no functional impact and
is mainly for debugging.
2. The 'young' counter is used to decide whether to place the current
PMD entry into the bloom filters by suitable_to_scan() (so that next
time we can check whether it has been accessed again), which may set
the hash bit in the bloom filters for a PMD entry that hasn't seen much
access. However, bloom filters inherently allow some error, so this
effect appears negligible.
Link: https://lkml.kernel.org/r/378f4acf7d07410aa7c2e4b49d56bb165918eb34.1772778858.git.baolin.wang@linux.alibaba.com
Signed-off-by: Baolin Wang <baolin.wang@linux.alibaba.com>
Reviewed-by: Rik van Riel <riel@surriel.com>
Acked-by: David Hildenbrand (Arm) <david@kernel.org>
Cc: Alistair Popple <apopple@nvidia.com>
Cc: Axel Rasmussen <axelrasmussen@google.com>
Cc: Barry Song <baohua@kernel.org>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Dev Jain <dev.jain@arm.com>
Cc: Jann Horn <jannh@google.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Liam Howlett <liam.howlett@oracle.com>
Cc: Lorenzo Stoakes (Oracle) <ljs@kernel.org>
Cc: Matthew Wilcox (Oracle) <willy@infradead.org>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Mike Rapoport <rppt@kernel.org>
Cc: Qi Zheng <zhengqi.arch@bytedance.com>
Cc: Ryan Roberts <ryan.roberts@arm.com>
Cc: Shakeel Butt <shakeel.butt@linux.dev>
Cc: Suren Baghdasaryan <surenb@google.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Wei Xu <weixugc@google.com>
Cc: Will Deacon <will@kernel.org>
Cc: Yuanchu Xie <yuanchu@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2026-03-06 14:43:41 +08:00
|
|
|
|
mm: rmap: support batched checks of the references for large folios
Patch series "support batch checking of references and unmapping for large
folios", v6.
Currently, folio_referenced_one() always checks the young flag for each
PTE sequentially, which is inefficient for large folios. This
inefficiency is especially noticeable when reclaiming clean file-backed
large folios, where folio_referenced() is observed as a significant
performance hotspot.
Moreover, on Arm architecture, which supports contiguous PTEs, there is
already an optimization to clear the young flags for PTEs within a
contiguous range. However, this is not sufficient. We can extend this to
perform batched operations for the entire large folio (which might exceed
the contiguous range: CONT_PTE_SIZE).
Similar to folio_referenced_one(), we can also apply batched unmapping for
large file folios to optimize the performance of file folio reclamation.
By supporting batched checking of the young flags, flushing TLB entries,
and unmapping, I can observed a significant performance improvements in my
performance tests for file folios reclamation. Please check the
performance data in the commit message of each patch.
This patch (of 5):
Currently, folio_referenced_one() always checks the young flag for each
PTE sequentially, which is inefficient for large folios. This
inefficiency is especially noticeable when reclaiming clean file-backed
large folios, where folio_referenced() is observed as a significant
performance hotspot.
Moreover, on Arm64 architecture, which supports contiguous PTEs, there is
already an optimization to clear the young flags for PTEs within a
contiguous range. However, this is not sufficient. We can extend this to
perform batched operations for the entire large folio (which might exceed
the contiguous range: CONT_PTE_SIZE).
Introduce a new API: clear_flush_young_ptes() to facilitate batched
checking of the young flags and flushing TLB entries, thereby improving
performance during large folio reclamation. And it will be overridden by
the architecture that implements a more efficient batch operation in the
following patches.
While we are at it, rename ptep_clear_flush_young_notify() to
clear_flush_young_ptes_notify() to indicate that this is a batch
operation.
Link: https://lkml.kernel.org/r/cover.1770645603.git.baolin.wang@linux.alibaba.com
Link: https://lkml.kernel.org/r/12132694536834262062d1fb304f8f8a064b6750.1770645603.git.baolin.wang@linux.alibaba.com
Signed-off-by: Baolin Wang <baolin.wang@linux.alibaba.com>
Reviewed-by: Harry Yoo <harry.yoo@oracle.com>
Reviewed-by: Ryan Roberts <ryan.roberts@arm.com>
Acked-by: David Hildenbrand (Arm) <david@kernel.org>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Jann Horn <jannh@google.com>
Cc: Liam Howlett <liam.howlett@oracle.com>
Cc: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: Matthew Wilcox (Oracle) <willy@infradead.org>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Mike Rapoport <rppt@kernel.org>
Cc: Rik van Riel <riel@surriel.com>
Cc: Suren Baghdasaryan <surenb@google.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Will Deacon <will@kernel.org>
Cc: Barry Song <baohua@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2026-02-09 22:07:24 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
mm: support batched checking of the young flag for MGLRU
Use the batched helper test_and_clear_young_ptes_notify() to check and
clear the young flag to improve the performance during large folio
reclamation when MGLRU is enabled.
Meanwhile, we can also support batched checking the young and dirty flag
when MGLRU walks the mm's pagetable to update the folios' generation
counter. Since MGLRU also checks the PTE dirty bit, use
folio_pte_batch_flags() with FPB_MERGE_YOUNG_DIRTY set to detect batches
of PTEs for a large folio.
Then we can remove the ptep_test_and_clear_young_notify() since it has no
users now.
Note that we also update the 'young' counter and 'mm_stats[MM_LEAF_YOUNG]'
counter with the batched count in the lru_gen_look_around() and
walk_pte_range(). However, the batched operations may inflate these two
counters, because in a large folio not all PTEs may have been accessed.
(Additionally, tracking how many PTEs have been accessed within a large
folio is not very meaningful, since the mm core actually tracks
access/dirty on a per-folio basis, not per page). The impact analysis is
as follows:
1. The 'mm_stats[MM_LEAF_YOUNG]' counter has no functional impact and
is mainly for debugging.
2. The 'young' counter is used to decide whether to place the current
PMD entry into the bloom filters by suitable_to_scan() (so that next
time we can check whether it has been accessed again), which may set
the hash bit in the bloom filters for a PMD entry that hasn't seen much
access. However, bloom filters inherently allow some error, so this
effect appears negligible.
Link: https://lkml.kernel.org/r/378f4acf7d07410aa7c2e4b49d56bb165918eb34.1772778858.git.baolin.wang@linux.alibaba.com
Signed-off-by: Baolin Wang <baolin.wang@linux.alibaba.com>
Reviewed-by: Rik van Riel <riel@surriel.com>
Acked-by: David Hildenbrand (Arm) <david@kernel.org>
Cc: Alistair Popple <apopple@nvidia.com>
Cc: Axel Rasmussen <axelrasmussen@google.com>
Cc: Barry Song <baohua@kernel.org>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Dev Jain <dev.jain@arm.com>
Cc: Jann Horn <jannh@google.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Liam Howlett <liam.howlett@oracle.com>
Cc: Lorenzo Stoakes (Oracle) <ljs@kernel.org>
Cc: Matthew Wilcox (Oracle) <willy@infradead.org>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Mike Rapoport <rppt@kernel.org>
Cc: Qi Zheng <zhengqi.arch@bytedance.com>
Cc: Ryan Roberts <ryan.roberts@arm.com>
Cc: Shakeel Butt <shakeel.butt@linux.dev>
Cc: Suren Baghdasaryan <surenb@google.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Wei Xu <weixugc@google.com>
Cc: Will Deacon <will@kernel.org>
Cc: Yuanchu Xie <yuanchu@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2026-03-06 14:43:41 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2016-01-15 16:54:37 -08:00
|
|
|
|
|
|
|
|
|
mm: introduce idle page tracking
Knowing the portion of memory that is not used by a certain application or
memory cgroup (idle memory) can be useful for partitioning the system
efficiently, e.g. by setting memory cgroup limits appropriately.
Currently, the only means to estimate the amount of idle memory provided
by the kernel is /proc/PID/{clear_refs,smaps}: the user can clear the
access bit for all pages mapped to a particular process by writing 1 to
clear_refs, wait for some time, and then count smaps:Referenced. However,
this method has two serious shortcomings:
- it does not count unmapped file pages
- it affects the reclaimer logic
To overcome these drawbacks, this patch introduces two new page flags,
Idle and Young, and a new sysfs file, /sys/kernel/mm/page_idle/bitmap.
A page's Idle flag can only be set from userspace by setting bit in
/sys/kernel/mm/page_idle/bitmap at the offset corresponding to the page,
and it is cleared whenever the page is accessed either through page tables
(it is cleared in page_referenced() in this case) or using the read(2)
system call (mark_page_accessed()). Thus by setting the Idle flag for
pages of a particular workload, which can be found e.g. by reading
/proc/PID/pagemap, waiting for some time to let the workload access its
working set, and then reading the bitmap file, one can estimate the amount
of pages that are not used by the workload.
The Young page flag is used to avoid interference with the memory
reclaimer. A page's Young flag is set whenever the Access bit of a page
table entry pointing to the page is cleared by writing to the bitmap file.
If page_referenced() is called on a Young page, it will add 1 to its
return value, therefore concealing the fact that the Access bit was
cleared.
Note, since there is no room for extra page flags on 32 bit, this feature
uses extended page flags when compiled on 32 bit.
[akpm@linux-foundation.org: fix build]
[akpm@linux-foundation.org: kpageidle requires an MMU]
[akpm@linux-foundation.org: decouple from page-flags rework]
Signed-off-by: Vladimir Davydov <vdavydov@parallels.com>
Reviewed-by: Andres Lagar-Cavilla <andreslc@google.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Raghavendra K T <raghavendra.kt@linux.vnet.ibm.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michal Hocko <mhocko@suse.cz>
Cc: Greg Thelen <gthelen@google.com>
Cc: Michel Lespinasse <walken@google.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Pavel Emelyanov <xemul@parallels.com>
Cc: Cyrill Gorcunov <gorcunov@openvz.org>
Cc: Jonathan Corbet <corbet@lwn.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2015-09-09 15:35:45 -07:00
|
|
|
|
2022-01-21 11:27:31 -05:00
|
|
|
|
|
|
|
|
|
mm: introduce idle page tracking
Knowing the portion of memory that is not used by a certain application or
memory cgroup (idle memory) can be useful for partitioning the system
efficiently, e.g. by setting memory cgroup limits appropriately.
Currently, the only means to estimate the amount of idle memory provided
by the kernel is /proc/PID/{clear_refs,smaps}: the user can clear the
access bit for all pages mapped to a particular process by writing 1 to
clear_refs, wait for some time, and then count smaps:Referenced. However,
this method has two serious shortcomings:
- it does not count unmapped file pages
- it affects the reclaimer logic
To overcome these drawbacks, this patch introduces two new page flags,
Idle and Young, and a new sysfs file, /sys/kernel/mm/page_idle/bitmap.
A page's Idle flag can only be set from userspace by setting bit in
/sys/kernel/mm/page_idle/bitmap at the offset corresponding to the page,
and it is cleared whenever the page is accessed either through page tables
(it is cleared in page_referenced() in this case) or using the read(2)
system call (mark_page_accessed()). Thus by setting the Idle flag for
pages of a particular workload, which can be found e.g. by reading
/proc/PID/pagemap, waiting for some time to let the workload access its
working set, and then reading the bitmap file, one can estimate the amount
of pages that are not used by the workload.
The Young page flag is used to avoid interference with the memory
reclaimer. A page's Young flag is set whenever the Access bit of a page
table entry pointing to the page is cleared by writing to the bitmap file.
If page_referenced() is called on a Young page, it will add 1 to its
return value, therefore concealing the fact that the Access bit was
cleared.
Note, since there is no room for extra page flags on 32 bit, this feature
uses extended page flags when compiled on 32 bit.
[akpm@linux-foundation.org: fix build]
[akpm@linux-foundation.org: kpageidle requires an MMU]
[akpm@linux-foundation.org: decouple from page-flags rework]
Signed-off-by: Vladimir Davydov <vdavydov@parallels.com>
Reviewed-by: Andres Lagar-Cavilla <andreslc@google.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Raghavendra K T <raghavendra.kt@linux.vnet.ibm.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Michal Hocko <mhocko@suse.cz>
Cc: Greg Thelen <gthelen@google.com>
Cc: Michel Lespinasse <walken@google.com>
Cc: David Rientjes <rientjes@google.com>
Cc: Pavel Emelyanov <xemul@parallels.com>
Cc: Cyrill Gorcunov <gorcunov@openvz.org>
Cc: Jonathan Corbet <corbet@lwn.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2015-09-09 15:35:45 -07:00
|
|
|
|
|
|
|
|
|
2014-01-21 15:49:53 -08:00
|
|
|
|
|
|
|
|
|
2022-02-14 18:42:33 -08:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2007-02-28 20:13:49 -08:00
|
|
|
|
2014-01-21 15:49:53 -08:00
|
|
|
|
2017-05-03 14:54:27 -07:00
|
|
|
|
2014-01-21 15:49:53 -08:00
|
|
|
|
2017-05-03 14:54:27 -07:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
2022-01-21 11:27:31 -05:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2022-01-21 11:27:31 -05:00
|
|
|
|
2014-01-21 15:49:53 -08:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2022-12-30 14:52:51 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2014-01-21 15:49:53 -08:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2014-01-21 15:49:53 -08:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-01-21 11:27:31 -05:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2012-01-12 17:18:32 -08:00
|
|
|
|
2022-01-21 11:27:31 -05:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2022-01-21 11:27:31 -05:00
|
|
|
|
|
|
|
|
|
mm: don't be stuck to rmap lock on reclaim path
The rmap locks(i_mmap_rwsem and anon_vma->root->rwsem) could be contended
under memory pressure if processes keep working on their vmas(e.g., fork,
mmap, munmap). It makes reclaim path stuck. In our real workload traces,
we see kswapd is waiting the lock for 300ms+(worst case, a sec) and it
makes other processes entering direct reclaim, which were also stuck on
the lock.
This patch makes lru aging path try_lock mode like shink_page_list so the
reclaim context will keep working with next lru pages without being stuck.
if it found the rmap lock contended, it rotates the page back to head of
lru in both active/inactive lrus to make them consistent behavior, which
is basic starting point rather than adding more heristic.
Since this patch introduces a new "contended" field as out-param along
with try_lock in-param in rmap_walk_control, it's not immutable any longer
if the try_lock is set so remove const keywords on rmap related functions.
Since rmap walking is already expensive operation, I doubt the const
would help sizable benefit( And we didn't have it until 5.17).
In a heavy app workload in Android, trace shows following statistics. It
almost removes rmap lock contention from reclaim path.
Martin Liu reported:
Before:
max_dur(ms) min_dur(ms) max-min(dur)ms avg_dur(ms) sum_dur(ms) count blocked_function
1632 0 1631 151.542173 31672 209 page_lock_anon_vma_read
601 0 601 145.544681 28817 198 rmap_walk_file
After:
max_dur(ms) min_dur(ms) max-min(dur)ms avg_dur(ms) sum_dur(ms) count blocked_function
NaN NaN NaN NaN NaN 0.0 NaN
0 0 0 0.127645 1 12 rmap_walk_file
[minchan@kernel.org: add comment, per Matthew]
Link: https://lkml.kernel.org/r/YnNqeB5tUf6LZ57b@google.com
Link: https://lkml.kernel.org/r/20220510215423.164547-1-minchan@kernel.org
Signed-off-by: Minchan Kim <minchan@kernel.org>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: Suren Baghdasaryan <surenb@google.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: John Dias <joaodias@google.com>
Cc: Tim Murray <timmurray@google.com>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Vladimir Davydov <vdavydov.dev@gmail.com>
Cc: Martin Liu <liumartin@google.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Matthew Wilcox <willy@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2022-05-19 14:08:54 -07:00
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2022-01-21 11:27:31 -05:00
|
|
|
|
2025-06-18 20:42:53 +01:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2024-04-28 09:20:48 +08:00
|
|
|
|
2022-01-21 11:27:31 -05:00
|
|
|
|
|
|
|
|
|
2014-01-21 15:49:53 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-01-21 11:27:31 -05:00
|
|
|
|
2014-01-21 15:49:53 -08:00
|
|
|
|
2022-01-29 16:06:53 -05:00
|
|
|
|
mm: don't be stuck to rmap lock on reclaim path
The rmap locks(i_mmap_rwsem and anon_vma->root->rwsem) could be contended
under memory pressure if processes keep working on their vmas(e.g., fork,
mmap, munmap). It makes reclaim path stuck. In our real workload traces,
we see kswapd is waiting the lock for 300ms+(worst case, a sec) and it
makes other processes entering direct reclaim, which were also stuck on
the lock.
This patch makes lru aging path try_lock mode like shink_page_list so the
reclaim context will keep working with next lru pages without being stuck.
if it found the rmap lock contended, it rotates the page back to head of
lru in both active/inactive lrus to make them consistent behavior, which
is basic starting point rather than adding more heristic.
Since this patch introduces a new "contended" field as out-param along
with try_lock in-param in rmap_walk_control, it's not immutable any longer
if the try_lock is set so remove const keywords on rmap related functions.
Since rmap walking is already expensive operation, I doubt the const
would help sizable benefit( And we didn't have it until 5.17).
In a heavy app workload in Android, trace shows following statistics. It
almost removes rmap lock contention from reclaim path.
Martin Liu reported:
Before:
max_dur(ms) min_dur(ms) max-min(dur)ms avg_dur(ms) sum_dur(ms) count blocked_function
1632 0 1631 151.542173 31672 209 page_lock_anon_vma_read
601 0 601 145.544681 28817 198 rmap_walk_file
After:
max_dur(ms) min_dur(ms) max-min(dur)ms avg_dur(ms) sum_dur(ms) count blocked_function
NaN NaN NaN NaN NaN 0.0 NaN
0 0 0 0.127645 1 12 rmap_walk_file
[minchan@kernel.org: add comment, per Matthew]
Link: https://lkml.kernel.org/r/YnNqeB5tUf6LZ57b@google.com
Link: https://lkml.kernel.org/r/20220510215423.164547-1-minchan@kernel.org
Signed-off-by: Minchan Kim <minchan@kernel.org>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: Suren Baghdasaryan <surenb@google.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: John Dias <joaodias@google.com>
Cc: Tim Murray <timmurray@google.com>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Vladimir Davydov <vdavydov.dev@gmail.com>
Cc: Martin Liu <liumartin@google.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Matthew Wilcox <willy@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2022-05-19 14:08:54 -07:00
|
|
|
|
2022-12-30 14:52:51 -07:00
|
|
|
|
2014-01-21 15:49:53 -08:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2026-03-06 14:43:39 +08:00
|
|
|
|
2009-06-16 15:33:05 -07:00
|
|
|
|
2019-05-13 17:21:07 -07:00
|
|
|
|
2014-01-21 15:49:53 -08:00
|
|
|
|
|
|
|
|
|
2022-01-21 11:27:31 -05:00
|
|
|
|
2014-01-21 15:49:53 -08:00
|
|
|
|
|
|
|
|
|
mm: always call rmap_walk() on locked folios
Patch series "Improve UFFDIO_MOVE scalability by removing anon_vma lock", v2.
Userfaultfd has a scalability issue in its UFFDIO_MOVE ioctl, which is
heavily used in Android as its java garbage collector uses it for
concurrent heap compaction.
The issue arises because UFFDIO_MOVE updates folio->mapping to an anon_vma
with a different root, in order to move the folio from a src VMA to dst
VMA. It performs the operation with the folio locked, but this is
insufficient, because rmap_walk() can be performed on non-KSM anonymous
folios without folio lock.
This means that UFFDIO_MOVE has to acquire the anon_vma write lock of the
root anon_vma belonging to the folio it wishes to move.
This causes scalability bottleneck when multiple threads perform
UFFDIO_MOVE simultanously on distinct pages of the same src VMA. In field
traces of arm64 android devices, we have observed janky user interactions
due to long (sometimes over ~50ms) uninterruptible sleeps on main UI
thread caused by anon_vma lock contention in UFFDIO_MOVE. This is
particularly severe during the beginning of GC's compaction phase when it
is likely to have multiple threads involved.
This patch resolves the issue by removing the exception in rmap_walk() for
non-KSM anon folios by ensuring that all folios are locked during rmap
walk. This is less problematic than it might seem, as the only major
caller which utilises this mode is shrink_active_list(), which is covered
in detail in the first patch of this series.
As a result of changing our approach to locking, we can remove all the
code that took steps to acquire an anon_vma write lock instead of a folio
lock. This results in a significant simplification and scalability
improvement of the code (currently only in UFFDIO_MOVE). Furthermore, as
a side-effect, folio_lock_anon_vma_read() gets simpler as we don't need to
worry that folio->mapping may have changed under us.
This patch (of 2):
Guarantee that rmap_walk() is called on locked folios so that threads
changing folio->mapping and folio->index for non-KSM anon folios can
serialize on fine-grained folio lock rather than anon_vma lock. Other
folio types are already always locked before rmap_walk(). With this, we
are going from 'not necessarily' locking the non-KSM anon folio to
'definitely' locking it during rmap walks.
This patch is in preparation for removing anon_vma write-lock from
UFFDIO_MOVE.
With this patch, three functions are now expected to be called with a
locked folio. To be careful of not missing any case, here is the
exhaustive list of all their callers.
1) rmap_walk() is called from:
a) folio_referenced()
b) damon_folio_mkold()
c) damon_folio_young()
d) page_idle_clear_pte_refs()
e) try_to_unmap()
f) try_to_migrate()
g) folio_mkclean()
h) remove_migration_ptes()
In the above list, first 4 are changed in this patch to try-lock non-KSM
anon folios, similar to other types of folios. The remaining functions in
the list already hold folio lock when calling rmap_walk().
2) folio_lock_anon_vma_read() is called from following functions:
a) collect_procs_anon()
b) page_idle_clear_pte_refs()
c) damon_folio_mkold()
d) damon_folio_young()
e) folio_referenced()
f) try_to_unmap()
g) try_to_migrate()
All the functions in above list, except collect_procs_anon(), are covered
by the rmap_walk() list above. For collect_procs_anon(), with
kill_procs_now() changed to take folio lock in this patch ensures that all
callers of folio_lock_anon_vma_read() now hold the lock.
3) folio_get_anon_vma() is called from following functions, all of which
already hold the folio lock:
a) move_pages_huge_pmd()
b) __folio_split()
c) move_pages_ptes()
d) migrate_folio_unmap()
e) unmap_and_move_huge_page()
Functionally, this patch doesn't break the logic because rmap walkers
generally do some other check to see if what is expected to mapped did
happen so it's fine, or otherwise treat things as best-effort.
Among the 4 functions changed in this patch, folio_referenced() is the
only core-mm function, and is also frequently accessed. To assess the
impact of locking non-KSM anon folios in
shrink_active_list()->folio_referenced() path, we performed an app cycle
test on an arm64 android device. During the whole duration of the test
there were over 140k invocations of shrink_active_list(), out of which
over 29k had at least one non-KSM anon folio on which folio_referenced()
was called. In none of these invocations folio_trylock() failed.
Of course, we now take a lock where we wouldn't previously have. In the
past it would have had a major impact in causing a CoW write fault to copy
a page in do_wp_page(), as commit 09854ba94c6a ("mm: do_wp_page()
simplification") caused a failure to obtain folio lock to result in a page
copy even if one wasn't necessary.
However, since commit 6c287605fd56 ("mm: remember exclusively mapped
anonymous pages with PG_anon_exclusive"), and the introduction of the
folio anon exclusive flag, this issue is significantly mitigated.
The only case remaining that we might worry about from this perspective is
that of read-only folios immediately after fork where the anon exclusive
bit will not have been set yet.
We note however in the case of read-only just-forked folios that
wp_can_reuse_anon_folio() will notice the raised reference count
established by shrink_active_list() via isolate_lru_folios() and refuse to
reuse in any case, so this will in fact have no impact - the folio lock is
ultimately immaterial here.
All-in-all it appears that there is little opportunity for meaningful
negative impact from this change.
Link: https://lkml.kernel.org/r/20250923071019.775806-1-lokeshgidra@google.com
Link: https://lkml.kernel.org/r/20250923071019.775806-2-lokeshgidra@google.com
Signed-off-by: Lokesh Gidra <lokeshgidra@google.com>
Acked-by: David Hildenbrand <david@redhat.com>
Acked-by: Peter Xu <peterx@redhat.com>
Reviewed-by: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: Harry Yoo <harry.yoo@oracle.com>
Cc: Suren Baghdasaryan <surenb@google.com>
Cc: Barry Song <baohua@kernel.org>
Cc: SeongJae Park <sj@kernel.org>
Cc: Jann Horn <jannh@google.com>
Cc: Kalesh Singh <kaleshsingh@google.com>
Cc: Nicolas Geoffray <ngeoffray@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2025-09-23 00:10:18 -07:00
|
|
|
|
2022-01-21 11:27:31 -05:00
|
|
|
|
2014-01-21 15:49:53 -08:00
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2014-01-21 15:49:53 -08:00
|
|
|
|
2022-01-29 16:06:53 -05:00
|
|
|
|
2014-01-21 15:49:53 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-01-21 11:27:31 -05:00
|
|
|
|
2014-01-21 15:49:53 -08:00
|
|
|
|
mm: don't be stuck to rmap lock on reclaim path
The rmap locks(i_mmap_rwsem and anon_vma->root->rwsem) could be contended
under memory pressure if processes keep working on their vmas(e.g., fork,
mmap, munmap). It makes reclaim path stuck. In our real workload traces,
we see kswapd is waiting the lock for 300ms+(worst case, a sec) and it
makes other processes entering direct reclaim, which were also stuck on
the lock.
This patch makes lru aging path try_lock mode like shink_page_list so the
reclaim context will keep working with next lru pages without being stuck.
if it found the rmap lock contended, it rotates the page back to head of
lru in both active/inactive lrus to make them consistent behavior, which
is basic starting point rather than adding more heristic.
Since this patch introduces a new "contended" field as out-param along
with try_lock in-param in rmap_walk_control, it's not immutable any longer
if the try_lock is set so remove const keywords on rmap related functions.
Since rmap walking is already expensive operation, I doubt the const
would help sizable benefit( And we didn't have it until 5.17).
In a heavy app workload in Android, trace shows following statistics. It
almost removes rmap lock contention from reclaim path.
Martin Liu reported:
Before:
max_dur(ms) min_dur(ms) max-min(dur)ms avg_dur(ms) sum_dur(ms) count blocked_function
1632 0 1631 151.542173 31672 209 page_lock_anon_vma_read
601 0 601 145.544681 28817 198 rmap_walk_file
After:
max_dur(ms) min_dur(ms) max-min(dur)ms avg_dur(ms) sum_dur(ms) count blocked_function
NaN NaN NaN NaN NaN 0.0 NaN
0 0 0 0.127645 1 12 rmap_walk_file
[minchan@kernel.org: add comment, per Matthew]
Link: https://lkml.kernel.org/r/YnNqeB5tUf6LZ57b@google.com
Link: https://lkml.kernel.org/r/20220510215423.164547-1-minchan@kernel.org
Signed-off-by: Minchan Kim <minchan@kernel.org>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: Suren Baghdasaryan <surenb@google.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: John Dias <joaodias@google.com>
Cc: Tim Murray <timmurray@google.com>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Vladimir Davydov <vdavydov.dev@gmail.com>
Cc: Martin Liu <liumartin@google.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Matthew Wilcox <willy@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2022-05-19 14:08:54 -07:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
2022-04-28 23:16:10 -07:00
|
|
|
|
2006-09-25 23:30:57 -07:00
|
|
|
|
2022-04-28 23:16:10 -07:00
|
|
|
|
|
|
|
|
|
2018-12-28 00:38:09 -08:00
|
|
|
|
2022-04-28 23:16:10 -07:00
|
|
|
|
2006-09-25 23:30:57 -07:00
|
|
|
|
2017-08-31 17:17:27 -04:00
|
|
|
|
|
|
|
|
|
2022-01-20 18:20:07 -05:00
|
|
|
|
2017-08-31 17:17:27 -04:00
|
|
|
|
2023-01-10 13:57:22 +11:00
|
|
|
|
|
|
|
|
|
2018-12-28 00:38:09 -08:00
|
|
|
|
2017-08-31 17:17:27 -04:00
|
|
|
|
2022-04-28 23:16:10 -07:00
|
|
|
|
2017-02-24 14:57:57 -08:00
|
|
|
|
2017-08-31 17:17:27 -04:00
|
|
|
|
2022-04-28 23:16:10 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
mm: ptep_get() conversion
Convert all instances of direct pte_t* dereferencing to instead use
ptep_get() helper. This means that by default, the accesses change from a
C dereference to a READ_ONCE(). This is technically the correct thing to
do since where pgtables are modified by HW (for access/dirty) they are
volatile and therefore we should always ensure READ_ONCE() semantics.
But more importantly, by always using the helper, it can be overridden by
the architecture to fully encapsulate the contents of the pte. Arch code
is deliberately not converted, as the arch code knows best. It is
intended that arch code (arm64) will override the default with its own
implementation that can (e.g.) hide certain bits from the core code, or
determine young/dirty status by mixing in state from another source.
Conversion was done using Coccinelle:
----
// $ make coccicheck \
// COCCI=ptepget.cocci \
// SPFLAGS="--include-headers" \
// MODE=patch
virtual patch
@ depends on patch @
pte_t *v;
@@
- *v
+ ptep_get(v)
----
Then reviewed and hand-edited to avoid multiple unnecessary calls to
ptep_get(), instead opting to store the result of a single call in a
variable, where it is correct to do so. This aims to negate any cost of
READ_ONCE() and will benefit arch-overrides that may be more complex.
Included is a fix for an issue in an earlier version of this patch that
was pointed out by kernel test robot. The issue arose because config
MMU=n elides definition of the ptep helper functions, including
ptep_get(). HUGETLB_PAGE=n configs still define a simple
huge_ptep_clear_flush() for linking purposes, which dereferences the ptep.
So when both configs are disabled, this caused a build error because
ptep_get() is not defined. Fix by continuing to do a direct dereference
when MMU=n. This is safe because for this config the arch code cannot be
trying to virtualize the ptes because none of the ptep helpers are
defined.
Link: https://lkml.kernel.org/r/20230612151545.3317766-4-ryan.roberts@arm.com
Reported-by: kernel test robot <lkp@intel.com>
Link: https://lore.kernel.org/oe-kbuild-all/202305120142.yXsNEo6H-lkp@intel.com/
Signed-off-by: Ryan Roberts <ryan.roberts@arm.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Alexander Potapenko <glider@google.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Alex Williamson <alex.williamson@redhat.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Andrey Konovalov <andreyknvl@gmail.com>
Cc: Andrey Ryabinin <ryabinin.a.a@gmail.com>
Cc: Christian Brauner <brauner@kernel.org>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: Daniel Vetter <daniel@ffwll.ch>
Cc: Dave Airlie <airlied@gmail.com>
Cc: Dimitri Sivanich <dimitri.sivanich@hpe.com>
Cc: Dmitry Vyukov <dvyukov@google.com>
Cc: Ian Rogers <irogers@google.com>
Cc: Jason Gunthorpe <jgg@ziepe.ca>
Cc: Jérôme Glisse <jglisse@redhat.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Lorenzo Stoakes <lstoakes@gmail.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Miaohe Lin <linmiaohe@huawei.com>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Mike Rapoport (IBM) <rppt@kernel.org>
Cc: Muchun Song <muchun.song@linux.dev>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Naoya Horiguchi <naoya.horiguchi@nec.com>
Cc: Oleksandr Tyshchenko <oleksandr_tyshchenko@epam.com>
Cc: Pavel Tatashin <pasha.tatashin@soleen.com>
Cc: Roman Gushchin <roman.gushchin@linux.dev>
Cc: SeongJae Park <sj@kernel.org>
Cc: Shakeel Butt <shakeelb@google.com>
Cc: Uladzislau Rezki (Sony) <urezki@gmail.com>
Cc: Vincenzo Frascino <vincenzo.frascino@arm.com>
Cc: Yu Zhao <yuzhao@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2023-06-12 16:15:45 +01:00
|
|
|
|
2006-09-25 23:30:57 -07:00
|
|
|
|
2025-02-10 20:37:54 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
mm: ptep_get() conversion
Convert all instances of direct pte_t* dereferencing to instead use
ptep_get() helper. This means that by default, the accesses change from a
C dereference to a READ_ONCE(). This is technically the correct thing to
do since where pgtables are modified by HW (for access/dirty) they are
volatile and therefore we should always ensure READ_ONCE() semantics.
But more importantly, by always using the helper, it can be overridden by
the architecture to fully encapsulate the contents of the pte. Arch code
is deliberately not converted, as the arch code knows best. It is
intended that arch code (arm64) will override the default with its own
implementation that can (e.g.) hide certain bits from the core code, or
determine young/dirty status by mixing in state from another source.
Conversion was done using Coccinelle:
----
// $ make coccicheck \
// COCCI=ptepget.cocci \
// SPFLAGS="--include-headers" \
// MODE=patch
virtual patch
@ depends on patch @
pte_t *v;
@@
- *v
+ ptep_get(v)
----
Then reviewed and hand-edited to avoid multiple unnecessary calls to
ptep_get(), instead opting to store the result of a single call in a
variable, where it is correct to do so. This aims to negate any cost of
READ_ONCE() and will benefit arch-overrides that may be more complex.
Included is a fix for an issue in an earlier version of this patch that
was pointed out by kernel test robot. The issue arose because config
MMU=n elides definition of the ptep helper functions, including
ptep_get(). HUGETLB_PAGE=n configs still define a simple
huge_ptep_clear_flush() for linking purposes, which dereferences the ptep.
So when both configs are disabled, this caused a build error because
ptep_get() is not defined. Fix by continuing to do a direct dereference
when MMU=n. This is safe because for this config the arch code cannot be
trying to virtualize the ptes because none of the ptep helpers are
defined.
Link: https://lkml.kernel.org/r/20230612151545.3317766-4-ryan.roberts@arm.com
Reported-by: kernel test robot <lkp@intel.com>
Link: https://lore.kernel.org/oe-kbuild-all/202305120142.yXsNEo6H-lkp@intel.com/
Signed-off-by: Ryan Roberts <ryan.roberts@arm.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Alexander Potapenko <glider@google.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Alex Williamson <alex.williamson@redhat.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Andrey Konovalov <andreyknvl@gmail.com>
Cc: Andrey Ryabinin <ryabinin.a.a@gmail.com>
Cc: Christian Brauner <brauner@kernel.org>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: Daniel Vetter <daniel@ffwll.ch>
Cc: Dave Airlie <airlied@gmail.com>
Cc: Dimitri Sivanich <dimitri.sivanich@hpe.com>
Cc: Dmitry Vyukov <dvyukov@google.com>
Cc: Ian Rogers <irogers@google.com>
Cc: Jason Gunthorpe <jgg@ziepe.ca>
Cc: Jérôme Glisse <jglisse@redhat.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Lorenzo Stoakes <lstoakes@gmail.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Miaohe Lin <linmiaohe@huawei.com>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Mike Rapoport (IBM) <rppt@kernel.org>
Cc: Muchun Song <muchun.song@linux.dev>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Naoya Horiguchi <naoya.horiguchi@nec.com>
Cc: Oleksandr Tyshchenko <oleksandr_tyshchenko@epam.com>
Cc: Pavel Tatashin <pasha.tatashin@soleen.com>
Cc: Roman Gushchin <roman.gushchin@linux.dev>
Cc: SeongJae Park <sj@kernel.org>
Cc: Shakeel Butt <shakeelb@google.com>
Cc: Uladzislau Rezki (Sony) <urezki@gmail.com>
Cc: Vincenzo Frascino <vincenzo.frascino@arm.com>
Cc: Yu Zhao <yuzhao@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2023-06-12 16:15:45 +01:00
|
|
|
|
2017-02-24 14:57:57 -08:00
|
|
|
|
2006-09-25 23:30:57 -07:00
|
|
|
|
mm: ptep_get() conversion
Convert all instances of direct pte_t* dereferencing to instead use
ptep_get() helper. This means that by default, the accesses change from a
C dereference to a READ_ONCE(). This is technically the correct thing to
do since where pgtables are modified by HW (for access/dirty) they are
volatile and therefore we should always ensure READ_ONCE() semantics.
But more importantly, by always using the helper, it can be overridden by
the architecture to fully encapsulate the contents of the pte. Arch code
is deliberately not converted, as the arch code knows best. It is
intended that arch code (arm64) will override the default with its own
implementation that can (e.g.) hide certain bits from the core code, or
determine young/dirty status by mixing in state from another source.
Conversion was done using Coccinelle:
----
// $ make coccicheck \
// COCCI=ptepget.cocci \
// SPFLAGS="--include-headers" \
// MODE=patch
virtual patch
@ depends on patch @
pte_t *v;
@@
- *v
+ ptep_get(v)
----
Then reviewed and hand-edited to avoid multiple unnecessary calls to
ptep_get(), instead opting to store the result of a single call in a
variable, where it is correct to do so. This aims to negate any cost of
READ_ONCE() and will benefit arch-overrides that may be more complex.
Included is a fix for an issue in an earlier version of this patch that
was pointed out by kernel test robot. The issue arose because config
MMU=n elides definition of the ptep helper functions, including
ptep_get(). HUGETLB_PAGE=n configs still define a simple
huge_ptep_clear_flush() for linking purposes, which dereferences the ptep.
So when both configs are disabled, this caused a build error because
ptep_get() is not defined. Fix by continuing to do a direct dereference
when MMU=n. This is safe because for this config the arch code cannot be
trying to virtualize the ptes because none of the ptep helpers are
defined.
Link: https://lkml.kernel.org/r/20230612151545.3317766-4-ryan.roberts@arm.com
Reported-by: kernel test robot <lkp@intel.com>
Link: https://lore.kernel.org/oe-kbuild-all/202305120142.yXsNEo6H-lkp@intel.com/
Signed-off-by: Ryan Roberts <ryan.roberts@arm.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Alexander Potapenko <glider@google.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Alex Williamson <alex.williamson@redhat.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Andrey Konovalov <andreyknvl@gmail.com>
Cc: Andrey Ryabinin <ryabinin.a.a@gmail.com>
Cc: Christian Brauner <brauner@kernel.org>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: Daniel Vetter <daniel@ffwll.ch>
Cc: Dave Airlie <airlied@gmail.com>
Cc: Dimitri Sivanich <dimitri.sivanich@hpe.com>
Cc: Dmitry Vyukov <dvyukov@google.com>
Cc: Ian Rogers <irogers@google.com>
Cc: Jason Gunthorpe <jgg@ziepe.ca>
Cc: Jérôme Glisse <jglisse@redhat.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Lorenzo Stoakes <lstoakes@gmail.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Miaohe Lin <linmiaohe@huawei.com>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Mike Rapoport (IBM) <rppt@kernel.org>
Cc: Muchun Song <muchun.song@linux.dev>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Naoya Horiguchi <naoya.horiguchi@nec.com>
Cc: Oleksandr Tyshchenko <oleksandr_tyshchenko@epam.com>
Cc: Pavel Tatashin <pasha.tatashin@soleen.com>
Cc: Roman Gushchin <roman.gushchin@linux.dev>
Cc: SeongJae Park <sj@kernel.org>
Cc: Shakeel Butt <shakeelb@google.com>
Cc: Uladzislau Rezki (Sony) <urezki@gmail.com>
Cc: Vincenzo Frascino <vincenzo.frascino@arm.com>
Cc: Yu Zhao <yuzhao@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2023-06-12 16:15:45 +01:00
|
|
|
|
Revert "rmap: do not call mmu_notifier_invalidate_page() under ptl"
This reverts commit aac2fea94f7a3df8ad1eeb477eb2643f81fd5393.
It turns out that that patch was complete and utter garbage, and broke
KVM, resulting in odd oopses.
Quoting Andrea Arcangeli:
"The aforementioned commit has 3 bugs.
1) mmu_notifier_invalidate_range cannot be used in replacement of
mmu_notifier_invalidate_range_start/end.
For KVM mmu_notifier_invalidate_range is a noop and rightfully so.
A MMU notifier implementation has to implement either
->invalidate_range method or the invalidate_range_start/end
methods, not both. And if you implement invalidate_range_start/end
like KVM is forced to do, calling mmu_notifier_invalidate_range in
common code is a noop for KVM.
For those MMU notifiers that can get away only implementing
->invalidate_range, the ->invalidate_range is implicitly called by
mmu_notifier_invalidate_range_end(). And only those secondary MMUs
that share the same pagetable with the primary MMU (like AMD
iommuv2) can get away only implementing ->invalidate_range.
So all cases (THP on/off) are broken right now.
To fix this is enough to replace mmu_notifier_invalidate_range with
mmu_notifier_invalidate_range_start;mmu_notifier_invalidate_range_end.
Either that or call multiple mmu_notifier_invalidate_page like
before.
2) address + (1UL << compound_order(page) is buggy, it should be
PAGE_SIZE << compound_order(page), it's bytes not pages, 2M not
512.
3) The whole invalidate_range thing was an attempt to call a single
invalidate while walking multiple 4k ptes that maps the same THP
(after a pmd virtual split without physical compound page THP
split).
It's unclear if the rmap_walk will always provide an address that
is 2M aligned as parameter to try_to_unmap_one, in presence of THP.
I think it needs also an address &= (PAGE_SIZE <<
compound_order(page)) - 1 to be safe"
In general, we should stop making excuses for horrible MMU notifier
users. It's much more important that the core VM is sane and safe, than
letting MMU notifiers sleep.
So if some MMU notifier is sleeping under a spinlock, we need to fix the
notifier, not try to make excuses for that garbage in the core VM.
Reported-and-tested-by: Bernhard Held <berny156@gmx.de>
Reported-and-tested-by: Adam Borowski <kilobyte@angband.pl>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Radim Krčmář <rkrcmar@redhat.com>
Cc: Wanpeng Li <kernellwp@gmail.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Takashi Iwai <tiwai@suse.de>
Cc: Nadav Amit <nadav.amit@gmail.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Jérôme Glisse <jglisse@redhat.com>
Cc: axie <axie@amd.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-08-29 09:11:06 -07:00
|
|
|
|
2017-02-24 14:57:57 -08:00
|
|
|
|
|
|
|
|
|
Revert "rmap: do not call mmu_notifier_invalidate_page() under ptl"
This reverts commit aac2fea94f7a3df8ad1eeb477eb2643f81fd5393.
It turns out that that patch was complete and utter garbage, and broke
KVM, resulting in odd oopses.
Quoting Andrea Arcangeli:
"The aforementioned commit has 3 bugs.
1) mmu_notifier_invalidate_range cannot be used in replacement of
mmu_notifier_invalidate_range_start/end.
For KVM mmu_notifier_invalidate_range is a noop and rightfully so.
A MMU notifier implementation has to implement either
->invalidate_range method or the invalidate_range_start/end
methods, not both. And if you implement invalidate_range_start/end
like KVM is forced to do, calling mmu_notifier_invalidate_range in
common code is a noop for KVM.
For those MMU notifiers that can get away only implementing
->invalidate_range, the ->invalidate_range is implicitly called by
mmu_notifier_invalidate_range_end(). And only those secondary MMUs
that share the same pagetable with the primary MMU (like AMD
iommuv2) can get away only implementing ->invalidate_range.
So all cases (THP on/off) are broken right now.
To fix this is enough to replace mmu_notifier_invalidate_range with
mmu_notifier_invalidate_range_start;mmu_notifier_invalidate_range_end.
Either that or call multiple mmu_notifier_invalidate_page like
before.
2) address + (1UL << compound_order(page) is buggy, it should be
PAGE_SIZE << compound_order(page), it's bytes not pages, 2M not
512.
3) The whole invalidate_range thing was an attempt to call a single
invalidate while walking multiple 4k ptes that maps the same THP
(after a pmd virtual split without physical compound page THP
split).
It's unclear if the rmap_walk will always provide an address that
is 2M aligned as parameter to try_to_unmap_one, in presence of THP.
I think it needs also an address &= (PAGE_SIZE <<
compound_order(page)) - 1 to be safe"
In general, we should stop making excuses for horrible MMU notifier
users. It's much more important that the core VM is sane and safe, than
letting MMU notifiers sleep.
So if some MMU notifier is sleeping under a spinlock, we need to fix the
notifier, not try to make excuses for that garbage in the core VM.
Reported-and-tested-by: Bernhard Held <berny156@gmx.de>
Reported-and-tested-by: Adam Borowski <kilobyte@angband.pl>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Radim Krčmář <rkrcmar@redhat.com>
Cc: Wanpeng Li <kernellwp@gmail.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Takashi Iwai <tiwai@suse.de>
Cc: Nadav Amit <nadav.amit@gmail.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Jérôme Glisse <jglisse@redhat.com>
Cc: axie <axie@amd.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-08-29 09:11:06 -07:00
|
|
|
|
2017-02-24 14:57:57 -08:00
|
|
|
|
|
|
|
|
|
2020-04-06 20:04:35 -07:00
|
|
|
|
2022-04-28 23:16:10 -07:00
|
|
|
|
2025-10-01 16:56:55 +10:00
|
|
|
|
2017-02-24 14:57:57 -08:00
|
|
|
|
2025-10-01 16:56:55 +10:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-02-24 14:57:57 -08:00
|
|
|
|
|
|
|
|
|
2022-04-28 23:16:09 -07:00
|
|
|
|
|
|
|
|
|
2019-05-13 17:19:11 -07:00
|
|
|
|
2017-02-24 14:57:57 -08:00
|
|
|
|
|
|
|
|
|
Revert "rmap: do not call mmu_notifier_invalidate_page() under ptl"
This reverts commit aac2fea94f7a3df8ad1eeb477eb2643f81fd5393.
It turns out that that patch was complete and utter garbage, and broke
KVM, resulting in odd oopses.
Quoting Andrea Arcangeli:
"The aforementioned commit has 3 bugs.
1) mmu_notifier_invalidate_range cannot be used in replacement of
mmu_notifier_invalidate_range_start/end.
For KVM mmu_notifier_invalidate_range is a noop and rightfully so.
A MMU notifier implementation has to implement either
->invalidate_range method or the invalidate_range_start/end
methods, not both. And if you implement invalidate_range_start/end
like KVM is forced to do, calling mmu_notifier_invalidate_range in
common code is a noop for KVM.
For those MMU notifiers that can get away only implementing
->invalidate_range, the ->invalidate_range is implicitly called by
mmu_notifier_invalidate_range_end(). And only those secondary MMUs
that share the same pagetable with the primary MMU (like AMD
iommuv2) can get away only implementing ->invalidate_range.
So all cases (THP on/off) are broken right now.
To fix this is enough to replace mmu_notifier_invalidate_range with
mmu_notifier_invalidate_range_start;mmu_notifier_invalidate_range_end.
Either that or call multiple mmu_notifier_invalidate_page like
before.
2) address + (1UL << compound_order(page) is buggy, it should be
PAGE_SIZE << compound_order(page), it's bytes not pages, 2M not
512.
3) The whole invalidate_range thing was an attempt to call a single
invalidate while walking multiple 4k ptes that maps the same THP
(after a pmd virtual split without physical compound page THP
split).
It's unclear if the rmap_walk will always provide an address that
is 2M aligned as parameter to try_to_unmap_one, in presence of THP.
I think it needs also an address &= (PAGE_SIZE <<
compound_order(page)) - 1 to be safe"
In general, we should stop making excuses for horrible MMU notifier
users. It's much more important that the core VM is sane and safe, than
letting MMU notifiers sleep.
So if some MMU notifier is sleeping under a spinlock, we need to fix the
notifier, not try to make excuses for that garbage in the core VM.
Reported-and-tested-by: Bernhard Held <berny156@gmx.de>
Reported-and-tested-by: Adam Borowski <kilobyte@angband.pl>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Radim Krčmář <rkrcmar@redhat.com>
Cc: Wanpeng Li <kernellwp@gmail.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Takashi Iwai <tiwai@suse.de>
Cc: Nadav Amit <nadav.amit@gmail.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Jérôme Glisse <jglisse@redhat.com>
Cc: axie <axie@amd.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-08-29 09:11:06 -07:00
|
|
|
|
2017-02-24 14:57:57 -08:00
|
|
|
|
|
|
|
|
|
2022-01-20 18:20:07 -05:00
|
|
|
|
2017-02-24 14:57:57 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-11-15 17:34:07 -08:00
|
|
|
|
2022-04-28 23:16:10 -07:00
|
|
|
|
2006-12-22 14:25:52 +01:00
|
|
|
|
2006-09-25 23:30:57 -07:00
|
|
|
|
2018-12-28 00:38:09 -08:00
|
|
|
|
2017-08-31 17:17:27 -04:00
|
|
|
|
2022-04-28 23:16:10 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-05-03 14:54:27 -07:00
|
|
|
|
2006-09-25 23:30:57 -07:00
|
|
|
|
|
|
|
|
|
2014-01-21 15:49:55 -08:00
|
|
|
|
2006-09-25 23:30:57 -07:00
|
|
|
|
2014-01-21 15:49:55 -08:00
|
|
|
|
2014-01-23 15:53:41 -08:00
|
|
|
|
2006-09-25 23:30:57 -07:00
|
|
|
|
2014-01-23 15:53:41 -08:00
|
|
|
|
2006-09-25 23:30:57 -07:00
|
|
|
|
|
|
|
|
|
2021-02-28 16:02:57 -05:00
|
|
|
|
2006-09-25 23:30:57 -07:00
|
|
|
|
2014-01-21 15:49:55 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2006-09-25 23:30:57 -07:00
|
|
|
|
2021-02-28 16:02:57 -05:00
|
|
|
|
2006-09-25 23:30:57 -07:00
|
|
|
|
2021-02-28 16:02:57 -05:00
|
|
|
|
2014-01-21 15:49:55 -08:00
|
|
|
|
2006-09-25 23:30:57 -07:00
|
|
|
|
2021-02-28 16:02:57 -05:00
|
|
|
|
2014-01-21 15:49:55 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-01-29 16:06:53 -05:00
|
|
|
|
2014-01-21 15:49:55 -08:00
|
|
|
|
|
|
|
|
|
2006-09-25 23:30:57 -07:00
|
|
|
|
2021-02-28 16:02:57 -05:00
|
|
|
|
2006-09-25 23:30:57 -07:00
|
|
|
|
mm: provide mapping_wrprotect_range() function
In the fb_defio video driver, page dirty state is used to determine when
frame buffer pages have been changed, allowing for batched, deferred I/O
to be performed for efficiency.
This implementation had only one means of doing so effectively - the use
of the folio_mkclean() function.
However, this use of the function is inappropriate, as the fb_defio
implementation allocates kernel memory to back the framebuffer, and then
is forced to specified page->index, mapping fields in order to permit the
folio_mkclean() rmap traversal to proceed correctly.
It is not correct to specify these fields on kernel-allocated memory, and
moreover since these are not folios, page->index, mapping are deprecated
fields, soon to be removed.
We therefore need to provide a means by which we can correctly traverse
the reverse mapping and write-protect mappings for a page backing an
address_space page cache object at a given offset.
This patch provides this - mapping_wrprotect_range() - which allows for
this operation to be performed for a specified address_space, offset, PFN
and size, without requiring a folio nor, of course, an inappropriate use
of page->index, mapping.
With this provided, we can subsequently adjust the fb_defio implementation
to make use of this function and avoid incorrect invocation of
folio_mkclean() and more importantly, incorrect manipulation of
page->index and mapping fields.
Link: https://lkml.kernel.org/r/e5bf969d64e7f2f2ae944d42341fc8994b736a81.1739029358.git.lorenzo.stoakes@oracle.com
Signed-off-by: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Helge Deller <deller@gmx.de>
Cc: Jaya Kumar <jayakumar.lkml@gmail.com>
Cc: Kajtar Zsolt <soci@c64.rulez.org>
Cc: Maíra Canal <mcanal@igalia.com>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Simona Vetter <simona.vetter@ffwll.ch>
Cc: Thomas Zimemrmann <tzimmermann@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2025-02-08 15:52:55 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-04-28 23:16:10 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-03-28 22:58:29 +00:00
|
|
|
|
2022-04-28 23:16:10 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-08-04 06:41:06 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-11-10 15:20:08 -08:00
|
|
|
|
2025-08-04 06:41:06 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-11-10 15:20:08 -08:00
|
|
|
|
2025-08-04 06:41:06 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-03-03 17:30:01 +01:00
|
|
|
|
2025-08-11 13:26:27 +02:00
|
|
|
|
2023-12-20 23:44:37 +01:00
|
|
|
|
|
|
|
|
|
mm: track mapcount of large folios in single value
Let's track the mapcount of large folios in a single value. The mapcount
of a large folio currently corresponds to the sum of the entire mapcount
and all page mapcounts.
This sum is what we actually want to know in folio_mapcount() and it is
also sufficient for implementing folio_mapped().
With PTE-mapped THP becoming more important and more widely used, we want
to avoid looping over all pages of a folio just to obtain the mapcount of
large folios. The comment "In the common case, avoid the loop when no
pages mapped by PTE" in folio_total_mapcount() does no longer hold for
mTHP that are always mapped by PTE.
Further, we are planning on using folio_mapcount() more frequently, and
might even want to remove page mapcounts for large folios in some kernel
configs. Therefore, allow for reading the mapcount of large folios
efficiently and atomically without looping over any pages.
Maintain the mapcount also for hugetlb pages for simplicity. Use the new
mapcount to implement folio_mapcount() and folio_mapped(). Make
page_mapped() simply call folio_mapped(). We can now get rid of
folio_large_is_mapped().
_nr_pages_mapped is now only used in rmap code and for debugging purposes.
Keep folio_nr_pages_mapped() around, but document that its use should be
limited to rmap internals and debugging purposes.
This change implies one additional atomic add/sub whenever
mapping/unmapping (parts of) a large folio.
As we now batch RMAP operations for PTE-mapped THP during fork(), during
unmap/zap, and when PTE-remapping a PMD-mapped THP, and we adjust the
large mapcount for a PTE batch only once, the added overhead in the common
case is small. Only when unmapping individual pages of a large folio
(e.g., during COW), the overhead might be bigger in comparison, but it's
essentially one additional atomic operation.
Note that before the new mapcount would overflow, already our refcount
would overflow: each mapping requires a folio reference. Extend the
focumentation of folio_mapcount().
Link: https://lkml.kernel.org/r/20240409192301.907377-5-david@redhat.com
Signed-off-by: David Hildenbrand <david@redhat.com>
Reviewed-by: Yin Fengwei <fengwei.yin@intel.com>
Cc: Chris Zankel <chris@zankel.net>
Cc: Hugh Dickins <hughd@google.com>
Cc: John Paul Adrian Glaubitz <glaubitz@physik.fu-berlin.de>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Matthew Wilcox (Oracle) <willy@infradead.org>
Cc: Max Filippov <jcmvbkbc@gmail.com>
Cc: Miaohe Lin <linmiaohe@huawei.com>
Cc: Muchun Song <muchun.song@linux.dev>
Cc: Naoya Horiguchi <nao.horiguchi@gmail.com>
Cc: Peter Xu <peterx@redhat.com>
Cc: Richard Chang <richardycc@google.com>
Cc: Rich Felker <dalias@libc.org>
Cc: Ryan Roberts <ryan.roberts@arm.com>
Cc: Yang Shi <shy828301@gmail.com>
Cc: Yoshinori Sato <ysato@users.sourceforge.jp>
Cc: Zi Yan <ziy@nvidia.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2024-04-09 21:22:47 +02:00
|
|
|
|
2025-08-04 06:41:06 +00:00
|
|
|
|
2023-12-20 23:44:37 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-08-11 13:26:27 +02:00
|
|
|
|
2024-04-09 21:22:46 +02:00
|
|
|
|
2024-08-16 12:32:46 +02:00
|
|
|
|
2024-04-09 21:22:46 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
mm: stop maintaining the per-page mapcount of large folios (CONFIG_NO_PAGE_MAPCOUNT)
Everything is in place to stop using the per-page mapcounts in large
folios: the mapcount of tail pages will always be logically 0 (-1 value),
just like it currently is for hugetlb folios already, and the page
mapcount of the head page is either 0 (-1 value) or contains a page type
(e.g., hugetlb).
Maintaining _nr_pages_mapped without per-page mapcounts is impossible, so
that one also has to go with CONFIG_NO_PAGE_MAPCOUNT.
There are two remaining implications:
(1) Per-node, per-cgroup and per-lruvec stats of "NR_ANON_MAPPED"
("mapped anonymous memory") and "NR_FILE_MAPPED"
("mapped file memory"):
As soon as any page of the folio is mapped -- folio_mapped() -- we
now account the complete folio as mapped. Once the last page is
unmapped -- !folio_mapped() -- we account the complete folio as
unmapped.
This implies that ...
* "AnonPages" and "Mapped" in /proc/meminfo and
/sys/devices/system/node/*/meminfo
* cgroup v2: "anon" and "file_mapped" in "memory.stat" and
"memory.numa_stat"
* cgroup v1: "rss" and "mapped_file" in "memory.stat" and
"memory.numa_stat
... can now appear higher than before. But note that these folios do
consume that memory, simply not all pages are actually currently
mapped.
It's worth nothing that other accounting in the kernel (esp. cgroup
charging on allocation) is not affected by this change.
[why oh why is "anon" called "rss" in cgroup v1]
(2) Detecting partial mappings
Detecting whether anon THPs are partially mapped gets a bit more
unreliable. As long as a single MM maps such a large folio
("exclusively mapped"), we can reliably detect it. Especially before
fork() / after a short-lived child process quit, we will detect
partial mappings reliably, which is the common case.
In essence, if the average per-page mapcount in an anon THP is < 1,
we know for sure that we have a partial mapping.
However, as soon as multiple MMs are involved, we might miss detecting
partial mappings: this might be relevant with long-lived child
processes. If we have a fully-mapped anon folio before fork(), once
our child processes and our parent all unmap (zap/COW) the same pages
(but not the complete folio), we might not detect the partial mapping.
However, once the child processes quit we would detect the partial
mapping.
How relevant this case is in practice remains to be seen.
Swapout/migration will likely mitigate this.
In the future, RMAP walkers could check for that for that case
(e.g., when collecting access bits during reclaim) and simply flag
them for deferred-splitting.
Link: https://lkml.kernel.org/r/20250303163014.1128035-21-david@redhat.com
Signed-off-by: David Hildenbrand <david@redhat.com>
Cc: Andy Lutomirks^H^Hski <luto@kernel.org>
Cc: Borislav Betkov <bp@alien8.de>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Jann Horn <jannh@google.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Lance Yang <ioworker0@gmail.com>
Cc: Liam Howlett <liam.howlett@oracle.com>
Cc: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: Matthew Wilcow (Oracle) <willy@infradead.org>
Cc: Michal Koutn <mkoutny@suse.com>
Cc: Muchun Song <muchun.song@linux.dev>
Cc: tejun heo <tj@kernel.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Zefan Li <lizefan.x@bytedance.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2025-03-03 17:30:13 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-12-20 23:44:37 +01:00
|
|
|
|
2024-08-07 13:55:15 +02:00
|
|
|
|
2023-12-20 23:44:37 +01:00
|
|
|
|
2024-08-07 13:55:15 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-03-03 17:30:02 +01:00
|
|
|
|
2023-12-20 23:44:37 +01:00
|
|
|
|
2025-08-11 13:26:27 +02:00
|
|
|
|
|
|
|
|
|
2023-12-20 23:44:37 +01:00
|
|
|
|
mm: stop maintaining the per-page mapcount of large folios (CONFIG_NO_PAGE_MAPCOUNT)
Everything is in place to stop using the per-page mapcounts in large
folios: the mapcount of tail pages will always be logically 0 (-1 value),
just like it currently is for hugetlb folios already, and the page
mapcount of the head page is either 0 (-1 value) or contains a page type
(e.g., hugetlb).
Maintaining _nr_pages_mapped without per-page mapcounts is impossible, so
that one also has to go with CONFIG_NO_PAGE_MAPCOUNT.
There are two remaining implications:
(1) Per-node, per-cgroup and per-lruvec stats of "NR_ANON_MAPPED"
("mapped anonymous memory") and "NR_FILE_MAPPED"
("mapped file memory"):
As soon as any page of the folio is mapped -- folio_mapped() -- we
now account the complete folio as mapped. Once the last page is
unmapped -- !folio_mapped() -- we account the complete folio as
unmapped.
This implies that ...
* "AnonPages" and "Mapped" in /proc/meminfo and
/sys/devices/system/node/*/meminfo
* cgroup v2: "anon" and "file_mapped" in "memory.stat" and
"memory.numa_stat"
* cgroup v1: "rss" and "mapped_file" in "memory.stat" and
"memory.numa_stat
... can now appear higher than before. But note that these folios do
consume that memory, simply not all pages are actually currently
mapped.
It's worth nothing that other accounting in the kernel (esp. cgroup
charging on allocation) is not affected by this change.
[why oh why is "anon" called "rss" in cgroup v1]
(2) Detecting partial mappings
Detecting whether anon THPs are partially mapped gets a bit more
unreliable. As long as a single MM maps such a large folio
("exclusively mapped"), we can reliably detect it. Especially before
fork() / after a short-lived child process quit, we will detect
partial mappings reliably, which is the common case.
In essence, if the average per-page mapcount in an anon THP is < 1,
we know for sure that we have a partial mapping.
However, as soon as multiple MMs are involved, we might miss detecting
partial mappings: this might be relevant with long-lived child
processes. If we have a fully-mapped anon folio before fork(), once
our child processes and our parent all unmap (zap/COW) the same pages
(but not the complete folio), we might not detect the partial mapping.
However, once the child processes quit we would detect the partial
mapping.
How relevant this case is in practice remains to be seen.
Swapout/migration will likely mitigate this.
In the future, RMAP walkers could check for that for that case
(e.g., when collecting access bits during reclaim) and simply flag
them for deferred-splitting.
Link: https://lkml.kernel.org/r/20250303163014.1128035-21-david@redhat.com
Signed-off-by: David Hildenbrand <david@redhat.com>
Cc: Andy Lutomirks^H^Hski <luto@kernel.org>
Cc: Borislav Betkov <bp@alien8.de>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Jann Horn <jannh@google.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Lance Yang <ioworker0@gmail.com>
Cc: Liam Howlett <liam.howlett@oracle.com>
Cc: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: Matthew Wilcow (Oracle) <willy@infradead.org>
Cc: Michal Koutn <mkoutny@suse.com>
Cc: Muchun Song <muchun.song@linux.dev>
Cc: tejun heo <tj@kernel.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Zefan Li <lizefan.x@bytedance.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2025-03-03 17:30:13 +01:00
|
|
|
|
2025-08-11 13:26:27 +02:00
|
|
|
|
2025-08-04 06:41:06 +00:00
|
|
|
|
mm: stop maintaining the per-page mapcount of large folios (CONFIG_NO_PAGE_MAPCOUNT)
Everything is in place to stop using the per-page mapcounts in large
folios: the mapcount of tail pages will always be logically 0 (-1 value),
just like it currently is for hugetlb folios already, and the page
mapcount of the head page is either 0 (-1 value) or contains a page type
(e.g., hugetlb).
Maintaining _nr_pages_mapped without per-page mapcounts is impossible, so
that one also has to go with CONFIG_NO_PAGE_MAPCOUNT.
There are two remaining implications:
(1) Per-node, per-cgroup and per-lruvec stats of "NR_ANON_MAPPED"
("mapped anonymous memory") and "NR_FILE_MAPPED"
("mapped file memory"):
As soon as any page of the folio is mapped -- folio_mapped() -- we
now account the complete folio as mapped. Once the last page is
unmapped -- !folio_mapped() -- we account the complete folio as
unmapped.
This implies that ...
* "AnonPages" and "Mapped" in /proc/meminfo and
/sys/devices/system/node/*/meminfo
* cgroup v2: "anon" and "file_mapped" in "memory.stat" and
"memory.numa_stat"
* cgroup v1: "rss" and "mapped_file" in "memory.stat" and
"memory.numa_stat
... can now appear higher than before. But note that these folios do
consume that memory, simply not all pages are actually currently
mapped.
It's worth nothing that other accounting in the kernel (esp. cgroup
charging on allocation) is not affected by this change.
[why oh why is "anon" called "rss" in cgroup v1]
(2) Detecting partial mappings
Detecting whether anon THPs are partially mapped gets a bit more
unreliable. As long as a single MM maps such a large folio
("exclusively mapped"), we can reliably detect it. Especially before
fork() / after a short-lived child process quit, we will detect
partial mappings reliably, which is the common case.
In essence, if the average per-page mapcount in an anon THP is < 1,
we know for sure that we have a partial mapping.
However, as soon as multiple MMs are involved, we might miss detecting
partial mappings: this might be relevant with long-lived child
processes. If we have a fully-mapped anon folio before fork(), once
our child processes and our parent all unmap (zap/COW) the same pages
(but not the complete folio), we might not detect the partial mapping.
However, once the child processes quit we would detect the partial
mapping.
How relevant this case is in practice remains to be seen.
Swapout/migration will likely mitigate this.
In the future, RMAP walkers could check for that for that case
(e.g., when collecting access bits during reclaim) and simply flag
them for deferred-splitting.
Link: https://lkml.kernel.org/r/20250303163014.1128035-21-david@redhat.com
Signed-off-by: David Hildenbrand <david@redhat.com>
Cc: Andy Lutomirks^H^Hski <luto@kernel.org>
Cc: Borislav Betkov <bp@alien8.de>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Jann Horn <jannh@google.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Lance Yang <ioworker0@gmail.com>
Cc: Liam Howlett <liam.howlett@oracle.com>
Cc: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: Matthew Wilcow (Oracle) <willy@infradead.org>
Cc: Michal Koutn <mkoutny@suse.com>
Cc: Muchun Song <muchun.song@linux.dev>
Cc: tejun heo <tj@kernel.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Zefan Li <lizefan.x@bytedance.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2025-03-03 17:30:13 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-12-20 23:44:37 +01:00
|
|
|
|
2023-12-20 23:45:03 +01:00
|
|
|
|
|
|
|
|
|
2025-03-03 17:30:04 +01:00
|
|
|
|
2025-02-28 14:31:09 +11:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-08-11 13:26:27 +02:00
|
|
|
|
2025-08-04 06:41:06 +00:00
|
|
|
|
2025-02-28 14:31:09 +11:00
|
|
|
|
2023-12-20 23:44:37 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-12-20 23:45:03 +01:00
|
|
|
|
2023-12-20 23:44:37 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-03-03 17:30:02 +01:00
|
|
|
|
2023-12-20 23:44:37 +01:00
|
|
|
|
2025-08-11 13:26:27 +02:00
|
|
|
|
|
|
|
|
|
2023-12-20 23:44:37 +01:00
|
|
|
|
2025-08-04 06:41:06 +00:00
|
|
|
|
2023-12-20 23:44:37 +01:00
|
|
|
|
|
|
|
|
|
2010-03-05 13:42:09 -08:00
|
|
|
|
2023-10-02 16:29:48 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2010-03-05 13:42:09 -08:00
|
|
|
|
2023-10-02 16:29:48 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2010-03-05 13:42:09 -08:00
|
|
|
|
2023-10-02 16:29:48 +02:00
|
|
|
|
2010-03-05 13:42:09 -08:00
|
|
|
|
2022-09-02 20:46:45 +01:00
|
|
|
|
2010-03-05 13:42:09 -08:00
|
|
|
|
2022-09-02 20:46:45 +01:00
|
|
|
|
2014-10-09 15:28:10 -07:00
|
|
|
|
2010-03-05 13:42:09 -08:00
|
|
|
|
2025-07-04 12:25:20 +02:00
|
|
|
|
2015-06-24 16:56:56 -07:00
|
|
|
|
2025-07-04 12:25:20 +02:00
|
|
|
|
2022-01-21 11:27:31 -05:00
|
|
|
|
|
|
|
|
|
2015-06-24 16:56:56 -07:00
|
|
|
|
2022-09-02 20:46:45 +01:00
|
|
|
|
2010-03-05 13:42:09 -08:00
|
|
|
|
|
|
|
|
|
2006-01-06 00:11:12 -08:00
|
|
|
|
2023-09-13 14:51:09 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
mm: ptep_get() conversion
Convert all instances of direct pte_t* dereferencing to instead use
ptep_get() helper. This means that by default, the accesses change from a
C dereference to a READ_ONCE(). This is technically the correct thing to
do since where pgtables are modified by HW (for access/dirty) they are
volatile and therefore we should always ensure READ_ONCE() semantics.
But more importantly, by always using the helper, it can be overridden by
the architecture to fully encapsulate the contents of the pte. Arch code
is deliberately not converted, as the arch code knows best. It is
intended that arch code (arm64) will override the default with its own
implementation that can (e.g.) hide certain bits from the core code, or
determine young/dirty status by mixing in state from another source.
Conversion was done using Coccinelle:
----
// $ make coccicheck \
// COCCI=ptepget.cocci \
// SPFLAGS="--include-headers" \
// MODE=patch
virtual patch
@ depends on patch @
pte_t *v;
@@
- *v
+ ptep_get(v)
----
Then reviewed and hand-edited to avoid multiple unnecessary calls to
ptep_get(), instead opting to store the result of a single call in a
variable, where it is correct to do so. This aims to negate any cost of
READ_ONCE() and will benefit arch-overrides that may be more complex.
Included is a fix for an issue in an earlier version of this patch that
was pointed out by kernel test robot. The issue arose because config
MMU=n elides definition of the ptep helper functions, including
ptep_get(). HUGETLB_PAGE=n configs still define a simple
huge_ptep_clear_flush() for linking purposes, which dereferences the ptep.
So when both configs are disabled, this caused a build error because
ptep_get() is not defined. Fix by continuing to do a direct dereference
when MMU=n. This is safe because for this config the arch code cannot be
trying to virtualize the ptes because none of the ptep helpers are
defined.
Link: https://lkml.kernel.org/r/20230612151545.3317766-4-ryan.roberts@arm.com
Reported-by: kernel test robot <lkp@intel.com>
Link: https://lore.kernel.org/oe-kbuild-all/202305120142.yXsNEo6H-lkp@intel.com/
Signed-off-by: Ryan Roberts <ryan.roberts@arm.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Alexander Potapenko <glider@google.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Alex Williamson <alex.williamson@redhat.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Andrey Konovalov <andreyknvl@gmail.com>
Cc: Andrey Ryabinin <ryabinin.a.a@gmail.com>
Cc: Christian Brauner <brauner@kernel.org>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: Daniel Vetter <daniel@ffwll.ch>
Cc: Dave Airlie <airlied@gmail.com>
Cc: Dimitri Sivanich <dimitri.sivanich@hpe.com>
Cc: Dmitry Vyukov <dvyukov@google.com>
Cc: Ian Rogers <irogers@google.com>
Cc: Jason Gunthorpe <jgg@ziepe.ca>
Cc: Jérôme Glisse <jglisse@redhat.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Lorenzo Stoakes <lstoakes@gmail.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Miaohe Lin <linmiaohe@huawei.com>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Mike Rapoport (IBM) <rppt@kernel.org>
Cc: Muchun Song <muchun.song@linux.dev>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Naoya Horiguchi <naoya.horiguchi@nec.com>
Cc: Oleksandr Tyshchenko <oleksandr_tyshchenko@epam.com>
Cc: Pavel Tatashin <pasha.tatashin@soleen.com>
Cc: Roman Gushchin <roman.gushchin@linux.dev>
Cc: SeongJae Park <sj@kernel.org>
Cc: Shakeel Butt <shakeelb@google.com>
Cc: Uladzislau Rezki (Sony) <urezki@gmail.com>
Cc: Vincenzo Frascino <vincenzo.frascino@arm.com>
Cc: Yu Zhao <yuzhao@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2023-06-12 16:15:45 +01:00
|
|
|
|
2023-09-13 14:51:09 +02:00
|
|
|
|
2006-01-06 00:11:12 -08:00
|
|
|
|
2023-09-13 14:51:09 +02:00
|
|
|
|
|
|
|
|
|
2006-01-06 00:11:12 -08:00
|
|
|
|
2010-04-14 17:59:28 -04:00
|
|
|
|
anonvma: when setting up page->mapping, we need to pick the _oldest_ anonvma
Otherwise we might be mapping in a page in a new mapping, but that page
(through the swapcache) would later be mapped into an old mapping too.
The page->mapping must be the case that works for everybody, not just
the mapping that happened to page it in first.
Here's the scenario:
- page gets allocated/mapped by process A. Let's call the anon_vma we
associate the page with 'A' to keep it easy to track.
- Process A forks, creating process B. The anon_vma in B is 'B', and has
a chain that looks like 'B' -> 'A'. Everything is fine.
- Swapping happens. The page (with mapping pointing to 'A') gets swapped
out (perhaps not to disk - it's enough to assume that it's just not
mapped any more, and lives entirely in the swap-cache)
- Process B pages it in, which goes like this:
do_swap_page ->
page = lookup_swap_cache(entry);
...
set_pte_at(mm, address, page_table, pte);
page_add_anon_rmap(page, vma, address);
And think about what happens here!
In particular, what happens is that this will now be the "first"
mapping of that page, so page_add_anon_rmap() used to do
if (first)
__page_set_anon_rmap(page, vma, address);
and notice what anon_vma it will use? It will use the anon_vma for
process B!
What happens then? Trivial: process 'A' also pages it in (nothing
happens, it's not the first mapping), and then process 'B' execve's
or exits or unmaps, making anon_vma B go away.
End result: process A has a page that points to anon_vma B, but
anon_vma B does not exist any more. This can go on forever. Forget
about RCU grace periods, forget about locking, forget anything like
that. The bug is simply that page->mapping points to an anon_vma
that was correct at one point, but was _not_ the one that was shared
by all users of that possible mapping.
Changing it to always use the deepest anon_vma in the anonvma chain gets
us to the safest model.
This can be improved in certain cases: if we know the page is private to
just this particular mapping (for example, it's a new page, or it is the
only swapcache entry), we could pick the top (most specific) anon_vma.
But that's a future optimization. Make it _work_ reliably first.
Reviewed-by: Rik van Riel <riel@redhat.com>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Tested-by: Borislav Petkov <bp@alien8.de> [ "What do you know, I think you fixed it!" ]
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-04-12 12:44:29 -07:00
|
|
|
|
2010-04-14 17:59:28 -04:00
|
|
|
|
anonvma: when setting up page->mapping, we need to pick the _oldest_ anonvma
Otherwise we might be mapping in a page in a new mapping, but that page
(through the swapcache) would later be mapped into an old mapping too.
The page->mapping must be the case that works for everybody, not just
the mapping that happened to page it in first.
Here's the scenario:
- page gets allocated/mapped by process A. Let's call the anon_vma we
associate the page with 'A' to keep it easy to track.
- Process A forks, creating process B. The anon_vma in B is 'B', and has
a chain that looks like 'B' -> 'A'. Everything is fine.
- Swapping happens. The page (with mapping pointing to 'A') gets swapped
out (perhaps not to disk - it's enough to assume that it's just not
mapped any more, and lives entirely in the swap-cache)
- Process B pages it in, which goes like this:
do_swap_page ->
page = lookup_swap_cache(entry);
...
set_pte_at(mm, address, page_table, pte);
page_add_anon_rmap(page, vma, address);
And think about what happens here!
In particular, what happens is that this will now be the "first"
mapping of that page, so page_add_anon_rmap() used to do
if (first)
__page_set_anon_rmap(page, vma, address);
and notice what anon_vma it will use? It will use the anon_vma for
process B!
What happens then? Trivial: process 'A' also pages it in (nothing
happens, it's not the first mapping), and then process 'B' execve's
or exits or unmaps, making anon_vma B go away.
End result: process A has a page that points to anon_vma B, but
anon_vma B does not exist any more. This can go on forever. Forget
about RCU grace periods, forget about locking, forget anything like
that. The bug is simply that page->mapping points to an anon_vma
that was correct at one point, but was _not_ the one that was shared
by all users of that possible mapping.
Changing it to always use the deepest anon_vma in the anonvma chain gets
us to the safest model.
This can be improved in certain cases: if we know the page is private to
just this particular mapping (for example, it's a new page, or it is the
only swapcache entry), we could pick the top (most specific) anon_vma.
But that's a future optimization. Make it _work_ reliably first.
Reviewed-by: Rik van Riel <riel@redhat.com>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Tested-by: Borislav Petkov <bp@alien8.de> [ "What do you know, I think you fixed it!" ]
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-04-12 12:44:29 -07:00
|
|
|
|
|
|
|
|
|
2023-09-13 14:51:09 +02:00
|
|
|
|
|
|
|
|
|
anonvma: when setting up page->mapping, we need to pick the _oldest_ anonvma
Otherwise we might be mapping in a page in a new mapping, but that page
(through the swapcache) would later be mapped into an old mapping too.
The page->mapping must be the case that works for everybody, not just
the mapping that happened to page it in first.
Here's the scenario:
- page gets allocated/mapped by process A. Let's call the anon_vma we
associate the page with 'A' to keep it easy to track.
- Process A forks, creating process B. The anon_vma in B is 'B', and has
a chain that looks like 'B' -> 'A'. Everything is fine.
- Swapping happens. The page (with mapping pointing to 'A') gets swapped
out (perhaps not to disk - it's enough to assume that it's just not
mapped any more, and lives entirely in the swap-cache)
- Process B pages it in, which goes like this:
do_swap_page ->
page = lookup_swap_cache(entry);
...
set_pte_at(mm, address, page_table, pte);
page_add_anon_rmap(page, vma, address);
And think about what happens here!
In particular, what happens is that this will now be the "first"
mapping of that page, so page_add_anon_rmap() used to do
if (first)
__page_set_anon_rmap(page, vma, address);
and notice what anon_vma it will use? It will use the anon_vma for
process B!
What happens then? Trivial: process 'A' also pages it in (nothing
happens, it's not the first mapping), and then process 'B' execve's
or exits or unmaps, making anon_vma B go away.
End result: process A has a page that points to anon_vma B, but
anon_vma B does not exist any more. This can go on forever. Forget
about RCU grace periods, forget about locking, forget anything like
that. The bug is simply that page->mapping points to an anon_vma
that was correct at one point, but was _not_ the one that was shared
by all users of that possible mapping.
Changing it to always use the deepest anon_vma in the anonvma chain gets
us to the safest model.
This can be improved in certain cases: if we know the page is private to
just this particular mapping (for example, it's a new page, or it is the
only swapcache entry), we could pick the top (most specific) anon_vma.
But that's a future optimization. Make it _work_ reliably first.
Reviewed-by: Rik van Riel <riel@redhat.com>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Tested-by: Borislav Petkov <bp@alien8.de> [ "What do you know, I think you fixed it!" ]
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-04-12 12:44:29 -07:00
|
|
|
|
2010-09-22 12:43:56 +02:00
|
|
|
|
2010-08-09 17:19:09 -07:00
|
|
|
|
2006-01-06 00:11:12 -08:00
|
|
|
|
2020-12-15 12:33:42 -08:00
|
|
|
|
2023-01-16 19:29:59 +00:00
|
|
|
|
2020-12-15 12:33:42 -08:00
|
|
|
|
2025-07-04 12:25:20 +02:00
|
|
|
|
2020-12-15 12:33:42 -08:00
|
|
|
|
|
|
|
|
|
2025-07-04 12:25:20 +02:00
|
|
|
|
2023-01-16 19:29:59 +00:00
|
|
|
|
|
|
|
|
|
2006-01-06 00:11:12 -08:00
|
|
|
|
|
|
|
|
|
2007-05-16 22:11:21 -07:00
|
|
|
|
2008-03-19 17:00:43 -07:00
|
|
|
|
2023-07-06 20:52:51 +01:00
|
|
|
|
|
|
|
|
|
2007-05-16 22:11:21 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-10-05 21:01:15 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2007-05-16 22:11:21 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-12-20 23:44:45 +01:00
|
|
|
|
2021-02-25 17:17:59 -08:00
|
|
|
|
2007-05-16 22:11:21 -07:00
|
|
|
|
2023-12-11 16:22:11 +00:00
|
|
|
|
2007-05-16 22:11:21 -07:00
|
|
|
|
2023-12-11 16:22:11 +00:00
|
|
|
|
2007-05-16 22:11:21 -07:00
|
|
|
|
2022-01-29 11:52:52 -05:00
|
|
|
|
|
|
|
|
|
2024-10-05 21:01:12 +01:00
|
|
|
|
2019-11-30 17:51:26 -08:00
|
|
|
|
2007-05-16 22:11:21 -07:00
|
|
|
|
|
|
|
|
|
2023-12-20 23:44:38 +01:00
|
|
|
|
|
|
|
|
|
2025-08-11 13:26:27 +02:00
|
|
|
|
2023-12-20 23:44:38 +01:00
|
|
|
|
2025-08-04 06:41:06 +00:00
|
|
|
|
mm,thp,rmap: simplify compound page mapcount handling
Compound page (folio) mapcount calculations have been different for anon
and file (or shmem) THPs, and involved the obscure PageDoubleMap flag.
And each huge mapping and unmapping of a file (or shmem) THP involved
atomically incrementing and decrementing the mapcount of every subpage of
that huge page, dirtying many struct page cachelines.
Add subpages_mapcount field to the struct folio and first tail page, so
that the total of subpage mapcounts is available in one place near the
head: then page_mapcount() and total_mapcount() and page_mapped(), and
their folio equivalents, are so quick that anon and file and hugetlb don't
need to be optimized differently. Delete the unloved PageDoubleMap.
page_add and page_remove rmap functions must now maintain the
subpages_mapcount as well as the subpage _mapcount, when dealing with pte
mappings of huge pages; and correct maintenance of NR_ANON_MAPPED and
NR_FILE_MAPPED statistics still needs reading through the subpages, using
nr_subpages_unmapped() - but only when first or last pmd mapping finds
subpages_mapcount raised (double-map case, not the common case).
But are those counts (used to decide when to split an anon THP, and in
vmscan's pagecache_reclaimable heuristic) correctly maintained? Not
quite: since page_remove_rmap() (and also split_huge_pmd()) is often
called without page lock, there can be races when a subpage pte mapcount
0<->1 while compound pmd mapcount 0<->1 is scanning - races which the
previous implementation had prevented. The statistics might become
inaccurate, and even drift down until they underflow through 0. That is
not good enough, but is better dealt with in a followup patch.
Update a few comments on first and second tail page overlaid fields.
hugepage_add_new_anon_rmap() has to "increment" compound_mapcount, but
subpages_mapcount and compound_pincount are already correctly at 0, so
delete its reinitialization of compound_pincount.
A simple 100 X munmap(mmap(2GB, MAP_SHARED|MAP_POPULATE, tmpfs), 2GB) took
18 seconds on small pages, and used to take 1 second on huge pages, but
now takes 119 milliseconds on huge pages. Mapping by pmds a second time
used to take 860ms and now takes 92ms; mapping by pmds after mapping by
ptes (when the scan is needed) used to take 870ms and now takes 495ms.
But there might be some benchmarks which would show a slowdown, because
tail struct pages now fall out of cache until final freeing checks them.
Link: https://lkml.kernel.org/r/47ad693-717-79c8-e1ba-46c3a6602e48@google.com
Signed-off-by: Hugh Dickins <hughd@google.com>
Acked-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: James Houghton <jthoughton@google.com>
Cc: John Hubbard <jhubbard@nvidia.com>
Cc: Matthew Wilcox (Oracle) <willy@infradead.org>
Cc: Miaohe Lin <linmiaohe@huawei.com>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Mina Almasry <almasrymina@google.com>
Cc: Muchun Song <songmuchun@bytedance.com>
Cc: Naoya Horiguchi <naoya.horiguchi@linux.dev>
Cc: Peter Xu <peterx@redhat.com>
Cc: Sidhartha Kumar <sidhartha.kumar@oracle.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Yang Shi <shy828301@gmail.com>
Cc: Zach O'Keefe <zokeefe@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2022-11-02 18:51:38 -07:00
|
|
|
|
2024-06-18 11:11:37 +12:00
|
|
|
|
|
|
|
|
|
2025-08-04 06:41:06 +00:00
|
|
|
|
ksm: let shared pages be swappable
Initial implementation for swapping out KSM's shared pages: add
page_referenced_ksm() and try_to_unmap_ksm(), which rmap.c calls when
faced with a PageKsm page.
Most of what's needed can be got from the rmap_items listed from the
stable_node of the ksm page, without discovering the actual vma: so in
this patch just fake up a struct vma for page_referenced_one() or
try_to_unmap_one(), then refine that in the next patch.
Add VM_NONLINEAR to ksm_madvise()'s list of exclusions: it has always been
implicit there (being only set with VM_SHARED, already excluded), but
let's make it explicit, to help justify the lack of nonlinear unmap.
Rely on the page lock to protect against concurrent modifications to that
page's node of the stable tree.
The awkward part is not swapout but swapin: do_swap_page() and
page_add_anon_rmap() now have to allow for new possibilities - perhaps a
ksm page still in swapcache, perhaps a swapcache page associated with one
location in one anon_vma now needed for another location or anon_vma.
(And the vma might even be no longer VM_MERGEABLE when that happens.)
ksm_might_need_to_copy() checks for that case, and supplies a duplicate
page when necessary, simply leaving it to a subsequent pass of ksmd to
rediscover the identity and merge them back into one ksm page.
Disappointingly primitive: but the alternative would have to accumulate
unswappable info about the swapped out ksm pages, limiting swappability.
Remove page_add_ksm_rmap(): page_add_anon_rmap() now has to allow for the
particular case it was handling, so just use it instead.
Signed-off-by: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Cc: Izik Eidus <ieidus@redhat.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Chris Wright <chrisw@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-12-14 17:59:24 -08:00
|
|
|
|
2024-06-18 11:11:37 +12:00
|
|
|
|
2023-09-13 14:51:10 +02:00
|
|
|
|
2023-12-20 23:44:38 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-08-11 13:26:27 +02:00
|
|
|
|
2023-12-20 23:44:38 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-08-11 13:26:27 +02:00
|
|
|
|
2023-12-20 23:44:38 +01:00
|
|
|
|
|
|
|
|
|
2025-08-11 13:26:27 +02:00
|
|
|
|
2025-02-28 14:31:09 +11:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-08-11 13:26:27 +02:00
|
|
|
|
|
|
|
|
|
2023-12-20 23:44:38 +01:00
|
|
|
|
|
|
|
|
|
mm: stop maintaining the per-page mapcount of large folios (CONFIG_NO_PAGE_MAPCOUNT)
Everything is in place to stop using the per-page mapcounts in large
folios: the mapcount of tail pages will always be logically 0 (-1 value),
just like it currently is for hugetlb folios already, and the page
mapcount of the head page is either 0 (-1 value) or contains a page type
(e.g., hugetlb).
Maintaining _nr_pages_mapped without per-page mapcounts is impossible, so
that one also has to go with CONFIG_NO_PAGE_MAPCOUNT.
There are two remaining implications:
(1) Per-node, per-cgroup and per-lruvec stats of "NR_ANON_MAPPED"
("mapped anonymous memory") and "NR_FILE_MAPPED"
("mapped file memory"):
As soon as any page of the folio is mapped -- folio_mapped() -- we
now account the complete folio as mapped. Once the last page is
unmapped -- !folio_mapped() -- we account the complete folio as
unmapped.
This implies that ...
* "AnonPages" and "Mapped" in /proc/meminfo and
/sys/devices/system/node/*/meminfo
* cgroup v2: "anon" and "file_mapped" in "memory.stat" and
"memory.numa_stat"
* cgroup v1: "rss" and "mapped_file" in "memory.stat" and
"memory.numa_stat
... can now appear higher than before. But note that these folios do
consume that memory, simply not all pages are actually currently
mapped.
It's worth nothing that other accounting in the kernel (esp. cgroup
charging on allocation) is not affected by this change.
[why oh why is "anon" called "rss" in cgroup v1]
(2) Detecting partial mappings
Detecting whether anon THPs are partially mapped gets a bit more
unreliable. As long as a single MM maps such a large folio
("exclusively mapped"), we can reliably detect it. Especially before
fork() / after a short-lived child process quit, we will detect
partial mappings reliably, which is the common case.
In essence, if the average per-page mapcount in an anon THP is < 1,
we know for sure that we have a partial mapping.
However, as soon as multiple MMs are involved, we might miss detecting
partial mappings: this might be relevant with long-lived child
processes. If we have a fully-mapped anon folio before fork(), once
our child processes and our parent all unmap (zap/COW) the same pages
(but not the complete folio), we might not detect the partial mapping.
However, once the child processes quit we would detect the partial
mapping.
How relevant this case is in practice remains to be seen.
Swapout/migration will likely mitigate this.
In the future, RMAP walkers could check for that for that case
(e.g., when collecting access bits during reclaim) and simply flag
them for deferred-splitting.
Link: https://lkml.kernel.org/r/20250303163014.1128035-21-david@redhat.com
Signed-off-by: David Hildenbrand <david@redhat.com>
Cc: Andy Lutomirks^H^Hski <luto@kernel.org>
Cc: Borislav Betkov <bp@alien8.de>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Jann Horn <jannh@google.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Lance Yang <ioworker0@gmail.com>
Cc: Liam Howlett <liam.howlett@oracle.com>
Cc: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: Matthew Wilcow (Oracle) <willy@infradead.org>
Cc: Michal Koutn <mkoutny@suse.com>
Cc: Muchun Song <muchun.song@linux.dev>
Cc: tejun heo <tj@kernel.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Zefan Li <lizefan.x@bytedance.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2025-03-03 17:30:13 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-12-20 23:44:38 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
mm: stop maintaining the per-page mapcount of large folios (CONFIG_NO_PAGE_MAPCOUNT)
Everything is in place to stop using the per-page mapcounts in large
folios: the mapcount of tail pages will always be logically 0 (-1 value),
just like it currently is for hugetlb folios already, and the page
mapcount of the head page is either 0 (-1 value) or contains a page type
(e.g., hugetlb).
Maintaining _nr_pages_mapped without per-page mapcounts is impossible, so
that one also has to go with CONFIG_NO_PAGE_MAPCOUNT.
There are two remaining implications:
(1) Per-node, per-cgroup and per-lruvec stats of "NR_ANON_MAPPED"
("mapped anonymous memory") and "NR_FILE_MAPPED"
("mapped file memory"):
As soon as any page of the folio is mapped -- folio_mapped() -- we
now account the complete folio as mapped. Once the last page is
unmapped -- !folio_mapped() -- we account the complete folio as
unmapped.
This implies that ...
* "AnonPages" and "Mapped" in /proc/meminfo and
/sys/devices/system/node/*/meminfo
* cgroup v2: "anon" and "file_mapped" in "memory.stat" and
"memory.numa_stat"
* cgroup v1: "rss" and "mapped_file" in "memory.stat" and
"memory.numa_stat
... can now appear higher than before. But note that these folios do
consume that memory, simply not all pages are actually currently
mapped.
It's worth nothing that other accounting in the kernel (esp. cgroup
charging on allocation) is not affected by this change.
[why oh why is "anon" called "rss" in cgroup v1]
(2) Detecting partial mappings
Detecting whether anon THPs are partially mapped gets a bit more
unreliable. As long as a single MM maps such a large folio
("exclusively mapped"), we can reliably detect it. Especially before
fork() / after a short-lived child process quit, we will detect
partial mappings reliably, which is the common case.
In essence, if the average per-page mapcount in an anon THP is < 1,
we know for sure that we have a partial mapping.
However, as soon as multiple MMs are involved, we might miss detecting
partial mappings: this might be relevant with long-lived child
processes. If we have a fully-mapped anon folio before fork(), once
our child processes and our parent all unmap (zap/COW) the same pages
(but not the complete folio), we might not detect the partial mapping.
However, once the child processes quit we would detect the partial
mapping.
How relevant this case is in practice remains to be seen.
Swapout/migration will likely mitigate this.
In the future, RMAP walkers could check for that for that case
(e.g., when collecting access bits during reclaim) and simply flag
them for deferred-splitting.
Link: https://lkml.kernel.org/r/20250303163014.1128035-21-david@redhat.com
Signed-off-by: David Hildenbrand <david@redhat.com>
Cc: Andy Lutomirks^H^Hski <luto@kernel.org>
Cc: Borislav Betkov <bp@alien8.de>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Jann Horn <jannh@google.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Lance Yang <ioworker0@gmail.com>
Cc: Liam Howlett <liam.howlett@oracle.com>
Cc: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: Matthew Wilcow (Oracle) <willy@infradead.org>
Cc: Michal Koutn <mkoutny@suse.com>
Cc: Muchun Song <muchun.song@linux.dev>
Cc: tejun heo <tj@kernel.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Zefan Li <lizefan.x@bytedance.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2025-03-03 17:30:13 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-12-20 23:44:38 +01:00
|
|
|
|
|
|
|
|
|
mm/munlock: rmap call mlock_vma_page() munlock_vma_page()
Add vma argument to mlock_vma_page() and munlock_vma_page(), make them
inline functions which check (vma->vm_flags & VM_LOCKED) before calling
mlock_page() and munlock_page() in mm/mlock.c.
Add bool compound to mlock_vma_page() and munlock_vma_page(): this is
because we have understandable difficulty in accounting pte maps of THPs,
and if passed a PageHead page, mlock_page() and munlock_page() cannot
tell whether it's a pmd map to be counted or a pte map to be ignored.
Add vma arg to page_add_file_rmap() and page_remove_rmap(), like the
others, and use that to call mlock_vma_page() at the end of the page
adds, and munlock_vma_page() at the end of page_remove_rmap() (end or
beginning? unimportant, but end was easier for assertions in testing).
No page lock is required (although almost all adds happen to hold it):
delete the "Serialize with page migration" BUG_ON(!PageLocked(page))s.
Certainly page lock did serialize with page migration, but I'm having
difficulty explaining why that was ever important.
Mlock accounting on THPs has been hard to define, differed between anon
and file, involved PageDoubleMap in some places and not others, required
clear_page_mlock() at some points. Keep it simple now: just count the
pmds and ignore the ptes, there is no reason for ptes to undo pmd mlocks.
page_add_new_anon_rmap() callers unchanged: they have long been calling
lru_cache_add_inactive_or_unevictable(), which does its own VM_LOCKED
handling (it also checks for not VM_SPECIAL: I think that's overcautious,
and inconsistent with other checks, that mmap_region() already prevents
VM_LOCKED on VM_SPECIAL; but haven't quite convinced myself to change it).
Signed-off-by: Hugh Dickins <hughd@google.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
2022-02-14 18:26:39 -08:00
|
|
|
|
2023-09-18 15:33:17 +08:00
|
|
|
|
2025-09-23 12:07:11 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-09-18 15:33:17 +08:00
|
|
|
|
2025-09-23 12:07:11 +01:00
|
|
|
|
2023-09-18 15:33:17 +08:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
2023-12-20 23:44:38 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-08-11 13:26:27 +02:00
|
|
|
|
2023-12-20 23:44:38 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-08-11 13:26:27 +02:00
|
|
|
|
2023-12-20 23:44:38 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2008-03-19 17:00:43 -07:00
|
|
|
|
2023-01-11 14:28:54 +00:00
|
|
|
|
|
|
|
|
|
2006-01-06 00:11:12 -08:00
|
|
|
|
|
|
|
|
|
2024-06-18 11:11:35 +12:00
|
|
|
|
2022-05-09 18:20:43 -07:00
|
|
|
|
2023-12-20 23:44:45 +01:00
|
|
|
|
2006-01-06 00:11:12 -08:00
|
|
|
|
2024-06-18 11:11:35 +12:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-01-11 14:28:54 +00:00
|
|
|
|
2024-06-18 11:11:35 +12:00
|
|
|
|
2006-01-06 00:11:12 -08:00
|
|
|
|
2023-01-11 14:28:54 +00:00
|
|
|
|
2024-06-18 11:11:35 +12:00
|
|
|
|
2006-01-06 00:11:12 -08:00
|
|
|
|
2024-06-18 11:11:35 +12:00
|
|
|
|
2025-03-03 17:30:04 +01:00
|
|
|
|
2016-01-15 16:52:16 -08:00
|
|
|
|
2023-12-20 23:44:30 +01:00
|
|
|
|
2024-06-18 11:11:35 +12:00
|
|
|
|
mm: use folio_add_new_anon_rmap() if folio_test_anon(folio)==false
For the !folio_test_anon(folio) case, we can now invoke
folio_add_new_anon_rmap() with the rmap flags set to either EXCLUSIVE or
non-EXCLUSIVE. This action will suppress the VM_WARN_ON_FOLIO check
within __folio_add_anon_rmap() while initiating the process of bringing up
mTHP swapin.
static __always_inline void __folio_add_anon_rmap(struct folio *folio,
struct page *page, int nr_pages, struct vm_area_struct *vma,
unsigned long address, rmap_t flags, enum rmap_level level)
{
...
if (unlikely(!folio_test_anon(folio))) {
VM_WARN_ON_FOLIO(folio_test_large(folio) &&
level != RMAP_LEVEL_PMD, folio);
}
...
}
It also improves the code's readability. Currently, all new anonymous
folios calling folio_add_anon_rmap_ptes() are order-0. This ensures that
new folios cannot be partially exclusive; they are either entirely
exclusive or entirely shared.
A useful comment from Hugh's fix:
: Commit "mm: use folio_add_new_anon_rmap() if folio_test_anon(folio)==
: false" has extended folio_add_new_anon_rmap() to use on non-exclusive
: folios, already visible to others in swap cache and on LRU.
:
: That renders its non-atomic __folio_set_swapbacked() unsafe: it risks
: overwriting concurrent atomic operations on folio->flags, losing bits
: added or restoring bits cleared. Since it's only used in this risky way
: when folio_test_locked and !folio_test_anon, many such races are excluded;
: but, for example, isolations by folio_test_clear_lru() are vulnerable, and
: setting or clearing active.
:
: It could just use the atomic folio_set_swapbacked(); but this function
: does try to avoid atomics where it can, so use a branch instead: just
: avoid setting swapbacked when it is already set, that is good enough.
: (Swapbacked is normally stable once set: lazyfree can undo it, but only
: later, when found anon in a page table.)
:
: This fixes a lot of instability under compaction and swapping loads:
: assorted "Bad page"s, VM_BUG_ON_FOLIO()s, apparently even page double
: frees - though I've not worked out what races could lead to the latter.
[akpm@linux-foundation.org: comment fixes, per David and akpm]
[v-songbaohua@oppo.com: lock the folio to avoid race]
Link: https://lkml.kernel.org/r/20240622032002.53033-1-21cnbao@gmail.com
[hughd@google.com: folio_add_new_anon_rmap() careful __folio_set_swapbacked()]
Link: https://lkml.kernel.org/r/f3599b1d-8323-0dc5-e9e0-fdb3cfc3dd5a@google.com
Link: https://lkml.kernel.org/r/20240617231137.80726-3-21cnbao@gmail.com
Signed-off-by: Barry Song <v-songbaohua@oppo.com>
Signed-off-by: Hugh Dickins <hughd@google.com>
Suggested-by: David Hildenbrand <david@redhat.com>
Tested-by: Shuai Yuan <yuanshuai@oppo.com>
Acked-by: David Hildenbrand <david@redhat.com>
Cc: Baolin Wang <baolin.wang@linux.alibaba.com>
Cc: Chris Li <chrisl@kernel.org>
Cc: "Huang, Ying" <ying.huang@intel.com>
Cc: Matthew Wilcox (Oracle) <willy@infradead.org>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Ryan Roberts <ryan.roberts@arm.com>
Cc: Suren Baghdasaryan <surenb@google.com>
Cc: Yang Shi <shy828301@gmail.com>
Cc: Yosry Ahmed <yosryahmed@google.com>
Cc: Yu Zhao <yuzhao@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2024-06-18 11:11:36 +12:00
|
|
|
|
mm: add MAP_DROPPABLE for designating always lazily freeable mappings
The vDSO getrandom() implementation works with a buffer allocated with a
new system call that has certain requirements:
- It shouldn't be written to core dumps.
* Easy: VM_DONTDUMP.
- It should be zeroed on fork.
* Easy: VM_WIPEONFORK.
- It shouldn't be written to swap.
* Uh-oh: mlock is rlimited.
* Uh-oh: mlock isn't inherited by forks.
- It shouldn't reserve actual memory, but it also shouldn't crash when
page faulting in memory if none is available
* Uh-oh: VM_NORESERVE means segfaults.
It turns out that the vDSO getrandom() function has three really nice
characteristics that we can exploit to solve this problem:
1) Due to being wiped during fork(), the vDSO code is already robust to
having the contents of the pages it reads zeroed out midway through
the function's execution.
2) In the absolute worst case of whatever contingency we're coding for,
we have the option to fallback to the getrandom() syscall, and
everything is fine.
3) The buffers the function uses are only ever useful for a maximum of
60 seconds -- a sort of cache, rather than a long term allocation.
These characteristics mean that we can introduce VM_DROPPABLE, which
has the following semantics:
a) It never is written out to swap.
b) Under memory pressure, mm can just drop the pages (so that they're
zero when read back again).
c) It is inherited by fork.
d) It doesn't count against the mlock budget, since nothing is locked.
e) If there's not enough memory to service a page fault, it's not fatal,
and no signal is sent.
This way, allocations used by vDSO getrandom() can use:
VM_DROPPABLE | VM_DONTDUMP | VM_WIPEONFORK | VM_NORESERVE
And there will be no problem with OOMing, crashing on overcommitment,
using memory when not in use, not wiping on fork(), coredumps, or
writing out to swap.
In order to let vDSO getrandom() use this, expose these via mmap(2) as
MAP_DROPPABLE.
Note that this involves removing the MADV_FREE special case from
sort_folio(), which according to Yu Zhao is unnecessary and will simply
result in an extra call to shrink_folio_list() in the worst case. The
chunk removed reenables the swapbacked flag, which we don't want for
VM_DROPPABLE, and we can't conditionalize it here because there isn't a
vma reference available.
Finally, the provided self test ensures that this is working as desired.
Cc: linux-mm@kvack.org
Acked-by: David Hildenbrand <david@redhat.com>
Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
2022-12-08 17:55:04 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-07-24 10:29:50 -07:00
|
|
|
|
mm: use folio_add_new_anon_rmap() if folio_test_anon(folio)==false
For the !folio_test_anon(folio) case, we can now invoke
folio_add_new_anon_rmap() with the rmap flags set to either EXCLUSIVE or
non-EXCLUSIVE. This action will suppress the VM_WARN_ON_FOLIO check
within __folio_add_anon_rmap() while initiating the process of bringing up
mTHP swapin.
static __always_inline void __folio_add_anon_rmap(struct folio *folio,
struct page *page, int nr_pages, struct vm_area_struct *vma,
unsigned long address, rmap_t flags, enum rmap_level level)
{
...
if (unlikely(!folio_test_anon(folio))) {
VM_WARN_ON_FOLIO(folio_test_large(folio) &&
level != RMAP_LEVEL_PMD, folio);
}
...
}
It also improves the code's readability. Currently, all new anonymous
folios calling folio_add_anon_rmap_ptes() are order-0. This ensures that
new folios cannot be partially exclusive; they are either entirely
exclusive or entirely shared.
A useful comment from Hugh's fix:
: Commit "mm: use folio_add_new_anon_rmap() if folio_test_anon(folio)==
: false" has extended folio_add_new_anon_rmap() to use on non-exclusive
: folios, already visible to others in swap cache and on LRU.
:
: That renders its non-atomic __folio_set_swapbacked() unsafe: it risks
: overwriting concurrent atomic operations on folio->flags, losing bits
: added or restoring bits cleared. Since it's only used in this risky way
: when folio_test_locked and !folio_test_anon, many such races are excluded;
: but, for example, isolations by folio_test_clear_lru() are vulnerable, and
: setting or clearing active.
:
: It could just use the atomic folio_set_swapbacked(); but this function
: does try to avoid atomics where it can, so use a branch instead: just
: avoid setting swapbacked when it is already set, that is good enough.
: (Swapbacked is normally stable once set: lazyfree can undo it, but only
: later, when found anon in a page table.)
:
: This fixes a lot of instability under compaction and swapping loads:
: assorted "Bad page"s, VM_BUG_ON_FOLIO()s, apparently even page double
: frees - though I've not worked out what races could lead to the latter.
[akpm@linux-foundation.org: comment fixes, per David and akpm]
[v-songbaohua@oppo.com: lock the folio to avoid race]
Link: https://lkml.kernel.org/r/20240622032002.53033-1-21cnbao@gmail.com
[hughd@google.com: folio_add_new_anon_rmap() careful __folio_set_swapbacked()]
Link: https://lkml.kernel.org/r/f3599b1d-8323-0dc5-e9e0-fdb3cfc3dd5a@google.com
Link: https://lkml.kernel.org/r/20240617231137.80726-3-21cnbao@gmail.com
Signed-off-by: Barry Song <v-songbaohua@oppo.com>
Signed-off-by: Hugh Dickins <hughd@google.com>
Suggested-by: David Hildenbrand <david@redhat.com>
Tested-by: Shuai Yuan <yuanshuai@oppo.com>
Acked-by: David Hildenbrand <david@redhat.com>
Cc: Baolin Wang <baolin.wang@linux.alibaba.com>
Cc: Chris Li <chrisl@kernel.org>
Cc: "Huang, Ying" <ying.huang@intel.com>
Cc: Matthew Wilcox (Oracle) <willy@infradead.org>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Ryan Roberts <ryan.roberts@arm.com>
Cc: Suren Baghdasaryan <surenb@google.com>
Cc: Yang Shi <shy828301@gmail.com>
Cc: Yosry Ahmed <yosryahmed@google.com>
Cc: Yu Zhao <yuzhao@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2024-06-18 11:11:36 +12:00
|
|
|
|
2024-06-18 11:11:35 +12:00
|
|
|
|
2022-11-09 18:18:49 -08:00
|
|
|
|
2023-12-07 16:12:03 +00:00
|
|
|
|
2022-11-09 18:18:49 -08:00
|
|
|
|
2023-01-11 14:28:54 +00:00
|
|
|
|
2024-06-18 11:11:35 +12:00
|
|
|
|
|
|
|
|
|
2023-12-07 16:12:03 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-03-03 17:30:04 +01:00
|
|
|
|
2023-12-07 16:12:03 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
mm: stop maintaining the per-page mapcount of large folios (CONFIG_NO_PAGE_MAPCOUNT)
Everything is in place to stop using the per-page mapcounts in large
folios: the mapcount of tail pages will always be logically 0 (-1 value),
just like it currently is for hugetlb folios already, and the page
mapcount of the head page is either 0 (-1 value) or contains a page type
(e.g., hugetlb).
Maintaining _nr_pages_mapped without per-page mapcounts is impossible, so
that one also has to go with CONFIG_NO_PAGE_MAPCOUNT.
There are two remaining implications:
(1) Per-node, per-cgroup and per-lruvec stats of "NR_ANON_MAPPED"
("mapped anonymous memory") and "NR_FILE_MAPPED"
("mapped file memory"):
As soon as any page of the folio is mapped -- folio_mapped() -- we
now account the complete folio as mapped. Once the last page is
unmapped -- !folio_mapped() -- we account the complete folio as
unmapped.
This implies that ...
* "AnonPages" and "Mapped" in /proc/meminfo and
/sys/devices/system/node/*/meminfo
* cgroup v2: "anon" and "file_mapped" in "memory.stat" and
"memory.numa_stat"
* cgroup v1: "rss" and "mapped_file" in "memory.stat" and
"memory.numa_stat
... can now appear higher than before. But note that these folios do
consume that memory, simply not all pages are actually currently
mapped.
It's worth nothing that other accounting in the kernel (esp. cgroup
charging on allocation) is not affected by this change.
[why oh why is "anon" called "rss" in cgroup v1]
(2) Detecting partial mappings
Detecting whether anon THPs are partially mapped gets a bit more
unreliable. As long as a single MM maps such a large folio
("exclusively mapped"), we can reliably detect it. Especially before
fork() / after a short-lived child process quit, we will detect
partial mappings reliably, which is the common case.
In essence, if the average per-page mapcount in an anon THP is < 1,
we know for sure that we have a partial mapping.
However, as soon as multiple MMs are involved, we might miss detecting
partial mappings: this might be relevant with long-lived child
processes. If we have a fully-mapped anon folio before fork(), once
our child processes and our parent all unmap (zap/COW) the same pages
(but not the complete folio), we might not detect the partial mapping.
However, once the child processes quit we would detect the partial
mapping.
How relevant this case is in practice remains to be seen.
Swapout/migration will likely mitigate this.
In the future, RMAP walkers could check for that for that case
(e.g., when collecting access bits during reclaim) and simply flag
them for deferred-splitting.
Link: https://lkml.kernel.org/r/20250303163014.1128035-21-david@redhat.com
Signed-off-by: David Hildenbrand <david@redhat.com>
Cc: Andy Lutomirks^H^Hski <luto@kernel.org>
Cc: Borislav Betkov <bp@alien8.de>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Jann Horn <jannh@google.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Lance Yang <ioworker0@gmail.com>
Cc: Liam Howlett <liam.howlett@oracle.com>
Cc: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: Matthew Wilcow (Oracle) <willy@infradead.org>
Cc: Michal Koutn <mkoutny@suse.com>
Cc: Muchun Song <muchun.song@linux.dev>
Cc: tejun heo <tj@kernel.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Zefan Li <lizefan.x@bytedance.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2025-03-03 17:30:13 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-06-18 11:11:35 +12:00
|
|
|
|
|
|
|
|
|
2023-12-07 16:12:03 +00:00
|
|
|
|
|
|
|
|
|
2025-03-03 17:30:02 +01:00
|
|
|
|
mm: stop maintaining the per-page mapcount of large folios (CONFIG_NO_PAGE_MAPCOUNT)
Everything is in place to stop using the per-page mapcounts in large
folios: the mapcount of tail pages will always be logically 0 (-1 value),
just like it currently is for hugetlb folios already, and the page
mapcount of the head page is either 0 (-1 value) or contains a page type
(e.g., hugetlb).
Maintaining _nr_pages_mapped without per-page mapcounts is impossible, so
that one also has to go with CONFIG_NO_PAGE_MAPCOUNT.
There are two remaining implications:
(1) Per-node, per-cgroup and per-lruvec stats of "NR_ANON_MAPPED"
("mapped anonymous memory") and "NR_FILE_MAPPED"
("mapped file memory"):
As soon as any page of the folio is mapped -- folio_mapped() -- we
now account the complete folio as mapped. Once the last page is
unmapped -- !folio_mapped() -- we account the complete folio as
unmapped.
This implies that ...
* "AnonPages" and "Mapped" in /proc/meminfo and
/sys/devices/system/node/*/meminfo
* cgroup v2: "anon" and "file_mapped" in "memory.stat" and
"memory.numa_stat"
* cgroup v1: "rss" and "mapped_file" in "memory.stat" and
"memory.numa_stat
... can now appear higher than before. But note that these folios do
consume that memory, simply not all pages are actually currently
mapped.
It's worth nothing that other accounting in the kernel (esp. cgroup
charging on allocation) is not affected by this change.
[why oh why is "anon" called "rss" in cgroup v1]
(2) Detecting partial mappings
Detecting whether anon THPs are partially mapped gets a bit more
unreliable. As long as a single MM maps such a large folio
("exclusively mapped"), we can reliably detect it. Especially before
fork() / after a short-lived child process quit, we will detect
partial mappings reliably, which is the common case.
In essence, if the average per-page mapcount in an anon THP is < 1,
we know for sure that we have a partial mapping.
However, as soon as multiple MMs are involved, we might miss detecting
partial mappings: this might be relevant with long-lived child
processes. If we have a fully-mapped anon folio before fork(), once
our child processes and our parent all unmap (zap/COW) the same pages
(but not the complete folio), we might not detect the partial mapping.
However, once the child processes quit we would detect the partial
mapping.
How relevant this case is in practice remains to be seen.
Swapout/migration will likely mitigate this.
In the future, RMAP walkers could check for that for that case
(e.g., when collecting access bits during reclaim) and simply flag
them for deferred-splitting.
Link: https://lkml.kernel.org/r/20250303163014.1128035-21-david@redhat.com
Signed-off-by: David Hildenbrand <david@redhat.com>
Cc: Andy Lutomirks^H^Hski <luto@kernel.org>
Cc: Borislav Betkov <bp@alien8.de>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Jann Horn <jannh@google.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Lance Yang <ioworker0@gmail.com>
Cc: Liam Howlett <liam.howlett@oracle.com>
Cc: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: Matthew Wilcow (Oracle) <willy@infradead.org>
Cc: Michal Koutn <mkoutny@suse.com>
Cc: Muchun Song <muchun.song@linux.dev>
Cc: tejun heo <tj@kernel.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Zefan Li <lizefan.x@bytedance.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2025-03-03 17:30:13 +01:00
|
|
|
|
|
|
|
|
|
2022-11-09 18:18:49 -08:00
|
|
|
|
2025-03-03 17:30:04 +01:00
|
|
|
|
2016-01-15 16:53:42 -08:00
|
|
|
|
2023-01-11 14:28:54 +00:00
|
|
|
|
2025-03-03 17:30:02 +01:00
|
|
|
|
mm: stop maintaining the per-page mapcount of large folios (CONFIG_NO_PAGE_MAPCOUNT)
Everything is in place to stop using the per-page mapcounts in large
folios: the mapcount of tail pages will always be logically 0 (-1 value),
just like it currently is for hugetlb folios already, and the page
mapcount of the head page is either 0 (-1 value) or contains a page type
(e.g., hugetlb).
Maintaining _nr_pages_mapped without per-page mapcounts is impossible, so
that one also has to go with CONFIG_NO_PAGE_MAPCOUNT.
There are two remaining implications:
(1) Per-node, per-cgroup and per-lruvec stats of "NR_ANON_MAPPED"
("mapped anonymous memory") and "NR_FILE_MAPPED"
("mapped file memory"):
As soon as any page of the folio is mapped -- folio_mapped() -- we
now account the complete folio as mapped. Once the last page is
unmapped -- !folio_mapped() -- we account the complete folio as
unmapped.
This implies that ...
* "AnonPages" and "Mapped" in /proc/meminfo and
/sys/devices/system/node/*/meminfo
* cgroup v2: "anon" and "file_mapped" in "memory.stat" and
"memory.numa_stat"
* cgroup v1: "rss" and "mapped_file" in "memory.stat" and
"memory.numa_stat
... can now appear higher than before. But note that these folios do
consume that memory, simply not all pages are actually currently
mapped.
It's worth nothing that other accounting in the kernel (esp. cgroup
charging on allocation) is not affected by this change.
[why oh why is "anon" called "rss" in cgroup v1]
(2) Detecting partial mappings
Detecting whether anon THPs are partially mapped gets a bit more
unreliable. As long as a single MM maps such a large folio
("exclusively mapped"), we can reliably detect it. Especially before
fork() / after a short-lived child process quit, we will detect
partial mappings reliably, which is the common case.
In essence, if the average per-page mapcount in an anon THP is < 1,
we know for sure that we have a partial mapping.
However, as soon as multiple MMs are involved, we might miss detecting
partial mappings: this might be relevant with long-lived child
processes. If we have a fully-mapped anon folio before fork(), once
our child processes and our parent all unmap (zap/COW) the same pages
(but not the complete folio), we might not detect the partial mapping.
However, once the child processes quit we would detect the partial
mapping.
How relevant this case is in practice remains to be seen.
Swapout/migration will likely mitigate this.
In the future, RMAP walkers could check for that for that case
(e.g., when collecting access bits during reclaim) and simply flag
them for deferred-splitting.
Link: https://lkml.kernel.org/r/20250303163014.1128035-21-david@redhat.com
Signed-off-by: David Hildenbrand <david@redhat.com>
Cc: Andy Lutomirks^H^Hski <luto@kernel.org>
Cc: Borislav Betkov <bp@alien8.de>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Jann Horn <jannh@google.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Lance Yang <ioworker0@gmail.com>
Cc: Liam Howlett <liam.howlett@oracle.com>
Cc: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: Matthew Wilcow (Oracle) <willy@infradead.org>
Cc: Michal Koutn <mkoutny@suse.com>
Cc: Muchun Song <muchun.song@linux.dev>
Cc: tejun heo <tj@kernel.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Zefan Li <lizefan.x@bytedance.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2025-03-03 17:30:13 +01:00
|
|
|
|
|
|
|
|
|
2024-06-18 11:11:35 +12:00
|
|
|
|
|
|
|
|
|
2024-05-06 21:13:33 +00:00
|
|
|
|
2016-01-15 16:52:16 -08:00
|
|
|
|
2022-11-09 18:18:49 -08:00
|
|
|
|
2025-03-03 17:30:04 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-05-06 21:13:33 +00:00
|
|
|
|
mm: count the number of anonymous THPs per size
Patch series "mm: count the number of anonymous THPs per size", v4.
Knowing the number of transparent anon THPs in the system is crucial
for performance analysis. It helps in understanding the ratio and
distribution of THPs versus small folios throughout the system.
Additionally, partial unmapping by userspace can lead to significant waste
of THPs over time and increase memory reclamation pressure. We need this
information for comprehensive system tuning.
This patch (of 2):
Let's track for each anonymous THP size, how many of them are currently
allocated. We'll track the complete lifespan of an anon THP, starting
when it becomes an anon THP ("large anon folio") (->mapping gets set),
until it gets freed (->mapping gets cleared).
Introduce a new "nr_anon" counter per THP size and adjust the
corresponding counter in the following cases:
* We allocate a new THP and call folio_add_new_anon_rmap() to map
it the first time and turn it into an anon THP.
* We split an anon THP into multiple smaller ones.
* We migrate an anon THP, when we prepare the destination.
* We free an anon THP back to the buddy.
Note that AnonPages in /proc/meminfo currently tracks the total number of
*mapped* anonymous *pages*, and therefore has slightly different
semantics. In the future, we might also want to track "nr_anon_mapped"
for each THP size, which might be helpful when comparing it to the number
of allocated anon THPs (long-term pinning, stuck in swapcache, memory
leaks, ...).
Further note that for now, we only track anon THPs after they got their
->mapping set, for example via folio_add_new_anon_rmap(). If we would
allocate some in the swapcache, they will only show up in the statistics
for now after they have been mapped to user space the first time, where we
call folio_add_new_anon_rmap().
[akpm@linux-foundation.org: documentation fixups, per David]
Link: https://lkml.kernel.org/r/3e8add35-e26b-443b-8a04-1078f4bc78f6@redhat.com
Link: https://lkml.kernel.org/r/20240824010441.21308-1-21cnbao@gmail.com
Link: https://lkml.kernel.org/r/20240824010441.21308-2-21cnbao@gmail.com
Signed-off-by: Barry Song <v-songbaohua@oppo.com>
Acked-by: David Hildenbrand <david@redhat.com>
Cc: Baolin Wang <baolin.wang@linux.alibaba.com>
Cc: Chris Li <chrisl@kernel.org>
Cc: Chuanhua Han <hanchuanhua@oppo.com>
Cc: Kairui Song <kasong@tencent.com>
Cc: Kalesh Singh <kaleshsingh@google.com>
Cc: Lance Yang <ioworker0@gmail.com>
Cc: Ryan Roberts <ryan.roberts@arm.com>
Cc: Shuai Yuan <yuanshuai@oppo.com>
Cc: Usama Arif <usamaarif642@gmail.com>
Cc: Zi Yan <ziy@nvidia.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2024-08-24 13:04:40 +12:00
|
|
|
|
2006-01-06 00:11:12 -08:00
|
|
|
|
|
|
|
|
|
2023-12-20 23:44:31 +01:00
|
|
|
|
|
|
|
|
|
2025-08-11 13:26:27 +02:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2023-12-20 23:44:31 +01:00
|
|
|
|
mm,thp,rmap: lock_compound_mapcounts() on THP mapcounts
Fix the races in maintaining compound_mapcount, subpages_mapcount and
subpage _mapcount by using PG_locked in the first tail of any compound
page for a bit_spin_lock() on such modifications; skipping the usual
atomic operations on those fields in this case.
Bring page_remove_file_rmap() and page_remove_anon_compound_rmap() back
into page_remove_rmap() itself. Rearrange page_add_anon_rmap() and
page_add_file_rmap() and page_remove_rmap() to follow the same "if
(compound) {lock} else if (PageCompound) {lock} else {atomic}" pattern
(with a PageTransHuge in the compound test, like before, to avoid BUG_ONs
and optimize away that block when THP is not configured). Move all the
stats updates outside, after the bit_spin_locked section, so that it is
sure to be a leaf lock.
Add page_dup_compound_rmap() to manage compound locking versus atomics in
sync with the rest. In particular, hugetlb pages are still using the
atomics: to avoid unnecessary interference there, and because they never
have subpage mappings; but this exception can easily be changed.
Conveniently, page_dup_compound_rmap() turns out to suit an anon THP's
__split_huge_pmd_locked() too.
bit_spin_lock() is not popular with PREEMPT_RT folks: but PREEMPT_RT
sensibly excludes TRANSPARENT_HUGEPAGE already, so its only exposure is to
the non-hugetlb non-THP pte-mapped compound pages (with large folios being
currently dependent on TRANSPARENT_HUGEPAGE). There is never any scan of
subpages in this case; but we have chosen to use PageCompound tests rather
than PageTransCompound tests to gate the use of lock_compound_mapcounts(),
so that page_mapped() is correct on all compound pages, whether or not
TRANSPARENT_HUGEPAGE is enabled: could that be a problem for PREEMPT_RT,
when there is contention on the lock - under heavy concurrent forking for
example? If so, then it can be turned into a sleeping lock (like
folio_lock()) when PREEMPT_RT.
A simple 100 X munmap(mmap(2GB, MAP_SHARED|MAP_POPULATE, tmpfs), 2GB) took
18 seconds on small pages, and used to take 1 second on huge pages, but
now takes 115 milliseconds on huge pages. Mapping by pmds a second time
used to take 860ms and now takes 86ms; mapping by pmds after mapping by
ptes (when the scan is needed) used to take 870ms and now takes 495ms.
Mapping huge pages by ptes is largely unaffected but variable: between 5%
faster and 5% slower in what I've recorded. Contention on the lock is
likely to behave worse than contention on the atomics behaved.
Link: https://lkml.kernel.org/r/1b42bd1a-8223-e827-602f-d466c2db7d3c@google.com
Signed-off-by: Hugh Dickins <hughd@google.com>
Acked-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: James Houghton <jthoughton@google.com>
Cc: John Hubbard <jhubbard@nvidia.com>
Cc: Matthew Wilcox (Oracle) <willy@infradead.org>
Cc: Miaohe Lin <linmiaohe@huawei.com>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Mina Almasry <almasrymina@google.com>
Cc: Muchun Song <songmuchun@bytedance.com>
Cc: Naoya Horiguchi <naoya.horiguchi@linux.dev>
Cc: Peter Xu <peterx@redhat.com>
Cc: Sidhartha Kumar <sidhartha.kumar@oracle.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Yang Shi <shy828301@gmail.com>
Cc: Zach O'Keefe <zokeefe@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2022-11-02 18:53:45 -07:00
|
|
|
|
2025-08-04 06:41:06 +00:00
|
|
|
|
mm/munlock: rmap call mlock_vma_page() munlock_vma_page()
Add vma argument to mlock_vma_page() and munlock_vma_page(), make them
inline functions which check (vma->vm_flags & VM_LOCKED) before calling
mlock_page() and munlock_page() in mm/mlock.c.
Add bool compound to mlock_vma_page() and munlock_vma_page(): this is
because we have understandable difficulty in accounting pte maps of THPs,
and if passed a PageHead page, mlock_page() and munlock_page() cannot
tell whether it's a pmd map to be counted or a pte map to be ignored.
Add vma arg to page_add_file_rmap() and page_remove_rmap(), like the
others, and use that to call mlock_vma_page() at the end of the page
adds, and munlock_vma_page() at the end of page_remove_rmap() (end or
beginning? unimportant, but end was easier for assertions in testing).
No page lock is required (although almost all adds happen to hold it):
delete the "Serialize with page migration" BUG_ON(!PageLocked(page))s.
Certainly page lock did serialize with page migration, but I'm having
difficulty explaining why that was ever important.
Mlock accounting on THPs has been hard to define, differed between anon
and file, involved PageDoubleMap in some places and not others, required
clear_page_mlock() at some points. Keep it simple now: just count the
pmds and ignore the ptes, there is no reason for ptes to undo pmd mlocks.
page_add_new_anon_rmap() callers unchanged: they have long been calling
lru_cache_add_inactive_or_unevictable(), which does its own VM_LOCKED
handling (it also checks for not VM_SPECIAL: I think that's overcautious,
and inconsistent with other checks, that mmap_region() already prevents
VM_LOCKED on VM_SPECIAL; but haven't quite convinced myself to change it).
Signed-off-by: Hugh Dickins <hughd@google.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
2022-02-14 18:26:39 -08:00
|
|
|
|
2025-09-23 12:07:11 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-09-18 15:33:17 +08:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
2023-12-20 23:44:31 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-08-11 13:26:27 +02:00
|
|
|
|
2023-12-20 23:44:31 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-08-11 13:26:27 +02:00
|
|
|
|
2023-12-20 23:44:31 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-02-28 14:31:09 +11:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-08-11 13:26:27 +02:00
|
|
|
|
2025-02-28 14:31:09 +11:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-12-20 23:44:47 +01:00
|
|
|
|
|
|
|
|
|
2025-08-11 13:26:27 +02:00
|
|
|
|
2023-12-20 23:44:47 +01:00
|
|
|
|
2023-01-11 14:28:51 +00:00
|
|
|
|
2024-08-07 13:55:15 +02:00
|
|
|
|
mm/rmap: do not add fully unmapped large folio to deferred split list
In __folio_remove_rmap(), a large folio is added to deferred split list if
any page in a folio loses its final mapping. But it is possible that the
folio is fully unmapped and adding it to deferred split list is
unnecessary.
For PMD-mapped THPs, that was not really an issue, because removing the
last PMD mapping in the absence of PTE mappings would not have added the
folio to the deferred split queue.
However, for PTE-mapped THPs, which are now more prominent due to mTHP,
they are always added to the deferred split queue. One side effect is
that the THP_DEFERRED_SPLIT_PAGE stat for a PTE-mapped folio can be
unintentionally increased, making it look like there are many partially
mapped folios -- although the whole folio is fully unmapped stepwise.
Core-mm now tries batch-unmapping consecutive PTEs of PTE-mapped THPs
where possible starting from commit b06dc281aa99 ("mm/rmap: introduce
folio_remove_rmap_[pte|ptes|pmd]()"). When it happens, a whole PTE-mapped
folio is unmapped in one go and can avoid being added to deferred split
list, reducing the THP_DEFERRED_SPLIT_PAGE noise. But there will still be
noise when we cannot batch-unmap a complete PTE-mapped folio in one go --
or where this type of batching is not implemented yet, e.g., migration.
To avoid the unnecessary addition, folio->_nr_pages_mapped is checked to
tell if the whole folio is unmapped. If the folio is already on deferred
split list, it will be skipped, too.
Note: commit 98046944a159 ("mm: huge_memory: add the missing
folio_test_pmd_mappable() for THP split statistics") tried to exclude mTHP
deferred split stats from THP_DEFERRED_SPLIT_PAGE, but it does not fix the
above issue. A fully unmapped PTE-mapped order-9 THP was still added to
deferred split list and counted as THP_DEFERRED_SPLIT_PAGE, since nr is
512 (non zero), level is RMAP_LEVEL_PTE, and inside deferred_split_folio()
the order-9 folio is folio_test_pmd_mappable().
Link: https://lkml.kernel.org/r/20240502132852.862138-1-zi.yan@sent.com
Signed-off-by: Zi Yan <ziy@nvidia.com>
Suggested-by: David Hildenbrand <david@redhat.com>
Reviewed-by: Yang Shi <shy828301@gmail.com>
Reviewed-by: David Hildenbrand <david@redhat.com>
Reviewed-by: Barry Song <baohua@kernel.org>
Reviewed-by: Lance Yang <ioworker0@gmail.com>
Cc: Alexander Gordeev <agordeev@linux.ibm.com>
Cc: Matthew Wilcox (Oracle) <willy@infradead.org>
Cc: Ryan Roberts <ryan.roberts@arm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2024-05-02 09:28:51 -04:00
|
|
|
|
mm,thp,rmap: lock_compound_mapcounts() on THP mapcounts
Fix the races in maintaining compound_mapcount, subpages_mapcount and
subpage _mapcount by using PG_locked in the first tail of any compound
page for a bit_spin_lock() on such modifications; skipping the usual
atomic operations on those fields in this case.
Bring page_remove_file_rmap() and page_remove_anon_compound_rmap() back
into page_remove_rmap() itself. Rearrange page_add_anon_rmap() and
page_add_file_rmap() and page_remove_rmap() to follow the same "if
(compound) {lock} else if (PageCompound) {lock} else {atomic}" pattern
(with a PageTransHuge in the compound test, like before, to avoid BUG_ONs
and optimize away that block when THP is not configured). Move all the
stats updates outside, after the bit_spin_locked section, so that it is
sure to be a leaf lock.
Add page_dup_compound_rmap() to manage compound locking versus atomics in
sync with the rest. In particular, hugetlb pages are still using the
atomics: to avoid unnecessary interference there, and because they never
have subpage mappings; but this exception can easily be changed.
Conveniently, page_dup_compound_rmap() turns out to suit an anon THP's
__split_huge_pmd_locked() too.
bit_spin_lock() is not popular with PREEMPT_RT folks: but PREEMPT_RT
sensibly excludes TRANSPARENT_HUGEPAGE already, so its only exposure is to
the non-hugetlb non-THP pte-mapped compound pages (with large folios being
currently dependent on TRANSPARENT_HUGEPAGE). There is never any scan of
subpages in this case; but we have chosen to use PageCompound tests rather
than PageTransCompound tests to gate the use of lock_compound_mapcounts(),
so that page_mapped() is correct on all compound pages, whether or not
TRANSPARENT_HUGEPAGE is enabled: could that be a problem for PREEMPT_RT,
when there is contention on the lock - under heavy concurrent forking for
example? If so, then it can be turned into a sleeping lock (like
folio_lock()) when PREEMPT_RT.
A simple 100 X munmap(mmap(2GB, MAP_SHARED|MAP_POPULATE, tmpfs), 2GB) took
18 seconds on small pages, and used to take 1 second on huge pages, but
now takes 115 milliseconds on huge pages. Mapping by pmds a second time
used to take 860ms and now takes 86ms; mapping by pmds after mapping by
ptes (when the scan is needed) used to take 870ms and now takes 495ms.
Mapping huge pages by ptes is largely unaffected but variable: between 5%
faster and 5% slower in what I've recorded. Contention on the lock is
likely to behave worse than contention on the atomics behaved.
Link: https://lkml.kernel.org/r/1b42bd1a-8223-e827-602f-d466c2db7d3c@google.com
Signed-off-by: Hugh Dickins <hughd@google.com>
Acked-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: James Houghton <jthoughton@google.com>
Cc: John Hubbard <jhubbard@nvidia.com>
Cc: Matthew Wilcox (Oracle) <willy@infradead.org>
Cc: Miaohe Lin <linmiaohe@huawei.com>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Mina Almasry <almasrymina@google.com>
Cc: Muchun Song <songmuchun@bytedance.com>
Cc: Naoya Horiguchi <naoya.horiguchi@linux.dev>
Cc: Peter Xu <peterx@redhat.com>
Cc: Sidhartha Kumar <sidhartha.kumar@oracle.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Yang Shi <shy828301@gmail.com>
Cc: Zach O'Keefe <zokeefe@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2022-11-02 18:53:45 -07:00
|
|
|
|
2023-12-20 23:44:47 +01:00
|
|
|
|
mm,thp,rmap: lock_compound_mapcounts() on THP mapcounts
Fix the races in maintaining compound_mapcount, subpages_mapcount and
subpage _mapcount by using PG_locked in the first tail of any compound
page for a bit_spin_lock() on such modifications; skipping the usual
atomic operations on those fields in this case.
Bring page_remove_file_rmap() and page_remove_anon_compound_rmap() back
into page_remove_rmap() itself. Rearrange page_add_anon_rmap() and
page_add_file_rmap() and page_remove_rmap() to follow the same "if
(compound) {lock} else if (PageCompound) {lock} else {atomic}" pattern
(with a PageTransHuge in the compound test, like before, to avoid BUG_ONs
and optimize away that block when THP is not configured). Move all the
stats updates outside, after the bit_spin_locked section, so that it is
sure to be a leaf lock.
Add page_dup_compound_rmap() to manage compound locking versus atomics in
sync with the rest. In particular, hugetlb pages are still using the
atomics: to avoid unnecessary interference there, and because they never
have subpage mappings; but this exception can easily be changed.
Conveniently, page_dup_compound_rmap() turns out to suit an anon THP's
__split_huge_pmd_locked() too.
bit_spin_lock() is not popular with PREEMPT_RT folks: but PREEMPT_RT
sensibly excludes TRANSPARENT_HUGEPAGE already, so its only exposure is to
the non-hugetlb non-THP pte-mapped compound pages (with large folios being
currently dependent on TRANSPARENT_HUGEPAGE). There is never any scan of
subpages in this case; but we have chosen to use PageCompound tests rather
than PageTransCompound tests to gate the use of lock_compound_mapcounts(),
so that page_mapped() is correct on all compound pages, whether or not
TRANSPARENT_HUGEPAGE is enabled: could that be a problem for PREEMPT_RT,
when there is contention on the lock - under heavy concurrent forking for
example? If so, then it can be turned into a sleeping lock (like
folio_lock()) when PREEMPT_RT.
A simple 100 X munmap(mmap(2GB, MAP_SHARED|MAP_POPULATE, tmpfs), 2GB) took
18 seconds on small pages, and used to take 1 second on huge pages, but
now takes 115 milliseconds on huge pages. Mapping by pmds a second time
used to take 860ms and now takes 86ms; mapping by pmds after mapping by
ptes (when the scan is needed) used to take 870ms and now takes 495ms.
Mapping huge pages by ptes is largely unaffected but variable: between 5%
faster and 5% slower in what I've recorded. Contention on the lock is
likely to behave worse than contention on the atomics behaved.
Link: https://lkml.kernel.org/r/1b42bd1a-8223-e827-602f-d466c2db7d3c@google.com
Signed-off-by: Hugh Dickins <hughd@google.com>
Acked-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: James Houghton <jthoughton@google.com>
Cc: John Hubbard <jhubbard@nvidia.com>
Cc: Matthew Wilcox (Oracle) <willy@infradead.org>
Cc: Miaohe Lin <linmiaohe@huawei.com>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Mina Almasry <almasrymina@google.com>
Cc: Muchun Song <songmuchun@bytedance.com>
Cc: Naoya Horiguchi <naoya.horiguchi@linux.dev>
Cc: Peter Xu <peterx@redhat.com>
Cc: Sidhartha Kumar <sidhartha.kumar@oracle.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Yang Shi <shy828301@gmail.com>
Cc: Zach O'Keefe <zokeefe@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2022-11-02 18:53:45 -07:00
|
|
|
|
2023-12-20 23:44:47 +01:00
|
|
|
|
2025-08-11 13:26:27 +02:00
|
|
|
|
2024-04-09 21:22:46 +02:00
|
|
|
|
2024-08-16 12:32:46 +02:00
|
|
|
|
2024-04-09 21:22:46 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
mm: stop maintaining the per-page mapcount of large folios (CONFIG_NO_PAGE_MAPCOUNT)
Everything is in place to stop using the per-page mapcounts in large
folios: the mapcount of tail pages will always be logically 0 (-1 value),
just like it currently is for hugetlb folios already, and the page
mapcount of the head page is either 0 (-1 value) or contains a page type
(e.g., hugetlb).
Maintaining _nr_pages_mapped without per-page mapcounts is impossible, so
that one also has to go with CONFIG_NO_PAGE_MAPCOUNT.
There are two remaining implications:
(1) Per-node, per-cgroup and per-lruvec stats of "NR_ANON_MAPPED"
("mapped anonymous memory") and "NR_FILE_MAPPED"
("mapped file memory"):
As soon as any page of the folio is mapped -- folio_mapped() -- we
now account the complete folio as mapped. Once the last page is
unmapped -- !folio_mapped() -- we account the complete folio as
unmapped.
This implies that ...
* "AnonPages" and "Mapped" in /proc/meminfo and
/sys/devices/system/node/*/meminfo
* cgroup v2: "anon" and "file_mapped" in "memory.stat" and
"memory.numa_stat"
* cgroup v1: "rss" and "mapped_file" in "memory.stat" and
"memory.numa_stat
... can now appear higher than before. But note that these folios do
consume that memory, simply not all pages are actually currently
mapped.
It's worth nothing that other accounting in the kernel (esp. cgroup
charging on allocation) is not affected by this change.
[why oh why is "anon" called "rss" in cgroup v1]
(2) Detecting partial mappings
Detecting whether anon THPs are partially mapped gets a bit more
unreliable. As long as a single MM maps such a large folio
("exclusively mapped"), we can reliably detect it. Especially before
fork() / after a short-lived child process quit, we will detect
partial mappings reliably, which is the common case.
In essence, if the average per-page mapcount in an anon THP is < 1,
we know for sure that we have a partial mapping.
However, as soon as multiple MMs are involved, we might miss detecting
partial mappings: this might be relevant with long-lived child
processes. If we have a fully-mapped anon folio before fork(), once
our child processes and our parent all unmap (zap/COW) the same pages
(but not the complete folio), we might not detect the partial mapping.
However, once the child processes quit we would detect the partial
mapping.
How relevant this case is in practice remains to be seen.
Swapout/migration will likely mitigate this.
In the future, RMAP walkers could check for that for that case
(e.g., when collecting access bits during reclaim) and simply flag
them for deferred-splitting.
Link: https://lkml.kernel.org/r/20250303163014.1128035-21-david@redhat.com
Signed-off-by: David Hildenbrand <david@redhat.com>
Cc: Andy Lutomirks^H^Hski <luto@kernel.org>
Cc: Borislav Betkov <bp@alien8.de>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Jann Horn <jannh@google.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Lance Yang <ioworker0@gmail.com>
Cc: Liam Howlett <liam.howlett@oracle.com>
Cc: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: Matthew Wilcow (Oracle) <willy@infradead.org>
Cc: Michal Koutn <mkoutny@suse.com>
Cc: Muchun Song <muchun.song@linux.dev>
Cc: tejun heo <tj@kernel.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Zefan Li <lizefan.x@bytedance.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2025-03-03 17:30:13 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-08-17 03:26:47 +00:00
|
|
|
|
mm: stop maintaining the per-page mapcount of large folios (CONFIG_NO_PAGE_MAPCOUNT)
Everything is in place to stop using the per-page mapcounts in large
folios: the mapcount of tail pages will always be logically 0 (-1 value),
just like it currently is for hugetlb folios already, and the page
mapcount of the head page is either 0 (-1 value) or contains a page type
(e.g., hugetlb).
Maintaining _nr_pages_mapped without per-page mapcounts is impossible, so
that one also has to go with CONFIG_NO_PAGE_MAPCOUNT.
There are two remaining implications:
(1) Per-node, per-cgroup and per-lruvec stats of "NR_ANON_MAPPED"
("mapped anonymous memory") and "NR_FILE_MAPPED"
("mapped file memory"):
As soon as any page of the folio is mapped -- folio_mapped() -- we
now account the complete folio as mapped. Once the last page is
unmapped -- !folio_mapped() -- we account the complete folio as
unmapped.
This implies that ...
* "AnonPages" and "Mapped" in /proc/meminfo and
/sys/devices/system/node/*/meminfo
* cgroup v2: "anon" and "file_mapped" in "memory.stat" and
"memory.numa_stat"
* cgroup v1: "rss" and "mapped_file" in "memory.stat" and
"memory.numa_stat
... can now appear higher than before. But note that these folios do
consume that memory, simply not all pages are actually currently
mapped.
It's worth nothing that other accounting in the kernel (esp. cgroup
charging on allocation) is not affected by this change.
[why oh why is "anon" called "rss" in cgroup v1]
(2) Detecting partial mappings
Detecting whether anon THPs are partially mapped gets a bit more
unreliable. As long as a single MM maps such a large folio
("exclusively mapped"), we can reliably detect it. Especially before
fork() / after a short-lived child process quit, we will detect
partial mappings reliably, which is the common case.
In essence, if the average per-page mapcount in an anon THP is < 1,
we know for sure that we have a partial mapping.
However, as soon as multiple MMs are involved, we might miss detecting
partial mappings: this might be relevant with long-lived child
processes. If we have a fully-mapped anon folio before fork(), once
our child processes and our parent all unmap (zap/COW) the same pages
(but not the complete folio), we might not detect the partial mapping.
However, once the child processes quit we would detect the partial
mapping.
How relevant this case is in practice remains to be seen.
Swapout/migration will likely mitigate this.
In the future, RMAP walkers could check for that for that case
(e.g., when collecting access bits during reclaim) and simply flag
them for deferred-splitting.
Link: https://lkml.kernel.org/r/20250303163014.1128035-21-david@redhat.com
Signed-off-by: David Hildenbrand <david@redhat.com>
Cc: Andy Lutomirks^H^Hski <luto@kernel.org>
Cc: Borislav Betkov <bp@alien8.de>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Jann Horn <jannh@google.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Lance Yang <ioworker0@gmail.com>
Cc: Liam Howlett <liam.howlett@oracle.com>
Cc: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: Matthew Wilcow (Oracle) <willy@infradead.org>
Cc: Michal Koutn <mkoutny@suse.com>
Cc: Muchun Song <muchun.song@linux.dev>
Cc: tejun heo <tj@kernel.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Zefan Li <lizefan.x@bytedance.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2025-03-03 17:30:13 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-03-03 17:30:02 +01:00
|
|
|
|
2023-12-20 23:44:47 +01:00
|
|
|
|
2024-08-07 13:55:15 +02:00
|
|
|
|
2023-12-20 23:44:47 +01:00
|
|
|
|
mm/rmap: do not add fully unmapped large folio to deferred split list
In __folio_remove_rmap(), a large folio is added to deferred split list if
any page in a folio loses its final mapping. But it is possible that the
folio is fully unmapped and adding it to deferred split list is
unnecessary.
For PMD-mapped THPs, that was not really an issue, because removing the
last PMD mapping in the absence of PTE mappings would not have added the
folio to the deferred split queue.
However, for PTE-mapped THPs, which are now more prominent due to mTHP,
they are always added to the deferred split queue. One side effect is
that the THP_DEFERRED_SPLIT_PAGE stat for a PTE-mapped folio can be
unintentionally increased, making it look like there are many partially
mapped folios -- although the whole folio is fully unmapped stepwise.
Core-mm now tries batch-unmapping consecutive PTEs of PTE-mapped THPs
where possible starting from commit b06dc281aa99 ("mm/rmap: introduce
folio_remove_rmap_[pte|ptes|pmd]()"). When it happens, a whole PTE-mapped
folio is unmapped in one go and can avoid being added to deferred split
list, reducing the THP_DEFERRED_SPLIT_PAGE noise. But there will still be
noise when we cannot batch-unmap a complete PTE-mapped folio in one go --
or where this type of batching is not implemented yet, e.g., migration.
To avoid the unnecessary addition, folio->_nr_pages_mapped is checked to
tell if the whole folio is unmapped. If the folio is already on deferred
split list, it will be skipped, too.
Note: commit 98046944a159 ("mm: huge_memory: add the missing
folio_test_pmd_mappable() for THP split statistics") tried to exclude mTHP
deferred split stats from THP_DEFERRED_SPLIT_PAGE, but it does not fix the
above issue. A fully unmapped PTE-mapped order-9 THP was still added to
deferred split list and counted as THP_DEFERRED_SPLIT_PAGE, since nr is
512 (non zero), level is RMAP_LEVEL_PTE, and inside deferred_split_folio()
the order-9 folio is folio_test_pmd_mappable().
Link: https://lkml.kernel.org/r/20240502132852.862138-1-zi.yan@sent.com
Signed-off-by: Zi Yan <ziy@nvidia.com>
Suggested-by: David Hildenbrand <david@redhat.com>
Reviewed-by: Yang Shi <shy828301@gmail.com>
Reviewed-by: David Hildenbrand <david@redhat.com>
Reviewed-by: Barry Song <baohua@kernel.org>
Reviewed-by: Lance Yang <ioworker0@gmail.com>
Cc: Alexander Gordeev <agordeev@linux.ibm.com>
Cc: Matthew Wilcox (Oracle) <willy@infradead.org>
Cc: Ryan Roberts <ryan.roberts@arm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2024-05-02 09:28:51 -04:00
|
|
|
|
2024-08-07 13:55:15 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
mm/rmap: do not add fully unmapped large folio to deferred split list
In __folio_remove_rmap(), a large folio is added to deferred split list if
any page in a folio loses its final mapping. But it is possible that the
folio is fully unmapped and adding it to deferred split list is
unnecessary.
For PMD-mapped THPs, that was not really an issue, because removing the
last PMD mapping in the absence of PTE mappings would not have added the
folio to the deferred split queue.
However, for PTE-mapped THPs, which are now more prominent due to mTHP,
they are always added to the deferred split queue. One side effect is
that the THP_DEFERRED_SPLIT_PAGE stat for a PTE-mapped folio can be
unintentionally increased, making it look like there are many partially
mapped folios -- although the whole folio is fully unmapped stepwise.
Core-mm now tries batch-unmapping consecutive PTEs of PTE-mapped THPs
where possible starting from commit b06dc281aa99 ("mm/rmap: introduce
folio_remove_rmap_[pte|ptes|pmd]()"). When it happens, a whole PTE-mapped
folio is unmapped in one go and can avoid being added to deferred split
list, reducing the THP_DEFERRED_SPLIT_PAGE noise. But there will still be
noise when we cannot batch-unmap a complete PTE-mapped folio in one go --
or where this type of batching is not implemented yet, e.g., migration.
To avoid the unnecessary addition, folio->_nr_pages_mapped is checked to
tell if the whole folio is unmapped. If the folio is already on deferred
split list, it will be skipped, too.
Note: commit 98046944a159 ("mm: huge_memory: add the missing
folio_test_pmd_mappable() for THP split statistics") tried to exclude mTHP
deferred split stats from THP_DEFERRED_SPLIT_PAGE, but it does not fix the
above issue. A fully unmapped PTE-mapped order-9 THP was still added to
deferred split list and counted as THP_DEFERRED_SPLIT_PAGE, since nr is
512 (non zero), level is RMAP_LEVEL_PTE, and inside deferred_split_folio()
the order-9 folio is folio_test_pmd_mappable().
Link: https://lkml.kernel.org/r/20240502132852.862138-1-zi.yan@sent.com
Signed-off-by: Zi Yan <ziy@nvidia.com>
Suggested-by: David Hildenbrand <david@redhat.com>
Reviewed-by: Yang Shi <shy828301@gmail.com>
Reviewed-by: David Hildenbrand <david@redhat.com>
Reviewed-by: Barry Song <baohua@kernel.org>
Reviewed-by: Lance Yang <ioworker0@gmail.com>
Cc: Alexander Gordeev <agordeev@linux.ibm.com>
Cc: Matthew Wilcox (Oracle) <willy@infradead.org>
Cc: Ryan Roberts <ryan.roberts@arm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2024-05-02 09:28:51 -04:00
|
|
|
|
2023-12-20 23:44:47 +01:00
|
|
|
|
2025-08-11 13:26:27 +02:00
|
|
|
|
|
|
|
|
|
mm: stop maintaining the per-page mapcount of large folios (CONFIG_NO_PAGE_MAPCOUNT)
Everything is in place to stop using the per-page mapcounts in large
folios: the mapcount of tail pages will always be logically 0 (-1 value),
just like it currently is for hugetlb folios already, and the page
mapcount of the head page is either 0 (-1 value) or contains a page type
(e.g., hugetlb).
Maintaining _nr_pages_mapped without per-page mapcounts is impossible, so
that one also has to go with CONFIG_NO_PAGE_MAPCOUNT.
There are two remaining implications:
(1) Per-node, per-cgroup and per-lruvec stats of "NR_ANON_MAPPED"
("mapped anonymous memory") and "NR_FILE_MAPPED"
("mapped file memory"):
As soon as any page of the folio is mapped -- folio_mapped() -- we
now account the complete folio as mapped. Once the last page is
unmapped -- !folio_mapped() -- we account the complete folio as
unmapped.
This implies that ...
* "AnonPages" and "Mapped" in /proc/meminfo and
/sys/devices/system/node/*/meminfo
* cgroup v2: "anon" and "file_mapped" in "memory.stat" and
"memory.numa_stat"
* cgroup v1: "rss" and "mapped_file" in "memory.stat" and
"memory.numa_stat
... can now appear higher than before. But note that these folios do
consume that memory, simply not all pages are actually currently
mapped.
It's worth nothing that other accounting in the kernel (esp. cgroup
charging on allocation) is not affected by this change.
[why oh why is "anon" called "rss" in cgroup v1]
(2) Detecting partial mappings
Detecting whether anon THPs are partially mapped gets a bit more
unreliable. As long as a single MM maps such a large folio
("exclusively mapped"), we can reliably detect it. Especially before
fork() / after a short-lived child process quit, we will detect
partial mappings reliably, which is the common case.
In essence, if the average per-page mapcount in an anon THP is < 1,
we know for sure that we have a partial mapping.
However, as soon as multiple MMs are involved, we might miss detecting
partial mappings: this might be relevant with long-lived child
processes. If we have a fully-mapped anon folio before fork(), once
our child processes and our parent all unmap (zap/COW) the same pages
(but not the complete folio), we might not detect the partial mapping.
However, once the child processes quit we would detect the partial
mapping.
How relevant this case is in practice remains to be seen.
Swapout/migration will likely mitigate this.
In the future, RMAP walkers could check for that for that case
(e.g., when collecting access bits during reclaim) and simply flag
them for deferred-splitting.
Link: https://lkml.kernel.org/r/20250303163014.1128035-21-david@redhat.com
Signed-off-by: David Hildenbrand <david@redhat.com>
Cc: Andy Lutomirks^H^Hski <luto@kernel.org>
Cc: Borislav Betkov <bp@alien8.de>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Jann Horn <jannh@google.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Lance Yang <ioworker0@gmail.com>
Cc: Liam Howlett <liam.howlett@oracle.com>
Cc: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: Matthew Wilcow (Oracle) <willy@infradead.org>
Cc: Michal Koutn <mkoutny@suse.com>
Cc: Muchun Song <muchun.song@linux.dev>
Cc: tejun heo <tj@kernel.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Zefan Li <lizefan.x@bytedance.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2025-03-03 17:30:13 +01:00
|
|
|
|
|
|
|
|
|
2025-08-11 13:26:27 +02:00
|
|
|
|
mm: stop maintaining the per-page mapcount of large folios (CONFIG_NO_PAGE_MAPCOUNT)
Everything is in place to stop using the per-page mapcounts in large
folios: the mapcount of tail pages will always be logically 0 (-1 value),
just like it currently is for hugetlb folios already, and the page
mapcount of the head page is either 0 (-1 value) or contains a page type
(e.g., hugetlb).
Maintaining _nr_pages_mapped without per-page mapcounts is impossible, so
that one also has to go with CONFIG_NO_PAGE_MAPCOUNT.
There are two remaining implications:
(1) Per-node, per-cgroup and per-lruvec stats of "NR_ANON_MAPPED"
("mapped anonymous memory") and "NR_FILE_MAPPED"
("mapped file memory"):
As soon as any page of the folio is mapped -- folio_mapped() -- we
now account the complete folio as mapped. Once the last page is
unmapped -- !folio_mapped() -- we account the complete folio as
unmapped.
This implies that ...
* "AnonPages" and "Mapped" in /proc/meminfo and
/sys/devices/system/node/*/meminfo
* cgroup v2: "anon" and "file_mapped" in "memory.stat" and
"memory.numa_stat"
* cgroup v1: "rss" and "mapped_file" in "memory.stat" and
"memory.numa_stat
... can now appear higher than before. But note that these folios do
consume that memory, simply not all pages are actually currently
mapped.
It's worth nothing that other accounting in the kernel (esp. cgroup
charging on allocation) is not affected by this change.
[why oh why is "anon" called "rss" in cgroup v1]
(2) Detecting partial mappings
Detecting whether anon THPs are partially mapped gets a bit more
unreliable. As long as a single MM maps such a large folio
("exclusively mapped"), we can reliably detect it. Especially before
fork() / after a short-lived child process quit, we will detect
partial mappings reliably, which is the common case.
In essence, if the average per-page mapcount in an anon THP is < 1,
we know for sure that we have a partial mapping.
However, as soon as multiple MMs are involved, we might miss detecting
partial mappings: this might be relevant with long-lived child
processes. If we have a fully-mapped anon folio before fork(), once
our child processes and our parent all unmap (zap/COW) the same pages
(but not the complete folio), we might not detect the partial mapping.
However, once the child processes quit we would detect the partial
mapping.
How relevant this case is in practice remains to be seen.
Swapout/migration will likely mitigate this.
In the future, RMAP walkers could check for that for that case
(e.g., when collecting access bits during reclaim) and simply flag
them for deferred-splitting.
Link: https://lkml.kernel.org/r/20250303163014.1128035-21-david@redhat.com
Signed-off-by: David Hildenbrand <david@redhat.com>
Cc: Andy Lutomirks^H^Hski <luto@kernel.org>
Cc: Borislav Betkov <bp@alien8.de>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Jann Horn <jannh@google.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Lance Yang <ioworker0@gmail.com>
Cc: Liam Howlett <liam.howlett@oracle.com>
Cc: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: Matthew Wilcow (Oracle) <willy@infradead.org>
Cc: Michal Koutn <mkoutny@suse.com>
Cc: Muchun Song <muchun.song@linux.dev>
Cc: tejun heo <tj@kernel.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Zefan Li <lizefan.x@bytedance.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2025-03-03 17:30:13 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-03-03 17:30:02 +01:00
|
|
|
|
2023-01-11 14:28:51 +00:00
|
|
|
|
mm,thp,rmap: lock_compound_mapcounts() on THP mapcounts
Fix the races in maintaining compound_mapcount, subpages_mapcount and
subpage _mapcount by using PG_locked in the first tail of any compound
page for a bit_spin_lock() on such modifications; skipping the usual
atomic operations on those fields in this case.
Bring page_remove_file_rmap() and page_remove_anon_compound_rmap() back
into page_remove_rmap() itself. Rearrange page_add_anon_rmap() and
page_add_file_rmap() and page_remove_rmap() to follow the same "if
(compound) {lock} else if (PageCompound) {lock} else {atomic}" pattern
(with a PageTransHuge in the compound test, like before, to avoid BUG_ONs
and optimize away that block when THP is not configured). Move all the
stats updates outside, after the bit_spin_locked section, so that it is
sure to be a leaf lock.
Add page_dup_compound_rmap() to manage compound locking versus atomics in
sync with the rest. In particular, hugetlb pages are still using the
atomics: to avoid unnecessary interference there, and because they never
have subpage mappings; but this exception can easily be changed.
Conveniently, page_dup_compound_rmap() turns out to suit an anon THP's
__split_huge_pmd_locked() too.
bit_spin_lock() is not popular with PREEMPT_RT folks: but PREEMPT_RT
sensibly excludes TRANSPARENT_HUGEPAGE already, so its only exposure is to
the non-hugetlb non-THP pte-mapped compound pages (with large folios being
currently dependent on TRANSPARENT_HUGEPAGE). There is never any scan of
subpages in this case; but we have chosen to use PageCompound tests rather
than PageTransCompound tests to gate the use of lock_compound_mapcounts(),
so that page_mapped() is correct on all compound pages, whether or not
TRANSPARENT_HUGEPAGE is enabled: could that be a problem for PREEMPT_RT,
when there is contention on the lock - under heavy concurrent forking for
example? If so, then it can be turned into a sleeping lock (like
folio_lock()) when PREEMPT_RT.
A simple 100 X munmap(mmap(2GB, MAP_SHARED|MAP_POPULATE, tmpfs), 2GB) took
18 seconds on small pages, and used to take 1 second on huge pages, but
now takes 115 milliseconds on huge pages. Mapping by pmds a second time
used to take 860ms and now takes 86ms; mapping by pmds after mapping by
ptes (when the scan is needed) used to take 870ms and now takes 495ms.
Mapping huge pages by ptes is largely unaffected but variable: between 5%
faster and 5% slower in what I've recorded. Contention on the lock is
likely to behave worse than contention on the atomics behaved.
Link: https://lkml.kernel.org/r/1b42bd1a-8223-e827-602f-d466c2db7d3c@google.com
Signed-off-by: Hugh Dickins <hughd@google.com>
Acked-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: James Houghton <jthoughton@google.com>
Cc: John Hubbard <jhubbard@nvidia.com>
Cc: Matthew Wilcox (Oracle) <willy@infradead.org>
Cc: Miaohe Lin <linmiaohe@huawei.com>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Mina Almasry <almasrymina@google.com>
Cc: Muchun Song <songmuchun@bytedance.com>
Cc: Naoya Horiguchi <naoya.horiguchi@linux.dev>
Cc: Peter Xu <peterx@redhat.com>
Cc: Sidhartha Kumar <sidhartha.kumar@oracle.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Yang Shi <shy828301@gmail.com>
Cc: Zach O'Keefe <zokeefe@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2022-11-02 18:53:45 -07:00
|
|
|
|
2023-12-20 23:45:03 +01:00
|
|
|
|
|
|
|
|
|
2025-03-03 17:30:04 +01:00
|
|
|
|
2025-08-11 13:26:27 +02:00
|
|
|
|
2025-02-28 14:31:09 +11:00
|
|
|
|
2025-08-17 03:26:46 +00:00
|
|
|
|
mm,thp,rmap: fix races between updates of subpages_mapcount
Commit 4b51634cd16a, introducing the COMPOUND_MAPPED bit, paid attention
to the impossibility of subpages_mapcount ever appearing negative; but did
not attend to those races in which it can momentarily appear larger than
thought possible.
These arise from how page_remove_rmap() first decrements page->_mapcount
or compound_mapcount, then, if that transition goes negative (logical 0),
decrements subpages_mapcount. The initial decrement lets a racing
page_add_*_rmap() reincrement _mapcount or compound_mapcount immediately,
and then in rare cases its corresponding increment of subpages_mapcount
may be completed before page_remove_rmap()'s decrement. There could even
(with increasing unlikelihood) be a series of increments intermixed with
the decrements.
In practice, checking subpages_mapcount with a temporary WARN on range,
has caught values of 0x1000000 (2*COMPOUND_MAPPED, when move_pages() was
using remove_migration_pmd()) and 0x800201 (do_huge_pmd_wp_page() using
__split_huge_pmd()): page_add_anon_rmap() racing page_remove_rmap(), as
predicted.
I certainly found it harder to reason about than when bit_spin_locked, but
the easy case gives a clue to how to handle the harder case. The easy
case being the three !(nr & COMPOUND_MAPPED) checks, which should
obviously be replaced by (nr < COMPOUND_MAPPED) checks - to count a page
as compound mapped, even while the bit in that position is 0.
The harder case is when trying to decide how many subpages are newly
covered or uncovered, when compound map is first added or last removed:
not knowing all that racily happened between first and second atomic ops.
But the easy way to handle that, is again to count the page as compound
mapped all the while that its subpages_mapcount indicates so - ignoring
the _mapcount or compound_mapcount transition while it is on the way to
being reversed.
Link: https://lkml.kernel.org/r/4388158-3092-a960-ff2d-55f2b0fe4ef8@google.com
Fixes: 4b51634cd16a ("mm,thp,rmap: subpages_mapcount COMPOUND_MAPPED if PMD-mapped")
Signed-off-by: Hugh Dickins <hughd@google.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: James Houghton <jthoughton@google.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: John Hubbard <jhubbard@nvidia.com>
Cc: "Kirill A . Shutemov" <kirill@shutemov.name>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Miaohe Lin <linmiaohe@huawei.com>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Mina Almasry <almasrymina@google.com>
Cc: Muchun Song <songmuchun@bytedance.com>
Cc: Naoya Horiguchi <naoya.horiguchi@linux.dev>
Cc: Peter Xu <peterx@redhat.com>
Cc: Sidhartha Kumar <sidhartha.kumar@oracle.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Yang Shi <shy828301@gmail.com>
Cc: Zach O'Keefe <zokeefe@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2022-12-04 17:57:07 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-12-20 23:45:03 +01:00
|
|
|
|
mm,thp,rmap: fix races between updates of subpages_mapcount
Commit 4b51634cd16a, introducing the COMPOUND_MAPPED bit, paid attention
to the impossibility of subpages_mapcount ever appearing negative; but did
not attend to those races in which it can momentarily appear larger than
thought possible.
These arise from how page_remove_rmap() first decrements page->_mapcount
or compound_mapcount, then, if that transition goes negative (logical 0),
decrements subpages_mapcount. The initial decrement lets a racing
page_add_*_rmap() reincrement _mapcount or compound_mapcount immediately,
and then in rare cases its corresponding increment of subpages_mapcount
may be completed before page_remove_rmap()'s decrement. There could even
(with increasing unlikelihood) be a series of increments intermixed with
the decrements.
In practice, checking subpages_mapcount with a temporary WARN on range,
has caught values of 0x1000000 (2*COMPOUND_MAPPED, when move_pages() was
using remove_migration_pmd()) and 0x800201 (do_huge_pmd_wp_page() using
__split_huge_pmd()): page_add_anon_rmap() racing page_remove_rmap(), as
predicted.
I certainly found it harder to reason about than when bit_spin_locked, but
the easy case gives a clue to how to handle the harder case. The easy
case being the three !(nr & COMPOUND_MAPPED) checks, which should
obviously be replaced by (nr < COMPOUND_MAPPED) checks - to count a page
as compound mapped, even while the bit in that position is 0.
The harder case is when trying to decide how many subpages are newly
covered or uncovered, when compound map is first added or last removed:
not knowing all that racily happened between first and second atomic ops.
But the easy way to handle that, is again to count the page as compound
mapped all the while that its subpages_mapcount indicates so - ignoring
the _mapcount or compound_mapcount transition while it is on the way to
being reversed.
Link: https://lkml.kernel.org/r/4388158-3092-a960-ff2d-55f2b0fe4ef8@google.com
Fixes: 4b51634cd16a ("mm,thp,rmap: subpages_mapcount COMPOUND_MAPPED if PMD-mapped")
Signed-off-by: Hugh Dickins <hughd@google.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: James Houghton <jthoughton@google.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: John Hubbard <jhubbard@nvidia.com>
Cc: "Kirill A . Shutemov" <kirill@shutemov.name>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Miaohe Lin <linmiaohe@huawei.com>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Mina Almasry <almasrymina@google.com>
Cc: Muchun Song <songmuchun@bytedance.com>
Cc: Naoya Horiguchi <naoya.horiguchi@linux.dev>
Cc: Peter Xu <peterx@redhat.com>
Cc: Sidhartha Kumar <sidhartha.kumar@oracle.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Yang Shi <shy828301@gmail.com>
Cc: Zach O'Keefe <zokeefe@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2022-12-04 17:57:07 -08:00
|
|
|
|
|
|
|
|
|
mm,thp,rmap: lock_compound_mapcounts() on THP mapcounts
Fix the races in maintaining compound_mapcount, subpages_mapcount and
subpage _mapcount by using PG_locked in the first tail of any compound
page for a bit_spin_lock() on such modifications; skipping the usual
atomic operations on those fields in this case.
Bring page_remove_file_rmap() and page_remove_anon_compound_rmap() back
into page_remove_rmap() itself. Rearrange page_add_anon_rmap() and
page_add_file_rmap() and page_remove_rmap() to follow the same "if
(compound) {lock} else if (PageCompound) {lock} else {atomic}" pattern
(with a PageTransHuge in the compound test, like before, to avoid BUG_ONs
and optimize away that block when THP is not configured). Move all the
stats updates outside, after the bit_spin_locked section, so that it is
sure to be a leaf lock.
Add page_dup_compound_rmap() to manage compound locking versus atomics in
sync with the rest. In particular, hugetlb pages are still using the
atomics: to avoid unnecessary interference there, and because they never
have subpage mappings; but this exception can easily be changed.
Conveniently, page_dup_compound_rmap() turns out to suit an anon THP's
__split_huge_pmd_locked() too.
bit_spin_lock() is not popular with PREEMPT_RT folks: but PREEMPT_RT
sensibly excludes TRANSPARENT_HUGEPAGE already, so its only exposure is to
the non-hugetlb non-THP pte-mapped compound pages (with large folios being
currently dependent on TRANSPARENT_HUGEPAGE). There is never any scan of
subpages in this case; but we have chosen to use PageCompound tests rather
than PageTransCompound tests to gate the use of lock_compound_mapcounts(),
so that page_mapped() is correct on all compound pages, whether or not
TRANSPARENT_HUGEPAGE is enabled: could that be a problem for PREEMPT_RT,
when there is contention on the lock - under heavy concurrent forking for
example? If so, then it can be turned into a sleeping lock (like
folio_lock()) when PREEMPT_RT.
A simple 100 X munmap(mmap(2GB, MAP_SHARED|MAP_POPULATE, tmpfs), 2GB) took
18 seconds on small pages, and used to take 1 second on huge pages, but
now takes 115 milliseconds on huge pages. Mapping by pmds a second time
used to take 860ms and now takes 86ms; mapping by pmds after mapping by
ptes (when the scan is needed) used to take 870ms and now takes 495ms.
Mapping huge pages by ptes is largely unaffected but variable: between 5%
faster and 5% slower in what I've recorded. Contention on the lock is
likely to behave worse than contention on the atomics behaved.
Link: https://lkml.kernel.org/r/1b42bd1a-8223-e827-602f-d466c2db7d3c@google.com
Signed-off-by: Hugh Dickins <hughd@google.com>
Acked-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: James Houghton <jthoughton@google.com>
Cc: John Hubbard <jhubbard@nvidia.com>
Cc: Matthew Wilcox (Oracle) <willy@infradead.org>
Cc: Miaohe Lin <linmiaohe@huawei.com>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Mina Almasry <almasrymina@google.com>
Cc: Muchun Song <songmuchun@bytedance.com>
Cc: Naoya Horiguchi <naoya.horiguchi@linux.dev>
Cc: Peter Xu <peterx@redhat.com>
Cc: Sidhartha Kumar <sidhartha.kumar@oracle.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Yang Shi <shy828301@gmail.com>
Cc: Zach O'Keefe <zokeefe@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2022-11-02 18:53:45 -07:00
|
|
|
|
mm/rmap: do not add fully unmapped large folio to deferred split list
In __folio_remove_rmap(), a large folio is added to deferred split list if
any page in a folio loses its final mapping. But it is possible that the
folio is fully unmapped and adding it to deferred split list is
unnecessary.
For PMD-mapped THPs, that was not really an issue, because removing the
last PMD mapping in the absence of PTE mappings would not have added the
folio to the deferred split queue.
However, for PTE-mapped THPs, which are now more prominent due to mTHP,
they are always added to the deferred split queue. One side effect is
that the THP_DEFERRED_SPLIT_PAGE stat for a PTE-mapped folio can be
unintentionally increased, making it look like there are many partially
mapped folios -- although the whole folio is fully unmapped stepwise.
Core-mm now tries batch-unmapping consecutive PTEs of PTE-mapped THPs
where possible starting from commit b06dc281aa99 ("mm/rmap: introduce
folio_remove_rmap_[pte|ptes|pmd]()"). When it happens, a whole PTE-mapped
folio is unmapped in one go and can avoid being added to deferred split
list, reducing the THP_DEFERRED_SPLIT_PAGE noise. But there will still be
noise when we cannot batch-unmap a complete PTE-mapped folio in one go --
or where this type of batching is not implemented yet, e.g., migration.
To avoid the unnecessary addition, folio->_nr_pages_mapped is checked to
tell if the whole folio is unmapped. If the folio is already on deferred
split list, it will be skipped, too.
Note: commit 98046944a159 ("mm: huge_memory: add the missing
folio_test_pmd_mappable() for THP split statistics") tried to exclude mTHP
deferred split stats from THP_DEFERRED_SPLIT_PAGE, but it does not fix the
above issue. A fully unmapped PTE-mapped order-9 THP was still added to
deferred split list and counted as THP_DEFERRED_SPLIT_PAGE, since nr is
512 (non zero), level is RMAP_LEVEL_PTE, and inside deferred_split_folio()
the order-9 folio is folio_test_pmd_mappable().
Link: https://lkml.kernel.org/r/20240502132852.862138-1-zi.yan@sent.com
Signed-off-by: Zi Yan <ziy@nvidia.com>
Suggested-by: David Hildenbrand <david@redhat.com>
Reviewed-by: Yang Shi <shy828301@gmail.com>
Reviewed-by: David Hildenbrand <david@redhat.com>
Reviewed-by: Barry Song <baohua@kernel.org>
Reviewed-by: Lance Yang <ioworker0@gmail.com>
Cc: Alexander Gordeev <agordeev@linux.ibm.com>
Cc: Matthew Wilcox (Oracle) <willy@infradead.org>
Cc: Ryan Roberts <ryan.roberts@arm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2024-05-02 09:28:51 -04:00
|
|
|
|
2024-07-10 23:43:50 +02:00
|
|
|
|
2023-12-20 23:44:47 +01:00
|
|
|
|
2025-08-11 13:26:27 +02:00
|
|
|
|
|
|
|
|
|
mm,thp,rmap: simplify compound page mapcount handling
Compound page (folio) mapcount calculations have been different for anon
and file (or shmem) THPs, and involved the obscure PageDoubleMap flag.
And each huge mapping and unmapping of a file (or shmem) THP involved
atomically incrementing and decrementing the mapcount of every subpage of
that huge page, dirtying many struct page cachelines.
Add subpages_mapcount field to the struct folio and first tail page, so
that the total of subpage mapcounts is available in one place near the
head: then page_mapcount() and total_mapcount() and page_mapped(), and
their folio equivalents, are so quick that anon and file and hugetlb don't
need to be optimized differently. Delete the unloved PageDoubleMap.
page_add and page_remove rmap functions must now maintain the
subpages_mapcount as well as the subpage _mapcount, when dealing with pte
mappings of huge pages; and correct maintenance of NR_ANON_MAPPED and
NR_FILE_MAPPED statistics still needs reading through the subpages, using
nr_subpages_unmapped() - but only when first or last pmd mapping finds
subpages_mapcount raised (double-map case, not the common case).
But are those counts (used to decide when to split an anon THP, and in
vmscan's pagecache_reclaimable heuristic) correctly maintained? Not
quite: since page_remove_rmap() (and also split_huge_pmd()) is often
called without page lock, there can be races when a subpage pte mapcount
0<->1 while compound pmd mapcount 0<->1 is scanning - races which the
previous implementation had prevented. The statistics might become
inaccurate, and even drift down until they underflow through 0. That is
not good enough, but is better dealt with in a followup patch.
Update a few comments on first and second tail page overlaid fields.
hugepage_add_new_anon_rmap() has to "increment" compound_mapcount, but
subpages_mapcount and compound_pincount are already correctly at 0, so
delete its reinitialization of compound_pincount.
A simple 100 X munmap(mmap(2GB, MAP_SHARED|MAP_POPULATE, tmpfs), 2GB) took
18 seconds on small pages, and used to take 1 second on huge pages, but
now takes 119 milliseconds on huge pages. Mapping by pmds a second time
used to take 860ms and now takes 92ms; mapping by pmds after mapping by
ptes (when the scan is needed) used to take 870ms and now takes 495ms.
But there might be some benchmarks which would show a slowdown, because
tail struct pages now fall out of cache until final freeing checks them.
Link: https://lkml.kernel.org/r/47ad693-717-79c8-e1ba-46c3a6602e48@google.com
Signed-off-by: Hugh Dickins <hughd@google.com>
Acked-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: James Houghton <jthoughton@google.com>
Cc: John Hubbard <jhubbard@nvidia.com>
Cc: Matthew Wilcox (Oracle) <willy@infradead.org>
Cc: Miaohe Lin <linmiaohe@huawei.com>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Mina Almasry <almasrymina@google.com>
Cc: Muchun Song <songmuchun@bytedance.com>
Cc: Naoya Horiguchi <naoya.horiguchi@linux.dev>
Cc: Peter Xu <peterx@redhat.com>
Cc: Sidhartha Kumar <sidhartha.kumar@oracle.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Yang Shi <shy828301@gmail.com>
Cc: Zach O'Keefe <zokeefe@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2022-11-02 18:51:38 -07:00
|
|
|
|
|
|
|
|
|
2024-07-10 23:43:50 +02:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
mm/zone_device: support large zone device private folios
Patch series "mm: support device-private THP", v7.
This patch series introduces support for Transparent Huge Page (THP)
migration in zone device-private memory. The implementation enables
efficient migration of large folios between system memory and
device-private memory
Background
Current zone device-private memory implementation only supports PAGE_SIZE
granularity, leading to:
- Increased TLB pressure
- Inefficient migration between CPU and device memory
This series extends the existing zone device-private infrastructure to
support THP, leading to:
- Reduced page table overhead
- Improved memory bandwidth utilization
- Seamless fallback to base pages when needed
In my local testing (using lib/test_hmm) and a throughput test, the series
shows a 350% improvement in data transfer throughput and a 80% improvement
in latency
These patches build on the earlier posts by Ralph Campbell [1]
Two new flags are added in vma_migration to select and mark compound
pages. migrate_vma_setup(), migrate_vma_pages() and
migrate_vma_finalize() support migration of these pages when
MIGRATE_VMA_SELECT_COMPOUND is passed in as arguments.
The series also adds zone device awareness to (m)THP pages along with
fault handling of large zone device private pages. page vma walk and the
rmap code is also zone device aware. Support has also been added for
folios that might need to be split in the middle of migration (when the
src and dst do not agree on MIGRATE_PFN_COMPOUND), that occurs when src
side of the migration can migrate large pages, but the destination has not
been able to allocate large pages. The code supported and used
folio_split() when migrating THP pages, this is used when
MIGRATE_VMA_SELECT_COMPOUND is not passed as an argument to
migrate_vma_setup().
The test infrastructure lib/test_hmm.c has been enhanced to support THP
migration. A new ioctl to emulate failure of large page allocations has
been added to test the folio split code path. hmm-tests.c has new test
cases for huge page migration and to test the folio split path. A new
throughput test has been added as well.
The nouveau dmem code has been enhanced to use the new THP migration
capability.
mTHP support:
The patches hard code, HPAGE_PMD_NR in a few places, but the code has been
kept generic to support various order sizes. With additional refactoring
of the code support of different order sizes should be possible.
The future plan is to post enhancements to support mTHP with a rough
design as follows:
1. Add the notion of allowable thp orders to the HMM based test driver
2. For non PMD based THP paths in migrate_device.c, check to see if
a suitable order is found and supported by the driver
3. Iterate across orders to check the highest supported order for migration
4. Migrate and finalize
The mTHP patches can be built on top of this series, the key design
elements that need to be worked out are infrastructure and driver support
for multiple ordered pages and their migration.
HMM support for large folios was added in 10b9feee2d0d ("mm/hmm:
populate PFNs from PMD swap entry").
This patch (of 16)
Add routines to support allocation of large order zone device folios and
helper functions for zone device folios, to check if a folio is device
private and helpers for setting zone device data.
When large folios are used, the existing page_free() callback in pgmap is
called when the folio is freed, this is true for both PAGE_SIZE and higher
order pages.
Zone device private large folios do not support deferred split and scan
like normal THP folios.
Link: https://lkml.kernel.org/r/20251001065707.920170-1-balbirs@nvidia.com
Link: https://lkml.kernel.org/r/20251001065707.920170-2-balbirs@nvidia.com
Link: https://lore.kernel.org/linux-mm/20201106005147.20113-1-rcampbell@nvidia.com/ [1]
Signed-off-by: Balbir Singh <balbirs@nvidia.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Zi Yan <ziy@nvidia.com>
Cc: Joshua Hahn <joshua.hahnjy@gmail.com>
Cc: Rakie Kim <rakie.kim@sk.com>
Cc: Byungchul Park <byungchul@sk.com>
Cc: Gregory Price <gourry@gourry.net>
Cc: Ying Huang <ying.huang@linux.alibaba.com>
Cc: Alistair Popple <apopple@nvidia.com>
Cc: Oscar Salvador <osalvador@suse.de>
Cc: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: Baolin Wang <baolin.wang@linux.alibaba.com>
Cc: "Liam R. Howlett" <Liam.Howlett@oracle.com>
Cc: Nico Pache <npache@redhat.com>
Cc: Ryan Roberts <ryan.roberts@arm.com>
Cc: Dev Jain <dev.jain@arm.com>
Cc: Barry Song <baohua@kernel.org>
Cc: Lyude Paul <lyude@redhat.com>
Cc: Danilo Krummrich <dakr@kernel.org>
Cc: David Airlie <airlied@gmail.com>
Cc: Simona Vetter <simona@ffwll.ch>
Cc: Ralph Campbell <rcampbell@nvidia.com>
Cc: Mika Penttilä <mpenttil@redhat.com>
Cc: Matthew Brost <matthew.brost@intel.com>
Cc: Francois Dugast <francois.dugast@intel.com>
Cc: Madhavan Srinivasan <maddy@linux.ibm.com>
Cc: Christophe Leroy <christophe.leroy@csgroup.eu>
Cc: Felix Kuehling <Felix.Kuehling@amd.com>
Cc: Alex Deucher <alexander.deucher@amd.com>
Cc: "Christian König" <christian.koenig@amd.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2025-10-01 16:56:52 +10:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-07-10 23:43:50 +02:00
|
|
|
|
|
|
|
|
|
mm/zone_device: support large zone device private folios
Patch series "mm: support device-private THP", v7.
This patch series introduces support for Transparent Huge Page (THP)
migration in zone device-private memory. The implementation enables
efficient migration of large folios between system memory and
device-private memory
Background
Current zone device-private memory implementation only supports PAGE_SIZE
granularity, leading to:
- Increased TLB pressure
- Inefficient migration between CPU and device memory
This series extends the existing zone device-private infrastructure to
support THP, leading to:
- Reduced page table overhead
- Improved memory bandwidth utilization
- Seamless fallback to base pages when needed
In my local testing (using lib/test_hmm) and a throughput test, the series
shows a 350% improvement in data transfer throughput and a 80% improvement
in latency
These patches build on the earlier posts by Ralph Campbell [1]
Two new flags are added in vma_migration to select and mark compound
pages. migrate_vma_setup(), migrate_vma_pages() and
migrate_vma_finalize() support migration of these pages when
MIGRATE_VMA_SELECT_COMPOUND is passed in as arguments.
The series also adds zone device awareness to (m)THP pages along with
fault handling of large zone device private pages. page vma walk and the
rmap code is also zone device aware. Support has also been added for
folios that might need to be split in the middle of migration (when the
src and dst do not agree on MIGRATE_PFN_COMPOUND), that occurs when src
side of the migration can migrate large pages, but the destination has not
been able to allocate large pages. The code supported and used
folio_split() when migrating THP pages, this is used when
MIGRATE_VMA_SELECT_COMPOUND is not passed as an argument to
migrate_vma_setup().
The test infrastructure lib/test_hmm.c has been enhanced to support THP
migration. A new ioctl to emulate failure of large page allocations has
been added to test the folio split code path. hmm-tests.c has new test
cases for huge page migration and to test the folio split path. A new
throughput test has been added as well.
The nouveau dmem code has been enhanced to use the new THP migration
capability.
mTHP support:
The patches hard code, HPAGE_PMD_NR in a few places, but the code has been
kept generic to support various order sizes. With additional refactoring
of the code support of different order sizes should be possible.
The future plan is to post enhancements to support mTHP with a rough
design as follows:
1. Add the notion of allowable thp orders to the HMM based test driver
2. For non PMD based THP paths in migrate_device.c, check to see if
a suitable order is found and supported by the driver
3. Iterate across orders to check the highest supported order for migration
4. Migrate and finalize
The mTHP patches can be built on top of this series, the key design
elements that need to be worked out are infrastructure and driver support
for multiple ordered pages and their migration.
HMM support for large folios was added in 10b9feee2d0d ("mm/hmm:
populate PFNs from PMD swap entry").
This patch (of 16)
Add routines to support allocation of large order zone device folios and
helper functions for zone device folios, to check if a folio is device
private and helpers for setting zone device data.
When large folios are used, the existing page_free() callback in pgmap is
called when the folio is freed, this is true for both PAGE_SIZE and higher
order pages.
Zone device private large folios do not support deferred split and scan
like normal THP folios.
Link: https://lkml.kernel.org/r/20251001065707.920170-1-balbirs@nvidia.com
Link: https://lkml.kernel.org/r/20251001065707.920170-2-balbirs@nvidia.com
Link: https://lore.kernel.org/linux-mm/20201106005147.20113-1-rcampbell@nvidia.com/ [1]
Signed-off-by: Balbir Singh <balbirs@nvidia.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Zi Yan <ziy@nvidia.com>
Cc: Joshua Hahn <joshua.hahnjy@gmail.com>
Cc: Rakie Kim <rakie.kim@sk.com>
Cc: Byungchul Park <byungchul@sk.com>
Cc: Gregory Price <gourry@gourry.net>
Cc: Ying Huang <ying.huang@linux.alibaba.com>
Cc: Alistair Popple <apopple@nvidia.com>
Cc: Oscar Salvador <osalvador@suse.de>
Cc: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: Baolin Wang <baolin.wang@linux.alibaba.com>
Cc: "Liam R. Howlett" <Liam.Howlett@oracle.com>
Cc: Nico Pache <npache@redhat.com>
Cc: Ryan Roberts <ryan.roberts@arm.com>
Cc: Dev Jain <dev.jain@arm.com>
Cc: Barry Song <baohua@kernel.org>
Cc: Lyude Paul <lyude@redhat.com>
Cc: Danilo Krummrich <dakr@kernel.org>
Cc: David Airlie <airlied@gmail.com>
Cc: Simona Vetter <simona@ffwll.ch>
Cc: Ralph Campbell <rcampbell@nvidia.com>
Cc: Mika Penttilä <mpenttil@redhat.com>
Cc: Matthew Brost <matthew.brost@intel.com>
Cc: Francois Dugast <francois.dugast@intel.com>
Cc: Madhavan Srinivasan <maddy@linux.ibm.com>
Cc: Christophe Leroy <christophe.leroy@csgroup.eu>
Cc: Felix Kuehling <Felix.Kuehling@amd.com>
Cc: Alex Deucher <alexander.deucher@amd.com>
Cc: "Christian König" <christian.koenig@amd.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2025-10-01 16:56:52 +10:00
|
|
|
|
|
|
|
|
|
2024-08-30 11:03:38 +01:00
|
|
|
|
|
|
|
|
|
2024-05-06 21:13:33 +00:00
|
|
|
|
2016-01-15 16:54:17 -08:00
|
|
|
|
2009-09-21 17:01:28 -07:00
|
|
|
|
2023-01-16 19:28:26 +00:00
|
|
|
|
2023-12-20 23:44:45 +01:00
|
|
|
|
2023-01-16 19:28:26 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2009-09-21 17:01:28 -07:00
|
|
|
|
mm,thp,rmap: lock_compound_mapcounts() on THP mapcounts
Fix the races in maintaining compound_mapcount, subpages_mapcount and
subpage _mapcount by using PG_locked in the first tail of any compound
page for a bit_spin_lock() on such modifications; skipping the usual
atomic operations on those fields in this case.
Bring page_remove_file_rmap() and page_remove_anon_compound_rmap() back
into page_remove_rmap() itself. Rearrange page_add_anon_rmap() and
page_add_file_rmap() and page_remove_rmap() to follow the same "if
(compound) {lock} else if (PageCompound) {lock} else {atomic}" pattern
(with a PageTransHuge in the compound test, like before, to avoid BUG_ONs
and optimize away that block when THP is not configured). Move all the
stats updates outside, after the bit_spin_locked section, so that it is
sure to be a leaf lock.
Add page_dup_compound_rmap() to manage compound locking versus atomics in
sync with the rest. In particular, hugetlb pages are still using the
atomics: to avoid unnecessary interference there, and because they never
have subpage mappings; but this exception can easily be changed.
Conveniently, page_dup_compound_rmap() turns out to suit an anon THP's
__split_huge_pmd_locked() too.
bit_spin_lock() is not popular with PREEMPT_RT folks: but PREEMPT_RT
sensibly excludes TRANSPARENT_HUGEPAGE already, so its only exposure is to
the non-hugetlb non-THP pte-mapped compound pages (with large folios being
currently dependent on TRANSPARENT_HUGEPAGE). There is never any scan of
subpages in this case; but we have chosen to use PageCompound tests rather
than PageTransCompound tests to gate the use of lock_compound_mapcounts(),
so that page_mapped() is correct on all compound pages, whether or not
TRANSPARENT_HUGEPAGE is enabled: could that be a problem for PREEMPT_RT,
when there is contention on the lock - under heavy concurrent forking for
example? If so, then it can be turned into a sleeping lock (like
folio_lock()) when PREEMPT_RT.
A simple 100 X munmap(mmap(2GB, MAP_SHARED|MAP_POPULATE, tmpfs), 2GB) took
18 seconds on small pages, and used to take 1 second on huge pages, but
now takes 115 milliseconds on huge pages. Mapping by pmds a second time
used to take 860ms and now takes 86ms; mapping by pmds after mapping by
ptes (when the scan is needed) used to take 870ms and now takes 495ms.
Mapping huge pages by ptes is largely unaffected but variable: between 5%
faster and 5% slower in what I've recorded. Contention on the lock is
likely to behave worse than contention on the atomics behaved.
Link: https://lkml.kernel.org/r/1b42bd1a-8223-e827-602f-d466c2db7d3c@google.com
Signed-off-by: Hugh Dickins <hughd@google.com>
Acked-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: James Houghton <jthoughton@google.com>
Cc: John Hubbard <jhubbard@nvidia.com>
Cc: Matthew Wilcox (Oracle) <willy@infradead.org>
Cc: Miaohe Lin <linmiaohe@huawei.com>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Mina Almasry <almasrymina@google.com>
Cc: Muchun Song <songmuchun@bytedance.com>
Cc: Naoya Horiguchi <naoya.horiguchi@linux.dev>
Cc: Peter Xu <peterx@redhat.com>
Cc: Sidhartha Kumar <sidhartha.kumar@oracle.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Yang Shi <shy828301@gmail.com>
Cc: Zach O'Keefe <zokeefe@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2022-11-02 18:53:45 -07:00
|
|
|
|
2023-09-18 15:33:17 +08:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
2023-12-20 23:44:47 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-08-11 13:26:27 +02:00
|
|
|
|
2023-12-20 23:44:47 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-08-11 13:26:27 +02:00
|
|
|
|
2023-12-20 23:44:47 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-02-28 14:31:09 +11:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-08-11 13:26:27 +02:00
|
|
|
|
2025-02-28 14:31:09 +11:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-06-27 14:23:19 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
mm: support batched unmap for lazyfree large folios during reclamation
Currently, the PTEs and rmap of a large folio are removed one at a time.
This is not only slow but also causes the large folio to be unnecessarily
added to deferred_split, which can lead to races between the
deferred_split shrinker callback and memory reclamation. This patch
releases all PTEs and rmap entries in a batch. Currently, it only handles
lazyfree large folios.
The below microbench tries to reclaim 128MB lazyfree large folios
whose sizes are 64KiB:
#include <stdio.h>
#include <sys/mman.h>
#include <string.h>
#include <time.h>
#define SIZE 128*1024*1024 // 128 MB
unsigned long read_split_deferred()
{
FILE *file = fopen("/sys/kernel/mm/transparent_hugepage"
"/hugepages-64kB/stats/split_deferred", "r");
if (!file) {
perror("Error opening file");
return 0;
}
unsigned long value;
if (fscanf(file, "%lu", &value) != 1) {
perror("Error reading value");
fclose(file);
return 0;
}
fclose(file);
return value;
}
int main(int argc, char *argv[])
{
while(1) {
volatile int *p = mmap(0, SIZE, PROT_READ | PROT_WRITE,
MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
memset((void *)p, 1, SIZE);
madvise((void *)p, SIZE, MADV_FREE);
clock_t start_time = clock();
unsigned long start_split = read_split_deferred();
madvise((void *)p, SIZE, MADV_PAGEOUT);
clock_t end_time = clock();
unsigned long end_split = read_split_deferred();
double elapsed_time = (double)(end_time - start_time) / CLOCKS_PER_SEC;
printf("Time taken by reclamation: %f seconds, split_deferred: %ld\n",
elapsed_time, end_split - start_split);
munmap((void *)p, SIZE);
}
return 0;
}
w/o patch:
~ # ./a.out
Time taken by reclamation: 0.177418 seconds, split_deferred: 2048
Time taken by reclamation: 0.178348 seconds, split_deferred: 2048
Time taken by reclamation: 0.174525 seconds, split_deferred: 2048
Time taken by reclamation: 0.171620 seconds, split_deferred: 2048
Time taken by reclamation: 0.172241 seconds, split_deferred: 2048
Time taken by reclamation: 0.174003 seconds, split_deferred: 2048
Time taken by reclamation: 0.171058 seconds, split_deferred: 2048
Time taken by reclamation: 0.171993 seconds, split_deferred: 2048
Time taken by reclamation: 0.169829 seconds, split_deferred: 2048
Time taken by reclamation: 0.172895 seconds, split_deferred: 2048
Time taken by reclamation: 0.176063 seconds, split_deferred: 2048
Time taken by reclamation: 0.172568 seconds, split_deferred: 2048
Time taken by reclamation: 0.171185 seconds, split_deferred: 2048
Time taken by reclamation: 0.170632 seconds, split_deferred: 2048
Time taken by reclamation: 0.170208 seconds, split_deferred: 2048
Time taken by reclamation: 0.174192 seconds, split_deferred: 2048
...
w/ patch:
~ # ./a.out
Time taken by reclamation: 0.074231 seconds, split_deferred: 0
Time taken by reclamation: 0.071026 seconds, split_deferred: 0
Time taken by reclamation: 0.072029 seconds, split_deferred: 0
Time taken by reclamation: 0.071873 seconds, split_deferred: 0
Time taken by reclamation: 0.073573 seconds, split_deferred: 0
Time taken by reclamation: 0.071906 seconds, split_deferred: 0
Time taken by reclamation: 0.073604 seconds, split_deferred: 0
Time taken by reclamation: 0.075903 seconds, split_deferred: 0
Time taken by reclamation: 0.073191 seconds, split_deferred: 0
Time taken by reclamation: 0.071228 seconds, split_deferred: 0
Time taken by reclamation: 0.071391 seconds, split_deferred: 0
Time taken by reclamation: 0.071468 seconds, split_deferred: 0
Time taken by reclamation: 0.071896 seconds, split_deferred: 0
Time taken by reclamation: 0.072508 seconds, split_deferred: 0
Time taken by reclamation: 0.071884 seconds, split_deferred: 0
Time taken by reclamation: 0.072433 seconds, split_deferred: 0
Time taken by reclamation: 0.071939 seconds, split_deferred: 0
...
Link: https://lkml.kernel.org/r/20250214093015.51024-4-21cnbao@gmail.com
Signed-off-by: Barry Song <v-songbaohua@oppo.com>
Cc: Albert Ou <aou@eecs.berkeley.edu>
Cc: Anshuman Khandual <anshuman.khandual@arm.com>
Cc: Baolin Wang <baolin.wang@linux.alibaba.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Chis Li <chrisl@kernel.org>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Gavin Shan <gshan@redhat.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: "Huang, Ying" <ying.huang@intel.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Kairui Song <kasong@tencent.com>
Cc: Kefeng Wang <wangkefeng.wang@huawei.com>
Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Lance Yang <ioworker0@gmail.com>
Cc: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Mauricio Faria de Oliveira <mfo@canonical.com>
Cc: Palmer Dabbelt <palmer@dabbelt.com>
Cc: Paul Walmsley <paul.walmsley@sifive.com>
Cc: Ryan Roberts <ryan.roberts@arm.com>
Cc: Shaoqin Huang <shahuang@redhat.com>
Cc: Tangquan Zheng <zhengtangquan@oppo.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Will Deacon <will@kernel.org>
Cc: Yicong Yang <yangyicong@hisilicon.com>
Cc: Yosry Ahmed <yosryahmed@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2025-02-14 22:30:14 +13:00
|
|
|
|
2025-06-27 14:23:19 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
mm: support batched unmap for lazyfree large folios during reclamation
Currently, the PTEs and rmap of a large folio are removed one at a time.
This is not only slow but also causes the large folio to be unnecessarily
added to deferred_split, which can lead to races between the
deferred_split shrinker callback and memory reclamation. This patch
releases all PTEs and rmap entries in a batch. Currently, it only handles
lazyfree large folios.
The below microbench tries to reclaim 128MB lazyfree large folios
whose sizes are 64KiB:
#include <stdio.h>
#include <sys/mman.h>
#include <string.h>
#include <time.h>
#define SIZE 128*1024*1024 // 128 MB
unsigned long read_split_deferred()
{
FILE *file = fopen("/sys/kernel/mm/transparent_hugepage"
"/hugepages-64kB/stats/split_deferred", "r");
if (!file) {
perror("Error opening file");
return 0;
}
unsigned long value;
if (fscanf(file, "%lu", &value) != 1) {
perror("Error reading value");
fclose(file);
return 0;
}
fclose(file);
return value;
}
int main(int argc, char *argv[])
{
while(1) {
volatile int *p = mmap(0, SIZE, PROT_READ | PROT_WRITE,
MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
memset((void *)p, 1, SIZE);
madvise((void *)p, SIZE, MADV_FREE);
clock_t start_time = clock();
unsigned long start_split = read_split_deferred();
madvise((void *)p, SIZE, MADV_PAGEOUT);
clock_t end_time = clock();
unsigned long end_split = read_split_deferred();
double elapsed_time = (double)(end_time - start_time) / CLOCKS_PER_SEC;
printf("Time taken by reclamation: %f seconds, split_deferred: %ld\n",
elapsed_time, end_split - start_split);
munmap((void *)p, SIZE);
}
return 0;
}
w/o patch:
~ # ./a.out
Time taken by reclamation: 0.177418 seconds, split_deferred: 2048
Time taken by reclamation: 0.178348 seconds, split_deferred: 2048
Time taken by reclamation: 0.174525 seconds, split_deferred: 2048
Time taken by reclamation: 0.171620 seconds, split_deferred: 2048
Time taken by reclamation: 0.172241 seconds, split_deferred: 2048
Time taken by reclamation: 0.174003 seconds, split_deferred: 2048
Time taken by reclamation: 0.171058 seconds, split_deferred: 2048
Time taken by reclamation: 0.171993 seconds, split_deferred: 2048
Time taken by reclamation: 0.169829 seconds, split_deferred: 2048
Time taken by reclamation: 0.172895 seconds, split_deferred: 2048
Time taken by reclamation: 0.176063 seconds, split_deferred: 2048
Time taken by reclamation: 0.172568 seconds, split_deferred: 2048
Time taken by reclamation: 0.171185 seconds, split_deferred: 2048
Time taken by reclamation: 0.170632 seconds, split_deferred: 2048
Time taken by reclamation: 0.170208 seconds, split_deferred: 2048
Time taken by reclamation: 0.174192 seconds, split_deferred: 2048
...
w/ patch:
~ # ./a.out
Time taken by reclamation: 0.074231 seconds, split_deferred: 0
Time taken by reclamation: 0.071026 seconds, split_deferred: 0
Time taken by reclamation: 0.072029 seconds, split_deferred: 0
Time taken by reclamation: 0.071873 seconds, split_deferred: 0
Time taken by reclamation: 0.073573 seconds, split_deferred: 0
Time taken by reclamation: 0.071906 seconds, split_deferred: 0
Time taken by reclamation: 0.073604 seconds, split_deferred: 0
Time taken by reclamation: 0.075903 seconds, split_deferred: 0
Time taken by reclamation: 0.073191 seconds, split_deferred: 0
Time taken by reclamation: 0.071228 seconds, split_deferred: 0
Time taken by reclamation: 0.071391 seconds, split_deferred: 0
Time taken by reclamation: 0.071468 seconds, split_deferred: 0
Time taken by reclamation: 0.071896 seconds, split_deferred: 0
Time taken by reclamation: 0.072508 seconds, split_deferred: 0
Time taken by reclamation: 0.071884 seconds, split_deferred: 0
Time taken by reclamation: 0.072433 seconds, split_deferred: 0
Time taken by reclamation: 0.071939 seconds, split_deferred: 0
...
Link: https://lkml.kernel.org/r/20250214093015.51024-4-21cnbao@gmail.com
Signed-off-by: Barry Song <v-songbaohua@oppo.com>
Cc: Albert Ou <aou@eecs.berkeley.edu>
Cc: Anshuman Khandual <anshuman.khandual@arm.com>
Cc: Baolin Wang <baolin.wang@linux.alibaba.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Chis Li <chrisl@kernel.org>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Gavin Shan <gshan@redhat.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: "Huang, Ying" <ying.huang@intel.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Kairui Song <kasong@tencent.com>
Cc: Kefeng Wang <wangkefeng.wang@huawei.com>
Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Lance Yang <ioworker0@gmail.com>
Cc: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Mauricio Faria de Oliveira <mfo@canonical.com>
Cc: Palmer Dabbelt <palmer@dabbelt.com>
Cc: Paul Walmsley <paul.walmsley@sifive.com>
Cc: Ryan Roberts <ryan.roberts@arm.com>
Cc: Shaoqin Huang <shahuang@redhat.com>
Cc: Tangquan Zheng <zhengtangquan@oppo.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Will Deacon <will@kernel.org>
Cc: Yicong Yang <yangyicong@hisilicon.com>
Cc: Yosry Ahmed <yosryahmed@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2025-02-14 22:30:14 +13:00
|
|
|
|
2025-06-27 14:23:19 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2026-02-09 22:07:28 +08:00
|
|
|
|
|
|
|
|
|
2025-06-27 14:23:19 +08:00
|
|
|
|
2026-02-09 22:07:28 +08:00
|
|
|
|
mm: support batched unmap for lazyfree large folios during reclamation
Currently, the PTEs and rmap of a large folio are removed one at a time.
This is not only slow but also causes the large folio to be unnecessarily
added to deferred_split, which can lead to races between the
deferred_split shrinker callback and memory reclamation. This patch
releases all PTEs and rmap entries in a batch. Currently, it only handles
lazyfree large folios.
The below microbench tries to reclaim 128MB lazyfree large folios
whose sizes are 64KiB:
#include <stdio.h>
#include <sys/mman.h>
#include <string.h>
#include <time.h>
#define SIZE 128*1024*1024 // 128 MB
unsigned long read_split_deferred()
{
FILE *file = fopen("/sys/kernel/mm/transparent_hugepage"
"/hugepages-64kB/stats/split_deferred", "r");
if (!file) {
perror("Error opening file");
return 0;
}
unsigned long value;
if (fscanf(file, "%lu", &value) != 1) {
perror("Error reading value");
fclose(file);
return 0;
}
fclose(file);
return value;
}
int main(int argc, char *argv[])
{
while(1) {
volatile int *p = mmap(0, SIZE, PROT_READ | PROT_WRITE,
MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
memset((void *)p, 1, SIZE);
madvise((void *)p, SIZE, MADV_FREE);
clock_t start_time = clock();
unsigned long start_split = read_split_deferred();
madvise((void *)p, SIZE, MADV_PAGEOUT);
clock_t end_time = clock();
unsigned long end_split = read_split_deferred();
double elapsed_time = (double)(end_time - start_time) / CLOCKS_PER_SEC;
printf("Time taken by reclamation: %f seconds, split_deferred: %ld\n",
elapsed_time, end_split - start_split);
munmap((void *)p, SIZE);
}
return 0;
}
w/o patch:
~ # ./a.out
Time taken by reclamation: 0.177418 seconds, split_deferred: 2048
Time taken by reclamation: 0.178348 seconds, split_deferred: 2048
Time taken by reclamation: 0.174525 seconds, split_deferred: 2048
Time taken by reclamation: 0.171620 seconds, split_deferred: 2048
Time taken by reclamation: 0.172241 seconds, split_deferred: 2048
Time taken by reclamation: 0.174003 seconds, split_deferred: 2048
Time taken by reclamation: 0.171058 seconds, split_deferred: 2048
Time taken by reclamation: 0.171993 seconds, split_deferred: 2048
Time taken by reclamation: 0.169829 seconds, split_deferred: 2048
Time taken by reclamation: 0.172895 seconds, split_deferred: 2048
Time taken by reclamation: 0.176063 seconds, split_deferred: 2048
Time taken by reclamation: 0.172568 seconds, split_deferred: 2048
Time taken by reclamation: 0.171185 seconds, split_deferred: 2048
Time taken by reclamation: 0.170632 seconds, split_deferred: 2048
Time taken by reclamation: 0.170208 seconds, split_deferred: 2048
Time taken by reclamation: 0.174192 seconds, split_deferred: 2048
...
w/ patch:
~ # ./a.out
Time taken by reclamation: 0.074231 seconds, split_deferred: 0
Time taken by reclamation: 0.071026 seconds, split_deferred: 0
Time taken by reclamation: 0.072029 seconds, split_deferred: 0
Time taken by reclamation: 0.071873 seconds, split_deferred: 0
Time taken by reclamation: 0.073573 seconds, split_deferred: 0
Time taken by reclamation: 0.071906 seconds, split_deferred: 0
Time taken by reclamation: 0.073604 seconds, split_deferred: 0
Time taken by reclamation: 0.075903 seconds, split_deferred: 0
Time taken by reclamation: 0.073191 seconds, split_deferred: 0
Time taken by reclamation: 0.071228 seconds, split_deferred: 0
Time taken by reclamation: 0.071391 seconds, split_deferred: 0
Time taken by reclamation: 0.071468 seconds, split_deferred: 0
Time taken by reclamation: 0.071896 seconds, split_deferred: 0
Time taken by reclamation: 0.072508 seconds, split_deferred: 0
Time taken by reclamation: 0.071884 seconds, split_deferred: 0
Time taken by reclamation: 0.072433 seconds, split_deferred: 0
Time taken by reclamation: 0.071939 seconds, split_deferred: 0
...
Link: https://lkml.kernel.org/r/20250214093015.51024-4-21cnbao@gmail.com
Signed-off-by: Barry Song <v-songbaohua@oppo.com>
Cc: Albert Ou <aou@eecs.berkeley.edu>
Cc: Anshuman Khandual <anshuman.khandual@arm.com>
Cc: Baolin Wang <baolin.wang@linux.alibaba.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Chis Li <chrisl@kernel.org>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Gavin Shan <gshan@redhat.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: "Huang, Ying" <ying.huang@intel.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Kairui Song <kasong@tencent.com>
Cc: Kefeng Wang <wangkefeng.wang@huawei.com>
Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Lance Yang <ioworker0@gmail.com>
Cc: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Mauricio Faria de Oliveira <mfo@canonical.com>
Cc: Palmer Dabbelt <palmer@dabbelt.com>
Cc: Paul Walmsley <paul.walmsley@sifive.com>
Cc: Ryan Roberts <ryan.roberts@arm.com>
Cc: Shaoqin Huang <shahuang@redhat.com>
Cc: Tangquan Zheng <zhengtangquan@oppo.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Will Deacon <will@kernel.org>
Cc: Yicong Yang <yangyicong@hisilicon.com>
Cc: Yosry Ahmed <yosryahmed@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2025-02-14 22:30:14 +13:00
|
|
|
|
2025-06-27 14:23:19 +08:00
|
|
|
|
mm: support batched unmap for lazyfree large folios during reclamation
Currently, the PTEs and rmap of a large folio are removed one at a time.
This is not only slow but also causes the large folio to be unnecessarily
added to deferred_split, which can lead to races between the
deferred_split shrinker callback and memory reclamation. This patch
releases all PTEs and rmap entries in a batch. Currently, it only handles
lazyfree large folios.
The below microbench tries to reclaim 128MB lazyfree large folios
whose sizes are 64KiB:
#include <stdio.h>
#include <sys/mman.h>
#include <string.h>
#include <time.h>
#define SIZE 128*1024*1024 // 128 MB
unsigned long read_split_deferred()
{
FILE *file = fopen("/sys/kernel/mm/transparent_hugepage"
"/hugepages-64kB/stats/split_deferred", "r");
if (!file) {
perror("Error opening file");
return 0;
}
unsigned long value;
if (fscanf(file, "%lu", &value) != 1) {
perror("Error reading value");
fclose(file);
return 0;
}
fclose(file);
return value;
}
int main(int argc, char *argv[])
{
while(1) {
volatile int *p = mmap(0, SIZE, PROT_READ | PROT_WRITE,
MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
memset((void *)p, 1, SIZE);
madvise((void *)p, SIZE, MADV_FREE);
clock_t start_time = clock();
unsigned long start_split = read_split_deferred();
madvise((void *)p, SIZE, MADV_PAGEOUT);
clock_t end_time = clock();
unsigned long end_split = read_split_deferred();
double elapsed_time = (double)(end_time - start_time) / CLOCKS_PER_SEC;
printf("Time taken by reclamation: %f seconds, split_deferred: %ld\n",
elapsed_time, end_split - start_split);
munmap((void *)p, SIZE);
}
return 0;
}
w/o patch:
~ # ./a.out
Time taken by reclamation: 0.177418 seconds, split_deferred: 2048
Time taken by reclamation: 0.178348 seconds, split_deferred: 2048
Time taken by reclamation: 0.174525 seconds, split_deferred: 2048
Time taken by reclamation: 0.171620 seconds, split_deferred: 2048
Time taken by reclamation: 0.172241 seconds, split_deferred: 2048
Time taken by reclamation: 0.174003 seconds, split_deferred: 2048
Time taken by reclamation: 0.171058 seconds, split_deferred: 2048
Time taken by reclamation: 0.171993 seconds, split_deferred: 2048
Time taken by reclamation: 0.169829 seconds, split_deferred: 2048
Time taken by reclamation: 0.172895 seconds, split_deferred: 2048
Time taken by reclamation: 0.176063 seconds, split_deferred: 2048
Time taken by reclamation: 0.172568 seconds, split_deferred: 2048
Time taken by reclamation: 0.171185 seconds, split_deferred: 2048
Time taken by reclamation: 0.170632 seconds, split_deferred: 2048
Time taken by reclamation: 0.170208 seconds, split_deferred: 2048
Time taken by reclamation: 0.174192 seconds, split_deferred: 2048
...
w/ patch:
~ # ./a.out
Time taken by reclamation: 0.074231 seconds, split_deferred: 0
Time taken by reclamation: 0.071026 seconds, split_deferred: 0
Time taken by reclamation: 0.072029 seconds, split_deferred: 0
Time taken by reclamation: 0.071873 seconds, split_deferred: 0
Time taken by reclamation: 0.073573 seconds, split_deferred: 0
Time taken by reclamation: 0.071906 seconds, split_deferred: 0
Time taken by reclamation: 0.073604 seconds, split_deferred: 0
Time taken by reclamation: 0.075903 seconds, split_deferred: 0
Time taken by reclamation: 0.073191 seconds, split_deferred: 0
Time taken by reclamation: 0.071228 seconds, split_deferred: 0
Time taken by reclamation: 0.071391 seconds, split_deferred: 0
Time taken by reclamation: 0.071468 seconds, split_deferred: 0
Time taken by reclamation: 0.071896 seconds, split_deferred: 0
Time taken by reclamation: 0.072508 seconds, split_deferred: 0
Time taken by reclamation: 0.071884 seconds, split_deferred: 0
Time taken by reclamation: 0.072433 seconds, split_deferred: 0
Time taken by reclamation: 0.071939 seconds, split_deferred: 0
...
Link: https://lkml.kernel.org/r/20250214093015.51024-4-21cnbao@gmail.com
Signed-off-by: Barry Song <v-songbaohua@oppo.com>
Cc: Albert Ou <aou@eecs.berkeley.edu>
Cc: Anshuman Khandual <anshuman.khandual@arm.com>
Cc: Baolin Wang <baolin.wang@linux.alibaba.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Chis Li <chrisl@kernel.org>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Gavin Shan <gshan@redhat.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: "Huang, Ying" <ying.huang@intel.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Kairui Song <kasong@tencent.com>
Cc: Kefeng Wang <wangkefeng.wang@huawei.com>
Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Lance Yang <ioworker0@gmail.com>
Cc: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Mauricio Faria de Oliveira <mfo@canonical.com>
Cc: Palmer Dabbelt <palmer@dabbelt.com>
Cc: Paul Walmsley <paul.walmsley@sifive.com>
Cc: Ryan Roberts <ryan.roberts@arm.com>
Cc: Shaoqin Huang <shahuang@redhat.com>
Cc: Tangquan Zheng <zhengtangquan@oppo.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Will Deacon <will@kernel.org>
Cc: Yicong Yang <yangyicong@hisilicon.com>
Cc: Yosry Ahmed <yosryahmed@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2025-02-14 22:30:14 +13:00
|
|
|
|
2026-02-09 22:07:28 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
mm/rmap: fix incorrect pte restoration for lazyfree folios
We batch unmap anonymous lazyfree folios by folio_unmap_pte_batch. If the
batch has a mix of writable and non-writable bits, we may end up setting
the entire batch writable. Fix this by respecting writable bit during
batching.
Although on a successful unmap of a lazyfree folio, the soft-dirty bit is
lost, preserve it on pte restoration by respecting the bit during
batching, to make the fix consistent w.r.t both writable bit and
soft-dirty bit.
I was able to write the below reproducer and crash the kernel.
Explanation of reproducer (set 64K mTHP to always):
Fault in a 64K large folio. Split the VMA at mid-point with
MADV_DONTFORK. fork() - parent points to the folio with 8 writable ptes
and 8 non-writable ptes. Merge the VMAs with MADV_DOFORK so that
folio_unmap_pte_batch() can determine all the 16 ptes as a batch. Do
MADV_FREE on the range to mark the folio as lazyfree. Write to the memory
to dirty the pte, eventually rmap will dirty the folio. Then trigger
reclaim, we will hit the pte restoration path, and the kernel will crash
with the trace given below.
The BUG happens at:
BUG_ON(atomic_inc_return(&ptc->anon_map_count) > 1 && rw);
The code path is asking for anonymous page to be mapped writable into the
pagetable. The BUG_ON() firing implies that such a writable page has been
mapped into the pagetables of more than one process, which breaks
anonymous memory/CoW semantics.
[ 21.134473] kernel BUG at mm/page_table_check.c:118!
[ 21.134497] Internal error: Oops - BUG: 00000000f2000800 [#1] SMP
[ 21.135917] Modules linked in:
[ 21.136085] CPU: 1 UID: 0 PID: 1735 Comm: dup-lazyfree Not tainted 7.0.0-rc1-00116-g018018a17770 #1028 PREEMPT
[ 21.136858] Hardware name: linux,dummy-virt (DT)
[ 21.137019] pstate: 21400005 (nzCv daif +PAN -UAO -TCO +DIT -SSBS BTYPE=--)
[ 21.137308] pc : page_table_check_set+0x28c/0x2a8
[ 21.137607] lr : page_table_check_set+0x134/0x2a8
[ 21.137885] sp : ffff80008a3b3340
[ 21.138124] x29: ffff80008a3b3340 x28: fffffdffc3d14400 x27: ffffd1a55e03d000
[ 21.138623] x26: 0040000000000040 x25: ffffd1a55f7dd000 x24: 0000000000000001
[ 21.139045] x23: 0000000000000001 x22: 0000000000000001 x21: ffffd1a55f217f30
[ 21.139629] x20: 0000000000134521 x19: 0000000000134519 x18: 005c43e000040000
[ 21.140027] x17: 0001400000000000 x16: 0001700000000000 x15: 000000000000ffff
[ 21.140578] x14: 000000000000000c x13: 005c006000000000 x12: 0000000000000020
[ 21.140828] x11: 0000000000000000 x10: 005c000000000000 x9 : ffffd1a55c079ee0
[ 21.141077] x8 : 0000000000000001 x7 : 005c03e000040000 x6 : 000000004000ffff
[ 21.141490] x5 : ffff00017fffce00 x4 : 0000000000000001 x3 : 0000000000000002
[ 21.141741] x2 : 0000000000134510 x1 : 0000000000000000 x0 : ffff0000c08228c0
[ 21.141991] Call trace:
[ 21.142093] page_table_check_set+0x28c/0x2a8 (P)
[ 21.142265] __page_table_check_ptes_set+0x144/0x1e8
[ 21.142441] __set_ptes_anysz.constprop.0+0x160/0x1a8
[ 21.142766] contpte_set_ptes+0xe8/0x140
[ 21.142907] try_to_unmap_one+0x10c4/0x10d0
[ 21.143177] rmap_walk_anon+0x100/0x250
[ 21.143315] try_to_unmap+0xa0/0xc8
[ 21.143441] shrink_folio_list+0x59c/0x18a8
[ 21.143759] shrink_lruvec+0x664/0xbf0
[ 21.144043] shrink_node+0x218/0x878
[ 21.144285] __node_reclaim.constprop.0+0x98/0x338
[ 21.144763] user_proactive_reclaim+0x2a4/0x340
[ 21.145056] reclaim_store+0x3c/0x60
[ 21.145216] dev_attr_store+0x20/0x40
[ 21.145585] sysfs_kf_write+0x84/0xa8
[ 21.145835] kernfs_fop_write_iter+0x130/0x1c8
[ 21.145994] vfs_write+0x2b8/0x368
[ 21.146119] ksys_write+0x70/0x110
[ 21.146240] __arm64_sys_write+0x24/0x38
[ 21.146380] invoke_syscall+0x50/0x120
[ 21.146513] el0_svc_common.constprop.0+0x48/0xf8
[ 21.146679] do_el0_svc+0x28/0x40
[ 21.146798] el0_svc+0x34/0x110
[ 21.146926] el0t_64_sync_handler+0xa0/0xe8
[ 21.147074] el0t_64_sync+0x198/0x1a0
[ 21.147225] Code: f9400441 b4fff241 17ffff94 d4210000 (d4210000)
[ 21.147440] ---[ end trace 0000000000000000 ]---
#define _GNU_SOURCE
#include <stdio.h>
#include <unistd.h>
#include <stdlib.h>
#include <sys/mman.h>
#include <string.h>
#include <sys/wait.h>
#include <sched.h>
#include <fcntl.h>
void write_to_reclaim() {
const char *path = "/sys/devices/system/node/node0/reclaim";
const char *value = "409600000000";
int fd = open(path, O_WRONLY);
if (fd == -1) {
perror("open");
exit(EXIT_FAILURE);
}
if (write(fd, value, sizeof("409600000000") - 1) == -1) {
perror("write");
close(fd);
exit(EXIT_FAILURE);
}
printf("Successfully wrote %s to %s\n", value, path);
close(fd);
}
int main()
{
char *ptr = mmap((void *)(1UL << 30), 1UL << 16, PROT_READ | PROT_WRITE,
MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
if ((unsigned long)ptr != (1UL << 30)) {
perror("mmap");
return 1;
}
/* a 64K folio gets faulted in */
memset(ptr, 0, 1UL << 16);
/* 32K half will not be shared into child */
if (madvise(ptr, 1UL << 15, MADV_DONTFORK)) {
perror("madvise madv dontfork");
return 1;
}
pid_t pid = fork();
if (pid < 0) {
perror("fork");
return 1;
} else if (pid == 0) {
sleep(15);
} else {
/* merge VMAs. now first half of the 16 ptes are writable, the other half not. */
if (madvise(ptr, 1UL << 15, MADV_DOFORK)) {
perror("madvise madv fork");
return 1;
}
if (madvise(ptr, (1UL << 16), MADV_FREE)) {
perror("madvise madv free");
return 1;
}
/* dirty the large folio */
(*ptr) += 10;
write_to_reclaim();
// sleep(10);
waitpid(pid, NULL, 0);
}
}
Link: https://lkml.kernel.org/r/20260303061528.2429162-1-dev.jain@arm.com
Fixes: 354dffd29575 ("mm: support batched unmap for lazyfree large folios during reclamation")
Signed-off-by: Dev Jain <dev.jain@arm.com>
Acked-by: David Hildenbrand (Arm) <david@kernel.org>
Reviewed-by: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Reviewed-by: Barry Song <baohua@kernel.org>
Reviewed-by: Wei Yang <richard.weiyang@gmail.com>
Tested-by: Lance Yang <lance.yang@linux.dev>
Cc: Anshuman Khandual <anshuman.khandual@arm.com>
Cc: Harry Yoo <harry.yoo@oracle.com>
Cc: Jann Horn <jannh@google.com>
Cc: Liam Howlett <liam.howlett@oracle.com>
Cc: Rik van Riel <riel@surriel.com>
Cc: Ryan Roberts <ryan.roberts@arm.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2026-03-03 11:45:28 +05:30
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
mm: support batched unmap for lazyfree large folios during reclamation
Currently, the PTEs and rmap of a large folio are removed one at a time.
This is not only slow but also causes the large folio to be unnecessarily
added to deferred_split, which can lead to races between the
deferred_split shrinker callback and memory reclamation. This patch
releases all PTEs and rmap entries in a batch. Currently, it only handles
lazyfree large folios.
The below microbench tries to reclaim 128MB lazyfree large folios
whose sizes are 64KiB:
#include <stdio.h>
#include <sys/mman.h>
#include <string.h>
#include <time.h>
#define SIZE 128*1024*1024 // 128 MB
unsigned long read_split_deferred()
{
FILE *file = fopen("/sys/kernel/mm/transparent_hugepage"
"/hugepages-64kB/stats/split_deferred", "r");
if (!file) {
perror("Error opening file");
return 0;
}
unsigned long value;
if (fscanf(file, "%lu", &value) != 1) {
perror("Error reading value");
fclose(file);
return 0;
}
fclose(file);
return value;
}
int main(int argc, char *argv[])
{
while(1) {
volatile int *p = mmap(0, SIZE, PROT_READ | PROT_WRITE,
MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
memset((void *)p, 1, SIZE);
madvise((void *)p, SIZE, MADV_FREE);
clock_t start_time = clock();
unsigned long start_split = read_split_deferred();
madvise((void *)p, SIZE, MADV_PAGEOUT);
clock_t end_time = clock();
unsigned long end_split = read_split_deferred();
double elapsed_time = (double)(end_time - start_time) / CLOCKS_PER_SEC;
printf("Time taken by reclamation: %f seconds, split_deferred: %ld\n",
elapsed_time, end_split - start_split);
munmap((void *)p, SIZE);
}
return 0;
}
w/o patch:
~ # ./a.out
Time taken by reclamation: 0.177418 seconds, split_deferred: 2048
Time taken by reclamation: 0.178348 seconds, split_deferred: 2048
Time taken by reclamation: 0.174525 seconds, split_deferred: 2048
Time taken by reclamation: 0.171620 seconds, split_deferred: 2048
Time taken by reclamation: 0.172241 seconds, split_deferred: 2048
Time taken by reclamation: 0.174003 seconds, split_deferred: 2048
Time taken by reclamation: 0.171058 seconds, split_deferred: 2048
Time taken by reclamation: 0.171993 seconds, split_deferred: 2048
Time taken by reclamation: 0.169829 seconds, split_deferred: 2048
Time taken by reclamation: 0.172895 seconds, split_deferred: 2048
Time taken by reclamation: 0.176063 seconds, split_deferred: 2048
Time taken by reclamation: 0.172568 seconds, split_deferred: 2048
Time taken by reclamation: 0.171185 seconds, split_deferred: 2048
Time taken by reclamation: 0.170632 seconds, split_deferred: 2048
Time taken by reclamation: 0.170208 seconds, split_deferred: 2048
Time taken by reclamation: 0.174192 seconds, split_deferred: 2048
...
w/ patch:
~ # ./a.out
Time taken by reclamation: 0.074231 seconds, split_deferred: 0
Time taken by reclamation: 0.071026 seconds, split_deferred: 0
Time taken by reclamation: 0.072029 seconds, split_deferred: 0
Time taken by reclamation: 0.071873 seconds, split_deferred: 0
Time taken by reclamation: 0.073573 seconds, split_deferred: 0
Time taken by reclamation: 0.071906 seconds, split_deferred: 0
Time taken by reclamation: 0.073604 seconds, split_deferred: 0
Time taken by reclamation: 0.075903 seconds, split_deferred: 0
Time taken by reclamation: 0.073191 seconds, split_deferred: 0
Time taken by reclamation: 0.071228 seconds, split_deferred: 0
Time taken by reclamation: 0.071391 seconds, split_deferred: 0
Time taken by reclamation: 0.071468 seconds, split_deferred: 0
Time taken by reclamation: 0.071896 seconds, split_deferred: 0
Time taken by reclamation: 0.072508 seconds, split_deferred: 0
Time taken by reclamation: 0.071884 seconds, split_deferred: 0
Time taken by reclamation: 0.072433 seconds, split_deferred: 0
Time taken by reclamation: 0.071939 seconds, split_deferred: 0
...
Link: https://lkml.kernel.org/r/20250214093015.51024-4-21cnbao@gmail.com
Signed-off-by: Barry Song <v-songbaohua@oppo.com>
Cc: Albert Ou <aou@eecs.berkeley.edu>
Cc: Anshuman Khandual <anshuman.khandual@arm.com>
Cc: Baolin Wang <baolin.wang@linux.alibaba.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Chis Li <chrisl@kernel.org>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Gavin Shan <gshan@redhat.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: "Huang, Ying" <ying.huang@intel.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Kairui Song <kasong@tencent.com>
Cc: Kefeng Wang <wangkefeng.wang@huawei.com>
Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Lance Yang <ioworker0@gmail.com>
Cc: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Mauricio Faria de Oliveira <mfo@canonical.com>
Cc: Palmer Dabbelt <palmer@dabbelt.com>
Cc: Paul Walmsley <paul.walmsley@sifive.com>
Cc: Ryan Roberts <ryan.roberts@arm.com>
Cc: Shaoqin Huang <shahuang@redhat.com>
Cc: Tangquan Zheng <zhengtangquan@oppo.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Will Deacon <will@kernel.org>
Cc: Yicong Yang <yangyicong@hisilicon.com>
Cc: Yosry Ahmed <yosryahmed@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2025-02-14 22:30:14 +13:00
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2014-01-21 15:49:50 -08:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2022-01-29 16:06:53 -05:00
|
|
|
|
2014-01-21 15:49:50 -08:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
2022-02-15 09:28:49 -05:00
|
|
|
|
2025-02-10 20:37:52 +01:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2017-02-24 14:58:01 -08:00
|
|
|
|
2018-12-28 00:38:09 -08:00
|
|
|
|
2020-04-06 20:08:00 -07:00
|
|
|
|
mm: support batched unmap for lazyfree large folios during reclamation
Currently, the PTEs and rmap of a large folio are removed one at a time.
This is not only slow but also causes the large folio to be unnecessarily
added to deferred_split, which can lead to races between the
deferred_split shrinker callback and memory reclamation. This patch
releases all PTEs and rmap entries in a batch. Currently, it only handles
lazyfree large folios.
The below microbench tries to reclaim 128MB lazyfree large folios
whose sizes are 64KiB:
#include <stdio.h>
#include <sys/mman.h>
#include <string.h>
#include <time.h>
#define SIZE 128*1024*1024 // 128 MB
unsigned long read_split_deferred()
{
FILE *file = fopen("/sys/kernel/mm/transparent_hugepage"
"/hugepages-64kB/stats/split_deferred", "r");
if (!file) {
perror("Error opening file");
return 0;
}
unsigned long value;
if (fscanf(file, "%lu", &value) != 1) {
perror("Error reading value");
fclose(file);
return 0;
}
fclose(file);
return value;
}
int main(int argc, char *argv[])
{
while(1) {
volatile int *p = mmap(0, SIZE, PROT_READ | PROT_WRITE,
MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
memset((void *)p, 1, SIZE);
madvise((void *)p, SIZE, MADV_FREE);
clock_t start_time = clock();
unsigned long start_split = read_split_deferred();
madvise((void *)p, SIZE, MADV_PAGEOUT);
clock_t end_time = clock();
unsigned long end_split = read_split_deferred();
double elapsed_time = (double)(end_time - start_time) / CLOCKS_PER_SEC;
printf("Time taken by reclamation: %f seconds, split_deferred: %ld\n",
elapsed_time, end_split - start_split);
munmap((void *)p, SIZE);
}
return 0;
}
w/o patch:
~ # ./a.out
Time taken by reclamation: 0.177418 seconds, split_deferred: 2048
Time taken by reclamation: 0.178348 seconds, split_deferred: 2048
Time taken by reclamation: 0.174525 seconds, split_deferred: 2048
Time taken by reclamation: 0.171620 seconds, split_deferred: 2048
Time taken by reclamation: 0.172241 seconds, split_deferred: 2048
Time taken by reclamation: 0.174003 seconds, split_deferred: 2048
Time taken by reclamation: 0.171058 seconds, split_deferred: 2048
Time taken by reclamation: 0.171993 seconds, split_deferred: 2048
Time taken by reclamation: 0.169829 seconds, split_deferred: 2048
Time taken by reclamation: 0.172895 seconds, split_deferred: 2048
Time taken by reclamation: 0.176063 seconds, split_deferred: 2048
Time taken by reclamation: 0.172568 seconds, split_deferred: 2048
Time taken by reclamation: 0.171185 seconds, split_deferred: 2048
Time taken by reclamation: 0.170632 seconds, split_deferred: 2048
Time taken by reclamation: 0.170208 seconds, split_deferred: 2048
Time taken by reclamation: 0.174192 seconds, split_deferred: 2048
...
w/ patch:
~ # ./a.out
Time taken by reclamation: 0.074231 seconds, split_deferred: 0
Time taken by reclamation: 0.071026 seconds, split_deferred: 0
Time taken by reclamation: 0.072029 seconds, split_deferred: 0
Time taken by reclamation: 0.071873 seconds, split_deferred: 0
Time taken by reclamation: 0.073573 seconds, split_deferred: 0
Time taken by reclamation: 0.071906 seconds, split_deferred: 0
Time taken by reclamation: 0.073604 seconds, split_deferred: 0
Time taken by reclamation: 0.075903 seconds, split_deferred: 0
Time taken by reclamation: 0.073191 seconds, split_deferred: 0
Time taken by reclamation: 0.071228 seconds, split_deferred: 0
Time taken by reclamation: 0.071391 seconds, split_deferred: 0
Time taken by reclamation: 0.071468 seconds, split_deferred: 0
Time taken by reclamation: 0.071896 seconds, split_deferred: 0
Time taken by reclamation: 0.072508 seconds, split_deferred: 0
Time taken by reclamation: 0.071884 seconds, split_deferred: 0
Time taken by reclamation: 0.072433 seconds, split_deferred: 0
Time taken by reclamation: 0.071939 seconds, split_deferred: 0
...
Link: https://lkml.kernel.org/r/20250214093015.51024-4-21cnbao@gmail.com
Signed-off-by: Barry Song <v-songbaohua@oppo.com>
Cc: Albert Ou <aou@eecs.berkeley.edu>
Cc: Anshuman Khandual <anshuman.khandual@arm.com>
Cc: Baolin Wang <baolin.wang@linux.alibaba.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Chis Li <chrisl@kernel.org>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Gavin Shan <gshan@redhat.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: "Huang, Ying" <ying.huang@intel.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Kairui Song <kasong@tencent.com>
Cc: Kefeng Wang <wangkefeng.wang@huawei.com>
Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Lance Yang <ioworker0@gmail.com>
Cc: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Mauricio Faria de Oliveira <mfo@canonical.com>
Cc: Palmer Dabbelt <palmer@dabbelt.com>
Cc: Paul Walmsley <paul.walmsley@sifive.com>
Cc: Ryan Roberts <ryan.roberts@arm.com>
Cc: Shaoqin Huang <shahuang@redhat.com>
Cc: Tangquan Zheng <zhengtangquan@oppo.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Will Deacon <will@kernel.org>
Cc: Yicong Yang <yangyicong@hisilicon.com>
Cc: Yosry Ahmed <yosryahmed@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2025-02-14 22:30:14 +13:00
|
|
|
|
mm: ptep_get() conversion
Convert all instances of direct pte_t* dereferencing to instead use
ptep_get() helper. This means that by default, the accesses change from a
C dereference to a READ_ONCE(). This is technically the correct thing to
do since where pgtables are modified by HW (for access/dirty) they are
volatile and therefore we should always ensure READ_ONCE() semantics.
But more importantly, by always using the helper, it can be overridden by
the architecture to fully encapsulate the contents of the pte. Arch code
is deliberately not converted, as the arch code knows best. It is
intended that arch code (arm64) will override the default with its own
implementation that can (e.g.) hide certain bits from the core code, or
determine young/dirty status by mixing in state from another source.
Conversion was done using Coccinelle:
----
// $ make coccicheck \
// COCCI=ptepget.cocci \
// SPFLAGS="--include-headers" \
// MODE=patch
virtual patch
@ depends on patch @
pte_t *v;
@@
- *v
+ ptep_get(v)
----
Then reviewed and hand-edited to avoid multiple unnecessary calls to
ptep_get(), instead opting to store the result of a single call in a
variable, where it is correct to do so. This aims to negate any cost of
READ_ONCE() and will benefit arch-overrides that may be more complex.
Included is a fix for an issue in an earlier version of this patch that
was pointed out by kernel test robot. The issue arose because config
MMU=n elides definition of the ptep helper functions, including
ptep_get(). HUGETLB_PAGE=n configs still define a simple
huge_ptep_clear_flush() for linking purposes, which dereferences the ptep.
So when both configs are disabled, this caused a build error because
ptep_get() is not defined. Fix by continuing to do a direct dereference
when MMU=n. This is safe because for this config the arch code cannot be
trying to virtualize the ptes because none of the ptep helpers are
defined.
Link: https://lkml.kernel.org/r/20230612151545.3317766-4-ryan.roberts@arm.com
Reported-by: kernel test robot <lkp@intel.com>
Link: https://lore.kernel.org/oe-kbuild-all/202305120142.yXsNEo6H-lkp@intel.com/
Signed-off-by: Ryan Roberts <ryan.roberts@arm.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Alexander Potapenko <glider@google.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Alex Williamson <alex.williamson@redhat.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Andrey Konovalov <andreyknvl@gmail.com>
Cc: Andrey Ryabinin <ryabinin.a.a@gmail.com>
Cc: Christian Brauner <brauner@kernel.org>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: Daniel Vetter <daniel@ffwll.ch>
Cc: Dave Airlie <airlied@gmail.com>
Cc: Dimitri Sivanich <dimitri.sivanich@hpe.com>
Cc: Dmitry Vyukov <dvyukov@google.com>
Cc: Ian Rogers <irogers@google.com>
Cc: Jason Gunthorpe <jgg@ziepe.ca>
Cc: Jérôme Glisse <jglisse@redhat.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Lorenzo Stoakes <lstoakes@gmail.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Miaohe Lin <linmiaohe@huawei.com>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Mike Rapoport (IBM) <rppt@kernel.org>
Cc: Muchun Song <muchun.song@linux.dev>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Naoya Horiguchi <naoya.horiguchi@nec.com>
Cc: Oleksandr Tyshchenko <oleksandr_tyshchenko@epam.com>
Cc: Pavel Tatashin <pasha.tatashin@soleen.com>
Cc: Roman Gushchin <roman.gushchin@linux.dev>
Cc: SeongJae Park <sj@kernel.org>
Cc: Shakeel Butt <shakeelb@google.com>
Cc: Uladzislau Rezki (Sony) <urezki@gmail.com>
Cc: Vincenzo Frascino <vincenzo.frascino@arm.com>
Cc: Yu Zhao <yuzhao@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2023-06-12 16:15:45 +01:00
|
|
|
|
mm: hugetlb: add huge page size param to set_huge_pte_at()
Patch series "Fix set_huge_pte_at() panic on arm64", v2.
This series fixes a bug in arm64's implementation of set_huge_pte_at(),
which can result in an unprivileged user causing a kernel panic. The
problem was triggered when running the new uffd poison mm selftest for
HUGETLB memory. This test (and the uffd poison feature) was merged for
v6.5-rc7.
Ideally, I'd like to get this fix in for v6.6 and I've cc'ed stable
(correctly this time) to get it backported to v6.5, where the issue first
showed up.
Description of Bug
==================
arm64's huge pte implementation supports multiple huge page sizes, some of
which are implemented in the page table with multiple contiguous entries.
So set_huge_pte_at() needs to work out how big the logical pte is, so that
it can also work out how many physical ptes (or pmds) need to be written.
It previously did this by grabbing the folio out of the pte and querying
its size.
However, there are cases when the pte being set is actually a swap entry.
But this also used to work fine, because for huge ptes, we only ever saw
migration entries and hwpoison entries. And both of these types of swap
entries have a PFN embedded, so the code would grab that and everything
still worked out.
But over time, more calls to set_huge_pte_at() have been added that set
swap entry types that do not embed a PFN. And this causes the code to go
bang. The triggering case is for the uffd poison test, commit
99aa77215ad0 ("selftests/mm: add uffd unit test for UFFDIO_POISON"), which
causes a PTE_MARKER_POISONED swap entry to be set, coutesey of commit
8a13897fb0da ("mm: userfaultfd: support UFFDIO_POISON for hugetlbfs") -
added in v6.5-rc7. Although review shows that there are other call sites
that set PTE_MARKER_UFFD_WP (which also has no PFN), these don't trigger
on arm64 because arm64 doesn't support UFFD WP.
If CONFIG_DEBUG_VM is enabled, we do at least get a BUG(), but otherwise,
it will dereference a bad pointer in page_folio():
static inline struct folio *hugetlb_swap_entry_to_folio(swp_entry_t entry)
{
VM_BUG_ON(!is_migration_entry(entry) && !is_hwpoison_entry(entry));
return page_folio(pfn_to_page(swp_offset_pfn(entry)));
}
Fix
===
The simplest fix would have been to revert the dodgy cleanup commit
18f3962953e4 ("mm: hugetlb: kill set_huge_swap_pte_at()"), but since
things have moved on, this would have required an audit of all the new
set_huge_pte_at() call sites to see if they should be converted to
set_huge_swap_pte_at(). As per the original intent of the change, it
would also leave us open to future bugs when people invariably get it
wrong and call the wrong helper.
So instead, I've added a huge page size parameter to set_huge_pte_at().
This means that the arm64 code has the size in all cases. It's a bigger
change, due to needing to touch the arches that implement the function,
but it is entirely mechanical, so in my view, low risk.
I've compile-tested all touched arches; arm64, parisc, powerpc, riscv,
s390, sparc (and additionally x86_64). I've additionally booted and run
mm selftests against arm64, where I observe the uffd poison test is fixed,
and there are no other regressions.
This patch (of 2):
In order to fix a bug, arm64 needs to be told the size of the huge page
for which the pte is being set in set_huge_pte_at(). Provide for this by
adding an `unsigned long sz` parameter to the function. This follows the
same pattern as huge_pte_clear().
This commit makes the required interface modifications to the core mm as
well as all arches that implement this function (arm64, parisc, powerpc,
riscv, s390, sparc). The actual arm64 bug will be fixed in a separate
commit.
No behavioral changes intended.
Link: https://lkml.kernel.org/r/20230922115804.2043771-1-ryan.roberts@arm.com
Link: https://lkml.kernel.org/r/20230922115804.2043771-2-ryan.roberts@arm.com
Fixes: 8a13897fb0da ("mm: userfaultfd: support UFFDIO_POISON for hugetlbfs")
Signed-off-by: Ryan Roberts <ryan.roberts@arm.com>
Reviewed-by: Christophe Leroy <christophe.leroy@csgroup.eu> [powerpc 8xx]
Reviewed-by: Lorenzo Stoakes <lstoakes@gmail.com> [vmalloc change]
Cc: Alexandre Ghiti <alex@ghiti.fr>
Cc: Albert Ou <aou@eecs.berkeley.edu>
Cc: Alexander Gordeev <agordeev@linux.ibm.com>
Cc: Anshuman Khandual <anshuman.khandual@arm.com>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Axel Rasmussen <axelrasmussen@google.com>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Christian Borntraeger <borntraeger@linux.ibm.com>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: David S. Miller <davem@davemloft.net>
Cc: Gerald Schaefer <gerald.schaefer@linux.ibm.com>
Cc: Heiko Carstens <hca@linux.ibm.com>
Cc: Helge Deller <deller@gmx.de>
Cc: "James E.J. Bottomley" <James.Bottomley@HansenPartnership.com>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Muchun Song <muchun.song@linux.dev>
Cc: Nicholas Piggin <npiggin@gmail.com>
Cc: Palmer Dabbelt <palmer@dabbelt.com>
Cc: Paul Walmsley <paul.walmsley@sifive.com>
Cc: Peter Xu <peterx@redhat.com>
Cc: Qi Zheng <zhengqi.arch@bytedance.com>
Cc: Ryan Roberts <ryan.roberts@arm.com>
Cc: SeongJae Park <sj@kernel.org>
Cc: Sven Schnelle <svens@linux.ibm.com>
Cc: Uladzislau Rezki (Sony) <urezki@gmail.com>
Cc: Vasily Gorbik <gor@linux.ibm.com>
Cc: Will Deacon <will@kernel.org>
Cc: <stable@vger.kernel.org> [6.5+]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2023-09-22 12:58:03 +01:00
|
|
|
|
2025-09-23 12:07:08 +01:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
mm/thp: try_to_unmap() use TTU_SYNC for safe splitting
Stressing huge tmpfs often crashed on unmap_page()'s VM_BUG_ON_PAGE
(!unmap_success): with dump_page() showing mapcount:1, but then its raw
struct page output showing _mapcount ffffffff i.e. mapcount 0.
And even if that particular VM_BUG_ON_PAGE(!unmap_success) is removed,
it is immediately followed by a VM_BUG_ON_PAGE(compound_mapcount(head)),
and further down an IS_ENABLED(CONFIG_DEBUG_VM) total_mapcount BUG():
all indicative of some mapcount difficulty in development here perhaps.
But the !CONFIG_DEBUG_VM path handles the failures correctly and
silently.
I believe the problem is that once a racing unmap has cleared pte or
pmd, try_to_unmap_one() may skip taking the page table lock, and emerge
from try_to_unmap() before the racing task has reached decrementing
mapcount.
Instead of abandoning the unsafe VM_BUG_ON_PAGE(), and the ones that
follow, use PVMW_SYNC in try_to_unmap_one() in this case: adding
TTU_SYNC to the options, and passing that from unmap_page().
When CONFIG_DEBUG_VM, or for non-debug too? Consensus is to do the same
for both: the slight overhead added should rarely matter, except perhaps
if splitting sparsely-populated multiply-mapped shmem. Once confident
that bugs are fixed, TTU_SYNC here can be removed, and the race
tolerated.
Link: https://lkml.kernel.org/r/c1e95853-8bcd-d8fd-55fa-e7f2488e78f@google.com
Fixes: fec89c109f3a ("thp: rewrite freeze_page()/unfreeze_page() with generic rmap walkers")
Signed-off-by: Hugh Dickins <hughd@google.com>
Cc: Alistair Popple <apopple@nvidia.com>
Cc: Jan Kara <jack@suse.cz>
Cc: Jue Wang <juew@google.com>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: "Matthew Wilcox (Oracle)" <willy@infradead.org>
Cc: Miaohe Lin <linmiaohe@huawei.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Naoya Horiguchi <naoya.horiguchi@nec.com>
Cc: Oscar Salvador <osalvador@suse.de>
Cc: Peter Xu <peterx@redhat.com>
Cc: Ralph Campbell <rcampbell@nvidia.com>
Cc: Shakeel Butt <shakeelb@google.com>
Cc: Wang Yugui <wangyugui@e16-tech.com>
Cc: Yang Shi <shy828301@gmail.com>
Cc: Zi Yan <ziy@nvidia.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2021-06-15 18:23:53 -07:00
|
|
|
|
|
|
|
|
|
2023-12-20 23:44:54 +01:00
|
|
|
|
2026-04-27 13:43:16 +02:00
|
|
|
|
mm/thp: try_to_unmap() use TTU_SYNC for safe splitting
Stressing huge tmpfs often crashed on unmap_page()'s VM_BUG_ON_PAGE
(!unmap_success): with dump_page() showing mapcount:1, but then its raw
struct page output showing _mapcount ffffffff i.e. mapcount 0.
And even if that particular VM_BUG_ON_PAGE(!unmap_success) is removed,
it is immediately followed by a VM_BUG_ON_PAGE(compound_mapcount(head)),
and further down an IS_ENABLED(CONFIG_DEBUG_VM) total_mapcount BUG():
all indicative of some mapcount difficulty in development here perhaps.
But the !CONFIG_DEBUG_VM path handles the failures correctly and
silently.
I believe the problem is that once a racing unmap has cleared pte or
pmd, try_to_unmap_one() may skip taking the page table lock, and emerge
from try_to_unmap() before the racing task has reached decrementing
mapcount.
Instead of abandoning the unsafe VM_BUG_ON_PAGE(), and the ones that
follow, use PVMW_SYNC in try_to_unmap_one() in this case: adding
TTU_SYNC to the options, and passing that from unmap_page().
When CONFIG_DEBUG_VM, or for non-debug too? Consensus is to do the same
for both: the slight overhead added should rarely matter, except perhaps
if splitting sparsely-populated multiply-mapped shmem. Once confident
that bugs are fixed, TTU_SYNC here can be removed, and the race
tolerated.
Link: https://lkml.kernel.org/r/c1e95853-8bcd-d8fd-55fa-e7f2488e78f@google.com
Fixes: fec89c109f3a ("thp: rewrite freeze_page()/unfreeze_page() with generic rmap walkers")
Signed-off-by: Hugh Dickins <hughd@google.com>
Cc: Alistair Popple <apopple@nvidia.com>
Cc: Jan Kara <jack@suse.cz>
Cc: Jue Wang <juew@google.com>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: "Matthew Wilcox (Oracle)" <willy@infradead.org>
Cc: Miaohe Lin <linmiaohe@huawei.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Naoya Horiguchi <naoya.horiguchi@nec.com>
Cc: Oscar Salvador <osalvador@suse.de>
Cc: Peter Xu <peterx@redhat.com>
Cc: Ralph Campbell <rcampbell@nvidia.com>
Cc: Shakeel Butt <shakeelb@google.com>
Cc: Wang Yugui <wangyugui@e16-tech.com>
Cc: Yang Shi <shy828301@gmail.com>
Cc: Zi Yan <ziy@nvidia.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2021-06-15 18:23:53 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-08-31 17:17:27 -04:00
|
|
|
|
2018-10-05 15:51:29 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-02-15 09:28:49 -05:00
|
|
|
|
|
|
|
|
|
2017-08-31 17:17:27 -04:00
|
|
|
|
2022-02-03 11:40:17 -05:00
|
|
|
|
2023-01-10 13:57:22 +11:00
|
|
|
|
mm/thp: fix vma_address() if virtual address below file offset
Running certain tests with a DEBUG_VM kernel would crash within hours,
on the total_mapcount BUG() in split_huge_page_to_list(), while trying
to free up some memory by punching a hole in a shmem huge page: split's
try_to_unmap() was unable to find all the mappings of the page (which,
on a !DEBUG_VM kernel, would then keep the huge page pinned in memory).
When that BUG() was changed to a WARN(), it would later crash on the
VM_BUG_ON_VMA(end < vma->vm_start || start >= vma->vm_end, vma) in
mm/internal.h:vma_address(), used by rmap_walk_file() for
try_to_unmap().
vma_address() is usually correct, but there's a wraparound case when the
vm_start address is unusually low, but vm_pgoff not so low:
vma_address() chooses max(start, vma->vm_start), but that decides on the
wrong address, because start has become almost ULONG_MAX.
Rewrite vma_address() to be more careful about vm_pgoff; move the
VM_BUG_ON_VMA() out of it, returning -EFAULT for errors, so that it can
be safely used from page_mapped_in_vma() and page_address_in_vma() too.
Add vma_address_end() to apply similar care to end address calculation,
in page_vma_mapped_walk() and page_mkclean_one() and try_to_unmap_one();
though it raises a question of whether callers would do better to supply
pvmw->end to page_vma_mapped_walk() - I chose not, for a smaller patch.
An irritation is that their apparent generality breaks down on KSM
pages, which cannot be located by the page->index that page_to_pgoff()
uses: as commit 4b0ece6fa016 ("mm: migrate: fix remove_migration_pte()
for ksm pages") once discovered. I dithered over the best thing to do
about that, and have ended up with a VM_BUG_ON_PAGE(PageKsm) in both
vma_address() and vma_address_end(); though the only place in danger of
using it on them was try_to_unmap_one().
Sidenote: vma_address() and vma_address_end() now use compound_nr() on a
head page, instead of thp_size(): to make the right calculation on a
hugetlbfs page, whether or not THPs are configured. try_to_unmap() is
used on hugetlbfs pages, but perhaps the wrong calculation never
mattered.
Link: https://lkml.kernel.org/r/caf1c1a3-7cfb-7f8f-1beb-ba816e932825@google.com
Fixes: a8fa41ad2f6f ("mm, rmap: check all VMAs that PTE-mapped THP can be part of")
Signed-off-by: Hugh Dickins <hughd@google.com>
Acked-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Alistair Popple <apopple@nvidia.com>
Cc: Jan Kara <jack@suse.cz>
Cc: Jue Wang <juew@google.com>
Cc: "Matthew Wilcox (Oracle)" <willy@infradead.org>
Cc: Miaohe Lin <linmiaohe@huawei.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Naoya Horiguchi <naoya.horiguchi@nec.com>
Cc: Oscar Salvador <osalvador@suse.de>
Cc: Peter Xu <peterx@redhat.com>
Cc: Ralph Campbell <rcampbell@nvidia.com>
Cc: Shakeel Butt <shakeelb@google.com>
Cc: Wang Yugui <wangyugui@e16-tech.com>
Cc: Yang Shi <shy828301@gmail.com>
Cc: Zi Yan <ziy@nvidia.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2021-06-15 18:23:56 -07:00
|
|
|
|
2022-02-15 09:28:49 -05:00
|
|
|
|
2018-10-05 15:51:29 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-12-28 00:38:09 -08:00
|
|
|
|
|
|
|
|
|
mm: hugetlb: add huge page size param to set_huge_pte_at()
Patch series "Fix set_huge_pte_at() panic on arm64", v2.
This series fixes a bug in arm64's implementation of set_huge_pte_at(),
which can result in an unprivileged user causing a kernel panic. The
problem was triggered when running the new uffd poison mm selftest for
HUGETLB memory. This test (and the uffd poison feature) was merged for
v6.5-rc7.
Ideally, I'd like to get this fix in for v6.6 and I've cc'ed stable
(correctly this time) to get it backported to v6.5, where the issue first
showed up.
Description of Bug
==================
arm64's huge pte implementation supports multiple huge page sizes, some of
which are implemented in the page table with multiple contiguous entries.
So set_huge_pte_at() needs to work out how big the logical pte is, so that
it can also work out how many physical ptes (or pmds) need to be written.
It previously did this by grabbing the folio out of the pte and querying
its size.
However, there are cases when the pte being set is actually a swap entry.
But this also used to work fine, because for huge ptes, we only ever saw
migration entries and hwpoison entries. And both of these types of swap
entries have a PFN embedded, so the code would grab that and everything
still worked out.
But over time, more calls to set_huge_pte_at() have been added that set
swap entry types that do not embed a PFN. And this causes the code to go
bang. The triggering case is for the uffd poison test, commit
99aa77215ad0 ("selftests/mm: add uffd unit test for UFFDIO_POISON"), which
causes a PTE_MARKER_POISONED swap entry to be set, coutesey of commit
8a13897fb0da ("mm: userfaultfd: support UFFDIO_POISON for hugetlbfs") -
added in v6.5-rc7. Although review shows that there are other call sites
that set PTE_MARKER_UFFD_WP (which also has no PFN), these don't trigger
on arm64 because arm64 doesn't support UFFD WP.
If CONFIG_DEBUG_VM is enabled, we do at least get a BUG(), but otherwise,
it will dereference a bad pointer in page_folio():
static inline struct folio *hugetlb_swap_entry_to_folio(swp_entry_t entry)
{
VM_BUG_ON(!is_migration_entry(entry) && !is_hwpoison_entry(entry));
return page_folio(pfn_to_page(swp_offset_pfn(entry)));
}
Fix
===
The simplest fix would have been to revert the dodgy cleanup commit
18f3962953e4 ("mm: hugetlb: kill set_huge_swap_pte_at()"), but since
things have moved on, this would have required an audit of all the new
set_huge_pte_at() call sites to see if they should be converted to
set_huge_swap_pte_at(). As per the original intent of the change, it
would also leave us open to future bugs when people invariably get it
wrong and call the wrong helper.
So instead, I've added a huge page size parameter to set_huge_pte_at().
This means that the arm64 code has the size in all cases. It's a bigger
change, due to needing to touch the arches that implement the function,
but it is entirely mechanical, so in my view, low risk.
I've compile-tested all touched arches; arm64, parisc, powerpc, riscv,
s390, sparc (and additionally x86_64). I've additionally booted and run
mm selftests against arm64, where I observe the uffd poison test is fixed,
and there are no other regressions.
This patch (of 2):
In order to fix a bug, arm64 needs to be told the size of the huge page
for which the pte is being set in set_huge_pte_at(). Provide for this by
adding an `unsigned long sz` parameter to the function. This follows the
same pattern as huge_pte_clear().
This commit makes the required interface modifications to the core mm as
well as all arches that implement this function (arm64, parisc, powerpc,
riscv, s390, sparc). The actual arm64 bug will be fixed in a separate
commit.
No behavioral changes intended.
Link: https://lkml.kernel.org/r/20230922115804.2043771-1-ryan.roberts@arm.com
Link: https://lkml.kernel.org/r/20230922115804.2043771-2-ryan.roberts@arm.com
Fixes: 8a13897fb0da ("mm: userfaultfd: support UFFDIO_POISON for hugetlbfs")
Signed-off-by: Ryan Roberts <ryan.roberts@arm.com>
Reviewed-by: Christophe Leroy <christophe.leroy@csgroup.eu> [powerpc 8xx]
Reviewed-by: Lorenzo Stoakes <lstoakes@gmail.com> [vmalloc change]
Cc: Alexandre Ghiti <alex@ghiti.fr>
Cc: Albert Ou <aou@eecs.berkeley.edu>
Cc: Alexander Gordeev <agordeev@linux.ibm.com>
Cc: Anshuman Khandual <anshuman.khandual@arm.com>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Axel Rasmussen <axelrasmussen@google.com>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Christian Borntraeger <borntraeger@linux.ibm.com>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: David S. Miller <davem@davemloft.net>
Cc: Gerald Schaefer <gerald.schaefer@linux.ibm.com>
Cc: Heiko Carstens <hca@linux.ibm.com>
Cc: Helge Deller <deller@gmx.de>
Cc: "James E.J. Bottomley" <James.Bottomley@HansenPartnership.com>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Muchun Song <muchun.song@linux.dev>
Cc: Nicholas Piggin <npiggin@gmail.com>
Cc: Palmer Dabbelt <palmer@dabbelt.com>
Cc: Paul Walmsley <paul.walmsley@sifive.com>
Cc: Peter Xu <peterx@redhat.com>
Cc: Qi Zheng <zhengqi.arch@bytedance.com>
Cc: Ryan Roberts <ryan.roberts@arm.com>
Cc: SeongJae Park <sj@kernel.org>
Cc: Sven Schnelle <svens@linux.ibm.com>
Cc: Uladzislau Rezki (Sony) <urezki@gmail.com>
Cc: Vasily Gorbik <gor@linux.ibm.com>
Cc: Will Deacon <will@kernel.org>
Cc: <stable@vger.kernel.org> [6.5+]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2023-09-22 12:58:03 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-10-05 15:51:29 -07:00
|
|
|
|
2018-12-28 00:38:09 -08:00
|
|
|
|
2017-08-31 17:17:27 -04:00
|
|
|
|
2017-02-24 14:58:01 -08:00
|
|
|
|
mm/rmap: initialize nr_pages to 1 at loop start in try_to_unmap_one
Initialize nr_pages to 1 at the start of each loop iteration, like
folio_referenced_one() does.
Without this, nr_pages computed by a previous folio_unmap_pte_batch() call
can be reused on a later iteration that does not run
folio_unmap_pte_batch() again.
mmap a 64K large folio with MAP_ANONYMOUS | MAP_DROPPABLE, then call
madvise(MADV_FREE), then make the last page device-exclusive via
HMM_DMIRROR_EXCLUSIVE.
Trigger node reclaim through sysfs. Now, in try_to_unmap_one(), we will
first clear the first 15 out of 16 entries mapping the lazyfree folio.
This will set nr_pages to 15. In the next pvmw walk, this nr_pages gets
reused on a device-exclusive pte, thus potentially corrupting folio
refcount/mapcount.
At the moment, I have a userspace program which can make the kernel spit
out a trace, but the blow up is in folio_referenced_one(), because there
are existing bugs in the interaction between device-private and rmap
(which too I am investigating). I did a one liner kernel change to avoid
going into folio_referenced_one(), and the kernel blows up at
folio_remove_rmap_ptes in try_to_unmap_one which is what I wanted.
Note that the bug is there not since file folio batching but lazyfree
folio batching, since device-exclusive only works for anonymous folios.
Userspace visible effect is simply kernel crashing somewhere due to
refcount/mapcount corruption.
Link: https://lore.kernel.org/20260518063656.3721056-1-dev.jain@arm.com
Fixes: 354dffd29575 ("mm: support batched unmap for lazyfree large folios during reclamation")
Signed-off-by: Dev Jain <dev.jain@arm.com>
Acked-by: Barry Song <baohua@kernel.org>
Acked-by: David Hildenbrand (Arm) <david@kernel.org>
Reviewed-by: Lorenzo Stoakes <ljs@kernel.org>
Cc: Anshuman Khandual <anshuman.khandual@arm.com>
Cc: Barry Song <baohua@kernel.org>
Cc: Dev Jain <dev.jain@arm.com>
Cc: Harry Yoo <harry@kernel.org>
Cc: Jann Horn <jannh@google.com>
Cc: Liam R. Howlett <liam@infradead.org>
Cc: Rik van Riel <riel@surriel.com>
Cc: Ryan Roberts <ryan.roberts@arm.com>
Cc: Vlastimil Babka <vbabka@kernel.org>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2026-05-18 12:06:56 +05:30
|
|
|
|
|
|
|
|
|
2017-02-24 14:58:01 -08:00
|
|
|
|
2022-02-15 09:28:49 -05:00
|
|
|
|
2017-02-24 14:58:01 -08:00
|
|
|
|
2021-07-11 20:10:49 -07:00
|
|
|
|
|
|
|
|
|
2025-09-23 12:07:08 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
mm/munlock: rmap call mlock_vma_page() munlock_vma_page()
Add vma argument to mlock_vma_page() and munlock_vma_page(), make them
inline functions which check (vma->vm_flags & VM_LOCKED) before calling
mlock_page() and munlock_page() in mm/mlock.c.
Add bool compound to mlock_vma_page() and munlock_vma_page(): this is
because we have understandable difficulty in accounting pte maps of THPs,
and if passed a PageHead page, mlock_page() and munlock_page() cannot
tell whether it's a pmd map to be counted or a pte map to be ignored.
Add vma arg to page_add_file_rmap() and page_remove_rmap(), like the
others, and use that to call mlock_vma_page() at the end of the page
adds, and munlock_vma_page() at the end of page_remove_rmap() (end or
beginning? unimportant, but end was easier for assertions in testing).
No page lock is required (although almost all adds happen to hold it):
delete the "Serialize with page migration" BUG_ON(!PageLocked(page))s.
Certainly page lock did serialize with page migration, but I'm having
difficulty explaining why that was ever important.
Mlock accounting on THPs has been hard to define, differed between anon
and file, involved PageDoubleMap in some places and not others, required
clear_page_mlock() at some points. Keep it simple now: just count the
pmds and ignore the ptes, there is no reason for ptes to undo pmd mlocks.
page_add_new_anon_rmap() callers unchanged: they have long been calling
lru_cache_add_inactive_or_unevictable(), which does its own VM_LOCKED
handling (it also checks for not VM_SPECIAL: I think that's overcautious,
and inconsistent with other checks, that mmap_region() already prevents
VM_LOCKED on VM_SPECIAL; but haven't quite convinced myself to change it).
Signed-off-by: Hugh Dickins <hughd@google.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
2022-02-14 18:26:39 -08:00
|
|
|
|
2025-09-23 12:07:08 +01:00
|
|
|
|
|
|
|
|
|
mm: rmap use pte lock not mmap_sem to set PageMlocked
KernelThreadSanitizer (ktsan) has shown that the down_read_trylock() of
mmap_sem in try_to_unmap_one() (when going to set PageMlocked on a page
found mapped in a VM_LOCKED vma) is ineffective against races with
exit_mmap()'s munlock_vma_pages_all(), because mmap_sem is not held when
tearing down an mm.
But that's okay, those races are benign; and although we've believed for
years in that ugly down_read_trylock(), it's unsuitable for the job, and
frustrates the good intention of setting PageMlocked when it fails.
It just doesn't matter if here we read vm_flags an instant before or after
a racing mlock() or munlock() or exit_mmap() sets or clears VM_LOCKED: the
syscalls (or exit) work their way up the address space (taking pt locks
after updating vm_flags) to establish the final state.
We do still need to be careful never to mark a page Mlocked (hence
unevictable) by any race that will not be corrected shortly after. The
page lock protects from many of the races, but not all (a page is not
necessarily locked when it's unmapped). But the pte lock we just dropped
is good to cover the rest (and serializes even with
munlock_vma_pages_all(), so no special barriers required): now hold on to
the pte lock while calling mlock_vma_page(). Is that lock ordering safe?
Yes, that's how follow_page_pte() calls it, and how page_remove_rmap()
calls the complementary clear_page_mlock().
This fixes the following case (though not a case which anyone has
complained of), which mmap_sem did not: truncation's preliminary
unmap_mapping_range() is supposed to remove even the anonymous COWs of
filecache pages, and that might race with try_to_unmap_one() on a
VM_LOCKED vma, so that mlock_vma_page() sets PageMlocked just after
zap_pte_range() unmaps the page, causing "Bad page state (mlocked)" when
freed. The pte lock protects against this.
You could say that it also protects against the more ordinary case, racing
with the preliminary unmapping of a filecache page itself: but in our
current tree, that's independently protected by i_mmap_rwsem; and that
race would be why "Bad page state (mlocked)" was seen before commit
48ec833b7851 ("Revert mm/memory.c: share the i_mmap_rwsem").
Vlastimil Babka points out another race which this patch protects against.
try_to_unmap_one() might reach its mlock_vma_page() TestSetPageMlocked a
moment after munlock_vma_pages_all() did its Phase 1 TestClearPageMlocked:
leaving PageMlocked and unevictable when it should be evictable. mmap_sem
is ineffective because exit_mmap() does not hold it; page lock ineffective
because __munlock_pagevec() only takes it afterwards, in Phase 2; pte lock
is effective because __munlock_pagevec_fill() takes it to get the page,
after VM_LOCKED was cleared from vm_flags, so visible to try_to_unmap_one.
Kirill Shutemov points out that if the compiler chooses to implement a
"vma->vm_flags &= VM_WHATEVER" or "vma->vm_flags |= VM_WHATEVER" operation
with an intermediate store of unrelated bits set, since I'm here foregoing
its usual protection by mmap_sem, try_to_unmap_one() might catch sight of
a spurious VM_LOCKED in vm_flags, and make the wrong decision. This does
not appear to be an immediate problem, but we may want to define vm_flags
accessors in future, to guard against such a possibility.
While we're here, make a related optimization in try_to_munmap_one(): if
it's doing TTU_MUNLOCK, then there's no point at all in descending the
page tables and getting the pt lock, unless the vma is VM_LOCKED. Yes,
that can change racily, but it can change racily even without the
optimization: it's not critical. Far better not to waste time here.
Stopped short of separating try_to_munlock_one() from try_to_munmap_one()
on this occasion, but that's probably the sensible next step - with a
rename, given that try_to_munlock()'s business is to try to set Mlocked.
Updated the unevictable-lru Documentation, to remove its reference to mmap
semaphore, but found a few more updates needed in just that area.
Signed-off-by: Hugh Dickins <hughd@google.com>
Cc: Christoph Lameter <cl@linux.com>
Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Rik van Riel <riel@redhat.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Davidlohr Bueso <dave@stgolabs.net>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Sasha Levin <sasha.levin@oracle.com>
Cc: Dmitry Vyukov <dvyukov@google.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2015-11-05 18:49:33 -08:00
|
|
|
|
2017-02-24 14:58:01 -08:00
|
|
|
|
2024-06-14 09:51:38 +08:00
|
|
|
|
2026-02-21 17:39:17 +08:00
|
|
|
|
mm: avoid splitting pmd for lazyfree pmd-mapped THP in try_to_unmap
The try_to_unmap_one() function currently handles PMD-mapped THPs
inefficiently. It first splits the PMD into PTEs, copies the dirty state
from the PMD to the PTEs, iterates over the PTEs to locate the dirty
state, and then marks the THP as swap-backed. This process involves
unnecessary PMD splitting and redundant iteration. Instead, this
functionality can be efficiently managed in
__discard_anon_folio_pmd_locked(), avoiding the extra steps and improving
performance.
The following microbenchmark redirties folios after invoking MADV_FREE,
then measures the time taken to perform memory reclamation (actually set
those folios swapbacked again) on the redirtied folios.
#include <stdio.h>
#include <sys/mman.h>
#include <string.h>
#include <time.h>
#define SIZE 128*1024*1024 // 128 MB
int main(int argc, char *argv[])
{
while(1) {
volatile int *p = mmap(0, SIZE, PROT_READ | PROT_WRITE,
MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
memset((void *)p, 1, SIZE);
madvise((void *)p, SIZE, MADV_FREE);
/* redirty after MADV_FREE */
memset((void *)p, 1, SIZE);
clock_t start_time = clock();
madvise((void *)p, SIZE, MADV_PAGEOUT);
clock_t end_time = clock();
double elapsed_time = (double)(end_time - start_time) / CLOCKS_PER_SEC;
printf("Time taken by reclamation: %f seconds\n", elapsed_time);
munmap((void *)p, SIZE);
}
return 0;
}
Testing results are as below,
w/o patch:
~ # ./a.out
Time taken by reclamation: 0.007300 seconds
Time taken by reclamation: 0.007226 seconds
Time taken by reclamation: 0.007295 seconds
Time taken by reclamation: 0.007731 seconds
Time taken by reclamation: 0.007134 seconds
Time taken by reclamation: 0.007285 seconds
Time taken by reclamation: 0.007720 seconds
Time taken by reclamation: 0.007128 seconds
Time taken by reclamation: 0.007710 seconds
Time taken by reclamation: 0.007712 seconds
Time taken by reclamation: 0.007236 seconds
Time taken by reclamation: 0.007690 seconds
Time taken by reclamation: 0.007174 seconds
Time taken by reclamation: 0.007670 seconds
Time taken by reclamation: 0.007169 seconds
Time taken by reclamation: 0.007305 seconds
Time taken by reclamation: 0.007432 seconds
Time taken by reclamation: 0.007158 seconds
Time taken by reclamation: 0.007133 seconds
…
w/ patch
~ # ./a.out
Time taken by reclamation: 0.002124 seconds
Time taken by reclamation: 0.002116 seconds
Time taken by reclamation: 0.002150 seconds
Time taken by reclamation: 0.002261 seconds
Time taken by reclamation: 0.002137 seconds
Time taken by reclamation: 0.002173 seconds
Time taken by reclamation: 0.002063 seconds
Time taken by reclamation: 0.002088 seconds
Time taken by reclamation: 0.002169 seconds
Time taken by reclamation: 0.002124 seconds
Time taken by reclamation: 0.002111 seconds
Time taken by reclamation: 0.002224 seconds
Time taken by reclamation: 0.002297 seconds
Time taken by reclamation: 0.002260 seconds
Time taken by reclamation: 0.002246 seconds
Time taken by reclamation: 0.002272 seconds
Time taken by reclamation: 0.002277 seconds
Time taken by reclamation: 0.002462 seconds
…
This patch significantly speeds up try_to_unmap_one() by allowing it
to skip redirtied THPs without splitting the PMD.
Link: https://lkml.kernel.org/r/20250214093015.51024-5-21cnbao@gmail.com
Signed-off-by: Barry Song <v-songbaohua@oppo.com>
Suggested-by: Baolin Wang <baolin.wang@linux.alibaba.com>
Suggested-by: Lance Yang <ioworker0@gmail.com>
Reviewed-by: Baolin Wang <baolin.wang@linux.alibaba.com>
Reviewed-by: Lance Yang <ioworker0@gmail.com>
Cc: Albert Ou <aou@eecs.berkeley.edu>
Cc: Anshuman Khandual <anshuman.khandual@arm.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Chis Li <chrisl@kernel.org>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Gavin Shan <gshan@redhat.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: "Huang, Ying" <ying.huang@intel.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Kairui Song <kasong@tencent.com>
Cc: Kefeng Wang <wangkefeng.wang@huawei.com>
Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Mauricio Faria de Oliveira <mfo@canonical.com>
Cc: Palmer Dabbelt <palmer@dabbelt.com>
Cc: Paul Walmsley <paul.walmsley@sifive.com>
Cc: Ryan Roberts <ryan.roberts@arm.com>
Cc: Shaoqin Huang <shahuang@redhat.com>
Cc: Tangquan Zheng <zhengtangquan@oppo.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Will Deacon <will@kernel.org>
Cc: Yicong Yang <yangyicong@hisilicon.com>
Cc: Yosry Ahmed <yosryahmed@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2025-02-14 22:30:15 +13:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-06-14 09:51:38 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-04-25 18:38:59 +08:00
|
|
|
|
2024-06-14 09:51:38 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-06-14 09:51:37 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-02-10 20:37:52 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-11-10 22:21:34 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-02-10 20:37:52 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
mm: ptep_get() conversion
Convert all instances of direct pte_t* dereferencing to instead use
ptep_get() helper. This means that by default, the accesses change from a
C dereference to a READ_ONCE(). This is technically the correct thing to
do since where pgtables are modified by HW (for access/dirty) they are
volatile and therefore we should always ensure READ_ONCE() semantics.
But more importantly, by always using the helper, it can be overridden by
the architecture to fully encapsulate the contents of the pte. Arch code
is deliberately not converted, as the arch code knows best. It is
intended that arch code (arm64) will override the default with its own
implementation that can (e.g.) hide certain bits from the core code, or
determine young/dirty status by mixing in state from another source.
Conversion was done using Coccinelle:
----
// $ make coccicheck \
// COCCI=ptepget.cocci \
// SPFLAGS="--include-headers" \
// MODE=patch
virtual patch
@ depends on patch @
pte_t *v;
@@
- *v
+ ptep_get(v)
----
Then reviewed and hand-edited to avoid multiple unnecessary calls to
ptep_get(), instead opting to store the result of a single call in a
variable, where it is correct to do so. This aims to negate any cost of
READ_ONCE() and will benefit arch-overrides that may be more complex.
Included is a fix for an issue in an earlier version of this patch that
was pointed out by kernel test robot. The issue arose because config
MMU=n elides definition of the ptep helper functions, including
ptep_get(). HUGETLB_PAGE=n configs still define a simple
huge_ptep_clear_flush() for linking purposes, which dereferences the ptep.
So when both configs are disabled, this caused a build error because
ptep_get() is not defined. Fix by continuing to do a direct dereference
when MMU=n. This is safe because for this config the arch code cannot be
trying to virtualize the ptes because none of the ptep helpers are
defined.
Link: https://lkml.kernel.org/r/20230612151545.3317766-4-ryan.roberts@arm.com
Reported-by: kernel test robot <lkp@intel.com>
Link: https://lore.kernel.org/oe-kbuild-all/202305120142.yXsNEo6H-lkp@intel.com/
Signed-off-by: Ryan Roberts <ryan.roberts@arm.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Alexander Potapenko <glider@google.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Alex Williamson <alex.williamson@redhat.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Andrey Konovalov <andreyknvl@gmail.com>
Cc: Andrey Ryabinin <ryabinin.a.a@gmail.com>
Cc: Christian Brauner <brauner@kernel.org>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: Daniel Vetter <daniel@ffwll.ch>
Cc: Dave Airlie <airlied@gmail.com>
Cc: Dimitri Sivanich <dimitri.sivanich@hpe.com>
Cc: Dmitry Vyukov <dvyukov@google.com>
Cc: Ian Rogers <irogers@google.com>
Cc: Jason Gunthorpe <jgg@ziepe.ca>
Cc: Jérôme Glisse <jglisse@redhat.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Lorenzo Stoakes <lstoakes@gmail.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Miaohe Lin <linmiaohe@huawei.com>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Mike Rapoport (IBM) <rppt@kernel.org>
Cc: Muchun Song <muchun.song@linux.dev>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Naoya Horiguchi <naoya.horiguchi@nec.com>
Cc: Oleksandr Tyshchenko <oleksandr_tyshchenko@epam.com>
Cc: Pavel Tatashin <pasha.tatashin@soleen.com>
Cc: Roman Gushchin <roman.gushchin@linux.dev>
Cc: SeongJae Park <sj@kernel.org>
Cc: Shakeel Butt <shakeelb@google.com>
Cc: Uladzislau Rezki (Sony) <urezki@gmail.com>
Cc: Vincenzo Frascino <vincenzo.frascino@arm.com>
Cc: Yu Zhao <yuzhao@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2023-06-12 16:15:45 +01:00
|
|
|
|
Revert "rmap: do not call mmu_notifier_invalidate_page() under ptl"
This reverts commit aac2fea94f7a3df8ad1eeb477eb2643f81fd5393.
It turns out that that patch was complete and utter garbage, and broke
KVM, resulting in odd oopses.
Quoting Andrea Arcangeli:
"The aforementioned commit has 3 bugs.
1) mmu_notifier_invalidate_range cannot be used in replacement of
mmu_notifier_invalidate_range_start/end.
For KVM mmu_notifier_invalidate_range is a noop and rightfully so.
A MMU notifier implementation has to implement either
->invalidate_range method or the invalidate_range_start/end
methods, not both. And if you implement invalidate_range_start/end
like KVM is forced to do, calling mmu_notifier_invalidate_range in
common code is a noop for KVM.
For those MMU notifiers that can get away only implementing
->invalidate_range, the ->invalidate_range is implicitly called by
mmu_notifier_invalidate_range_end(). And only those secondary MMUs
that share the same pagetable with the primary MMU (like AMD
iommuv2) can get away only implementing ->invalidate_range.
So all cases (THP on/off) are broken right now.
To fix this is enough to replace mmu_notifier_invalidate_range with
mmu_notifier_invalidate_range_start;mmu_notifier_invalidate_range_end.
Either that or call multiple mmu_notifier_invalidate_page like
before.
2) address + (1UL << compound_order(page) is buggy, it should be
PAGE_SIZE << compound_order(page), it's bytes not pages, 2M not
512.
3) The whole invalidate_range thing was an attempt to call a single
invalidate while walking multiple 4k ptes that maps the same THP
(after a pmd virtual split without physical compound page THP
split).
It's unclear if the rmap_walk will always provide an address that
is 2M aligned as parameter to try_to_unmap_one, in presence of THP.
I think it needs also an address &= (PAGE_SIZE <<
compound_order(page)) - 1 to be safe"
In general, we should stop making excuses for horrible MMU notifier
users. It's much more important that the core VM is sane and safe, than
letting MMU notifiers sleep.
So if some MMU notifier is sleeping under a spinlock, we need to fix the
notifier, not try to make excuses for that garbage in the core VM.
Reported-and-tested-by: Bernhard Held <berny156@gmx.de>
Reported-and-tested-by: Adam Borowski <kilobyte@angband.pl>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Radim Krčmář <rkrcmar@redhat.com>
Cc: Wanpeng Li <kernellwp@gmail.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Takashi Iwai <tiwai@suse.de>
Cc: Nadav Amit <nadav.amit@gmail.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Jérôme Glisse <jglisse@redhat.com>
Cc: axie <axie@amd.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-08-29 09:11:06 -07:00
|
|
|
|
mm: remember exclusively mapped anonymous pages with PG_anon_exclusive
Let's mark exclusively mapped anonymous pages with PG_anon_exclusive as
exclusive, and use that information to make GUP pins reliable and stay
consistent with the page mapped into the page table even if the page table
entry gets write-protected.
With that information at hand, we can extend our COW logic to always reuse
anonymous pages that are exclusive. For anonymous pages that might be
shared, the existing logic applies.
As already documented, PG_anon_exclusive is usually only expressive in
combination with a page table entry. Especially PTE vs. PMD-mapped
anonymous pages require more thought, some examples: due to mremap() we
can easily have a single compound page PTE-mapped into multiple page
tables exclusively in a single process -- multiple page table locks apply.
Further, due to MADV_WIPEONFORK we might not necessarily write-protect
all PTEs, and only some subpages might be pinned. Long story short: once
PTE-mapped, we have to track information about exclusivity per sub-page,
but until then, we can just track it for the compound page in the head
page and not having to update a whole bunch of subpages all of the time
for a simple PMD mapping of a THP.
For simplicity, this commit mostly talks about "anonymous pages", while
it's for THP actually "the part of an anonymous folio referenced via a
page table entry".
To not spill PG_anon_exclusive code all over the mm code-base, we let the
anon rmap code to handle all PG_anon_exclusive logic it can easily handle.
If a writable, present page table entry points at an anonymous (sub)page,
that (sub)page must be PG_anon_exclusive. If GUP wants to take a reliably
pin (FOLL_PIN) on an anonymous page references via a present page table
entry, it must only pin if PG_anon_exclusive is set for the mapped
(sub)page.
This commit doesn't adjust GUP, so this is only implicitly handled for
FOLL_WRITE, follow-up commits will teach GUP to also respect it for
FOLL_PIN without FOLL_WRITE, to make all GUP pins of anonymous pages fully
reliable.
Whenever an anonymous page is to be shared (fork(), KSM), or when
temporarily unmapping an anonymous page (swap, migration), the relevant
PG_anon_exclusive bit has to be cleared to mark the anonymous page
possibly shared. Clearing will fail if there are GUP pins on the page:
* For fork(), this means having to copy the page and not being able to
share it. fork() protects against concurrent GUP using the PT lock and
the src_mm->write_protect_seq.
* For KSM, this means sharing will fail. For swap this means, unmapping
will fail, For migration this means, migration will fail early. All
three cases protect against concurrent GUP using the PT lock and a
proper clear/invalidate+flush of the relevant page table entry.
This fixes memory corruptions reported for FOLL_PIN | FOLL_WRITE, when a
pinned page gets mapped R/O and the successive write fault ends up
replacing the page instead of reusing it. It improves the situation for
O_DIRECT/vmsplice/... that still use FOLL_GET instead of FOLL_PIN, if
fork() is *not* involved, however swapout and fork() are still
problematic. Properly using FOLL_PIN instead of FOLL_GET for these GUP
users will fix the issue for them.
I. Details about basic handling
I.1. Fresh anonymous pages
page_add_new_anon_rmap() and hugepage_add_new_anon_rmap() will mark the
given page exclusive via __page_set_anon_rmap(exclusive=1). As that is
the mechanism fresh anonymous pages come into life (besides migration code
where we copy the page->mapping), all fresh anonymous pages will start out
as exclusive.
I.2. COW reuse handling of anonymous pages
When a COW handler stumbles over a (sub)page that's marked exclusive, it
simply reuses it. Otherwise, the handler tries harder under page lock to
detect if the (sub)page is exclusive and can be reused. If exclusive,
page_move_anon_rmap() will mark the given (sub)page exclusive.
Note that hugetlb code does not yet check for PageAnonExclusive(), as it
still uses the old COW logic that is prone to the COW security issue
because hugetlb code cannot really tolerate unnecessary/wrong COW as huge
pages are a scarce resource.
I.3. Migration handling
try_to_migrate() has to try marking an exclusive anonymous page shared via
page_try_share_anon_rmap(). If it fails because there are GUP pins on the
page, unmap fails. migrate_vma_collect_pmd() and
__split_huge_pmd_locked() are handled similarly.
Writable migration entries implicitly point at shared anonymous pages.
For readable migration entries that information is stored via a new
"readable-exclusive" migration entry, specific to anonymous pages.
When restoring a migration entry in remove_migration_pte(), information
about exlusivity is detected via the migration entry type, and
RMAP_EXCLUSIVE is set accordingly for
page_add_anon_rmap()/hugepage_add_anon_rmap() to restore that information.
I.4. Swapout handling
try_to_unmap() has to try marking the mapped page possibly shared via
page_try_share_anon_rmap(). If it fails because there are GUP pins on the
page, unmap fails. For now, information about exclusivity is lost. In
the future, we might want to remember that information in the swap entry
in some cases, however, it requires more thought, care, and a way to store
that information in swap entries.
I.5. Swapin handling
do_swap_page() will never stumble over exclusive anonymous pages in the
swap cache, as try_to_migrate() prohibits that. do_swap_page() always has
to detect manually if an anonymous page is exclusive and has to set
RMAP_EXCLUSIVE for page_add_anon_rmap() accordingly.
I.6. THP handling
__split_huge_pmd_locked() has to move the information about exclusivity
from the PMD to the PTEs.
a) In case we have a readable-exclusive PMD migration entry, simply
insert readable-exclusive PTE migration entries.
b) In case we have a present PMD entry and we don't want to freeze
("convert to migration entries"), simply forward PG_anon_exclusive to
all sub-pages, no need to temporarily clear the bit.
c) In case we have a present PMD entry and want to freeze, handle it
similar to try_to_migrate(): try marking the page shared first. In
case we fail, we ignore the "freeze" instruction and simply split
ordinarily. try_to_migrate() will properly fail because the THP is
still mapped via PTEs.
When splitting a compound anonymous folio (THP), the information about
exclusivity is implicitly handled via the migration entries: no need to
replicate PG_anon_exclusive manually.
I.7. fork() handling fork() handling is relatively easy, because
PG_anon_exclusive is only expressive for some page table entry types.
a) Present anonymous pages
page_try_dup_anon_rmap() will mark the given subpage shared -- which will
fail if the page is pinned. If it failed, we have to copy (or PTE-map a
PMD to handle it on the PTE level).
Note that device exclusive entries are just a pointer at a PageAnon()
page. fork() will first convert a device exclusive entry to a present
page table and handle it just like present anonymous pages.
b) Device private entry
Device private entries point at PageAnon() pages that cannot be mapped
directly and, therefore, cannot get pinned.
page_try_dup_anon_rmap() will mark the given subpage shared, which cannot
fail because they cannot get pinned.
c) HW poison entries
PG_anon_exclusive will remain untouched and is stale -- the page table
entry is just a placeholder after all.
d) Migration entries
Writable and readable-exclusive entries are converted to readable entries:
possibly shared.
I.8. mprotect() handling
mprotect() only has to properly handle the new readable-exclusive
migration entry:
When write-protecting a migration entry that points at an anonymous page,
remember the information about exclusivity via the "readable-exclusive"
migration entry type.
II. Migration and GUP-fast
Whenever replacing a present page table entry that maps an exclusive
anonymous page by a migration entry, we have to mark the page possibly
shared and synchronize against GUP-fast by a proper clear/invalidate+flush
to make the following scenario impossible:
1. try_to_migrate() places a migration entry after checking for GUP pins
and marks the page possibly shared.
2. GUP-fast pins the page due to lack of synchronization
3. fork() converts the "writable/readable-exclusive" migration entry into a
readable migration entry
4. Migration fails due to the GUP pin (failing to freeze the refcount)
5. Migration entries are restored. PG_anon_exclusive is lost
-> We have a pinned page that is not marked exclusive anymore.
Note that we move information about exclusivity from the page to the
migration entry as it otherwise highly overcomplicates fork() and
PTE-mapping a THP.
III. Swapout and GUP-fast
Whenever replacing a present page table entry that maps an exclusive
anonymous page by a swap entry, we have to mark the page possibly shared
and synchronize against GUP-fast by a proper clear/invalidate+flush to
make the following scenario impossible:
1. try_to_unmap() places a swap entry after checking for GUP pins and
clears exclusivity information on the page.
2. GUP-fast pins the page due to lack of synchronization.
-> We have a pinned page that is not marked exclusive anymore.
If we'd ever store information about exclusivity in the swap entry,
similar to migration handling, the same considerations as in II would
apply. This is future work.
Link: https://lkml.kernel.org/r/20220428083441.37290-13-david@redhat.com
Signed-off-by: David Hildenbrand <david@redhat.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Christoph Hellwig <hch@lst.de>
Cc: David Rientjes <rientjes@google.com>
Cc: Don Dutile <ddutile@redhat.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Jan Kara <jack@suse.cz>
Cc: Jann Horn <jannh@google.com>
Cc: Jason Gunthorpe <jgg@nvidia.com>
Cc: John Hubbard <jhubbard@nvidia.com>
Cc: Khalid Aziz <khalid.aziz@oracle.com>
Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Liang Zhang <zhangliang5@huawei.com>
Cc: "Matthew Wilcox (Oracle)" <willy@infradead.org>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Mike Rapoport <rppt@linux.ibm.com>
Cc: Nadav Amit <namit@vmware.com>
Cc: Oded Gabbay <oded.gabbay@gmail.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Pedro Demarchi Gomes <pedrodemargomes@gmail.com>
Cc: Peter Xu <peterx@redhat.com>
Cc: Rik van Riel <riel@surriel.com>
Cc: Roman Gushchin <guro@fb.com>
Cc: Shakeel Butt <shakeelb@google.com>
Cc: Yang Shi <shy828301@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2022-05-09 18:20:44 -07:00
|
|
|
|
|
|
|
|
|
Revert "rmap: do not call mmu_notifier_invalidate_page() under ptl"
This reverts commit aac2fea94f7a3df8ad1eeb477eb2643f81fd5393.
It turns out that that patch was complete and utter garbage, and broke
KVM, resulting in odd oopses.
Quoting Andrea Arcangeli:
"The aforementioned commit has 3 bugs.
1) mmu_notifier_invalidate_range cannot be used in replacement of
mmu_notifier_invalidate_range_start/end.
For KVM mmu_notifier_invalidate_range is a noop and rightfully so.
A MMU notifier implementation has to implement either
->invalidate_range method or the invalidate_range_start/end
methods, not both. And if you implement invalidate_range_start/end
like KVM is forced to do, calling mmu_notifier_invalidate_range in
common code is a noop for KVM.
For those MMU notifiers that can get away only implementing
->invalidate_range, the ->invalidate_range is implicitly called by
mmu_notifier_invalidate_range_end(). And only those secondary MMUs
that share the same pagetable with the primary MMU (like AMD
iommuv2) can get away only implementing ->invalidate_range.
So all cases (THP on/off) are broken right now.
To fix this is enough to replace mmu_notifier_invalidate_range with
mmu_notifier_invalidate_range_start;mmu_notifier_invalidate_range_end.
Either that or call multiple mmu_notifier_invalidate_page like
before.
2) address + (1UL << compound_order(page) is buggy, it should be
PAGE_SIZE << compound_order(page), it's bytes not pages, 2M not
512.
3) The whole invalidate_range thing was an attempt to call a single
invalidate while walking multiple 4k ptes that maps the same THP
(after a pmd virtual split without physical compound page THP
split).
It's unclear if the rmap_walk will always provide an address that
is 2M aligned as parameter to try_to_unmap_one, in presence of THP.
I think it needs also an address &= (PAGE_SIZE <<
compound_order(page)) - 1 to be safe"
In general, we should stop making excuses for horrible MMU notifier
users. It's much more important that the core VM is sane and safe, than
letting MMU notifiers sleep.
So if some MMU notifier is sleeping under a spinlock, we need to fix the
notifier, not try to make excuses for that garbage in the core VM.
Reported-and-tested-by: Bernhard Held <berny156@gmx.de>
Reported-and-tested-by: Adam Borowski <kilobyte@angband.pl>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Radim Krčmář <rkrcmar@redhat.com>
Cc: Wanpeng Li <kernellwp@gmail.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Takashi Iwai <tiwai@suse.de>
Cc: Nadav Amit <nadav.amit@gmail.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Jérôme Glisse <jglisse@redhat.com>
Cc: axie <axie@amd.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-08-29 09:11:06 -07:00
|
|
|
|
2022-05-09 18:20:53 -07:00
|
|
|
|
2022-06-20 19:47:15 +08:00
|
|
|
|
|
|
|
|
|
mm: rmap: fix CONT-PTE/PMD size hugetlb issue when unmapping
On some architectures (like ARM64), it can support CONT-PTE/PMD size
hugetlb, which means it can support not only PMD/PUD size hugetlb: 2M and
1G, but also CONT-PTE/PMD size: 64K and 32M if a 4K page size specified.
When unmapping a hugetlb page, we will get the relevant page table entry
by huge_pte_offset() only once to nuke it. This is correct for PMD or PUD
size hugetlb, since they always contain only one pmd entry or pud entry in
the page table.
However this is incorrect for CONT-PTE and CONT-PMD size hugetlb, since
they can contain several continuous pte or pmd entry with same page table
attributes, so we will nuke only one pte or pmd entry for this
CONT-PTE/PMD size hugetlb page.
And now try_to_unmap() is only passed a hugetlb page in the case where the
hugetlb page is poisoned. Which means now we will unmap only one pte
entry for a CONT-PTE or CONT-PMD size poisoned hugetlb page, and we can
still access other subpages of a CONT-PTE or CONT-PMD size poisoned
hugetlb page, which will cause serious issues possibly.
So we should change to use huge_ptep_clear_flush() to nuke the hugetlb
page table to fix this issue, which already considered CONT-PTE and
CONT-PMD size hugetlb.
We've already used set_huge_swap_pte_at() to set a poisoned swap entry for
a poisoned hugetlb page. Meanwhile adding a VM_BUG_ON() to make sure the
passed hugetlb page is poisoned in try_to_unmap().
Link: https://lkml.kernel.org/r/0a2e547238cad5bc153a85c3e9658cb9d55f9cac.1652270205.git.baolin.wang@linux.alibaba.com
Link: https://lkml.kernel.org/r/730ea4b6d292f32fb10b7a4e87dad49b0eb30474.1652147571.git.baolin.wang@linux.alibaba.com
Signed-off-by: Baolin Wang <baolin.wang@linux.alibaba.com>
Reviewed-by: Muchun Song <songmuchun@bytedance.com>
Reviewed-by: Mike Kravetz <mike.kravetz@oracle.com>
Acked-by: David Hildenbrand <david@redhat.com>
Cc: Alexander Gordeev <agordeev@linux.ibm.com>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Christian Borntraeger <borntraeger@linux.ibm.com>
Cc: David S. Miller <davem@davemloft.net>
Cc: Gerald Schaefer <gerald.schaefer@linux.ibm.com>
Cc: Heiko Carstens <hca@linux.ibm.com>
Cc: Helge Deller <deller@gmx.de>
Cc: James Bottomley <James.Bottomley@HansenPartnership.com>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Rich Felker <dalias@libc.org>
Cc: Sven Schnelle <svens@linux.ibm.com>
Cc: Thomas Bogendoerfer <tsbogend@alpha.franken.de>
Cc: Vasily Gorbik <gor@linux.ibm.com>
Cc: Will Deacon <will@kernel.org>
Cc: Yoshinori Sato <ysato@users.osdn.me>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2022-05-13 16:48:56 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-05-09 18:20:53 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-06-20 19:47:15 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-09-14 15:18:09 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-06-20 19:47:15 +08:00
|
|
|
|
2022-09-14 15:18:09 -07:00
|
|
|
|
mm/hugetlb: fix excessive IPI broadcasts when unsharing PMD tables using mmu_gather
As reported, ever since commit 1013af4f585f ("mm/hugetlb: fix
huge_pmd_unshare() vs GUP-fast race") we can end up in some situations
where we perform so many IPI broadcasts when unsharing hugetlb PMD page
tables that it severely regresses some workloads.
In particular, when we fork()+exit(), or when we munmap() a large
area backed by many shared PMD tables, we perform one IPI broadcast per
unshared PMD table.
There are two optimizations to be had:
(1) When we process (unshare) multiple such PMD tables, such as during
exit(), it is sufficient to send a single IPI broadcast (as long as
we respect locking rules) instead of one per PMD table.
Locking prevents that any of these PMD tables could get reused before
we drop the lock.
(2) When we are not the last sharer (> 2 users including us), there is
no need to send the IPI broadcast. The shared PMD tables cannot
become exclusive (fully unshared) before an IPI will be broadcasted
by the last sharer.
Concurrent GUP-fast could walk into a PMD table just before we
unshared it. It could then succeed in grabbing a page from the
shared page table even after munmap() etc succeeded (and supressed
an IPI). But there is not difference compared to GUP-fast just
sleeping for a while after grabbing the page and re-enabling IRQs.
Most importantly, GUP-fast will never walk into page tables that are
no-longer shared, because the last sharer will issue an IPI
broadcast.
(if ever required, checking whether the PUD changed in GUP-fast
after grabbing the page like we do in the PTE case could handle
this)
So let's rework PMD sharing TLB flushing + IPI sync to use the mmu_gather
infrastructure so we can implement these optimizations and demystify the
code at least a bit. Extend the mmu_gather infrastructure to be able to
deal with our special hugetlb PMD table sharing implementation.
To make initialization of the mmu_gather easier when working on a single
VMA (in particular, when dealing with hugetlb), provide
tlb_gather_mmu_vma().
We'll consolidate the handling for (full) unsharing of PMD tables in
tlb_unshare_pmd_ptdesc() and tlb_flush_unshared_tables(), and track
in "struct mmu_gather" whether we had (full) unsharing of PMD tables.
Because locking is very special (concurrent unsharing+reuse must be
prevented), we disallow deferring flushing to tlb_finish_mmu() and instead
require an explicit earlier call to tlb_flush_unshared_tables().
From hugetlb code, we call huge_pmd_unshare_flush() where we make sure
that the expected lock protecting us from concurrent unsharing+reuse is
still held.
Check with a VM_WARN_ON_ONCE() in tlb_finish_mmu() that
tlb_flush_unshared_tables() was properly called earlier.
Document it all properly.
Notes about tlb_remove_table_sync_one() interaction with unsharing:
There are two fairly tricky things:
(1) tlb_remove_table_sync_one() is a NOP on architectures without
CONFIG_MMU_GATHER_RCU_TABLE_FREE.
Here, the assumption is that the previous TLB flush would send an
IPI to all relevant CPUs. Careful: some architectures like x86 only
send IPIs to all relevant CPUs when tlb->freed_tables is set.
The relevant architectures should be selecting
MMU_GATHER_RCU_TABLE_FREE, but x86 might not do that in stable
kernels and it might have been problematic before this patch.
Also, the arch flushing behavior (independent of IPIs) is different
when tlb->freed_tables is set. Do we have to enlighten them to also
take care of tlb->unshared_tables? So far we didn't care, so
hopefully we are fine. Of course, we could be setting
tlb->freed_tables as well, but that might then unnecessarily flush
too much, because the semantics of tlb->freed_tables are a bit
fuzzy.
This patch changes nothing in this regard.
(2) tlb_remove_table_sync_one() is not a NOP on architectures with
CONFIG_MMU_GATHER_RCU_TABLE_FREE that actually don't need a sync.
Take x86 as an example: in the common case (!pv, !X86_FEATURE_INVLPGB)
we still issue IPIs during TLB flushes and don't actually need the
second tlb_remove_table_sync_one().
This optimized can be implemented on top of this, by checking e.g., in
tlb_remove_table_sync_one() whether we really need IPIs. But as
described in (1), it really must honor tlb->freed_tables then to
send IPIs to all relevant CPUs.
Notes on TLB flushing changes:
(1) Flushing for non-shared PMD tables
We're converting from flush_hugetlb_tlb_range() to
tlb_remove_huge_tlb_entry(). Given that we properly initialize the
MMU gather in tlb_gather_mmu_vma() to be hugetlb aware, similar to
__unmap_hugepage_range(), that should be fine.
(2) Flushing for shared PMD tables
We're converting from various things (flush_hugetlb_tlb_range(),
tlb_flush_pmd_range(), flush_tlb_range()) to tlb_flush_pmd_range().
tlb_flush_pmd_range() achieves the same that
tlb_remove_huge_tlb_entry() would achieve in these scenarios.
Note that tlb_remove_huge_tlb_entry() also calls
__tlb_remove_tlb_entry(), however that is only implemented on
powerpc, which does not support PMD table sharing.
Similar to (1), tlb_gather_mmu_vma() should make sure that TLB
flushing keeps on working as expected.
Further, note that the ptdesc_pmd_pts_dec() in huge_pmd_share() is not a
concern, as we are holding the i_mmap_lock the whole time, preventing
concurrent unsharing. That ptdesc_pmd_pts_dec() usage will be removed
separately as a cleanup later.
There are plenty more cleanups to be had, but they have to wait until
this is fixed.
[david@kernel.org: fix kerneldoc]
Link: https://lkml.kernel.org/r/f223dd74-331c-412d-93fc-69e360a5006c@kernel.org
Link: https://lkml.kernel.org/r/20251223214037.580860-5-david@kernel.org
Fixes: 1013af4f585f ("mm/hugetlb: fix huge_pmd_unshare() vs GUP-fast race")
Signed-off-by: David Hildenbrand (Red Hat) <david@kernel.org>
Reported-by: Uschakow, Stanislav" <suschako@amazon.de>
Closes: https://lore.kernel.org/all/4d3878531c76479d9f8ca9789dc6485d@amazon.de/
Tested-by: Laurence Oberman <loberman@redhat.com>
Acked-by: Harry Yoo <harry.yoo@oracle.com>
Reviewed-by: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: Lance Yang <lance.yang@linux.dev>
Cc: Liu Shixin <liushixin2@huawei.com>
Cc: Oscar Salvador <osalvador@suse.de>
Cc: Rik van Riel <riel@surriel.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2025-12-23 22:40:37 +01:00
|
|
|
|
|
|
|
|
|
2022-09-14 15:18:09 -07:00
|
|
|
|
2024-06-14 09:51:36 +08:00
|
|
|
|
|
|
|
|
|
mm/hugetlb: fix excessive IPI broadcasts when unsharing PMD tables using mmu_gather
As reported, ever since commit 1013af4f585f ("mm/hugetlb: fix
huge_pmd_unshare() vs GUP-fast race") we can end up in some situations
where we perform so many IPI broadcasts when unsharing hugetlb PMD page
tables that it severely regresses some workloads.
In particular, when we fork()+exit(), or when we munmap() a large
area backed by many shared PMD tables, we perform one IPI broadcast per
unshared PMD table.
There are two optimizations to be had:
(1) When we process (unshare) multiple such PMD tables, such as during
exit(), it is sufficient to send a single IPI broadcast (as long as
we respect locking rules) instead of one per PMD table.
Locking prevents that any of these PMD tables could get reused before
we drop the lock.
(2) When we are not the last sharer (> 2 users including us), there is
no need to send the IPI broadcast. The shared PMD tables cannot
become exclusive (fully unshared) before an IPI will be broadcasted
by the last sharer.
Concurrent GUP-fast could walk into a PMD table just before we
unshared it. It could then succeed in grabbing a page from the
shared page table even after munmap() etc succeeded (and supressed
an IPI). But there is not difference compared to GUP-fast just
sleeping for a while after grabbing the page and re-enabling IRQs.
Most importantly, GUP-fast will never walk into page tables that are
no-longer shared, because the last sharer will issue an IPI
broadcast.
(if ever required, checking whether the PUD changed in GUP-fast
after grabbing the page like we do in the PTE case could handle
this)
So let's rework PMD sharing TLB flushing + IPI sync to use the mmu_gather
infrastructure so we can implement these optimizations and demystify the
code at least a bit. Extend the mmu_gather infrastructure to be able to
deal with our special hugetlb PMD table sharing implementation.
To make initialization of the mmu_gather easier when working on a single
VMA (in particular, when dealing with hugetlb), provide
tlb_gather_mmu_vma().
We'll consolidate the handling for (full) unsharing of PMD tables in
tlb_unshare_pmd_ptdesc() and tlb_flush_unshared_tables(), and track
in "struct mmu_gather" whether we had (full) unsharing of PMD tables.
Because locking is very special (concurrent unsharing+reuse must be
prevented), we disallow deferring flushing to tlb_finish_mmu() and instead
require an explicit earlier call to tlb_flush_unshared_tables().
From hugetlb code, we call huge_pmd_unshare_flush() where we make sure
that the expected lock protecting us from concurrent unsharing+reuse is
still held.
Check with a VM_WARN_ON_ONCE() in tlb_finish_mmu() that
tlb_flush_unshared_tables() was properly called earlier.
Document it all properly.
Notes about tlb_remove_table_sync_one() interaction with unsharing:
There are two fairly tricky things:
(1) tlb_remove_table_sync_one() is a NOP on architectures without
CONFIG_MMU_GATHER_RCU_TABLE_FREE.
Here, the assumption is that the previous TLB flush would send an
IPI to all relevant CPUs. Careful: some architectures like x86 only
send IPIs to all relevant CPUs when tlb->freed_tables is set.
The relevant architectures should be selecting
MMU_GATHER_RCU_TABLE_FREE, but x86 might not do that in stable
kernels and it might have been problematic before this patch.
Also, the arch flushing behavior (independent of IPIs) is different
when tlb->freed_tables is set. Do we have to enlighten them to also
take care of tlb->unshared_tables? So far we didn't care, so
hopefully we are fine. Of course, we could be setting
tlb->freed_tables as well, but that might then unnecessarily flush
too much, because the semantics of tlb->freed_tables are a bit
fuzzy.
This patch changes nothing in this regard.
(2) tlb_remove_table_sync_one() is not a NOP on architectures with
CONFIG_MMU_GATHER_RCU_TABLE_FREE that actually don't need a sync.
Take x86 as an example: in the common case (!pv, !X86_FEATURE_INVLPGB)
we still issue IPIs during TLB flushes and don't actually need the
second tlb_remove_table_sync_one().
This optimized can be implemented on top of this, by checking e.g., in
tlb_remove_table_sync_one() whether we really need IPIs. But as
described in (1), it really must honor tlb->freed_tables then to
send IPIs to all relevant CPUs.
Notes on TLB flushing changes:
(1) Flushing for non-shared PMD tables
We're converting from flush_hugetlb_tlb_range() to
tlb_remove_huge_tlb_entry(). Given that we properly initialize the
MMU gather in tlb_gather_mmu_vma() to be hugetlb aware, similar to
__unmap_hugepage_range(), that should be fine.
(2) Flushing for shared PMD tables
We're converting from various things (flush_hugetlb_tlb_range(),
tlb_flush_pmd_range(), flush_tlb_range()) to tlb_flush_pmd_range().
tlb_flush_pmd_range() achieves the same that
tlb_remove_huge_tlb_entry() would achieve in these scenarios.
Note that tlb_remove_huge_tlb_entry() also calls
__tlb_remove_tlb_entry(), however that is only implemented on
powerpc, which does not support PMD table sharing.
Similar to (1), tlb_gather_mmu_vma() should make sure that TLB
flushing keeps on working as expected.
Further, note that the ptdesc_pmd_pts_dec() in huge_pmd_share() is not a
concern, as we are holding the i_mmap_lock the whole time, preventing
concurrent unsharing. That ptdesc_pmd_pts_dec() usage will be removed
separately as a cleanup later.
There are plenty more cleanups to be had, but they have to wait until
this is fixed.
[david@kernel.org: fix kerneldoc]
Link: https://lkml.kernel.org/r/f223dd74-331c-412d-93fc-69e360a5006c@kernel.org
Link: https://lkml.kernel.org/r/20251223214037.580860-5-david@kernel.org
Fixes: 1013af4f585f ("mm/hugetlb: fix huge_pmd_unshare() vs GUP-fast race")
Signed-off-by: David Hildenbrand (Red Hat) <david@kernel.org>
Reported-by: Uschakow, Stanislav" <suschako@amazon.de>
Closes: https://lore.kernel.org/all/4d3878531c76479d9f8ca9789dc6485d@amazon.de/
Tested-by: Laurence Oberman <loberman@redhat.com>
Acked-by: Harry Yoo <harry.yoo@oracle.com>
Reviewed-by: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: Lance Yang <lance.yang@linux.dev>
Cc: Liu Shixin <liushixin2@huawei.com>
Cc: Oscar Salvador <osalvador@suse.de>
Cc: Rik van Riel <riel@surriel.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2025-12-23 22:40:37 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-09-14 15:18:09 -07:00
|
|
|
|
mm/hugetlb: fix excessive IPI broadcasts when unsharing PMD tables using mmu_gather
As reported, ever since commit 1013af4f585f ("mm/hugetlb: fix
huge_pmd_unshare() vs GUP-fast race") we can end up in some situations
where we perform so many IPI broadcasts when unsharing hugetlb PMD page
tables that it severely regresses some workloads.
In particular, when we fork()+exit(), or when we munmap() a large
area backed by many shared PMD tables, we perform one IPI broadcast per
unshared PMD table.
There are two optimizations to be had:
(1) When we process (unshare) multiple such PMD tables, such as during
exit(), it is sufficient to send a single IPI broadcast (as long as
we respect locking rules) instead of one per PMD table.
Locking prevents that any of these PMD tables could get reused before
we drop the lock.
(2) When we are not the last sharer (> 2 users including us), there is
no need to send the IPI broadcast. The shared PMD tables cannot
become exclusive (fully unshared) before an IPI will be broadcasted
by the last sharer.
Concurrent GUP-fast could walk into a PMD table just before we
unshared it. It could then succeed in grabbing a page from the
shared page table even after munmap() etc succeeded (and supressed
an IPI). But there is not difference compared to GUP-fast just
sleeping for a while after grabbing the page and re-enabling IRQs.
Most importantly, GUP-fast will never walk into page tables that are
no-longer shared, because the last sharer will issue an IPI
broadcast.
(if ever required, checking whether the PUD changed in GUP-fast
after grabbing the page like we do in the PTE case could handle
this)
So let's rework PMD sharing TLB flushing + IPI sync to use the mmu_gather
infrastructure so we can implement these optimizations and demystify the
code at least a bit. Extend the mmu_gather infrastructure to be able to
deal with our special hugetlb PMD table sharing implementation.
To make initialization of the mmu_gather easier when working on a single
VMA (in particular, when dealing with hugetlb), provide
tlb_gather_mmu_vma().
We'll consolidate the handling for (full) unsharing of PMD tables in
tlb_unshare_pmd_ptdesc() and tlb_flush_unshared_tables(), and track
in "struct mmu_gather" whether we had (full) unsharing of PMD tables.
Because locking is very special (concurrent unsharing+reuse must be
prevented), we disallow deferring flushing to tlb_finish_mmu() and instead
require an explicit earlier call to tlb_flush_unshared_tables().
From hugetlb code, we call huge_pmd_unshare_flush() where we make sure
that the expected lock protecting us from concurrent unsharing+reuse is
still held.
Check with a VM_WARN_ON_ONCE() in tlb_finish_mmu() that
tlb_flush_unshared_tables() was properly called earlier.
Document it all properly.
Notes about tlb_remove_table_sync_one() interaction with unsharing:
There are two fairly tricky things:
(1) tlb_remove_table_sync_one() is a NOP on architectures without
CONFIG_MMU_GATHER_RCU_TABLE_FREE.
Here, the assumption is that the previous TLB flush would send an
IPI to all relevant CPUs. Careful: some architectures like x86 only
send IPIs to all relevant CPUs when tlb->freed_tables is set.
The relevant architectures should be selecting
MMU_GATHER_RCU_TABLE_FREE, but x86 might not do that in stable
kernels and it might have been problematic before this patch.
Also, the arch flushing behavior (independent of IPIs) is different
when tlb->freed_tables is set. Do we have to enlighten them to also
take care of tlb->unshared_tables? So far we didn't care, so
hopefully we are fine. Of course, we could be setting
tlb->freed_tables as well, but that might then unnecessarily flush
too much, because the semantics of tlb->freed_tables are a bit
fuzzy.
This patch changes nothing in this regard.
(2) tlb_remove_table_sync_one() is not a NOP on architectures with
CONFIG_MMU_GATHER_RCU_TABLE_FREE that actually don't need a sync.
Take x86 as an example: in the common case (!pv, !X86_FEATURE_INVLPGB)
we still issue IPIs during TLB flushes and don't actually need the
second tlb_remove_table_sync_one().
This optimized can be implemented on top of this, by checking e.g., in
tlb_remove_table_sync_one() whether we really need IPIs. But as
described in (1), it really must honor tlb->freed_tables then to
send IPIs to all relevant CPUs.
Notes on TLB flushing changes:
(1) Flushing for non-shared PMD tables
We're converting from flush_hugetlb_tlb_range() to
tlb_remove_huge_tlb_entry(). Given that we properly initialize the
MMU gather in tlb_gather_mmu_vma() to be hugetlb aware, similar to
__unmap_hugepage_range(), that should be fine.
(2) Flushing for shared PMD tables
We're converting from various things (flush_hugetlb_tlb_range(),
tlb_flush_pmd_range(), flush_tlb_range()) to tlb_flush_pmd_range().
tlb_flush_pmd_range() achieves the same that
tlb_remove_huge_tlb_entry() would achieve in these scenarios.
Note that tlb_remove_huge_tlb_entry() also calls
__tlb_remove_tlb_entry(), however that is only implemented on
powerpc, which does not support PMD table sharing.
Similar to (1), tlb_gather_mmu_vma() should make sure that TLB
flushing keeps on working as expected.
Further, note that the ptdesc_pmd_pts_dec() in huge_pmd_share() is not a
concern, as we are holding the i_mmap_lock the whole time, preventing
concurrent unsharing. That ptdesc_pmd_pts_dec() usage will be removed
separately as a cleanup later.
There are plenty more cleanups to be had, but they have to wait until
this is fixed.
[david@kernel.org: fix kerneldoc]
Link: https://lkml.kernel.org/r/f223dd74-331c-412d-93fc-69e360a5006c@kernel.org
Link: https://lkml.kernel.org/r/20251223214037.580860-5-david@kernel.org
Fixes: 1013af4f585f ("mm/hugetlb: fix huge_pmd_unshare() vs GUP-fast race")
Signed-off-by: David Hildenbrand (Red Hat) <david@kernel.org>
Reported-by: Uschakow, Stanislav" <suschako@amazon.de>
Closes: https://lore.kernel.org/all/4d3878531c76479d9f8ca9789dc6485d@amazon.de/
Tested-by: Laurence Oberman <loberman@redhat.com>
Acked-by: Harry Yoo <harry.yoo@oracle.com>
Reviewed-by: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: Lance Yang <lance.yang@linux.dev>
Cc: Liu Shixin <liushixin2@huawei.com>
Cc: Oscar Salvador <osalvador@suse.de>
Cc: Rik van Riel <riel@surriel.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2025-12-23 22:40:37 +01:00
|
|
|
|
|
|
|
|
|
2022-09-14 15:18:09 -07:00
|
|
|
|
2025-12-23 22:40:36 +01:00
|
|
|
|
|
|
|
|
|
2022-09-14 15:18:09 -07:00
|
|
|
|
2024-06-14 09:51:36 +08:00
|
|
|
|
2022-09-14 15:18:09 -07:00
|
|
|
|
|
|
|
|
|
mm/hugetlb: fix excessive IPI broadcasts when unsharing PMD tables using mmu_gather
As reported, ever since commit 1013af4f585f ("mm/hugetlb: fix
huge_pmd_unshare() vs GUP-fast race") we can end up in some situations
where we perform so many IPI broadcasts when unsharing hugetlb PMD page
tables that it severely regresses some workloads.
In particular, when we fork()+exit(), or when we munmap() a large
area backed by many shared PMD tables, we perform one IPI broadcast per
unshared PMD table.
There are two optimizations to be had:
(1) When we process (unshare) multiple such PMD tables, such as during
exit(), it is sufficient to send a single IPI broadcast (as long as
we respect locking rules) instead of one per PMD table.
Locking prevents that any of these PMD tables could get reused before
we drop the lock.
(2) When we are not the last sharer (> 2 users including us), there is
no need to send the IPI broadcast. The shared PMD tables cannot
become exclusive (fully unshared) before an IPI will be broadcasted
by the last sharer.
Concurrent GUP-fast could walk into a PMD table just before we
unshared it. It could then succeed in grabbing a page from the
shared page table even after munmap() etc succeeded (and supressed
an IPI). But there is not difference compared to GUP-fast just
sleeping for a while after grabbing the page and re-enabling IRQs.
Most importantly, GUP-fast will never walk into page tables that are
no-longer shared, because the last sharer will issue an IPI
broadcast.
(if ever required, checking whether the PUD changed in GUP-fast
after grabbing the page like we do in the PTE case could handle
this)
So let's rework PMD sharing TLB flushing + IPI sync to use the mmu_gather
infrastructure so we can implement these optimizations and demystify the
code at least a bit. Extend the mmu_gather infrastructure to be able to
deal with our special hugetlb PMD table sharing implementation.
To make initialization of the mmu_gather easier when working on a single
VMA (in particular, when dealing with hugetlb), provide
tlb_gather_mmu_vma().
We'll consolidate the handling for (full) unsharing of PMD tables in
tlb_unshare_pmd_ptdesc() and tlb_flush_unshared_tables(), and track
in "struct mmu_gather" whether we had (full) unsharing of PMD tables.
Because locking is very special (concurrent unsharing+reuse must be
prevented), we disallow deferring flushing to tlb_finish_mmu() and instead
require an explicit earlier call to tlb_flush_unshared_tables().
From hugetlb code, we call huge_pmd_unshare_flush() where we make sure
that the expected lock protecting us from concurrent unsharing+reuse is
still held.
Check with a VM_WARN_ON_ONCE() in tlb_finish_mmu() that
tlb_flush_unshared_tables() was properly called earlier.
Document it all properly.
Notes about tlb_remove_table_sync_one() interaction with unsharing:
There are two fairly tricky things:
(1) tlb_remove_table_sync_one() is a NOP on architectures without
CONFIG_MMU_GATHER_RCU_TABLE_FREE.
Here, the assumption is that the previous TLB flush would send an
IPI to all relevant CPUs. Careful: some architectures like x86 only
send IPIs to all relevant CPUs when tlb->freed_tables is set.
The relevant architectures should be selecting
MMU_GATHER_RCU_TABLE_FREE, but x86 might not do that in stable
kernels and it might have been problematic before this patch.
Also, the arch flushing behavior (independent of IPIs) is different
when tlb->freed_tables is set. Do we have to enlighten them to also
take care of tlb->unshared_tables? So far we didn't care, so
hopefully we are fine. Of course, we could be setting
tlb->freed_tables as well, but that might then unnecessarily flush
too much, because the semantics of tlb->freed_tables are a bit
fuzzy.
This patch changes nothing in this regard.
(2) tlb_remove_table_sync_one() is not a NOP on architectures with
CONFIG_MMU_GATHER_RCU_TABLE_FREE that actually don't need a sync.
Take x86 as an example: in the common case (!pv, !X86_FEATURE_INVLPGB)
we still issue IPIs during TLB flushes and don't actually need the
second tlb_remove_table_sync_one().
This optimized can be implemented on top of this, by checking e.g., in
tlb_remove_table_sync_one() whether we really need IPIs. But as
described in (1), it really must honor tlb->freed_tables then to
send IPIs to all relevant CPUs.
Notes on TLB flushing changes:
(1) Flushing for non-shared PMD tables
We're converting from flush_hugetlb_tlb_range() to
tlb_remove_huge_tlb_entry(). Given that we properly initialize the
MMU gather in tlb_gather_mmu_vma() to be hugetlb aware, similar to
__unmap_hugepage_range(), that should be fine.
(2) Flushing for shared PMD tables
We're converting from various things (flush_hugetlb_tlb_range(),
tlb_flush_pmd_range(), flush_tlb_range()) to tlb_flush_pmd_range().
tlb_flush_pmd_range() achieves the same that
tlb_remove_huge_tlb_entry() would achieve in these scenarios.
Note that tlb_remove_huge_tlb_entry() also calls
__tlb_remove_tlb_entry(), however that is only implemented on
powerpc, which does not support PMD table sharing.
Similar to (1), tlb_gather_mmu_vma() should make sure that TLB
flushing keeps on working as expected.
Further, note that the ptdesc_pmd_pts_dec() in huge_pmd_share() is not a
concern, as we are holding the i_mmap_lock the whole time, preventing
concurrent unsharing. That ptdesc_pmd_pts_dec() usage will be removed
separately as a cleanup later.
There are plenty more cleanups to be had, but they have to wait until
this is fixed.
[david@kernel.org: fix kerneldoc]
Link: https://lkml.kernel.org/r/f223dd74-331c-412d-93fc-69e360a5006c@kernel.org
Link: https://lkml.kernel.org/r/20251223214037.580860-5-david@kernel.org
Fixes: 1013af4f585f ("mm/hugetlb: fix huge_pmd_unshare() vs GUP-fast race")
Signed-off-by: David Hildenbrand (Red Hat) <david@kernel.org>
Reported-by: Uschakow, Stanislav" <suschako@amazon.de>
Closes: https://lore.kernel.org/all/4d3878531c76479d9f8ca9789dc6485d@amazon.de/
Tested-by: Laurence Oberman <loberman@redhat.com>
Acked-by: Harry Yoo <harry.yoo@oracle.com>
Reviewed-by: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: Lance Yang <lance.yang@linux.dev>
Cc: Liu Shixin <liushixin2@huawei.com>
Cc: Oscar Salvador <osalvador@suse.de>
Cc: Rik van Riel <riel@surriel.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2025-12-23 22:40:37 +01:00
|
|
|
|
2018-10-05 15:51:29 -07:00
|
|
|
|
mm: rmap: fix CONT-PTE/PMD size hugetlb issue when unmapping
On some architectures (like ARM64), it can support CONT-PTE/PMD size
hugetlb, which means it can support not only PMD/PUD size hugetlb: 2M and
1G, but also CONT-PTE/PMD size: 64K and 32M if a 4K page size specified.
When unmapping a hugetlb page, we will get the relevant page table entry
by huge_pte_offset() only once to nuke it. This is correct for PMD or PUD
size hugetlb, since they always contain only one pmd entry or pud entry in
the page table.
However this is incorrect for CONT-PTE and CONT-PMD size hugetlb, since
they can contain several continuous pte or pmd entry with same page table
attributes, so we will nuke only one pte or pmd entry for this
CONT-PTE/PMD size hugetlb page.
And now try_to_unmap() is only passed a hugetlb page in the case where the
hugetlb page is poisoned. Which means now we will unmap only one pte
entry for a CONT-PTE or CONT-PMD size poisoned hugetlb page, and we can
still access other subpages of a CONT-PTE or CONT-PMD size poisoned
hugetlb page, which will cause serious issues possibly.
So we should change to use huge_ptep_clear_flush() to nuke the hugetlb
page table to fix this issue, which already considered CONT-PTE and
CONT-PMD size hugetlb.
We've already used set_huge_swap_pte_at() to set a poisoned swap entry for
a poisoned hugetlb page. Meanwhile adding a VM_BUG_ON() to make sure the
passed hugetlb page is poisoned in try_to_unmap().
Link: https://lkml.kernel.org/r/0a2e547238cad5bc153a85c3e9658cb9d55f9cac.1652270205.git.baolin.wang@linux.alibaba.com
Link: https://lkml.kernel.org/r/730ea4b6d292f32fb10b7a4e87dad49b0eb30474.1652147571.git.baolin.wang@linux.alibaba.com
Signed-off-by: Baolin Wang <baolin.wang@linux.alibaba.com>
Reviewed-by: Muchun Song <songmuchun@bytedance.com>
Reviewed-by: Mike Kravetz <mike.kravetz@oracle.com>
Acked-by: David Hildenbrand <david@redhat.com>
Cc: Alexander Gordeev <agordeev@linux.ibm.com>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Christian Borntraeger <borntraeger@linux.ibm.com>
Cc: David S. Miller <davem@davemloft.net>
Cc: Gerald Schaefer <gerald.schaefer@linux.ibm.com>
Cc: Heiko Carstens <hca@linux.ibm.com>
Cc: Helge Deller <deller@gmx.de>
Cc: James Bottomley <James.Bottomley@HansenPartnership.com>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Rich Felker <dalias@libc.org>
Cc: Sven Schnelle <svens@linux.ibm.com>
Cc: Thomas Bogendoerfer <tsbogend@alpha.franken.de>
Cc: Vasily Gorbik <gor@linux.ibm.com>
Cc: Will Deacon <will@kernel.org>
Cc: Yoshinori Sato <ysato@users.osdn.me>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2022-05-13 16:48:56 -07:00
|
|
|
|
2025-02-10 20:37:52 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-06-27 14:23:19 +08:00
|
|
|
|
mm: support batched unmap for lazyfree large folios during reclamation
Currently, the PTEs and rmap of a large folio are removed one at a time.
This is not only slow but also causes the large folio to be unnecessarily
added to deferred_split, which can lead to races between the
deferred_split shrinker callback and memory reclamation. This patch
releases all PTEs and rmap entries in a batch. Currently, it only handles
lazyfree large folios.
The below microbench tries to reclaim 128MB lazyfree large folios
whose sizes are 64KiB:
#include <stdio.h>
#include <sys/mman.h>
#include <string.h>
#include <time.h>
#define SIZE 128*1024*1024 // 128 MB
unsigned long read_split_deferred()
{
FILE *file = fopen("/sys/kernel/mm/transparent_hugepage"
"/hugepages-64kB/stats/split_deferred", "r");
if (!file) {
perror("Error opening file");
return 0;
}
unsigned long value;
if (fscanf(file, "%lu", &value) != 1) {
perror("Error reading value");
fclose(file);
return 0;
}
fclose(file);
return value;
}
int main(int argc, char *argv[])
{
while(1) {
volatile int *p = mmap(0, SIZE, PROT_READ | PROT_WRITE,
MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
memset((void *)p, 1, SIZE);
madvise((void *)p, SIZE, MADV_FREE);
clock_t start_time = clock();
unsigned long start_split = read_split_deferred();
madvise((void *)p, SIZE, MADV_PAGEOUT);
clock_t end_time = clock();
unsigned long end_split = read_split_deferred();
double elapsed_time = (double)(end_time - start_time) / CLOCKS_PER_SEC;
printf("Time taken by reclamation: %f seconds, split_deferred: %ld\n",
elapsed_time, end_split - start_split);
munmap((void *)p, SIZE);
}
return 0;
}
w/o patch:
~ # ./a.out
Time taken by reclamation: 0.177418 seconds, split_deferred: 2048
Time taken by reclamation: 0.178348 seconds, split_deferred: 2048
Time taken by reclamation: 0.174525 seconds, split_deferred: 2048
Time taken by reclamation: 0.171620 seconds, split_deferred: 2048
Time taken by reclamation: 0.172241 seconds, split_deferred: 2048
Time taken by reclamation: 0.174003 seconds, split_deferred: 2048
Time taken by reclamation: 0.171058 seconds, split_deferred: 2048
Time taken by reclamation: 0.171993 seconds, split_deferred: 2048
Time taken by reclamation: 0.169829 seconds, split_deferred: 2048
Time taken by reclamation: 0.172895 seconds, split_deferred: 2048
Time taken by reclamation: 0.176063 seconds, split_deferred: 2048
Time taken by reclamation: 0.172568 seconds, split_deferred: 2048
Time taken by reclamation: 0.171185 seconds, split_deferred: 2048
Time taken by reclamation: 0.170632 seconds, split_deferred: 2048
Time taken by reclamation: 0.170208 seconds, split_deferred: 2048
Time taken by reclamation: 0.174192 seconds, split_deferred: 2048
...
w/ patch:
~ # ./a.out
Time taken by reclamation: 0.074231 seconds, split_deferred: 0
Time taken by reclamation: 0.071026 seconds, split_deferred: 0
Time taken by reclamation: 0.072029 seconds, split_deferred: 0
Time taken by reclamation: 0.071873 seconds, split_deferred: 0
Time taken by reclamation: 0.073573 seconds, split_deferred: 0
Time taken by reclamation: 0.071906 seconds, split_deferred: 0
Time taken by reclamation: 0.073604 seconds, split_deferred: 0
Time taken by reclamation: 0.075903 seconds, split_deferred: 0
Time taken by reclamation: 0.073191 seconds, split_deferred: 0
Time taken by reclamation: 0.071228 seconds, split_deferred: 0
Time taken by reclamation: 0.071391 seconds, split_deferred: 0
Time taken by reclamation: 0.071468 seconds, split_deferred: 0
Time taken by reclamation: 0.071896 seconds, split_deferred: 0
Time taken by reclamation: 0.072508 seconds, split_deferred: 0
Time taken by reclamation: 0.071884 seconds, split_deferred: 0
Time taken by reclamation: 0.072433 seconds, split_deferred: 0
Time taken by reclamation: 0.071939 seconds, split_deferred: 0
...
Link: https://lkml.kernel.org/r/20250214093015.51024-4-21cnbao@gmail.com
Signed-off-by: Barry Song <v-songbaohua@oppo.com>
Cc: Albert Ou <aou@eecs.berkeley.edu>
Cc: Anshuman Khandual <anshuman.khandual@arm.com>
Cc: Baolin Wang <baolin.wang@linux.alibaba.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Chis Li <chrisl@kernel.org>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Gavin Shan <gshan@redhat.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: "Huang, Ying" <ying.huang@intel.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Kairui Song <kasong@tencent.com>
Cc: Kefeng Wang <wangkefeng.wang@huawei.com>
Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Lance Yang <ioworker0@gmail.com>
Cc: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Mauricio Faria de Oliveira <mfo@canonical.com>
Cc: Palmer Dabbelt <palmer@dabbelt.com>
Cc: Paul Walmsley <paul.walmsley@sifive.com>
Cc: Ryan Roberts <ryan.roberts@arm.com>
Cc: Shaoqin Huang <shahuang@redhat.com>
Cc: Tangquan Zheng <zhengtangquan@oppo.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Will Deacon <will@kernel.org>
Cc: Yicong Yang <yangyicong@hisilicon.com>
Cc: Yosry Ahmed <yosryahmed@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2025-02-14 22:30:14 +13:00
|
|
|
|
|
|
|
|
|
2015-09-04 15:47:32 -07:00
|
|
|
|
mm: support batched unmap for lazyfree large folios during reclamation
Currently, the PTEs and rmap of a large folio are removed one at a time.
This is not only slow but also causes the large folio to be unnecessarily
added to deferred_split, which can lead to races between the
deferred_split shrinker callback and memory reclamation. This patch
releases all PTEs and rmap entries in a batch. Currently, it only handles
lazyfree large folios.
The below microbench tries to reclaim 128MB lazyfree large folios
whose sizes are 64KiB:
#include <stdio.h>
#include <sys/mman.h>
#include <string.h>
#include <time.h>
#define SIZE 128*1024*1024 // 128 MB
unsigned long read_split_deferred()
{
FILE *file = fopen("/sys/kernel/mm/transparent_hugepage"
"/hugepages-64kB/stats/split_deferred", "r");
if (!file) {
perror("Error opening file");
return 0;
}
unsigned long value;
if (fscanf(file, "%lu", &value) != 1) {
perror("Error reading value");
fclose(file);
return 0;
}
fclose(file);
return value;
}
int main(int argc, char *argv[])
{
while(1) {
volatile int *p = mmap(0, SIZE, PROT_READ | PROT_WRITE,
MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
memset((void *)p, 1, SIZE);
madvise((void *)p, SIZE, MADV_FREE);
clock_t start_time = clock();
unsigned long start_split = read_split_deferred();
madvise((void *)p, SIZE, MADV_PAGEOUT);
clock_t end_time = clock();
unsigned long end_split = read_split_deferred();
double elapsed_time = (double)(end_time - start_time) / CLOCKS_PER_SEC;
printf("Time taken by reclamation: %f seconds, split_deferred: %ld\n",
elapsed_time, end_split - start_split);
munmap((void *)p, SIZE);
}
return 0;
}
w/o patch:
~ # ./a.out
Time taken by reclamation: 0.177418 seconds, split_deferred: 2048
Time taken by reclamation: 0.178348 seconds, split_deferred: 2048
Time taken by reclamation: 0.174525 seconds, split_deferred: 2048
Time taken by reclamation: 0.171620 seconds, split_deferred: 2048
Time taken by reclamation: 0.172241 seconds, split_deferred: 2048
Time taken by reclamation: 0.174003 seconds, split_deferred: 2048
Time taken by reclamation: 0.171058 seconds, split_deferred: 2048
Time taken by reclamation: 0.171993 seconds, split_deferred: 2048
Time taken by reclamation: 0.169829 seconds, split_deferred: 2048
Time taken by reclamation: 0.172895 seconds, split_deferred: 2048
Time taken by reclamation: 0.176063 seconds, split_deferred: 2048
Time taken by reclamation: 0.172568 seconds, split_deferred: 2048
Time taken by reclamation: 0.171185 seconds, split_deferred: 2048
Time taken by reclamation: 0.170632 seconds, split_deferred: 2048
Time taken by reclamation: 0.170208 seconds, split_deferred: 2048
Time taken by reclamation: 0.174192 seconds, split_deferred: 2048
...
w/ patch:
~ # ./a.out
Time taken by reclamation: 0.074231 seconds, split_deferred: 0
Time taken by reclamation: 0.071026 seconds, split_deferred: 0
Time taken by reclamation: 0.072029 seconds, split_deferred: 0
Time taken by reclamation: 0.071873 seconds, split_deferred: 0
Time taken by reclamation: 0.073573 seconds, split_deferred: 0
Time taken by reclamation: 0.071906 seconds, split_deferred: 0
Time taken by reclamation: 0.073604 seconds, split_deferred: 0
Time taken by reclamation: 0.075903 seconds, split_deferred: 0
Time taken by reclamation: 0.073191 seconds, split_deferred: 0
Time taken by reclamation: 0.071228 seconds, split_deferred: 0
Time taken by reclamation: 0.071391 seconds, split_deferred: 0
Time taken by reclamation: 0.071468 seconds, split_deferred: 0
Time taken by reclamation: 0.071896 seconds, split_deferred: 0
Time taken by reclamation: 0.072508 seconds, split_deferred: 0
Time taken by reclamation: 0.071884 seconds, split_deferred: 0
Time taken by reclamation: 0.072433 seconds, split_deferred: 0
Time taken by reclamation: 0.071939 seconds, split_deferred: 0
...
Link: https://lkml.kernel.org/r/20250214093015.51024-4-21cnbao@gmail.com
Signed-off-by: Barry Song <v-songbaohua@oppo.com>
Cc: Albert Ou <aou@eecs.berkeley.edu>
Cc: Anshuman Khandual <anshuman.khandual@arm.com>
Cc: Baolin Wang <baolin.wang@linux.alibaba.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Chis Li <chrisl@kernel.org>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Gavin Shan <gshan@redhat.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: "Huang, Ying" <ying.huang@intel.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Kairui Song <kasong@tencent.com>
Cc: Kefeng Wang <wangkefeng.wang@huawei.com>
Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Lance Yang <ioworker0@gmail.com>
Cc: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Mauricio Faria de Oliveira <mfo@canonical.com>
Cc: Palmer Dabbelt <palmer@dabbelt.com>
Cc: Paul Walmsley <paul.walmsley@sifive.com>
Cc: Ryan Roberts <ryan.roberts@arm.com>
Cc: Shaoqin Huang <shahuang@redhat.com>
Cc: Tangquan Zheng <zhengtangquan@oppo.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Will Deacon <will@kernel.org>
Cc: Yicong Yang <yangyicong@hisilicon.com>
Cc: Yosry Ahmed <yosryahmed@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2025-02-14 22:30:14 +13:00
|
|
|
|
mm: add get_and_clear_ptes() and clear_ptes()
Patch series "Optimizations for khugepaged", v4.
If the underlying folio mapped by the ptes is large, we can process those
ptes in a batch using folio_pte_batch().
For arm64 specifically, this results in a 16x reduction in the number of
ptep_get() calls, since on a contig block, ptep_get() on arm64 will
iterate through all 16 entries to collect a/d bits. Next, ptep_clear()
will cause a TLBI for every contig block in the range via
contpte_try_unfold(). Instead, use clear_ptes() to only do the TLBI at
the first and last contig block of the range.
For split folios, there will be no pte batching; the batch size returned
by folio_pte_batch() will be 1. For pagetable split folios, the ptes will
still point to the same large folio; for arm64, this results in the
optimization described above, and for other arches, a minor improvement is
expected due to a reduction in the number of function calls and batching
atomic operations.
This patch (of 3):
Let's add variants to be used where "full" does not apply -- which will
be the majority of cases in the future. "full" really only applies if
we are about to tear down a full MM.
Use get_and_clear_ptes() in existing code, clear_ptes() users will
be added next.
Link: https://lkml.kernel.org/r/20250724052301.23844-2-dev.jain@arm.com
Signed-off-by: David Hildenbrand <david@redhat.com>
Signed-off-by: Dev Jain <dev.jain@arm.com>
Reviewed-by: Baolin Wang <baolin.wang@linux.alibaba.com>
Reviewed-by: Barry Song <baohua@kernel.org>
Reviewed-by: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Reviewed-by: Zi Yan <ziy@nvidia.com>
Cc: Liam Howlett <liam.howlett@oracle.com>
Cc: Mariano Pache <npache@redhat.com>
Cc: Ryan Roberts <ryan.roberts@arm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2025-07-24 10:52:59 +05:30
|
|
|
|
mm: support batched unmap for lazyfree large folios during reclamation
Currently, the PTEs and rmap of a large folio are removed one at a time.
This is not only slow but also causes the large folio to be unnecessarily
added to deferred_split, which can lead to races between the
deferred_split shrinker callback and memory reclamation. This patch
releases all PTEs and rmap entries in a batch. Currently, it only handles
lazyfree large folios.
The below microbench tries to reclaim 128MB lazyfree large folios
whose sizes are 64KiB:
#include <stdio.h>
#include <sys/mman.h>
#include <string.h>
#include <time.h>
#define SIZE 128*1024*1024 // 128 MB
unsigned long read_split_deferred()
{
FILE *file = fopen("/sys/kernel/mm/transparent_hugepage"
"/hugepages-64kB/stats/split_deferred", "r");
if (!file) {
perror("Error opening file");
return 0;
}
unsigned long value;
if (fscanf(file, "%lu", &value) != 1) {
perror("Error reading value");
fclose(file);
return 0;
}
fclose(file);
return value;
}
int main(int argc, char *argv[])
{
while(1) {
volatile int *p = mmap(0, SIZE, PROT_READ | PROT_WRITE,
MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
memset((void *)p, 1, SIZE);
madvise((void *)p, SIZE, MADV_FREE);
clock_t start_time = clock();
unsigned long start_split = read_split_deferred();
madvise((void *)p, SIZE, MADV_PAGEOUT);
clock_t end_time = clock();
unsigned long end_split = read_split_deferred();
double elapsed_time = (double)(end_time - start_time) / CLOCKS_PER_SEC;
printf("Time taken by reclamation: %f seconds, split_deferred: %ld\n",
elapsed_time, end_split - start_split);
munmap((void *)p, SIZE);
}
return 0;
}
w/o patch:
~ # ./a.out
Time taken by reclamation: 0.177418 seconds, split_deferred: 2048
Time taken by reclamation: 0.178348 seconds, split_deferred: 2048
Time taken by reclamation: 0.174525 seconds, split_deferred: 2048
Time taken by reclamation: 0.171620 seconds, split_deferred: 2048
Time taken by reclamation: 0.172241 seconds, split_deferred: 2048
Time taken by reclamation: 0.174003 seconds, split_deferred: 2048
Time taken by reclamation: 0.171058 seconds, split_deferred: 2048
Time taken by reclamation: 0.171993 seconds, split_deferred: 2048
Time taken by reclamation: 0.169829 seconds, split_deferred: 2048
Time taken by reclamation: 0.172895 seconds, split_deferred: 2048
Time taken by reclamation: 0.176063 seconds, split_deferred: 2048
Time taken by reclamation: 0.172568 seconds, split_deferred: 2048
Time taken by reclamation: 0.171185 seconds, split_deferred: 2048
Time taken by reclamation: 0.170632 seconds, split_deferred: 2048
Time taken by reclamation: 0.170208 seconds, split_deferred: 2048
Time taken by reclamation: 0.174192 seconds, split_deferred: 2048
...
w/ patch:
~ # ./a.out
Time taken by reclamation: 0.074231 seconds, split_deferred: 0
Time taken by reclamation: 0.071026 seconds, split_deferred: 0
Time taken by reclamation: 0.072029 seconds, split_deferred: 0
Time taken by reclamation: 0.071873 seconds, split_deferred: 0
Time taken by reclamation: 0.073573 seconds, split_deferred: 0
Time taken by reclamation: 0.071906 seconds, split_deferred: 0
Time taken by reclamation: 0.073604 seconds, split_deferred: 0
Time taken by reclamation: 0.075903 seconds, split_deferred: 0
Time taken by reclamation: 0.073191 seconds, split_deferred: 0
Time taken by reclamation: 0.071228 seconds, split_deferred: 0
Time taken by reclamation: 0.071391 seconds, split_deferred: 0
Time taken by reclamation: 0.071468 seconds, split_deferred: 0
Time taken by reclamation: 0.071896 seconds, split_deferred: 0
Time taken by reclamation: 0.072508 seconds, split_deferred: 0
Time taken by reclamation: 0.071884 seconds, split_deferred: 0
Time taken by reclamation: 0.072433 seconds, split_deferred: 0
Time taken by reclamation: 0.071939 seconds, split_deferred: 0
...
Link: https://lkml.kernel.org/r/20250214093015.51024-4-21cnbao@gmail.com
Signed-off-by: Barry Song <v-songbaohua@oppo.com>
Cc: Albert Ou <aou@eecs.berkeley.edu>
Cc: Anshuman Khandual <anshuman.khandual@arm.com>
Cc: Baolin Wang <baolin.wang@linux.alibaba.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Chis Li <chrisl@kernel.org>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Gavin Shan <gshan@redhat.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: "Huang, Ying" <ying.huang@intel.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Kairui Song <kasong@tencent.com>
Cc: Kefeng Wang <wangkefeng.wang@huawei.com>
Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Lance Yang <ioworker0@gmail.com>
Cc: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Mauricio Faria de Oliveira <mfo@canonical.com>
Cc: Palmer Dabbelt <palmer@dabbelt.com>
Cc: Paul Walmsley <paul.walmsley@sifive.com>
Cc: Ryan Roberts <ryan.roberts@arm.com>
Cc: Shaoqin Huang <shahuang@redhat.com>
Cc: Tangquan Zheng <zhengtangquan@oppo.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Will Deacon <will@kernel.org>
Cc: Yicong Yang <yangyicong@hisilicon.com>
Cc: Yosry Ahmed <yosryahmed@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2025-02-14 22:30:14 +13:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-02-10 20:37:52 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-02-24 14:58:01 -08:00
|
|
|
|
|
|
|
|
|
2022-05-12 20:22:53 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-02-24 14:58:01 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-02-21 17:59:05 +09:00
|
|
|
|
2017-07-06 15:39:53 -07:00
|
|
|
|
2022-02-15 09:28:49 -05:00
|
|
|
|
|
|
|
|
|
mm: hugetlb: add huge page size param to set_huge_pte_at()
Patch series "Fix set_huge_pte_at() panic on arm64", v2.
This series fixes a bug in arm64's implementation of set_huge_pte_at(),
which can result in an unprivileged user causing a kernel panic. The
problem was triggered when running the new uffd poison mm selftest for
HUGETLB memory. This test (and the uffd poison feature) was merged for
v6.5-rc7.
Ideally, I'd like to get this fix in for v6.6 and I've cc'ed stable
(correctly this time) to get it backported to v6.5, where the issue first
showed up.
Description of Bug
==================
arm64's huge pte implementation supports multiple huge page sizes, some of
which are implemented in the page table with multiple contiguous entries.
So set_huge_pte_at() needs to work out how big the logical pte is, so that
it can also work out how many physical ptes (or pmds) need to be written.
It previously did this by grabbing the folio out of the pte and querying
its size.
However, there are cases when the pte being set is actually a swap entry.
But this also used to work fine, because for huge ptes, we only ever saw
migration entries and hwpoison entries. And both of these types of swap
entries have a PFN embedded, so the code would grab that and everything
still worked out.
But over time, more calls to set_huge_pte_at() have been added that set
swap entry types that do not embed a PFN. And this causes the code to go
bang. The triggering case is for the uffd poison test, commit
99aa77215ad0 ("selftests/mm: add uffd unit test for UFFDIO_POISON"), which
causes a PTE_MARKER_POISONED swap entry to be set, coutesey of commit
8a13897fb0da ("mm: userfaultfd: support UFFDIO_POISON for hugetlbfs") -
added in v6.5-rc7. Although review shows that there are other call sites
that set PTE_MARKER_UFFD_WP (which also has no PFN), these don't trigger
on arm64 because arm64 doesn't support UFFD WP.
If CONFIG_DEBUG_VM is enabled, we do at least get a BUG(), but otherwise,
it will dereference a bad pointer in page_folio():
static inline struct folio *hugetlb_swap_entry_to_folio(swp_entry_t entry)
{
VM_BUG_ON(!is_migration_entry(entry) && !is_hwpoison_entry(entry));
return page_folio(pfn_to_page(swp_offset_pfn(entry)));
}
Fix
===
The simplest fix would have been to revert the dodgy cleanup commit
18f3962953e4 ("mm: hugetlb: kill set_huge_swap_pte_at()"), but since
things have moved on, this would have required an audit of all the new
set_huge_pte_at() call sites to see if they should be converted to
set_huge_swap_pte_at(). As per the original intent of the change, it
would also leave us open to future bugs when people invariably get it
wrong and call the wrong helper.
So instead, I've added a huge page size parameter to set_huge_pte_at().
This means that the arm64 code has the size in all cases. It's a bigger
change, due to needing to touch the arches that implement the function,
but it is entirely mechanical, so in my view, low risk.
I've compile-tested all touched arches; arm64, parisc, powerpc, riscv,
s390, sparc (and additionally x86_64). I've additionally booted and run
mm selftests against arm64, where I observe the uffd poison test is fixed,
and there are no other regressions.
This patch (of 2):
In order to fix a bug, arm64 needs to be told the size of the huge page
for which the pte is being set in set_huge_pte_at(). Provide for this by
adding an `unsigned long sz` parameter to the function. This follows the
same pattern as huge_pte_clear().
This commit makes the required interface modifications to the core mm as
well as all arches that implement this function (arm64, parisc, powerpc,
riscv, s390, sparc). The actual arm64 bug will be fixed in a separate
commit.
No behavioral changes intended.
Link: https://lkml.kernel.org/r/20230922115804.2043771-1-ryan.roberts@arm.com
Link: https://lkml.kernel.org/r/20230922115804.2043771-2-ryan.roberts@arm.com
Fixes: 8a13897fb0da ("mm: userfaultfd: support UFFDIO_POISON for hugetlbfs")
Signed-off-by: Ryan Roberts <ryan.roberts@arm.com>
Reviewed-by: Christophe Leroy <christophe.leroy@csgroup.eu> [powerpc 8xx]
Reviewed-by: Lorenzo Stoakes <lstoakes@gmail.com> [vmalloc change]
Cc: Alexandre Ghiti <alex@ghiti.fr>
Cc: Albert Ou <aou@eecs.berkeley.edu>
Cc: Alexander Gordeev <agordeev@linux.ibm.com>
Cc: Anshuman Khandual <anshuman.khandual@arm.com>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Axel Rasmussen <axelrasmussen@google.com>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Christian Borntraeger <borntraeger@linux.ibm.com>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: David S. Miller <davem@davemloft.net>
Cc: Gerald Schaefer <gerald.schaefer@linux.ibm.com>
Cc: Heiko Carstens <hca@linux.ibm.com>
Cc: Helge Deller <deller@gmx.de>
Cc: "James E.J. Bottomley" <James.Bottomley@HansenPartnership.com>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Muchun Song <muchun.song@linux.dev>
Cc: Nicholas Piggin <npiggin@gmail.com>
Cc: Palmer Dabbelt <palmer@dabbelt.com>
Cc: Paul Walmsley <paul.walmsley@sifive.com>
Cc: Peter Xu <peterx@redhat.com>
Cc: Qi Zheng <zhengqi.arch@bytedance.com>
Cc: Ryan Roberts <ryan.roberts@arm.com>
Cc: SeongJae Park <sj@kernel.org>
Cc: Sven Schnelle <svens@linux.ibm.com>
Cc: Uladzislau Rezki (Sony) <urezki@gmail.com>
Cc: Vasily Gorbik <gor@linux.ibm.com>
Cc: Will Deacon <will@kernel.org>
Cc: <stable@vger.kernel.org> [6.5+]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2023-09-22 12:58:03 +01:00
|
|
|
|
|
|
|
|
|
2017-02-24 14:58:01 -08:00
|
|
|
|
2024-01-11 15:24:28 +00:00
|
|
|
|
Revert "rmap: do not call mmu_notifier_invalidate_page() under ptl"
This reverts commit aac2fea94f7a3df8ad1eeb477eb2643f81fd5393.
It turns out that that patch was complete and utter garbage, and broke
KVM, resulting in odd oopses.
Quoting Andrea Arcangeli:
"The aforementioned commit has 3 bugs.
1) mmu_notifier_invalidate_range cannot be used in replacement of
mmu_notifier_invalidate_range_start/end.
For KVM mmu_notifier_invalidate_range is a noop and rightfully so.
A MMU notifier implementation has to implement either
->invalidate_range method or the invalidate_range_start/end
methods, not both. And if you implement invalidate_range_start/end
like KVM is forced to do, calling mmu_notifier_invalidate_range in
common code is a noop for KVM.
For those MMU notifiers that can get away only implementing
->invalidate_range, the ->invalidate_range is implicitly called by
mmu_notifier_invalidate_range_end(). And only those secondary MMUs
that share the same pagetable with the primary MMU (like AMD
iommuv2) can get away only implementing ->invalidate_range.
So all cases (THP on/off) are broken right now.
To fix this is enough to replace mmu_notifier_invalidate_range with
mmu_notifier_invalidate_range_start;mmu_notifier_invalidate_range_end.
Either that or call multiple mmu_notifier_invalidate_page like
before.
2) address + (1UL << compound_order(page) is buggy, it should be
PAGE_SIZE << compound_order(page), it's bytes not pages, 2M not
512.
3) The whole invalidate_range thing was an attempt to call a single
invalidate while walking multiple 4k ptes that maps the same THP
(after a pmd virtual split without physical compound page THP
split).
It's unclear if the rmap_walk will always provide an address that
is 2M aligned as parameter to try_to_unmap_one, in presence of THP.
I think it needs also an address &= (PAGE_SIZE <<
compound_order(page)) - 1 to be safe"
In general, we should stop making excuses for horrible MMU notifier
users. It's much more important that the core VM is sane and safe, than
letting MMU notifiers sleep.
So if some MMU notifier is sleeping under a spinlock, we need to fix the
notifier, not try to make excuses for that garbage in the core VM.
Reported-and-tested-by: Bernhard Held <berny156@gmx.de>
Reported-and-tested-by: Adam Borowski <kilobyte@angband.pl>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Radim Krčmář <rkrcmar@redhat.com>
Cc: Wanpeng Li <kernellwp@gmail.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Takashi Iwai <tiwai@suse.de>
Cc: Nadav Amit <nadav.amit@gmail.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Jérôme Glisse <jglisse@redhat.com>
Cc: axie <axie@amd.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-08-29 09:11:06 -07:00
|
|
|
|
2017-02-24 14:58:01 -08:00
|
|
|
|
2025-02-10 20:37:52 +01:00
|
|
|
|
|
|
|
|
|
2017-02-24 14:58:01 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-07-13 16:58:52 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-02-24 14:58:01 -08:00
|
|
|
|
2024-01-11 15:24:28 +00:00
|
|
|
|
2022-02-15 09:28:49 -05:00
|
|
|
|
2023-08-21 18:08:46 +02:00
|
|
|
|
2017-02-24 14:58:01 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-02-15 09:28:49 -05:00
|
|
|
|
|
|
|
|
|
2017-05-03 14:52:36 -07:00
|
|
|
|
2024-06-14 09:51:36 +08:00
|
|
|
|
2017-05-03 14:52:36 -07:00
|
|
|
|
mm: support madvise(MADV_FREE)
Linux doesn't have an ability to free pages lazy while other OS already
have been supported that named by madvise(MADV_FREE).
The gain is clear that kernel can discard freed pages rather than
swapping out or OOM if memory pressure happens.
Without memory pressure, freed pages would be reused by userspace
without another additional overhead(ex, page fault + allocation +
zeroing).
Jason Evans said:
: Facebook has been using MAP_UNINITIALIZED
: (https://lkml.org/lkml/2012/1/18/308) in some of its applications for
: several years, but there are operational costs to maintaining this
: out-of-tree in our kernel and in jemalloc, and we are anxious to retire it
: in favor of MADV_FREE. When we first enabled MAP_UNINITIALIZED it
: increased throughput for much of our workload by ~5%, and although the
: benefit has decreased using newer hardware and kernels, there is still
: enough benefit that we cannot reasonably retire it without a replacement.
:
: Aside from Facebook operations, there are numerous broadly used
: applications that would benefit from MADV_FREE. The ones that immediately
: come to mind are redis, varnish, and MariaDB. I don't have much insight
: into Android internals and development process, but I would hope to see
: MADV_FREE support eventually end up there as well to benefit applications
: linked with the integrated jemalloc.
:
: jemalloc will use MADV_FREE once it becomes available in the Linux kernel.
: In fact, jemalloc already uses MADV_FREE or equivalent everywhere it's
: available: *BSD, OS X, Windows, and Solaris -- every platform except Linux
: (and AIX, but I'm not sure it even compiles on AIX). The lack of
: MADV_FREE on Linux forced me down a long series of increasingly
: sophisticated heuristics for madvise() volume reduction, and even so this
: remains a common performance issue for people using jemalloc on Linux.
: Please integrate MADV_FREE; many people will benefit substantially.
How it works:
When madvise syscall is called, VM clears dirty bit of ptes of the
range. If memory pressure happens, VM checks dirty bit of page table
and if it found still "clean", it means it's a "lazyfree pages" so VM
could discard the page instead of swapping out. Once there was store
operation for the page before VM peek a page to reclaim, dirty bit is
set so VM can swap out the page instead of discarding.
One thing we should notice is that basically, MADV_FREE relies on dirty
bit in page table entry to decide whether VM allows to discard the page
or not. IOW, if page table entry includes marked dirty bit, VM
shouldn't discard the page.
However, as a example, if swap-in by read fault happens, page table
entry doesn't have dirty bit so MADV_FREE could discard the page
wrongly.
For avoiding the problem, MADV_FREE did more checks with PageDirty and
PageSwapCache. It worked out because swapped-in page lives on swap
cache and since it is evicted from the swap cache, the page has PG_dirty
flag. So both page flags check effectively prevent wrong discarding by
MADV_FREE.
However, a problem in above logic is that swapped-in page has PG_dirty
still after they are removed from swap cache so VM cannot consider the
page as freeable any more even if madvise_free is called in future.
Look at below example for detail.
ptr = malloc();
memset(ptr);
..
..
.. heavy memory pressure so all of pages are swapped out
..
..
var = *ptr; -> a page swapped-in and could be removed from
swapcache. Then, page table doesn't mark
dirty bit and page descriptor includes PG_dirty
..
..
madvise_free(ptr); -> It doesn't clear PG_dirty of the page.
..
..
..
.. heavy memory pressure again.
.. In this time, VM cannot discard the page because the page
.. has *PG_dirty*
To solve the problem, this patch clears PG_dirty if only the page is
owned exclusively by current process when madvise is called because
PG_dirty represents ptes's dirtiness in several processes so we could
clear it only if we own it exclusively.
Firstly, heavy users would be general allocators(ex, jemalloc, tcmalloc
and hope glibc supports it) and jemalloc/tcmalloc already have supported
the feature for other OS(ex, FreeBSD)
barrios@blaptop:~/benchmark/ebizzy$ lscpu
Architecture: x86_64
CPU op-mode(s): 32-bit, 64-bit
Byte Order: Little Endian
CPU(s): 12
On-line CPU(s) list: 0-11
Thread(s) per core: 1
Core(s) per socket: 1
Socket(s): 12
NUMA node(s): 1
Vendor ID: GenuineIntel
CPU family: 6
Model: 2
Stepping: 3
CPU MHz: 3200.185
BogoMIPS: 6400.53
Virtualization: VT-x
Hypervisor vendor: KVM
Virtualization type: full
L1d cache: 32K
L1i cache: 32K
L2 cache: 4096K
NUMA node0 CPU(s): 0-11
ebizzy benchmark(./ebizzy -S 10 -n 512)
Higher avg is better.
vanilla-jemalloc MADV_free-jemalloc
1 thread
records: 10 records: 10
avg: 2961.90 avg: 12069.70
std: 71.96(2.43%) std: 186.68(1.55%)
max: 3070.00 max: 12385.00
min: 2796.00 min: 11746.00
2 thread
records: 10 records: 10
avg: 5020.00 avg: 17827.00
std: 264.87(5.28%) std: 358.52(2.01%)
max: 5244.00 max: 18760.00
min: 4251.00 min: 17382.00
4 thread
records: 10 records: 10
avg: 8988.80 avg: 27930.80
std: 1175.33(13.08%) std: 3317.33(11.88%)
max: 9508.00 max: 30879.00
min: 5477.00 min: 21024.00
8 thread
records: 10 records: 10
avg: 13036.50 avg: 33739.40
std: 170.67(1.31%) std: 5146.22(15.25%)
max: 13371.00 max: 40572.00
min: 12785.00 min: 24088.00
16 thread
records: 10 records: 10
avg: 11092.40 avg: 31424.20
std: 710.60(6.41%) std: 3763.89(11.98%)
max: 12446.00 max: 36635.00
min: 9949.00 min: 25669.00
32 thread
records: 10 records: 10
avg: 11067.00 avg: 34495.80
std: 971.06(8.77%) std: 2721.36(7.89%)
max: 12010.00 max: 38598.00
min: 9002.00 min: 30636.00
In summary, MADV_FREE is about much faster than MADV_DONTNEED.
This patch (of 12):
Add core MADV_FREE implementation.
[akpm@linux-foundation.org: small cleanups]
Signed-off-by: Minchan Kim <minchan@kernel.org>
Acked-by: Michal Hocko <mhocko@suse.com>
Acked-by: Hugh Dickins <hughd@google.com>
Cc: Mika Penttil <mika.penttila@nextfour.com>
Cc: Michael Kerrisk <mtk.manpages@gmail.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Rik van Riel <riel@redhat.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Jason Evans <je@fb.com>
Cc: Daniel Micay <danielmicay@gmail.com>
Cc: "Kirill A. Shutemov" <kirill@shutemov.name>
Cc: Shaohua Li <shli@kernel.org>
Cc: <yalin.wang2010@gmail.com>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: "James E.J. Bottomley" <jejb@parisc-linux.org>
Cc: "Kirill A. Shutemov" <kirill@shutemov.name>
Cc: "Shaohua Li" <shli@kernel.org>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Chen Gang <gang.chen.5i5j@gmail.com>
Cc: Chris Zankel <chris@zankel.net>
Cc: Darrick J. Wong <darrick.wong@oracle.com>
Cc: David S. Miller <davem@davemloft.net>
Cc: Helge Deller <deller@gmx.de>
Cc: Ivan Kokshaysky <ink@jurassic.park.msu.ru>
Cc: Matt Turner <mattst88@gmail.com>
Cc: Max Filippov <jcmvbkbc@gmail.com>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: Richard Henderson <rth@twiddle.net>
Cc: Roland Dreier <roland@kernel.org>
Cc: Russell King <rmk@arm.linux.org.uk>
Cc: Shaohua Li <shli@kernel.org>
Cc: Will Deacon <will.deacon@arm.com>
Cc: Wu Fengguang <fengguang.wu@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-01-15 16:54:53 -08:00
|
|
|
|
2017-05-03 14:52:32 -07:00
|
|
|
|
2022-02-15 09:28:49 -05:00
|
|
|
|
mm: fix race between MADV_FREE reclaim and blkdev direct IO read
Problem:
=======
Userspace might read the zero-page instead of actual data from a direct IO
read on a block device if the buffers have been called madvise(MADV_FREE)
on earlier (this is discussed below) due to a race between page reclaim on
MADV_FREE and blkdev direct IO read.
- Race condition:
==============
During page reclaim, the MADV_FREE page check in try_to_unmap_one() checks
if the page is not dirty, then discards its rmap PTE(s) (vs. remap back
if the page is dirty).
However, after try_to_unmap_one() returns to shrink_page_list(), it might
keep the page _anyway_ if page_ref_freeze() fails (it expects exactly
_one_ page reference, from the isolation for page reclaim).
Well, blkdev_direct_IO() gets references for all pages, and on READ
operations it only sets them dirty _later_.
So, if MADV_FREE'd pages (i.e., not dirty) are used as buffers for direct
IO read from block devices, and page reclaim happens during
__blkdev_direct_IO[_simple]() exactly AFTER bio_iov_iter_get_pages()
returns, but BEFORE the pages are set dirty, the situation happens.
The direct IO read eventually completes. Now, when userspace reads the
buffers, the PTE is no longer there and the page fault handler
do_anonymous_page() services that with the zero-page, NOT the data!
A synthetic reproducer is provided.
- Page faults:
===========
If page reclaim happens BEFORE bio_iov_iter_get_pages() the issue doesn't
happen, because that faults-in all pages as writeable, so
do_anonymous_page() sets up a new page/rmap/PTE, and that is used by
direct IO. The userspace reads don't fault as the PTE is there (thus
zero-page is not used/setup).
But if page reclaim happens AFTER it / BEFORE setting pages dirty, the PTE
is no longer there; the subsequent page faults can't help:
The data-read from the block device probably won't generate faults due to
DMA (no MMU) but even in the case it wouldn't use DMA, that happens on
different virtual addresses (not user-mapped addresses) because `struct
bio_vec` stores `struct page` to figure addresses out (which are different
from user-mapped addresses) for the read.
Thus userspace reads (to user-mapped addresses) still fault, then
do_anonymous_page() gets another `struct page` that would address/ map to
other memory than the `struct page` used by `struct bio_vec` for the read.
(The original `struct page` is not available, since it wasn't freed, as
page_ref_freeze() failed due to more page refs. And even if it were
available, its data cannot be trusted anymore.)
Solution:
========
One solution is to check for the expected page reference count in
try_to_unmap_one().
There should be one reference from the isolation (that is also checked in
shrink_page_list() with page_ref_freeze()) plus one or more references
from page mapping(s) (put in discard: label). Further references mean
that rmap/PTE cannot be unmapped/nuked.
(Note: there might be more than one reference from mapping due to
fork()/clone() without CLONE_VM, which use the same `struct page` for
references, until the copy-on-write page gets copied.)
So, additional page references (e.g., from direct IO read) now prevent the
rmap/PTE from being unmapped/dropped; similarly to the page is not freed
per shrink_page_list()/page_ref_freeze()).
- Races and Barriers:
==================
The new check in try_to_unmap_one() should be safe in races with
bio_iov_iter_get_pages() in get_user_pages() fast and slow paths, as it's
done under the PTE lock.
The fast path doesn't take the lock, but it checks if the PTE has changed
and if so, it drops the reference and leaves the page for the slow path
(which does take that lock).
The fast path requires synchronization w/ full memory barrier: it writes
the page reference count first then it reads the PTE later, while
try_to_unmap() writes PTE first then it reads page refcount.
And a second barrier is needed, as the page dirty flag should not be read
before the page reference count (as in __remove_mapping()). (This can be
a load memory barrier only; no writes are involved.)
Call stack/comments:
- try_to_unmap_one()
- page_vma_mapped_walk()
- map_pte() # see pte_offset_map_lock():
pte_offset_map()
spin_lock()
- ptep_get_and_clear() # write PTE
- smp_mb() # (new barrier) GUP fast path
- page_ref_count() # (new check) read refcount
- page_vma_mapped_walk_done() # see pte_unmap_unlock():
pte_unmap()
spin_unlock()
- bio_iov_iter_get_pages()
- __bio_iov_iter_get_pages()
- iov_iter_get_pages()
- get_user_pages_fast()
- internal_get_user_pages_fast()
# fast path
- lockless_pages_from_mm()
- gup_{pgd,p4d,pud,pmd,pte}_range()
ptep = pte_offset_map() # not _lock()
pte = ptep_get_lockless(ptep)
page = pte_page(pte)
try_grab_compound_head(page) # inc refcount
# (RMW/barrier
# on success)
if (pte_val(pte) != pte_val(*ptep)) # read PTE
put_compound_head(page) # dec refcount
# go slow path
# slow path
- __gup_longterm_unlocked()
- get_user_pages_unlocked()
- __get_user_pages_locked()
- __get_user_pages()
- follow_{page,p4d,pud,pmd}_mask()
- follow_page_pte()
ptep = pte_offset_map_lock()
pte = *ptep
page = vm_normal_page(pte)
try_grab_page(page) # inc refcount
pte_unmap_unlock()
- Huge Pages:
==========
Regarding transparent hugepages, that logic shouldn't change, as MADV_FREE
(aka lazyfree) pages are PageAnon() && !PageSwapBacked()
(madvise_free_pte_range() -> mark_page_lazyfree() -> lru_lazyfree_fn())
thus should reach shrink_page_list() -> split_huge_page_to_list() before
try_to_unmap[_one](), so it deals with normal pages only.
(And in case unlikely/TTU_SPLIT_HUGE_PMD/split_huge_pmd_address() happens,
which should not or be rare, the page refcount should be greater than
mapcount: the head page is referenced by tail pages. That also prevents
checking the head `page` then incorrectly call page_remove_rmap(subpage)
for a tail page, that isn't even in the shrink_page_list()'s page_list (an
effect of split huge pmd/pmvw), as it might happen today in this unlikely
scenario.)
MADV_FREE'd buffers:
===================
So, back to the "if MADV_FREE pages are used as buffers" note. The case
is arguable, and subject to multiple interpretations.
The madvise(2) manual page on the MADV_FREE advice value says:
1) 'After a successful MADV_FREE ... data will be lost when
the kernel frees the pages.'
2) 'the free operation will be canceled if the caller writes
into the page' / 'subsequent writes ... will succeed and
then [the] kernel cannot free those dirtied pages'
3) 'If there is no subsequent write, the kernel can free the
pages at any time.'
Thoughts, questions, considerations... respectively:
1) Since the kernel didn't actually free the page (page_ref_freeze()
failed), should the data not have been lost? (on userspace read.)
2) Should writes performed by the direct IO read be able to cancel
the free operation?
- Should the direct IO read be considered as 'the caller' too,
as it's been requested by 'the caller'?
- Should the bio technique to dirty pages on return to userspace
(bio_check_pages_dirty() is called/used by __blkdev_direct_IO())
be considered in another/special way here?
3) Should an upcoming write from a previously requested direct IO
read be considered as a subsequent write, so the kernel should
not free the pages? (as it's known at the time of page reclaim.)
And lastly:
Technically, the last point would seem a reasonable consideration and
balance, as the madvise(2) manual page apparently (and fairly) seem to
assume that 'writes' are memory access from the userspace process (not
explicitly considering writes from the kernel or its corner cases; again,
fairly).. plus the kernel fix implementation for the corner case of the
largely 'non-atomic write' encompassed by a direct IO read operation, is
relatively simple; and it helps.
Reproducer:
==========
@ test.c (simplified, but works)
#define _GNU_SOURCE
#include <fcntl.h>
#include <stdio.h>
#include <unistd.h>
#include <sys/mman.h>
int main() {
int fd, i;
char *buf;
fd = open(DEV, O_RDONLY | O_DIRECT);
buf = mmap(NULL, BUF_SIZE, PROT_READ | PROT_WRITE,
MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
for (i = 0; i < BUF_SIZE; i += PAGE_SIZE)
buf[i] = 1; // init to non-zero
madvise(buf, BUF_SIZE, MADV_FREE);
read(fd, buf, BUF_SIZE);
for (i = 0; i < BUF_SIZE; i += PAGE_SIZE)
printf("%p: 0x%x\n", &buf[i], buf[i]);
return 0;
}
@ block/fops.c (formerly fs/block_dev.c)
+#include <linux/swap.h>
...
... __blkdev_direct_IO[_simple](...)
{
...
+ if (!strcmp(current->comm, "good"))
+ shrink_all_memory(ULONG_MAX);
+
ret = bio_iov_iter_get_pages(...);
+
+ if (!strcmp(current->comm, "bad"))
+ shrink_all_memory(ULONG_MAX);
...
}
@ shell
# NUM_PAGES=4
# PAGE_SIZE=$(getconf PAGE_SIZE)
# yes | dd of=test.img bs=${PAGE_SIZE} count=${NUM_PAGES}
# DEV=$(losetup -f --show test.img)
# gcc -DDEV=\"$DEV\" \
-DBUF_SIZE=$((PAGE_SIZE * NUM_PAGES)) \
-DPAGE_SIZE=${PAGE_SIZE} \
test.c -o test
# od -tx1 $DEV
0000000 79 0a 79 0a 79 0a 79 0a 79 0a 79 0a 79 0a 79 0a
*
0040000
# mv test good
# ./good
0x7f7c10418000: 0x79
0x7f7c10419000: 0x79
0x7f7c1041a000: 0x79
0x7f7c1041b000: 0x79
# mv good bad
# ./bad
0x7fa1b8050000: 0x0
0x7fa1b8051000: 0x0
0x7fa1b8052000: 0x0
0x7fa1b8053000: 0x0
Note: the issue is consistent on v5.17-rc3, but it's intermittent with the
support of MADV_FREE on v4.5 (60%-70% error; needs swap). [wrap
do_direct_IO() in do_blockdev_direct_IO() @ fs/direct-io.c].
- v5.17-rc3:
# for i in {1..1000}; do ./good; done \
| cut -d: -f2 | sort | uniq -c
4000 0x79
# mv good bad
# for i in {1..1000}; do ./bad; done \
| cut -d: -f2 | sort | uniq -c
4000 0x0
# free | grep Swap
Swap: 0 0 0
- v4.5:
# for i in {1..1000}; do ./good; done \
| cut -d: -f2 | sort | uniq -c
4000 0x79
# mv good bad
# for i in {1..1000}; do ./bad; done \
| cut -d: -f2 | sort | uniq -c
2702 0x0
1298 0x79
# swapoff -av
swapoff /swap
# for i in {1..1000}; do ./bad; done \
| cut -d: -f2 | sort | uniq -c
4000 0x79
Ceph/TCMalloc:
=============
For documentation purposes, the use case driving the analysis/fix is Ceph
on Ubuntu 18.04, as the TCMalloc library there still uses MADV_FREE to
release unused memory to the system from the mmap'ed page heap (might be
committed back/used again; it's not munmap'ed.) - PageHeap::DecommitSpan()
-> TCMalloc_SystemRelease() -> madvise() - PageHeap::CommitSpan() ->
TCMalloc_SystemCommit() -> do nothing.
Note: TCMalloc switched back to MADV_DONTNEED a few commits after the
release in Ubuntu 18.04 (google-perftools/gperftools 2.5), so the issue
just 'disappeared' on Ceph on later Ubuntu releases but is still present
in the kernel, and can be hit by other use cases.
The observed issue seems to be the old Ceph bug #22464 [1], where checksum
mismatches are observed (and instrumentation with buffer dumps shows
zero-pages read from mmap'ed/MADV_FREE'd page ranges).
The issue in Ceph was reasonably deemed a kernel bug (comment #50) and
mostly worked around with a retry mechanism, but other parts of Ceph could
still hit that (rocksdb). Anyway, it's less likely to be hit again as
TCMalloc switched out of MADV_FREE by default.
(Some kernel versions/reports from the Ceph bug, and relation with
the MADV_FREE introduction/changes; TCMalloc versions not checked.)
- 4.4 good
- 4.5 (madv_free: introduction)
- 4.9 bad
- 4.10 good? maybe a swapless system
- 4.12 (madv_free: no longer free instantly on swapless systems)
- 4.13 bad
[1] https://tracker.ceph.com/issues/22464
Thanks:
======
Several people contributed to analysis/discussions/tests/reproducers in
the first stages when drilling down on ceph/tcmalloc/linux kernel:
- Dan Hill
- Dan Streetman
- Dongdong Tao
- Gavin Guo
- Gerald Yang
- Heitor Alves de Siqueira
- Ioanna Alifieraki
- Jay Vosburgh
- Matthew Ruffell
- Ponnuvel Palaniyappan
Reviews, suggestions, corrections, comments:
- Minchan Kim
- Yu Zhao
- Huang, Ying
- John Hubbard
- Christoph Hellwig
[mfo@canonical.com: v4]
Link: https://lkml.kernel.org/r/20220209202659.183418-1-mfo@canonical.comLink: https://lkml.kernel.org/r/20220131230255.789059-1-mfo@canonical.com
Fixes: 802a3a92ad7a ("mm: reclaim MADV_FREE pages")
Signed-off-by: Mauricio Faria de Oliveira <mfo@canonical.com>
Reviewed-by: "Huang, Ying" <ying.huang@intel.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Yu Zhao <yuzhao@google.com>
Cc: Yang Shi <shy828301@gmail.com>
Cc: Miaohe Lin <linmiaohe@huawei.com>
Cc: Dan Hill <daniel.hill@canonical.com>
Cc: Dan Streetman <dan.streetman@canonical.com>
Cc: Dongdong Tao <dongdong.tao@canonical.com>
Cc: Gavin Guo <gavin.guo@canonical.com>
Cc: Gerald Yang <gerald.yang@canonical.com>
Cc: Heitor Alves de Siqueira <halves@canonical.com>
Cc: Ioanna Alifieraki <ioanna-maria.alifieraki@canonical.com>
Cc: Jay Vosburgh <jay.vosburgh@canonical.com>
Cc: Matthew Ruffell <matthew.ruffell@canonical.com>
Cc: Ponnuvel Palaniyappan <ponnuvel.palaniyappan@canonical.com>
Cc: <stable@vger.kernel.org>
Cc: Christoph Hellwig <hch@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2022-03-24 18:14:09 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
mm: set folio swapbacked iff folios are dirty in try_to_unmap_one
Patch series "mm: batched unmap lazyfree large folios during reclamation",
v4.
Commit 735ecdfaf4e8 ("mm/vmscan: avoid split lazyfree THP during
shrink_folio_list()") prevents the splitting of MADV_FREE'd THP in
madvise.c.
However, those folios are still added to the deferred_split list in
try_to_unmap_one() because we are unmapping PTEs and removing rmap entries
one by one.
Firstly, this has rendered the following counter somewhat confusing,
/sys/kernel/mm/transparent_hugepage/hugepages-size/stats/split_deferred
The split_deferred counter was originally designed to track operations
such as partial unmap or madvise of large folios. However, in practice,
most split_deferred cases arise from memory reclamation of aligned
lazyfree mTHPs as observed by Tangquan. This discrepancy has made the
split_deferred counter highly misleading.
Secondly, this approach is slow because it requires iterating through each
PTE and removing the rmap one by one for a large folio. In fact, all PTEs
of a pte-mapped large folio should be unmapped at once, and the entire
folio should be removed from the rmap as a whole.
Thirdly, it also increases the risk of a race condition where lazyfree
folios are incorrectly set back to swapbacked, as a speculative folio_get
may occur in the shrinker's callback.
deferred_split_scan() might call folio_try_get(folio) since we have added
the folio to split_deferred list while removing rmap for the 1st subpage,
and while we are scanning the 2nd to nr_pages PTEs of this folio in
try_to_unmap_one(), the entire mTHP could be transitioned back to
swap-backed because the reference count is incremented, which can make
"ref_count == 1 + map_count" within try_to_unmap_one() false.
/*
* The only page refs must be one from isolation
* plus the rmap(s) (dropped by discard:).
*/
if (ref_count == 1 + map_count &&
(!folio_test_dirty(folio) ||
...
(vma->vm_flags & VM_DROPPABLE))) {
dec_mm_counter(mm, MM_ANONPAGES);
goto discard;
}
This patchset resolves the issue by marking only genuinely dirty folios as
swap-backed, as suggested by David, and transitioning to batched unmapping
of entire folios in try_to_unmap_one(). Consequently, the deferred_split
count drops to zero, and memory reclamation performance improves
significantly — reclaiming 64KiB lazyfree large folios is now 2.5x
faster(The specific data is embedded in the changelog of patch 3/4).
By the way, while the patchset is primarily aimed at PTE-mapped large
folios, Baolin and Lance also found that try_to_unmap_one() handles
lazyfree redirtied PMD-mapped large folios inefficiently — it splits the
PMD into PTEs and iterates over them. This patchset removes the
unnecessary splitting, enabling us to skip redirtied PMD-mapped large
folios 3.5X faster during memory reclamation. (The specific data can be
found in the changelog of patch 4/4).
This patch (of 4):
The refcount may be temporarily or long-term increased, but this does not
change the fundamental nature of the folio already being lazy- freed.
Therefore, we only reset 'swapbacked' when we are certain the folio is
dirty and not droppable.
Link: https://lkml.kernel.org/r/20250214093015.51024-1-21cnbao@gmail.com
Link: https://lkml.kernel.org/r/20250214093015.51024-2-21cnbao@gmail.com
Fixes: 6c8e2a256915 ("mm: fix race between MADV_FREE reclaim and blkdev direct IO read")
Signed-off-by: Barry Song <v-songbaohua@oppo.com>
Suggested-by: David Hildenbrand <david@redhat.com>
Acked-by: David Hildenbrand <david@redhat.com>
Reviewed-by: Baolin Wang <baolin.wang@linux.alibaba.com>
Reviewed-by: Lance Yang <ioworker0@gmail.com>
Cc: Mauricio Faria de Oliveira <mfo@canonical.com>
Cc: Chis Li <chrisl@kernel.org> (Google)
Cc: "Huang, Ying" <ying.huang@intel.com>
Cc: Kairui Song <kasong@tencent.com>
Cc: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: Ryan Roberts <ryan.roberts@arm.com>
Cc: Tangquan Zheng <zhengtangquan@oppo.com>
Cc: Albert Ou <aou@eecs.berkeley.edu>
Cc: Anshuman Khandual <anshuman.khandual@arm.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Gavin Shan <gshan@redhat.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Kefeng Wang <wangkefeng.wang@huawei.com>
Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Palmer Dabbelt <palmer@dabbelt.com>
Cc: Paul Walmsley <paul.walmsley@sifive.com>
Cc: Shaoqin Huang <shahuang@redhat.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Will Deacon <will@kernel.org>
Cc: Yicong Yang <yangyicong@hisilicon.com>
Cc: Yosry Ahmed <yosryahmed@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2025-02-14 22:30:12 +13:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
mm: support batched unmap for lazyfree large folios during reclamation
Currently, the PTEs and rmap of a large folio are removed one at a time.
This is not only slow but also causes the large folio to be unnecessarily
added to deferred_split, which can lead to races between the
deferred_split shrinker callback and memory reclamation. This patch
releases all PTEs and rmap entries in a batch. Currently, it only handles
lazyfree large folios.
The below microbench tries to reclaim 128MB lazyfree large folios
whose sizes are 64KiB:
#include <stdio.h>
#include <sys/mman.h>
#include <string.h>
#include <time.h>
#define SIZE 128*1024*1024 // 128 MB
unsigned long read_split_deferred()
{
FILE *file = fopen("/sys/kernel/mm/transparent_hugepage"
"/hugepages-64kB/stats/split_deferred", "r");
if (!file) {
perror("Error opening file");
return 0;
}
unsigned long value;
if (fscanf(file, "%lu", &value) != 1) {
perror("Error reading value");
fclose(file);
return 0;
}
fclose(file);
return value;
}
int main(int argc, char *argv[])
{
while(1) {
volatile int *p = mmap(0, SIZE, PROT_READ | PROT_WRITE,
MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
memset((void *)p, 1, SIZE);
madvise((void *)p, SIZE, MADV_FREE);
clock_t start_time = clock();
unsigned long start_split = read_split_deferred();
madvise((void *)p, SIZE, MADV_PAGEOUT);
clock_t end_time = clock();
unsigned long end_split = read_split_deferred();
double elapsed_time = (double)(end_time - start_time) / CLOCKS_PER_SEC;
printf("Time taken by reclamation: %f seconds, split_deferred: %ld\n",
elapsed_time, end_split - start_split);
munmap((void *)p, SIZE);
}
return 0;
}
w/o patch:
~ # ./a.out
Time taken by reclamation: 0.177418 seconds, split_deferred: 2048
Time taken by reclamation: 0.178348 seconds, split_deferred: 2048
Time taken by reclamation: 0.174525 seconds, split_deferred: 2048
Time taken by reclamation: 0.171620 seconds, split_deferred: 2048
Time taken by reclamation: 0.172241 seconds, split_deferred: 2048
Time taken by reclamation: 0.174003 seconds, split_deferred: 2048
Time taken by reclamation: 0.171058 seconds, split_deferred: 2048
Time taken by reclamation: 0.171993 seconds, split_deferred: 2048
Time taken by reclamation: 0.169829 seconds, split_deferred: 2048
Time taken by reclamation: 0.172895 seconds, split_deferred: 2048
Time taken by reclamation: 0.176063 seconds, split_deferred: 2048
Time taken by reclamation: 0.172568 seconds, split_deferred: 2048
Time taken by reclamation: 0.171185 seconds, split_deferred: 2048
Time taken by reclamation: 0.170632 seconds, split_deferred: 2048
Time taken by reclamation: 0.170208 seconds, split_deferred: 2048
Time taken by reclamation: 0.174192 seconds, split_deferred: 2048
...
w/ patch:
~ # ./a.out
Time taken by reclamation: 0.074231 seconds, split_deferred: 0
Time taken by reclamation: 0.071026 seconds, split_deferred: 0
Time taken by reclamation: 0.072029 seconds, split_deferred: 0
Time taken by reclamation: 0.071873 seconds, split_deferred: 0
Time taken by reclamation: 0.073573 seconds, split_deferred: 0
Time taken by reclamation: 0.071906 seconds, split_deferred: 0
Time taken by reclamation: 0.073604 seconds, split_deferred: 0
Time taken by reclamation: 0.075903 seconds, split_deferred: 0
Time taken by reclamation: 0.073191 seconds, split_deferred: 0
Time taken by reclamation: 0.071228 seconds, split_deferred: 0
Time taken by reclamation: 0.071391 seconds, split_deferred: 0
Time taken by reclamation: 0.071468 seconds, split_deferred: 0
Time taken by reclamation: 0.071896 seconds, split_deferred: 0
Time taken by reclamation: 0.072508 seconds, split_deferred: 0
Time taken by reclamation: 0.071884 seconds, split_deferred: 0
Time taken by reclamation: 0.072433 seconds, split_deferred: 0
Time taken by reclamation: 0.071939 seconds, split_deferred: 0
...
Link: https://lkml.kernel.org/r/20250214093015.51024-4-21cnbao@gmail.com
Signed-off-by: Barry Song <v-songbaohua@oppo.com>
Cc: Albert Ou <aou@eecs.berkeley.edu>
Cc: Anshuman Khandual <anshuman.khandual@arm.com>
Cc: Baolin Wang <baolin.wang@linux.alibaba.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Chis Li <chrisl@kernel.org>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Gavin Shan <gshan@redhat.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: "Huang, Ying" <ying.huang@intel.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Kairui Song <kasong@tencent.com>
Cc: Kefeng Wang <wangkefeng.wang@huawei.com>
Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Lance Yang <ioworker0@gmail.com>
Cc: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Mauricio Faria de Oliveira <mfo@canonical.com>
Cc: Palmer Dabbelt <palmer@dabbelt.com>
Cc: Paul Walmsley <paul.walmsley@sifive.com>
Cc: Ryan Roberts <ryan.roberts@arm.com>
Cc: Shaoqin Huang <shahuang@redhat.com>
Cc: Tangquan Zheng <zhengtangquan@oppo.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Will Deacon <will@kernel.org>
Cc: Yicong Yang <yangyicong@hisilicon.com>
Cc: Yosry Ahmed <yosryahmed@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2025-02-14 22:30:14 +13:00
|
|
|
|
mm: add MAP_DROPPABLE for designating always lazily freeable mappings
The vDSO getrandom() implementation works with a buffer allocated with a
new system call that has certain requirements:
- It shouldn't be written to core dumps.
* Easy: VM_DONTDUMP.
- It should be zeroed on fork.
* Easy: VM_WIPEONFORK.
- It shouldn't be written to swap.
* Uh-oh: mlock is rlimited.
* Uh-oh: mlock isn't inherited by forks.
- It shouldn't reserve actual memory, but it also shouldn't crash when
page faulting in memory if none is available
* Uh-oh: VM_NORESERVE means segfaults.
It turns out that the vDSO getrandom() function has three really nice
characteristics that we can exploit to solve this problem:
1) Due to being wiped during fork(), the vDSO code is already robust to
having the contents of the pages it reads zeroed out midway through
the function's execution.
2) In the absolute worst case of whatever contingency we're coding for,
we have the option to fallback to the getrandom() syscall, and
everything is fine.
3) The buffers the function uses are only ever useful for a maximum of
60 seconds -- a sort of cache, rather than a long term allocation.
These characteristics mean that we can introduce VM_DROPPABLE, which
has the following semantics:
a) It never is written out to swap.
b) Under memory pressure, mm can just drop the pages (so that they're
zero when read back again).
c) It is inherited by fork.
d) It doesn't count against the mlock budget, since nothing is locked.
e) If there's not enough memory to service a page fault, it's not fatal,
and no signal is sent.
This way, allocations used by vDSO getrandom() can use:
VM_DROPPABLE | VM_DONTDUMP | VM_WIPEONFORK | VM_NORESERVE
And there will be no problem with OOMing, crashing on overcommitment,
using memory when not in use, not wiping on fork(), coredumps, or
writing out to swap.
In order to let vDSO getrandom() use this, expose these via mmap(2) as
MAP_DROPPABLE.
Note that this involves removing the MADV_FREE special case from
sort_folio(), which according to Yu Zhao is unnecessary and will simply
result in an extra call to shrink_folio_list() in the worst case. The
chunk removed reenables the swapbacked flag, which we don't want for
VM_DROPPABLE, and we can't conditionalize it here because there isn't a
vma reference available.
Finally, the provided self test ensures that this is working as desired.
Cc: linux-mm@kvack.org
Acked-by: David Hildenbrand <david@redhat.com>
Signed-off-by: Jason A. Donenfeld <Jason@zx2c4.com>
2022-12-08 17:55:04 +01:00
|
|
|
|
mm: set folio swapbacked iff folios are dirty in try_to_unmap_one
Patch series "mm: batched unmap lazyfree large folios during reclamation",
v4.
Commit 735ecdfaf4e8 ("mm/vmscan: avoid split lazyfree THP during
shrink_folio_list()") prevents the splitting of MADV_FREE'd THP in
madvise.c.
However, those folios are still added to the deferred_split list in
try_to_unmap_one() because we are unmapping PTEs and removing rmap entries
one by one.
Firstly, this has rendered the following counter somewhat confusing,
/sys/kernel/mm/transparent_hugepage/hugepages-size/stats/split_deferred
The split_deferred counter was originally designed to track operations
such as partial unmap or madvise of large folios. However, in practice,
most split_deferred cases arise from memory reclamation of aligned
lazyfree mTHPs as observed by Tangquan. This discrepancy has made the
split_deferred counter highly misleading.
Secondly, this approach is slow because it requires iterating through each
PTE and removing the rmap one by one for a large folio. In fact, all PTEs
of a pte-mapped large folio should be unmapped at once, and the entire
folio should be removed from the rmap as a whole.
Thirdly, it also increases the risk of a race condition where lazyfree
folios are incorrectly set back to swapbacked, as a speculative folio_get
may occur in the shrinker's callback.
deferred_split_scan() might call folio_try_get(folio) since we have added
the folio to split_deferred list while removing rmap for the 1st subpage,
and while we are scanning the 2nd to nr_pages PTEs of this folio in
try_to_unmap_one(), the entire mTHP could be transitioned back to
swap-backed because the reference count is incremented, which can make
"ref_count == 1 + map_count" within try_to_unmap_one() false.
/*
* The only page refs must be one from isolation
* plus the rmap(s) (dropped by discard:).
*/
if (ref_count == 1 + map_count &&
(!folio_test_dirty(folio) ||
...
(vma->vm_flags & VM_DROPPABLE))) {
dec_mm_counter(mm, MM_ANONPAGES);
goto discard;
}
This patchset resolves the issue by marking only genuinely dirty folios as
swap-backed, as suggested by David, and transitioning to batched unmapping
of entire folios in try_to_unmap_one(). Consequently, the deferred_split
count drops to zero, and memory reclamation performance improves
significantly — reclaiming 64KiB lazyfree large folios is now 2.5x
faster(The specific data is embedded in the changelog of patch 3/4).
By the way, while the patchset is primarily aimed at PTE-mapped large
folios, Baolin and Lance also found that try_to_unmap_one() handles
lazyfree redirtied PMD-mapped large folios inefficiently — it splits the
PMD into PTEs and iterates over them. This patchset removes the
unnecessary splitting, enabling us to skip redirtied PMD-mapped large
folios 3.5X faster during memory reclamation. (The specific data can be
found in the changelog of patch 4/4).
This patch (of 4):
The refcount may be temporarily or long-term increased, but this does not
change the fundamental nature of the folio already being lazy- freed.
Therefore, we only reset 'swapbacked' when we are certain the folio is
dirty and not droppable.
Link: https://lkml.kernel.org/r/20250214093015.51024-1-21cnbao@gmail.com
Link: https://lkml.kernel.org/r/20250214093015.51024-2-21cnbao@gmail.com
Fixes: 6c8e2a256915 ("mm: fix race between MADV_FREE reclaim and blkdev direct IO read")
Signed-off-by: Barry Song <v-songbaohua@oppo.com>
Suggested-by: David Hildenbrand <david@redhat.com>
Acked-by: David Hildenbrand <david@redhat.com>
Reviewed-by: Baolin Wang <baolin.wang@linux.alibaba.com>
Reviewed-by: Lance Yang <ioworker0@gmail.com>
Cc: Mauricio Faria de Oliveira <mfo@canonical.com>
Cc: Chis Li <chrisl@kernel.org> (Google)
Cc: "Huang, Ying" <ying.huang@intel.com>
Cc: Kairui Song <kasong@tencent.com>
Cc: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: Ryan Roberts <ryan.roberts@arm.com>
Cc: Tangquan Zheng <zhengtangquan@oppo.com>
Cc: Albert Ou <aou@eecs.berkeley.edu>
Cc: Anshuman Khandual <anshuman.khandual@arm.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Gavin Shan <gshan@redhat.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Kefeng Wang <wangkefeng.wang@huawei.com>
Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Palmer Dabbelt <palmer@dabbelt.com>
Cc: Paul Walmsley <paul.walmsley@sifive.com>
Cc: Shaoqin Huang <shahuang@redhat.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Will Deacon <will@kernel.org>
Cc: Yicong Yang <yangyicong@hisilicon.com>
Cc: Yosry Ahmed <yosryahmed@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2025-02-14 22:30:12 +13:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
mm: support batched unmap for lazyfree large folios during reclamation
Currently, the PTEs and rmap of a large folio are removed one at a time.
This is not only slow but also causes the large folio to be unnecessarily
added to deferred_split, which can lead to races between the
deferred_split shrinker callback and memory reclamation. This patch
releases all PTEs and rmap entries in a batch. Currently, it only handles
lazyfree large folios.
The below microbench tries to reclaim 128MB lazyfree large folios
whose sizes are 64KiB:
#include <stdio.h>
#include <sys/mman.h>
#include <string.h>
#include <time.h>
#define SIZE 128*1024*1024 // 128 MB
unsigned long read_split_deferred()
{
FILE *file = fopen("/sys/kernel/mm/transparent_hugepage"
"/hugepages-64kB/stats/split_deferred", "r");
if (!file) {
perror("Error opening file");
return 0;
}
unsigned long value;
if (fscanf(file, "%lu", &value) != 1) {
perror("Error reading value");
fclose(file);
return 0;
}
fclose(file);
return value;
}
int main(int argc, char *argv[])
{
while(1) {
volatile int *p = mmap(0, SIZE, PROT_READ | PROT_WRITE,
MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
memset((void *)p, 1, SIZE);
madvise((void *)p, SIZE, MADV_FREE);
clock_t start_time = clock();
unsigned long start_split = read_split_deferred();
madvise((void *)p, SIZE, MADV_PAGEOUT);
clock_t end_time = clock();
unsigned long end_split = read_split_deferred();
double elapsed_time = (double)(end_time - start_time) / CLOCKS_PER_SEC;
printf("Time taken by reclamation: %f seconds, split_deferred: %ld\n",
elapsed_time, end_split - start_split);
munmap((void *)p, SIZE);
}
return 0;
}
w/o patch:
~ # ./a.out
Time taken by reclamation: 0.177418 seconds, split_deferred: 2048
Time taken by reclamation: 0.178348 seconds, split_deferred: 2048
Time taken by reclamation: 0.174525 seconds, split_deferred: 2048
Time taken by reclamation: 0.171620 seconds, split_deferred: 2048
Time taken by reclamation: 0.172241 seconds, split_deferred: 2048
Time taken by reclamation: 0.174003 seconds, split_deferred: 2048
Time taken by reclamation: 0.171058 seconds, split_deferred: 2048
Time taken by reclamation: 0.171993 seconds, split_deferred: 2048
Time taken by reclamation: 0.169829 seconds, split_deferred: 2048
Time taken by reclamation: 0.172895 seconds, split_deferred: 2048
Time taken by reclamation: 0.176063 seconds, split_deferred: 2048
Time taken by reclamation: 0.172568 seconds, split_deferred: 2048
Time taken by reclamation: 0.171185 seconds, split_deferred: 2048
Time taken by reclamation: 0.170632 seconds, split_deferred: 2048
Time taken by reclamation: 0.170208 seconds, split_deferred: 2048
Time taken by reclamation: 0.174192 seconds, split_deferred: 2048
...
w/ patch:
~ # ./a.out
Time taken by reclamation: 0.074231 seconds, split_deferred: 0
Time taken by reclamation: 0.071026 seconds, split_deferred: 0
Time taken by reclamation: 0.072029 seconds, split_deferred: 0
Time taken by reclamation: 0.071873 seconds, split_deferred: 0
Time taken by reclamation: 0.073573 seconds, split_deferred: 0
Time taken by reclamation: 0.071906 seconds, split_deferred: 0
Time taken by reclamation: 0.073604 seconds, split_deferred: 0
Time taken by reclamation: 0.075903 seconds, split_deferred: 0
Time taken by reclamation: 0.073191 seconds, split_deferred: 0
Time taken by reclamation: 0.071228 seconds, split_deferred: 0
Time taken by reclamation: 0.071391 seconds, split_deferred: 0
Time taken by reclamation: 0.071468 seconds, split_deferred: 0
Time taken by reclamation: 0.071896 seconds, split_deferred: 0
Time taken by reclamation: 0.072508 seconds, split_deferred: 0
Time taken by reclamation: 0.071884 seconds, split_deferred: 0
Time taken by reclamation: 0.072433 seconds, split_deferred: 0
Time taken by reclamation: 0.071939 seconds, split_deferred: 0
...
Link: https://lkml.kernel.org/r/20250214093015.51024-4-21cnbao@gmail.com
Signed-off-by: Barry Song <v-songbaohua@oppo.com>
Cc: Albert Ou <aou@eecs.berkeley.edu>
Cc: Anshuman Khandual <anshuman.khandual@arm.com>
Cc: Baolin Wang <baolin.wang@linux.alibaba.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Chis Li <chrisl@kernel.org>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Gavin Shan <gshan@redhat.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: "Huang, Ying" <ying.huang@intel.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Kairui Song <kasong@tencent.com>
Cc: Kefeng Wang <wangkefeng.wang@huawei.com>
Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Lance Yang <ioworker0@gmail.com>
Cc: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Mauricio Faria de Oliveira <mfo@canonical.com>
Cc: Palmer Dabbelt <palmer@dabbelt.com>
Cc: Paul Walmsley <paul.walmsley@sifive.com>
Cc: Ryan Roberts <ryan.roberts@arm.com>
Cc: Shaoqin Huang <shahuang@redhat.com>
Cc: Tangquan Zheng <zhengtangquan@oppo.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Will Deacon <will@kernel.org>
Cc: Yicong Yang <yangyicong@hisilicon.com>
Cc: Yosry Ahmed <yosryahmed@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2025-02-14 22:30:14 +13:00
|
|
|
|
mm: set folio swapbacked iff folios are dirty in try_to_unmap_one
Patch series "mm: batched unmap lazyfree large folios during reclamation",
v4.
Commit 735ecdfaf4e8 ("mm/vmscan: avoid split lazyfree THP during
shrink_folio_list()") prevents the splitting of MADV_FREE'd THP in
madvise.c.
However, those folios are still added to the deferred_split list in
try_to_unmap_one() because we are unmapping PTEs and removing rmap entries
one by one.
Firstly, this has rendered the following counter somewhat confusing,
/sys/kernel/mm/transparent_hugepage/hugepages-size/stats/split_deferred
The split_deferred counter was originally designed to track operations
such as partial unmap or madvise of large folios. However, in practice,
most split_deferred cases arise from memory reclamation of aligned
lazyfree mTHPs as observed by Tangquan. This discrepancy has made the
split_deferred counter highly misleading.
Secondly, this approach is slow because it requires iterating through each
PTE and removing the rmap one by one for a large folio. In fact, all PTEs
of a pte-mapped large folio should be unmapped at once, and the entire
folio should be removed from the rmap as a whole.
Thirdly, it also increases the risk of a race condition where lazyfree
folios are incorrectly set back to swapbacked, as a speculative folio_get
may occur in the shrinker's callback.
deferred_split_scan() might call folio_try_get(folio) since we have added
the folio to split_deferred list while removing rmap for the 1st subpage,
and while we are scanning the 2nd to nr_pages PTEs of this folio in
try_to_unmap_one(), the entire mTHP could be transitioned back to
swap-backed because the reference count is incremented, which can make
"ref_count == 1 + map_count" within try_to_unmap_one() false.
/*
* The only page refs must be one from isolation
* plus the rmap(s) (dropped by discard:).
*/
if (ref_count == 1 + map_count &&
(!folio_test_dirty(folio) ||
...
(vma->vm_flags & VM_DROPPABLE))) {
dec_mm_counter(mm, MM_ANONPAGES);
goto discard;
}
This patchset resolves the issue by marking only genuinely dirty folios as
swap-backed, as suggested by David, and transitioning to batched unmapping
of entire folios in try_to_unmap_one(). Consequently, the deferred_split
count drops to zero, and memory reclamation performance improves
significantly — reclaiming 64KiB lazyfree large folios is now 2.5x
faster(The specific data is embedded in the changelog of patch 3/4).
By the way, while the patchset is primarily aimed at PTE-mapped large
folios, Baolin and Lance also found that try_to_unmap_one() handles
lazyfree redirtied PMD-mapped large folios inefficiently — it splits the
PMD into PTEs and iterates over them. This patchset removes the
unnecessary splitting, enabling us to skip redirtied PMD-mapped large
folios 3.5X faster during memory reclamation. (The specific data can be
found in the changelog of patch 4/4).
This patch (of 4):
The refcount may be temporarily or long-term increased, but this does not
change the fundamental nature of the folio already being lazy- freed.
Therefore, we only reset 'swapbacked' when we are certain the folio is
dirty and not droppable.
Link: https://lkml.kernel.org/r/20250214093015.51024-1-21cnbao@gmail.com
Link: https://lkml.kernel.org/r/20250214093015.51024-2-21cnbao@gmail.com
Fixes: 6c8e2a256915 ("mm: fix race between MADV_FREE reclaim and blkdev direct IO read")
Signed-off-by: Barry Song <v-songbaohua@oppo.com>
Suggested-by: David Hildenbrand <david@redhat.com>
Acked-by: David Hildenbrand <david@redhat.com>
Reviewed-by: Baolin Wang <baolin.wang@linux.alibaba.com>
Reviewed-by: Lance Yang <ioworker0@gmail.com>
Cc: Mauricio Faria de Oliveira <mfo@canonical.com>
Cc: Chis Li <chrisl@kernel.org> (Google)
Cc: "Huang, Ying" <ying.huang@intel.com>
Cc: Kairui Song <kasong@tencent.com>
Cc: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: Ryan Roberts <ryan.roberts@arm.com>
Cc: Tangquan Zheng <zhengtangquan@oppo.com>
Cc: Albert Ou <aou@eecs.berkeley.edu>
Cc: Anshuman Khandual <anshuman.khandual@arm.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Gavin Shan <gshan@redhat.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Kefeng Wang <wangkefeng.wang@huawei.com>
Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Palmer Dabbelt <palmer@dabbelt.com>
Cc: Paul Walmsley <paul.walmsley@sifive.com>
Cc: Shaoqin Huang <shahuang@redhat.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Will Deacon <will@kernel.org>
Cc: Yicong Yang <yangyicong@hisilicon.com>
Cc: Yosry Ahmed <yosryahmed@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2025-02-14 22:30:12 +13:00
|
|
|
|
|
|
|
|
|
mm: support batched unmap for lazyfree large folios during reclamation
Currently, the PTEs and rmap of a large folio are removed one at a time.
This is not only slow but also causes the large folio to be unnecessarily
added to deferred_split, which can lead to races between the
deferred_split shrinker callback and memory reclamation. This patch
releases all PTEs and rmap entries in a batch. Currently, it only handles
lazyfree large folios.
The below microbench tries to reclaim 128MB lazyfree large folios
whose sizes are 64KiB:
#include <stdio.h>
#include <sys/mman.h>
#include <string.h>
#include <time.h>
#define SIZE 128*1024*1024 // 128 MB
unsigned long read_split_deferred()
{
FILE *file = fopen("/sys/kernel/mm/transparent_hugepage"
"/hugepages-64kB/stats/split_deferred", "r");
if (!file) {
perror("Error opening file");
return 0;
}
unsigned long value;
if (fscanf(file, "%lu", &value) != 1) {
perror("Error reading value");
fclose(file);
return 0;
}
fclose(file);
return value;
}
int main(int argc, char *argv[])
{
while(1) {
volatile int *p = mmap(0, SIZE, PROT_READ | PROT_WRITE,
MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
memset((void *)p, 1, SIZE);
madvise((void *)p, SIZE, MADV_FREE);
clock_t start_time = clock();
unsigned long start_split = read_split_deferred();
madvise((void *)p, SIZE, MADV_PAGEOUT);
clock_t end_time = clock();
unsigned long end_split = read_split_deferred();
double elapsed_time = (double)(end_time - start_time) / CLOCKS_PER_SEC;
printf("Time taken by reclamation: %f seconds, split_deferred: %ld\n",
elapsed_time, end_split - start_split);
munmap((void *)p, SIZE);
}
return 0;
}
w/o patch:
~ # ./a.out
Time taken by reclamation: 0.177418 seconds, split_deferred: 2048
Time taken by reclamation: 0.178348 seconds, split_deferred: 2048
Time taken by reclamation: 0.174525 seconds, split_deferred: 2048
Time taken by reclamation: 0.171620 seconds, split_deferred: 2048
Time taken by reclamation: 0.172241 seconds, split_deferred: 2048
Time taken by reclamation: 0.174003 seconds, split_deferred: 2048
Time taken by reclamation: 0.171058 seconds, split_deferred: 2048
Time taken by reclamation: 0.171993 seconds, split_deferred: 2048
Time taken by reclamation: 0.169829 seconds, split_deferred: 2048
Time taken by reclamation: 0.172895 seconds, split_deferred: 2048
Time taken by reclamation: 0.176063 seconds, split_deferred: 2048
Time taken by reclamation: 0.172568 seconds, split_deferred: 2048
Time taken by reclamation: 0.171185 seconds, split_deferred: 2048
Time taken by reclamation: 0.170632 seconds, split_deferred: 2048
Time taken by reclamation: 0.170208 seconds, split_deferred: 2048
Time taken by reclamation: 0.174192 seconds, split_deferred: 2048
...
w/ patch:
~ # ./a.out
Time taken by reclamation: 0.074231 seconds, split_deferred: 0
Time taken by reclamation: 0.071026 seconds, split_deferred: 0
Time taken by reclamation: 0.072029 seconds, split_deferred: 0
Time taken by reclamation: 0.071873 seconds, split_deferred: 0
Time taken by reclamation: 0.073573 seconds, split_deferred: 0
Time taken by reclamation: 0.071906 seconds, split_deferred: 0
Time taken by reclamation: 0.073604 seconds, split_deferred: 0
Time taken by reclamation: 0.075903 seconds, split_deferred: 0
Time taken by reclamation: 0.073191 seconds, split_deferred: 0
Time taken by reclamation: 0.071228 seconds, split_deferred: 0
Time taken by reclamation: 0.071391 seconds, split_deferred: 0
Time taken by reclamation: 0.071468 seconds, split_deferred: 0
Time taken by reclamation: 0.071896 seconds, split_deferred: 0
Time taken by reclamation: 0.072508 seconds, split_deferred: 0
Time taken by reclamation: 0.071884 seconds, split_deferred: 0
Time taken by reclamation: 0.072433 seconds, split_deferred: 0
Time taken by reclamation: 0.071939 seconds, split_deferred: 0
...
Link: https://lkml.kernel.org/r/20250214093015.51024-4-21cnbao@gmail.com
Signed-off-by: Barry Song <v-songbaohua@oppo.com>
Cc: Albert Ou <aou@eecs.berkeley.edu>
Cc: Anshuman Khandual <anshuman.khandual@arm.com>
Cc: Baolin Wang <baolin.wang@linux.alibaba.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Chis Li <chrisl@kernel.org>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Gavin Shan <gshan@redhat.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: "Huang, Ying" <ying.huang@intel.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Kairui Song <kasong@tencent.com>
Cc: Kefeng Wang <wangkefeng.wang@huawei.com>
Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Lance Yang <ioworker0@gmail.com>
Cc: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Mauricio Faria de Oliveira <mfo@canonical.com>
Cc: Palmer Dabbelt <palmer@dabbelt.com>
Cc: Paul Walmsley <paul.walmsley@sifive.com>
Cc: Ryan Roberts <ryan.roberts@arm.com>
Cc: Shaoqin Huang <shahuang@redhat.com>
Cc: Tangquan Zheng <zhengtangquan@oppo.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Will Deacon <will@kernel.org>
Cc: Yicong Yang <yangyicong@hisilicon.com>
Cc: Yosry Ahmed <yosryahmed@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2025-02-14 22:30:14 +13:00
|
|
|
|
mm: set folio swapbacked iff folios are dirty in try_to_unmap_one
Patch series "mm: batched unmap lazyfree large folios during reclamation",
v4.
Commit 735ecdfaf4e8 ("mm/vmscan: avoid split lazyfree THP during
shrink_folio_list()") prevents the splitting of MADV_FREE'd THP in
madvise.c.
However, those folios are still added to the deferred_split list in
try_to_unmap_one() because we are unmapping PTEs and removing rmap entries
one by one.
Firstly, this has rendered the following counter somewhat confusing,
/sys/kernel/mm/transparent_hugepage/hugepages-size/stats/split_deferred
The split_deferred counter was originally designed to track operations
such as partial unmap or madvise of large folios. However, in practice,
most split_deferred cases arise from memory reclamation of aligned
lazyfree mTHPs as observed by Tangquan. This discrepancy has made the
split_deferred counter highly misleading.
Secondly, this approach is slow because it requires iterating through each
PTE and removing the rmap one by one for a large folio. In fact, all PTEs
of a pte-mapped large folio should be unmapped at once, and the entire
folio should be removed from the rmap as a whole.
Thirdly, it also increases the risk of a race condition where lazyfree
folios are incorrectly set back to swapbacked, as a speculative folio_get
may occur in the shrinker's callback.
deferred_split_scan() might call folio_try_get(folio) since we have added
the folio to split_deferred list while removing rmap for the 1st subpage,
and while we are scanning the 2nd to nr_pages PTEs of this folio in
try_to_unmap_one(), the entire mTHP could be transitioned back to
swap-backed because the reference count is incremented, which can make
"ref_count == 1 + map_count" within try_to_unmap_one() false.
/*
* The only page refs must be one from isolation
* plus the rmap(s) (dropped by discard:).
*/
if (ref_count == 1 + map_count &&
(!folio_test_dirty(folio) ||
...
(vma->vm_flags & VM_DROPPABLE))) {
dec_mm_counter(mm, MM_ANONPAGES);
goto discard;
}
This patchset resolves the issue by marking only genuinely dirty folios as
swap-backed, as suggested by David, and transitioning to batched unmapping
of entire folios in try_to_unmap_one(). Consequently, the deferred_split
count drops to zero, and memory reclamation performance improves
significantly — reclaiming 64KiB lazyfree large folios is now 2.5x
faster(The specific data is embedded in the changelog of patch 3/4).
By the way, while the patchset is primarily aimed at PTE-mapped large
folios, Baolin and Lance also found that try_to_unmap_one() handles
lazyfree redirtied PMD-mapped large folios inefficiently — it splits the
PMD into PTEs and iterates over them. This patchset removes the
unnecessary splitting, enabling us to skip redirtied PMD-mapped large
folios 3.5X faster during memory reclamation. (The specific data can be
found in the changelog of patch 4/4).
This patch (of 4):
The refcount may be temporarily or long-term increased, but this does not
change the fundamental nature of the folio already being lazy- freed.
Therefore, we only reset 'swapbacked' when we are certain the folio is
dirty and not droppable.
Link: https://lkml.kernel.org/r/20250214093015.51024-1-21cnbao@gmail.com
Link: https://lkml.kernel.org/r/20250214093015.51024-2-21cnbao@gmail.com
Fixes: 6c8e2a256915 ("mm: fix race between MADV_FREE reclaim and blkdev direct IO read")
Signed-off-by: Barry Song <v-songbaohua@oppo.com>
Suggested-by: David Hildenbrand <david@redhat.com>
Acked-by: David Hildenbrand <david@redhat.com>
Reviewed-by: Baolin Wang <baolin.wang@linux.alibaba.com>
Reviewed-by: Lance Yang <ioworker0@gmail.com>
Cc: Mauricio Faria de Oliveira <mfo@canonical.com>
Cc: Chis Li <chrisl@kernel.org> (Google)
Cc: "Huang, Ying" <ying.huang@intel.com>
Cc: Kairui Song <kasong@tencent.com>
Cc: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: Ryan Roberts <ryan.roberts@arm.com>
Cc: Tangquan Zheng <zhengtangquan@oppo.com>
Cc: Albert Ou <aou@eecs.berkeley.edu>
Cc: Anshuman Khandual <anshuman.khandual@arm.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Gavin Shan <gshan@redhat.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Kefeng Wang <wangkefeng.wang@huawei.com>
Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Palmer Dabbelt <palmer@dabbelt.com>
Cc: Paul Walmsley <paul.walmsley@sifive.com>
Cc: Shaoqin Huang <shahuang@redhat.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Will Deacon <will@kernel.org>
Cc: Yicong Yang <yangyicong@hisilicon.com>
Cc: Yosry Ahmed <yosryahmed@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2025-02-14 22:30:12 +13:00
|
|
|
|
2017-02-24 14:58:01 -08:00
|
|
|
|
|
|
|
|
|
mm, swap: cleanup swap entry management workflow
The current swap entry allocation/freeing workflow has never had a clear
definition. This makes it hard to debug or add new optimizations.
This commit introduces a proper definition of how swap entries would be
allocated and freed. Now, most operations are folio based, so they will
never exceed one swap cluster, and we now have a cleaner border between
swap and the rest of mm, making it much easier to follow and debug,
especially with new added sanity checks. Also making more optimization
possible.
Swap entry will be mostly freed and free with a folio bound. The folio
lock will be useful for resolving many swap related races.
Now swap allocation (except hibernation) always starts with a folio in the
swap cache, and gets duped/freed protected by the folio lock:
- folio_alloc_swap() - The only allocation entry point now.
Context: The folio must be locked.
This allocates one or a set of continuous swap slots for a folio and
binds them to the folio by adding the folio to the swap cache. The
swap slots' swap count start with zero value.
- folio_dup_swap() - Increase the swap count of one or more entries.
Context: The folio must be locked and in the swap cache. For now, the
caller still has to lock the new swap entry owner (e.g., PTL).
This increases the ref count of swap entries allocated to a folio.
Newly allocated swap slots' count has to be increased by this helper
as the folio got unmapped (and swap entries got installed).
- folio_put_swap() - Decrease the swap count of one or more entries.
Context: The folio must be locked and in the swap cache. For now, the
caller still has to lock the new swap entry owner (e.g., PTL).
This decreases the ref count of swap entries allocated to a folio.
Typically, swapin will decrease the swap count as the folio got
installed back and the swap entry got uninstalled
This won't remove the folio from the swap cache and free the
slot. Lazy freeing of swap cache is helpful for reducing IO.
There is already a folio_free_swap() for immediate cache reclaim.
This part could be further optimized later.
The above locking constraints could be further relaxed when the swap table
is fully implemented. Currently dup still needs the caller to lock the
swap entry container (e.g. PTL), or a concurrent zap may underflow the
swap count.
Some swap users need to interact with swap count without involving folio
(e.g. forking/zapping the page table or mapping truncate without swapin).
In such cases, the caller has to ensure there is no race condition on
whatever owns the swap count and use the below helpers:
- swap_put_entries_direct() - Decrease the swap count directly.
Context: The caller must lock whatever is referencing the slots to
avoid a race.
Typically the page table zapping or shmem mapping truncate will need
to free swap slots directly. If a slot is cached (has a folio bound),
this will also try to release the swap cache.
- swap_dup_entry_direct() - Increase the swap count directly.
Context: The caller must lock whatever is referencing the entries to
avoid race, and the entries must already have a swap count > 1.
Typically, forking will need to copy the page table and hence needs to
increase the swap count of the entries in the table. The page table is
locked while referencing the swap entries, so the entries all have a
swap count > 1 and can't be freed.
Hibernation subsystem is a bit different, so two special wrappers are here:
- swap_alloc_hibernation_slot() - Allocate one entry from one device.
- swap_free_hibernation_slot() - Free one entry allocated by the above
helper.
All hibernation entries are exclusive to the hibernation subsystem and
should not interact with ordinary swap routines.
By separating the workflows, it will be possible to bind folio more
tightly with swap cache and get rid of the SWAP_HAS_CACHE as a temporary
pin.
This commit should not introduce any behavior change
[kasong@tencent.com: fix leak, per Chris Mason. Remove WARN_ON, per Lai Yi]
Link: https://lkml.kernel.org/r/CAMgjq7AUz10uETVm8ozDWcB3XohkOqf0i33KGrAquvEVvfp5cg@mail.gmail.com
[ryncsn@gmail.com: fix KSM copy pages for swapoff, per Chris]
Link: https://lkml.kernel.org/r/aXxkANcET3l2Xu6J@KASONG-MC4
Link: https://lkml.kernel.org/r/20251220-swap-table-p2-v5-14-8862a265a033@tencent.com
Signed-off-by: Kairui Song <kasong@tencent.com>
Signed-off-by: Kairui Song <ryncsn@gmail.com>
Acked-by: Rafael J. Wysocki (Intel) <rafael@kernel.org>
Reviewed-by: Baoquan He <bhe@redhat.com>
Cc: Baolin Wang <baolin.wang@linux.alibaba.com>
Cc: Barry Song <baohua@kernel.org>
Cc: Chris Li <chrisl@kernel.org>
Cc: Nhat Pham <nphamcs@gmail.com>
Cc: Yosry Ahmed <yosry.ahmed@linux.dev>
Cc: Deepanshu Kartikey <kartikey406@gmail.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Kairui Song <ryncsn@gmail.com>
Cc: Chris Mason <clm@fb.com>
Cc: Chris Mason <clm@meta.com>
Cc: Lai Yi <yi1.lai@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2025-12-20 03:43:43 +08:00
|
|
|
|
Revert "rmap: do not call mmu_notifier_invalidate_page() under ptl"
This reverts commit aac2fea94f7a3df8ad1eeb477eb2643f81fd5393.
It turns out that that patch was complete and utter garbage, and broke
KVM, resulting in odd oopses.
Quoting Andrea Arcangeli:
"The aforementioned commit has 3 bugs.
1) mmu_notifier_invalidate_range cannot be used in replacement of
mmu_notifier_invalidate_range_start/end.
For KVM mmu_notifier_invalidate_range is a noop and rightfully so.
A MMU notifier implementation has to implement either
->invalidate_range method or the invalidate_range_start/end
methods, not both. And if you implement invalidate_range_start/end
like KVM is forced to do, calling mmu_notifier_invalidate_range in
common code is a noop for KVM.
For those MMU notifiers that can get away only implementing
->invalidate_range, the ->invalidate_range is implicitly called by
mmu_notifier_invalidate_range_end(). And only those secondary MMUs
that share the same pagetable with the primary MMU (like AMD
iommuv2) can get away only implementing ->invalidate_range.
So all cases (THP on/off) are broken right now.
To fix this is enough to replace mmu_notifier_invalidate_range with
mmu_notifier_invalidate_range_start;mmu_notifier_invalidate_range_end.
Either that or call multiple mmu_notifier_invalidate_page like
before.
2) address + (1UL << compound_order(page) is buggy, it should be
PAGE_SIZE << compound_order(page), it's bytes not pages, 2M not
512.
3) The whole invalidate_range thing was an attempt to call a single
invalidate while walking multiple 4k ptes that maps the same THP
(after a pmd virtual split without physical compound page THP
split).
It's unclear if the rmap_walk will always provide an address that
is 2M aligned as parameter to try_to_unmap_one, in presence of THP.
I think it needs also an address &= (PAGE_SIZE <<
compound_order(page)) - 1 to be safe"
In general, we should stop making excuses for horrible MMU notifier
users. It's much more important that the core VM is sane and safe, than
letting MMU notifiers sleep.
So if some MMU notifier is sleeping under a spinlock, we need to fix the
notifier, not try to make excuses for that garbage in the core VM.
Reported-and-tested-by: Bernhard Held <berny156@gmx.de>
Reported-and-tested-by: Adam Borowski <kilobyte@angband.pl>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Radim Krčmář <rkrcmar@redhat.com>
Cc: Wanpeng Li <kernellwp@gmail.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Takashi Iwai <tiwai@suse.de>
Cc: Nadav Amit <nadav.amit@gmail.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Jérôme Glisse <jglisse@redhat.com>
Cc: axie <axie@amd.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-08-29 09:11:06 -07:00
|
|
|
|
2024-06-14 09:51:36 +08:00
|
|
|
|
2017-02-24 14:58:01 -08:00
|
|
|
|
2025-02-10 20:37:52 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2018-02-21 10:15:44 -07:00
|
|
|
|
mm, swap: cleanup swap entry management workflow
The current swap entry allocation/freeing workflow has never had a clear
definition. This makes it hard to debug or add new optimizations.
This commit introduces a proper definition of how swap entries would be
allocated and freed. Now, most operations are folio based, so they will
never exceed one swap cluster, and we now have a cleaner border between
swap and the rest of mm, making it much easier to follow and debug,
especially with new added sanity checks. Also making more optimization
possible.
Swap entry will be mostly freed and free with a folio bound. The folio
lock will be useful for resolving many swap related races.
Now swap allocation (except hibernation) always starts with a folio in the
swap cache, and gets duped/freed protected by the folio lock:
- folio_alloc_swap() - The only allocation entry point now.
Context: The folio must be locked.
This allocates one or a set of continuous swap slots for a folio and
binds them to the folio by adding the folio to the swap cache. The
swap slots' swap count start with zero value.
- folio_dup_swap() - Increase the swap count of one or more entries.
Context: The folio must be locked and in the swap cache. For now, the
caller still has to lock the new swap entry owner (e.g., PTL).
This increases the ref count of swap entries allocated to a folio.
Newly allocated swap slots' count has to be increased by this helper
as the folio got unmapped (and swap entries got installed).
- folio_put_swap() - Decrease the swap count of one or more entries.
Context: The folio must be locked and in the swap cache. For now, the
caller still has to lock the new swap entry owner (e.g., PTL).
This decreases the ref count of swap entries allocated to a folio.
Typically, swapin will decrease the swap count as the folio got
installed back and the swap entry got uninstalled
This won't remove the folio from the swap cache and free the
slot. Lazy freeing of swap cache is helpful for reducing IO.
There is already a folio_free_swap() for immediate cache reclaim.
This part could be further optimized later.
The above locking constraints could be further relaxed when the swap table
is fully implemented. Currently dup still needs the caller to lock the
swap entry container (e.g. PTL), or a concurrent zap may underflow the
swap count.
Some swap users need to interact with swap count without involving folio
(e.g. forking/zapping the page table or mapping truncate without swapin).
In such cases, the caller has to ensure there is no race condition on
whatever owns the swap count and use the below helpers:
- swap_put_entries_direct() - Decrease the swap count directly.
Context: The caller must lock whatever is referencing the slots to
avoid a race.
Typically the page table zapping or shmem mapping truncate will need
to free swap slots directly. If a slot is cached (has a folio bound),
this will also try to release the swap cache.
- swap_dup_entry_direct() - Increase the swap count directly.
Context: The caller must lock whatever is referencing the entries to
avoid race, and the entries must already have a swap count > 1.
Typically, forking will need to copy the page table and hence needs to
increase the swap count of the entries in the table. The page table is
locked while referencing the swap entries, so the entries all have a
swap count > 1 and can't be freed.
Hibernation subsystem is a bit different, so two special wrappers are here:
- swap_alloc_hibernation_slot() - Allocate one entry from one device.
- swap_free_hibernation_slot() - Free one entry allocated by the above
helper.
All hibernation entries are exclusive to the hibernation subsystem and
should not interact with ordinary swap routines.
By separating the workflows, it will be possible to bind folio more
tightly with swap cache and get rid of the SWAP_HAS_CACHE as a temporary
pin.
This commit should not introduce any behavior change
[kasong@tencent.com: fix leak, per Chris Mason. Remove WARN_ON, per Lai Yi]
Link: https://lkml.kernel.org/r/CAMgjq7AUz10uETVm8ozDWcB3XohkOqf0i33KGrAquvEVvfp5cg@mail.gmail.com
[ryncsn@gmail.com: fix KSM copy pages for swapoff, per Chris]
Link: https://lkml.kernel.org/r/aXxkANcET3l2Xu6J@KASONG-MC4
Link: https://lkml.kernel.org/r/20251220-swap-table-p2-v5-14-8862a265a033@tencent.com
Signed-off-by: Kairui Song <kasong@tencent.com>
Signed-off-by: Kairui Song <ryncsn@gmail.com>
Acked-by: Rafael J. Wysocki (Intel) <rafael@kernel.org>
Reviewed-by: Baoquan He <bhe@redhat.com>
Cc: Baolin Wang <baolin.wang@linux.alibaba.com>
Cc: Barry Song <baohua@kernel.org>
Cc: Chris Li <chrisl@kernel.org>
Cc: Nhat Pham <nphamcs@gmail.com>
Cc: Yosry Ahmed <yosry.ahmed@linux.dev>
Cc: Deepanshu Kartikey <kartikey406@gmail.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Kairui Song <ryncsn@gmail.com>
Cc: Chris Mason <clm@fb.com>
Cc: Chris Mason <clm@meta.com>
Cc: Lai Yi <yi1.lai@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2025-12-20 03:43:43 +08:00
|
|
|
|
2018-02-21 10:15:44 -07:00
|
|
|
|
2024-06-14 09:51:36 +08:00
|
|
|
|
2018-02-21 10:15:44 -07:00
|
|
|
|
mm: fix PageAnonExclusive clearing racing with concurrent RCU GUP-fast
commit 6c287605fd56 ("mm: remember exclusively mapped anonymous pages with
PG_anon_exclusive") made sure that when PageAnonExclusive() has to be
cleared during temporary unmapping of a page, that the PTE is
cleared/invalidated and that the TLB is flushed.
What we want to achieve in all cases is that we cannot end up with a pin on
an anonymous page that may be shared, because such pins would be
unreliable and could result in memory corruptions when the mapped page
and the pin go out of sync due to a write fault.
That TLB flush handling was inspired by an outdated comment in
mm/ksm.c:write_protect_page(), which similarly required the TLB flush in
the past to synchronize with GUP-fast. However, ever since general RCU GUP
fast was introduced in commit 2667f50e8b81 ("mm: introduce a general RCU
get_user_pages_fast()"), a TLB flush is no longer sufficient to handle
concurrent GUP-fast in all cases -- it only handles traditional IPI-based
GUP-fast correctly.
Peter Xu (thankfully) questioned whether that TLB flush is really
required. On architectures that send an IPI broadcast on TLB flush,
it works as expected. To synchronize with RCU GUP-fast properly, we're
conceptually fine, however, we have to enforce a certain memory order and
are missing memory barriers.
Let's document that, avoid the TLB flush where possible and use proper
explicit memory barriers where required. We shouldn't really care about the
additional memory barriers here, as we're not on extremely hot paths --
and we're getting rid of some TLB flushes.
We use a smp_mb() pair for handling concurrent pinning and a
smp_rmb()/smp_wmb() pair for handling the corner case of only temporary
PTE changes but permanent PageAnonExclusive changes.
One extreme example, whereby GUP-fast takes a R/O pin and KSM wants to
convert an exclusive anonymous page to a KSM page, and that page is already
mapped write-protected (-> no PTE change) would be:
Thread 0 (KSM) Thread 1 (GUP-fast)
(B1) Read the PTE
# (B2) skipped without FOLL_WRITE
(A1) Clear PTE
smp_mb()
(A2) Check pinned
(B3) Pin the mapped page
smp_mb()
(A3) Clear PageAnonExclusive
smp_wmb()
(A4) Restore PTE
(B4) Check if the PTE changed
smp_rmb()
(B5) Check PageAnonExclusive
Thread 1 will properly detect that PageAnonExclusive was cleared and
back off.
Note that we don't need a memory barrier between checking if the page is
pinned and clearing PageAnonExclusive, because stores are not
speculated.
The possible issues due to reordering are of theoretical nature so far
and attempts to reproduce the race failed.
Especially the "no PTE change" case isn't the common case, because we'd
need an exclusive anonymous page that's mapped R/O and the PTE is clean
in KSM code -- and using KSM with page pinning isn't extremely common.
Further, the clear+TLB flush we used for now implies a memory barrier.
So the problematic missing part should be the missing memory barrier
after pinning but before checking if the PTE changed.
Link: https://lkml.kernel.org/r/20220901083559.67446-1-david@redhat.com
Fixes: 6c287605fd56 ("mm: remember exclusively mapped anonymous pages with PG_anon_exclusive")
Signed-off-by: David Hildenbrand <david@redhat.com>
Cc: Jason Gunthorpe <jgg@nvidia.com>
Cc: John Hubbard <jhubbard@nvidia.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Peter Xu <peterx@redhat.com>
Cc: Alistair Popple <apopple@nvidia.com>
Cc: Nadav Amit <namit@vmware.com>
Cc: Yang Shi <shy828301@gmail.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Andrea Parri <parri.andrea@gmail.com>
Cc: Will Deacon <will@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: "Paul E. McKenney" <paulmck@kernel.org>
Cc: Christoph von Recklinghausen <crecklin@redhat.com>
Cc: Don Dutile <ddutile@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2022-09-01 10:35:59 +02:00
|
|
|
|
2023-12-20 23:45:02 +01:00
|
|
|
|
mm: remember exclusively mapped anonymous pages with PG_anon_exclusive
Let's mark exclusively mapped anonymous pages with PG_anon_exclusive as
exclusive, and use that information to make GUP pins reliable and stay
consistent with the page mapped into the page table even if the page table
entry gets write-protected.
With that information at hand, we can extend our COW logic to always reuse
anonymous pages that are exclusive. For anonymous pages that might be
shared, the existing logic applies.
As already documented, PG_anon_exclusive is usually only expressive in
combination with a page table entry. Especially PTE vs. PMD-mapped
anonymous pages require more thought, some examples: due to mremap() we
can easily have a single compound page PTE-mapped into multiple page
tables exclusively in a single process -- multiple page table locks apply.
Further, due to MADV_WIPEONFORK we might not necessarily write-protect
all PTEs, and only some subpages might be pinned. Long story short: once
PTE-mapped, we have to track information about exclusivity per sub-page,
but until then, we can just track it for the compound page in the head
page and not having to update a whole bunch of subpages all of the time
for a simple PMD mapping of a THP.
For simplicity, this commit mostly talks about "anonymous pages", while
it's for THP actually "the part of an anonymous folio referenced via a
page table entry".
To not spill PG_anon_exclusive code all over the mm code-base, we let the
anon rmap code to handle all PG_anon_exclusive logic it can easily handle.
If a writable, present page table entry points at an anonymous (sub)page,
that (sub)page must be PG_anon_exclusive. If GUP wants to take a reliably
pin (FOLL_PIN) on an anonymous page references via a present page table
entry, it must only pin if PG_anon_exclusive is set for the mapped
(sub)page.
This commit doesn't adjust GUP, so this is only implicitly handled for
FOLL_WRITE, follow-up commits will teach GUP to also respect it for
FOLL_PIN without FOLL_WRITE, to make all GUP pins of anonymous pages fully
reliable.
Whenever an anonymous page is to be shared (fork(), KSM), or when
temporarily unmapping an anonymous page (swap, migration), the relevant
PG_anon_exclusive bit has to be cleared to mark the anonymous page
possibly shared. Clearing will fail if there are GUP pins on the page:
* For fork(), this means having to copy the page and not being able to
share it. fork() protects against concurrent GUP using the PT lock and
the src_mm->write_protect_seq.
* For KSM, this means sharing will fail. For swap this means, unmapping
will fail, For migration this means, migration will fail early. All
three cases protect against concurrent GUP using the PT lock and a
proper clear/invalidate+flush of the relevant page table entry.
This fixes memory corruptions reported for FOLL_PIN | FOLL_WRITE, when a
pinned page gets mapped R/O and the successive write fault ends up
replacing the page instead of reusing it. It improves the situation for
O_DIRECT/vmsplice/... that still use FOLL_GET instead of FOLL_PIN, if
fork() is *not* involved, however swapout and fork() are still
problematic. Properly using FOLL_PIN instead of FOLL_GET for these GUP
users will fix the issue for them.
I. Details about basic handling
I.1. Fresh anonymous pages
page_add_new_anon_rmap() and hugepage_add_new_anon_rmap() will mark the
given page exclusive via __page_set_anon_rmap(exclusive=1). As that is
the mechanism fresh anonymous pages come into life (besides migration code
where we copy the page->mapping), all fresh anonymous pages will start out
as exclusive.
I.2. COW reuse handling of anonymous pages
When a COW handler stumbles over a (sub)page that's marked exclusive, it
simply reuses it. Otherwise, the handler tries harder under page lock to
detect if the (sub)page is exclusive and can be reused. If exclusive,
page_move_anon_rmap() will mark the given (sub)page exclusive.
Note that hugetlb code does not yet check for PageAnonExclusive(), as it
still uses the old COW logic that is prone to the COW security issue
because hugetlb code cannot really tolerate unnecessary/wrong COW as huge
pages are a scarce resource.
I.3. Migration handling
try_to_migrate() has to try marking an exclusive anonymous page shared via
page_try_share_anon_rmap(). If it fails because there are GUP pins on the
page, unmap fails. migrate_vma_collect_pmd() and
__split_huge_pmd_locked() are handled similarly.
Writable migration entries implicitly point at shared anonymous pages.
For readable migration entries that information is stored via a new
"readable-exclusive" migration entry, specific to anonymous pages.
When restoring a migration entry in remove_migration_pte(), information
about exlusivity is detected via the migration entry type, and
RMAP_EXCLUSIVE is set accordingly for
page_add_anon_rmap()/hugepage_add_anon_rmap() to restore that information.
I.4. Swapout handling
try_to_unmap() has to try marking the mapped page possibly shared via
page_try_share_anon_rmap(). If it fails because there are GUP pins on the
page, unmap fails. For now, information about exclusivity is lost. In
the future, we might want to remember that information in the swap entry
in some cases, however, it requires more thought, care, and a way to store
that information in swap entries.
I.5. Swapin handling
do_swap_page() will never stumble over exclusive anonymous pages in the
swap cache, as try_to_migrate() prohibits that. do_swap_page() always has
to detect manually if an anonymous page is exclusive and has to set
RMAP_EXCLUSIVE for page_add_anon_rmap() accordingly.
I.6. THP handling
__split_huge_pmd_locked() has to move the information about exclusivity
from the PMD to the PTEs.
a) In case we have a readable-exclusive PMD migration entry, simply
insert readable-exclusive PTE migration entries.
b) In case we have a present PMD entry and we don't want to freeze
("convert to migration entries"), simply forward PG_anon_exclusive to
all sub-pages, no need to temporarily clear the bit.
c) In case we have a present PMD entry and want to freeze, handle it
similar to try_to_migrate(): try marking the page shared first. In
case we fail, we ignore the "freeze" instruction and simply split
ordinarily. try_to_migrate() will properly fail because the THP is
still mapped via PTEs.
When splitting a compound anonymous folio (THP), the information about
exclusivity is implicitly handled via the migration entries: no need to
replicate PG_anon_exclusive manually.
I.7. fork() handling fork() handling is relatively easy, because
PG_anon_exclusive is only expressive for some page table entry types.
a) Present anonymous pages
page_try_dup_anon_rmap() will mark the given subpage shared -- which will
fail if the page is pinned. If it failed, we have to copy (or PTE-map a
PMD to handle it on the PTE level).
Note that device exclusive entries are just a pointer at a PageAnon()
page. fork() will first convert a device exclusive entry to a present
page table and handle it just like present anonymous pages.
b) Device private entry
Device private entries point at PageAnon() pages that cannot be mapped
directly and, therefore, cannot get pinned.
page_try_dup_anon_rmap() will mark the given subpage shared, which cannot
fail because they cannot get pinned.
c) HW poison entries
PG_anon_exclusive will remain untouched and is stale -- the page table
entry is just a placeholder after all.
d) Migration entries
Writable and readable-exclusive entries are converted to readable entries:
possibly shared.
I.8. mprotect() handling
mprotect() only has to properly handle the new readable-exclusive
migration entry:
When write-protecting a migration entry that points at an anonymous page,
remember the information about exclusivity via the "readable-exclusive"
migration entry type.
II. Migration and GUP-fast
Whenever replacing a present page table entry that maps an exclusive
anonymous page by a migration entry, we have to mark the page possibly
shared and synchronize against GUP-fast by a proper clear/invalidate+flush
to make the following scenario impossible:
1. try_to_migrate() places a migration entry after checking for GUP pins
and marks the page possibly shared.
2. GUP-fast pins the page due to lack of synchronization
3. fork() converts the "writable/readable-exclusive" migration entry into a
readable migration entry
4. Migration fails due to the GUP pin (failing to freeze the refcount)
5. Migration entries are restored. PG_anon_exclusive is lost
-> We have a pinned page that is not marked exclusive anymore.
Note that we move information about exclusivity from the page to the
migration entry as it otherwise highly overcomplicates fork() and
PTE-mapping a THP.
III. Swapout and GUP-fast
Whenever replacing a present page table entry that maps an exclusive
anonymous page by a swap entry, we have to mark the page possibly shared
and synchronize against GUP-fast by a proper clear/invalidate+flush to
make the following scenario impossible:
1. try_to_unmap() places a swap entry after checking for GUP pins and
clears exclusivity information on the page.
2. GUP-fast pins the page due to lack of synchronization.
-> We have a pinned page that is not marked exclusive anymore.
If we'd ever store information about exclusivity in the swap entry,
similar to migration handling, the same considerations as in II would
apply. This is future work.
Link: https://lkml.kernel.org/r/20220428083441.37290-13-david@redhat.com
Signed-off-by: David Hildenbrand <david@redhat.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Christoph Hellwig <hch@lst.de>
Cc: David Rientjes <rientjes@google.com>
Cc: Don Dutile <ddutile@redhat.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Jan Kara <jack@suse.cz>
Cc: Jann Horn <jannh@google.com>
Cc: Jason Gunthorpe <jgg@nvidia.com>
Cc: John Hubbard <jhubbard@nvidia.com>
Cc: Khalid Aziz <khalid.aziz@oracle.com>
Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Liang Zhang <zhangliang5@huawei.com>
Cc: "Matthew Wilcox (Oracle)" <willy@infradead.org>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Mike Rapoport <rppt@linux.ibm.com>
Cc: Nadav Amit <namit@vmware.com>
Cc: Oded Gabbay <oded.gabbay@gmail.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Pedro Demarchi Gomes <pedrodemargomes@gmail.com>
Cc: Peter Xu <peterx@redhat.com>
Cc: Rik van Riel <riel@surriel.com>
Cc: Roman Gushchin <guro@fb.com>
Cc: Shakeel Butt <shakeelb@google.com>
Cc: Yang Shi <shy828301@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2022-05-09 18:20:44 -07:00
|
|
|
|
2023-12-20 23:45:02 +01:00
|
|
|
|
mm, swap: cleanup swap entry management workflow
The current swap entry allocation/freeing workflow has never had a clear
definition. This makes it hard to debug or add new optimizations.
This commit introduces a proper definition of how swap entries would be
allocated and freed. Now, most operations are folio based, so they will
never exceed one swap cluster, and we now have a cleaner border between
swap and the rest of mm, making it much easier to follow and debug,
especially with new added sanity checks. Also making more optimization
possible.
Swap entry will be mostly freed and free with a folio bound. The folio
lock will be useful for resolving many swap related races.
Now swap allocation (except hibernation) always starts with a folio in the
swap cache, and gets duped/freed protected by the folio lock:
- folio_alloc_swap() - The only allocation entry point now.
Context: The folio must be locked.
This allocates one or a set of continuous swap slots for a folio and
binds them to the folio by adding the folio to the swap cache. The
swap slots' swap count start with zero value.
- folio_dup_swap() - Increase the swap count of one or more entries.
Context: The folio must be locked and in the swap cache. For now, the
caller still has to lock the new swap entry owner (e.g., PTL).
This increases the ref count of swap entries allocated to a folio.
Newly allocated swap slots' count has to be increased by this helper
as the folio got unmapped (and swap entries got installed).
- folio_put_swap() - Decrease the swap count of one or more entries.
Context: The folio must be locked and in the swap cache. For now, the
caller still has to lock the new swap entry owner (e.g., PTL).
This decreases the ref count of swap entries allocated to a folio.
Typically, swapin will decrease the swap count as the folio got
installed back and the swap entry got uninstalled
This won't remove the folio from the swap cache and free the
slot. Lazy freeing of swap cache is helpful for reducing IO.
There is already a folio_free_swap() for immediate cache reclaim.
This part could be further optimized later.
The above locking constraints could be further relaxed when the swap table
is fully implemented. Currently dup still needs the caller to lock the
swap entry container (e.g. PTL), or a concurrent zap may underflow the
swap count.
Some swap users need to interact with swap count without involving folio
(e.g. forking/zapping the page table or mapping truncate without swapin).
In such cases, the caller has to ensure there is no race condition on
whatever owns the swap count and use the below helpers:
- swap_put_entries_direct() - Decrease the swap count directly.
Context: The caller must lock whatever is referencing the slots to
avoid a race.
Typically the page table zapping or shmem mapping truncate will need
to free swap slots directly. If a slot is cached (has a folio bound),
this will also try to release the swap cache.
- swap_dup_entry_direct() - Increase the swap count directly.
Context: The caller must lock whatever is referencing the entries to
avoid race, and the entries must already have a swap count > 1.
Typically, forking will need to copy the page table and hence needs to
increase the swap count of the entries in the table. The page table is
locked while referencing the swap entries, so the entries all have a
swap count > 1 and can't be freed.
Hibernation subsystem is a bit different, so two special wrappers are here:
- swap_alloc_hibernation_slot() - Allocate one entry from one device.
- swap_free_hibernation_slot() - Free one entry allocated by the above
helper.
All hibernation entries are exclusive to the hibernation subsystem and
should not interact with ordinary swap routines.
By separating the workflows, it will be possible to bind folio more
tightly with swap cache and get rid of the SWAP_HAS_CACHE as a temporary
pin.
This commit should not introduce any behavior change
[kasong@tencent.com: fix leak, per Chris Mason. Remove WARN_ON, per Lai Yi]
Link: https://lkml.kernel.org/r/CAMgjq7AUz10uETVm8ozDWcB3XohkOqf0i33KGrAquvEVvfp5cg@mail.gmail.com
[ryncsn@gmail.com: fix KSM copy pages for swapoff, per Chris]
Link: https://lkml.kernel.org/r/aXxkANcET3l2Xu6J@KASONG-MC4
Link: https://lkml.kernel.org/r/20251220-swap-table-p2-v5-14-8862a265a033@tencent.com
Signed-off-by: Kairui Song <kasong@tencent.com>
Signed-off-by: Kairui Song <ryncsn@gmail.com>
Acked-by: Rafael J. Wysocki (Intel) <rafael@kernel.org>
Reviewed-by: Baoquan He <bhe@redhat.com>
Cc: Baolin Wang <baolin.wang@linux.alibaba.com>
Cc: Barry Song <baohua@kernel.org>
Cc: Chris Li <chrisl@kernel.org>
Cc: Nhat Pham <nphamcs@gmail.com>
Cc: Yosry Ahmed <yosry.ahmed@linux.dev>
Cc: Deepanshu Kartikey <kartikey406@gmail.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Kairui Song <ryncsn@gmail.com>
Cc: Chris Mason <clm@fb.com>
Cc: Chris Mason <clm@meta.com>
Cc: Lai Yi <yi1.lai@linux.intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2025-12-20 03:43:43 +08:00
|
|
|
|
mm: remember exclusively mapped anonymous pages with PG_anon_exclusive
Let's mark exclusively mapped anonymous pages with PG_anon_exclusive as
exclusive, and use that information to make GUP pins reliable and stay
consistent with the page mapped into the page table even if the page table
entry gets write-protected.
With that information at hand, we can extend our COW logic to always reuse
anonymous pages that are exclusive. For anonymous pages that might be
shared, the existing logic applies.
As already documented, PG_anon_exclusive is usually only expressive in
combination with a page table entry. Especially PTE vs. PMD-mapped
anonymous pages require more thought, some examples: due to mremap() we
can easily have a single compound page PTE-mapped into multiple page
tables exclusively in a single process -- multiple page table locks apply.
Further, due to MADV_WIPEONFORK we might not necessarily write-protect
all PTEs, and only some subpages might be pinned. Long story short: once
PTE-mapped, we have to track information about exclusivity per sub-page,
but until then, we can just track it for the compound page in the head
page and not having to update a whole bunch of subpages all of the time
for a simple PMD mapping of a THP.
For simplicity, this commit mostly talks about "anonymous pages", while
it's for THP actually "the part of an anonymous folio referenced via a
page table entry".
To not spill PG_anon_exclusive code all over the mm code-base, we let the
anon rmap code to handle all PG_anon_exclusive logic it can easily handle.
If a writable, present page table entry points at an anonymous (sub)page,
that (sub)page must be PG_anon_exclusive. If GUP wants to take a reliably
pin (FOLL_PIN) on an anonymous page references via a present page table
entry, it must only pin if PG_anon_exclusive is set for the mapped
(sub)page.
This commit doesn't adjust GUP, so this is only implicitly handled for
FOLL_WRITE, follow-up commits will teach GUP to also respect it for
FOLL_PIN without FOLL_WRITE, to make all GUP pins of anonymous pages fully
reliable.
Whenever an anonymous page is to be shared (fork(), KSM), or when
temporarily unmapping an anonymous page (swap, migration), the relevant
PG_anon_exclusive bit has to be cleared to mark the anonymous page
possibly shared. Clearing will fail if there are GUP pins on the page:
* For fork(), this means having to copy the page and not being able to
share it. fork() protects against concurrent GUP using the PT lock and
the src_mm->write_protect_seq.
* For KSM, this means sharing will fail. For swap this means, unmapping
will fail, For migration this means, migration will fail early. All
three cases protect against concurrent GUP using the PT lock and a
proper clear/invalidate+flush of the relevant page table entry.
This fixes memory corruptions reported for FOLL_PIN | FOLL_WRITE, when a
pinned page gets mapped R/O and the successive write fault ends up
replacing the page instead of reusing it. It improves the situation for
O_DIRECT/vmsplice/... that still use FOLL_GET instead of FOLL_PIN, if
fork() is *not* involved, however swapout and fork() are still
problematic. Properly using FOLL_PIN instead of FOLL_GET for these GUP
users will fix the issue for them.
I. Details about basic handling
I.1. Fresh anonymous pages
page_add_new_anon_rmap() and hugepage_add_new_anon_rmap() will mark the
given page exclusive via __page_set_anon_rmap(exclusive=1). As that is
the mechanism fresh anonymous pages come into life (besides migration code
where we copy the page->mapping), all fresh anonymous pages will start out
as exclusive.
I.2. COW reuse handling of anonymous pages
When a COW handler stumbles over a (sub)page that's marked exclusive, it
simply reuses it. Otherwise, the handler tries harder under page lock to
detect if the (sub)page is exclusive and can be reused. If exclusive,
page_move_anon_rmap() will mark the given (sub)page exclusive.
Note that hugetlb code does not yet check for PageAnonExclusive(), as it
still uses the old COW logic that is prone to the COW security issue
because hugetlb code cannot really tolerate unnecessary/wrong COW as huge
pages are a scarce resource.
I.3. Migration handling
try_to_migrate() has to try marking an exclusive anonymous page shared via
page_try_share_anon_rmap(). If it fails because there are GUP pins on the
page, unmap fails. migrate_vma_collect_pmd() and
__split_huge_pmd_locked() are handled similarly.
Writable migration entries implicitly point at shared anonymous pages.
For readable migration entries that information is stored via a new
"readable-exclusive" migration entry, specific to anonymous pages.
When restoring a migration entry in remove_migration_pte(), information
about exlusivity is detected via the migration entry type, and
RMAP_EXCLUSIVE is set accordingly for
page_add_anon_rmap()/hugepage_add_anon_rmap() to restore that information.
I.4. Swapout handling
try_to_unmap() has to try marking the mapped page possibly shared via
page_try_share_anon_rmap(). If it fails because there are GUP pins on the
page, unmap fails. For now, information about exclusivity is lost. In
the future, we might want to remember that information in the swap entry
in some cases, however, it requires more thought, care, and a way to store
that information in swap entries.
I.5. Swapin handling
do_swap_page() will never stumble over exclusive anonymous pages in the
swap cache, as try_to_migrate() prohibits that. do_swap_page() always has
to detect manually if an anonymous page is exclusive and has to set
RMAP_EXCLUSIVE for page_add_anon_rmap() accordingly.
I.6. THP handling
__split_huge_pmd_locked() has to move the information about exclusivity
from the PMD to the PTEs.
a) In case we have a readable-exclusive PMD migration entry, simply
insert readable-exclusive PTE migration entries.
b) In case we have a present PMD entry and we don't want to freeze
("convert to migration entries"), simply forward PG_anon_exclusive to
all sub-pages, no need to temporarily clear the bit.
c) In case we have a present PMD entry and want to freeze, handle it
similar to try_to_migrate(): try marking the page shared first. In
case we fail, we ignore the "freeze" instruction and simply split
ordinarily. try_to_migrate() will properly fail because the THP is
still mapped via PTEs.
When splitting a compound anonymous folio (THP), the information about
exclusivity is implicitly handled via the migration entries: no need to
replicate PG_anon_exclusive manually.
I.7. fork() handling fork() handling is relatively easy, because
PG_anon_exclusive is only expressive for some page table entry types.
a) Present anonymous pages
page_try_dup_anon_rmap() will mark the given subpage shared -- which will
fail if the page is pinned. If it failed, we have to copy (or PTE-map a
PMD to handle it on the PTE level).
Note that device exclusive entries are just a pointer at a PageAnon()
page. fork() will first convert a device exclusive entry to a present
page table and handle it just like present anonymous pages.
b) Device private entry
Device private entries point at PageAnon() pages that cannot be mapped
directly and, therefore, cannot get pinned.
page_try_dup_anon_rmap() will mark the given subpage shared, which cannot
fail because they cannot get pinned.
c) HW poison entries
PG_anon_exclusive will remain untouched and is stale -- the page table
entry is just a placeholder after all.
d) Migration entries
Writable and readable-exclusive entries are converted to readable entries:
possibly shared.
I.8. mprotect() handling
mprotect() only has to properly handle the new readable-exclusive
migration entry:
When write-protecting a migration entry that points at an anonymous page,
remember the information about exclusivity via the "readable-exclusive"
migration entry type.
II. Migration and GUP-fast
Whenever replacing a present page table entry that maps an exclusive
anonymous page by a migration entry, we have to mark the page possibly
shared and synchronize against GUP-fast by a proper clear/invalidate+flush
to make the following scenario impossible:
1. try_to_migrate() places a migration entry after checking for GUP pins
and marks the page possibly shared.
2. GUP-fast pins the page due to lack of synchronization
3. fork() converts the "writable/readable-exclusive" migration entry into a
readable migration entry
4. Migration fails due to the GUP pin (failing to freeze the refcount)
5. Migration entries are restored. PG_anon_exclusive is lost
-> We have a pinned page that is not marked exclusive anymore.
Note that we move information about exclusivity from the page to the
migration entry as it otherwise highly overcomplicates fork() and
PTE-mapping a THP.
III. Swapout and GUP-fast
Whenever replacing a present page table entry that maps an exclusive
anonymous page by a swap entry, we have to mark the page possibly shared
and synchronize against GUP-fast by a proper clear/invalidate+flush to
make the following scenario impossible:
1. try_to_unmap() places a swap entry after checking for GUP pins and
clears exclusivity information on the page.
2. GUP-fast pins the page due to lack of synchronization.
-> We have a pinned page that is not marked exclusive anymore.
If we'd ever store information about exclusivity in the swap entry,
similar to migration handling, the same considerations as in II would
apply. This is future work.
Link: https://lkml.kernel.org/r/20220428083441.37290-13-david@redhat.com
Signed-off-by: David Hildenbrand <david@redhat.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Christoph Hellwig <hch@lst.de>
Cc: David Rientjes <rientjes@google.com>
Cc: Don Dutile <ddutile@redhat.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Jan Kara <jack@suse.cz>
Cc: Jann Horn <jannh@google.com>
Cc: Jason Gunthorpe <jgg@nvidia.com>
Cc: John Hubbard <jhubbard@nvidia.com>
Cc: Khalid Aziz <khalid.aziz@oracle.com>
Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Liang Zhang <zhangliang5@huawei.com>
Cc: "Matthew Wilcox (Oracle)" <willy@infradead.org>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Mike Rapoport <rppt@linux.ibm.com>
Cc: Nadav Amit <namit@vmware.com>
Cc: Oded Gabbay <oded.gabbay@gmail.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Pedro Demarchi Gomes <pedrodemargomes@gmail.com>
Cc: Peter Xu <peterx@redhat.com>
Cc: Rik van Riel <riel@surriel.com>
Cc: Roman Gushchin <guro@fb.com>
Cc: Shakeel Butt <shakeelb@google.com>
Cc: Yang Shi <shy828301@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2022-05-09 18:20:44 -07:00
|
|
|
|
2024-06-14 09:51:36 +08:00
|
|
|
|
mm: remember exclusively mapped anonymous pages with PG_anon_exclusive
Let's mark exclusively mapped anonymous pages with PG_anon_exclusive as
exclusive, and use that information to make GUP pins reliable and stay
consistent with the page mapped into the page table even if the page table
entry gets write-protected.
With that information at hand, we can extend our COW logic to always reuse
anonymous pages that are exclusive. For anonymous pages that might be
shared, the existing logic applies.
As already documented, PG_anon_exclusive is usually only expressive in
combination with a page table entry. Especially PTE vs. PMD-mapped
anonymous pages require more thought, some examples: due to mremap() we
can easily have a single compound page PTE-mapped into multiple page
tables exclusively in a single process -- multiple page table locks apply.
Further, due to MADV_WIPEONFORK we might not necessarily write-protect
all PTEs, and only some subpages might be pinned. Long story short: once
PTE-mapped, we have to track information about exclusivity per sub-page,
but until then, we can just track it for the compound page in the head
page and not having to update a whole bunch of subpages all of the time
for a simple PMD mapping of a THP.
For simplicity, this commit mostly talks about "anonymous pages", while
it's for THP actually "the part of an anonymous folio referenced via a
page table entry".
To not spill PG_anon_exclusive code all over the mm code-base, we let the
anon rmap code to handle all PG_anon_exclusive logic it can easily handle.
If a writable, present page table entry points at an anonymous (sub)page,
that (sub)page must be PG_anon_exclusive. If GUP wants to take a reliably
pin (FOLL_PIN) on an anonymous page references via a present page table
entry, it must only pin if PG_anon_exclusive is set for the mapped
(sub)page.
This commit doesn't adjust GUP, so this is only implicitly handled for
FOLL_WRITE, follow-up commits will teach GUP to also respect it for
FOLL_PIN without FOLL_WRITE, to make all GUP pins of anonymous pages fully
reliable.
Whenever an anonymous page is to be shared (fork(), KSM), or when
temporarily unmapping an anonymous page (swap, migration), the relevant
PG_anon_exclusive bit has to be cleared to mark the anonymous page
possibly shared. Clearing will fail if there are GUP pins on the page:
* For fork(), this means having to copy the page and not being able to
share it. fork() protects against concurrent GUP using the PT lock and
the src_mm->write_protect_seq.
* For KSM, this means sharing will fail. For swap this means, unmapping
will fail, For migration this means, migration will fail early. All
three cases protect against concurrent GUP using the PT lock and a
proper clear/invalidate+flush of the relevant page table entry.
This fixes memory corruptions reported for FOLL_PIN | FOLL_WRITE, when a
pinned page gets mapped R/O and the successive write fault ends up
replacing the page instead of reusing it. It improves the situation for
O_DIRECT/vmsplice/... that still use FOLL_GET instead of FOLL_PIN, if
fork() is *not* involved, however swapout and fork() are still
problematic. Properly using FOLL_PIN instead of FOLL_GET for these GUP
users will fix the issue for them.
I. Details about basic handling
I.1. Fresh anonymous pages
page_add_new_anon_rmap() and hugepage_add_new_anon_rmap() will mark the
given page exclusive via __page_set_anon_rmap(exclusive=1). As that is
the mechanism fresh anonymous pages come into life (besides migration code
where we copy the page->mapping), all fresh anonymous pages will start out
as exclusive.
I.2. COW reuse handling of anonymous pages
When a COW handler stumbles over a (sub)page that's marked exclusive, it
simply reuses it. Otherwise, the handler tries harder under page lock to
detect if the (sub)page is exclusive and can be reused. If exclusive,
page_move_anon_rmap() will mark the given (sub)page exclusive.
Note that hugetlb code does not yet check for PageAnonExclusive(), as it
still uses the old COW logic that is prone to the COW security issue
because hugetlb code cannot really tolerate unnecessary/wrong COW as huge
pages are a scarce resource.
I.3. Migration handling
try_to_migrate() has to try marking an exclusive anonymous page shared via
page_try_share_anon_rmap(). If it fails because there are GUP pins on the
page, unmap fails. migrate_vma_collect_pmd() and
__split_huge_pmd_locked() are handled similarly.
Writable migration entries implicitly point at shared anonymous pages.
For readable migration entries that information is stored via a new
"readable-exclusive" migration entry, specific to anonymous pages.
When restoring a migration entry in remove_migration_pte(), information
about exlusivity is detected via the migration entry type, and
RMAP_EXCLUSIVE is set accordingly for
page_add_anon_rmap()/hugepage_add_anon_rmap() to restore that information.
I.4. Swapout handling
try_to_unmap() has to try marking the mapped page possibly shared via
page_try_share_anon_rmap(). If it fails because there are GUP pins on the
page, unmap fails. For now, information about exclusivity is lost. In
the future, we might want to remember that information in the swap entry
in some cases, however, it requires more thought, care, and a way to store
that information in swap entries.
I.5. Swapin handling
do_swap_page() will never stumble over exclusive anonymous pages in the
swap cache, as try_to_migrate() prohibits that. do_swap_page() always has
to detect manually if an anonymous page is exclusive and has to set
RMAP_EXCLUSIVE for page_add_anon_rmap() accordingly.
I.6. THP handling
__split_huge_pmd_locked() has to move the information about exclusivity
from the PMD to the PTEs.
a) In case we have a readable-exclusive PMD migration entry, simply
insert readable-exclusive PTE migration entries.
b) In case we have a present PMD entry and we don't want to freeze
("convert to migration entries"), simply forward PG_anon_exclusive to
all sub-pages, no need to temporarily clear the bit.
c) In case we have a present PMD entry and want to freeze, handle it
similar to try_to_migrate(): try marking the page shared first. In
case we fail, we ignore the "freeze" instruction and simply split
ordinarily. try_to_migrate() will properly fail because the THP is
still mapped via PTEs.
When splitting a compound anonymous folio (THP), the information about
exclusivity is implicitly handled via the migration entries: no need to
replicate PG_anon_exclusive manually.
I.7. fork() handling fork() handling is relatively easy, because
PG_anon_exclusive is only expressive for some page table entry types.
a) Present anonymous pages
page_try_dup_anon_rmap() will mark the given subpage shared -- which will
fail if the page is pinned. If it failed, we have to copy (or PTE-map a
PMD to handle it on the PTE level).
Note that device exclusive entries are just a pointer at a PageAnon()
page. fork() will first convert a device exclusive entry to a present
page table and handle it just like present anonymous pages.
b) Device private entry
Device private entries point at PageAnon() pages that cannot be mapped
directly and, therefore, cannot get pinned.
page_try_dup_anon_rmap() will mark the given subpage shared, which cannot
fail because they cannot get pinned.
c) HW poison entries
PG_anon_exclusive will remain untouched and is stale -- the page table
entry is just a placeholder after all.
d) Migration entries
Writable and readable-exclusive entries are converted to readable entries:
possibly shared.
I.8. mprotect() handling
mprotect() only has to properly handle the new readable-exclusive
migration entry:
When write-protecting a migration entry that points at an anonymous page,
remember the information about exclusivity via the "readable-exclusive"
migration entry type.
II. Migration and GUP-fast
Whenever replacing a present page table entry that maps an exclusive
anonymous page by a migration entry, we have to mark the page possibly
shared and synchronize against GUP-fast by a proper clear/invalidate+flush
to make the following scenario impossible:
1. try_to_migrate() places a migration entry after checking for GUP pins
and marks the page possibly shared.
2. GUP-fast pins the page due to lack of synchronization
3. fork() converts the "writable/readable-exclusive" migration entry into a
readable migration entry
4. Migration fails due to the GUP pin (failing to freeze the refcount)
5. Migration entries are restored. PG_anon_exclusive is lost
-> We have a pinned page that is not marked exclusive anymore.
Note that we move information about exclusivity from the page to the
migration entry as it otherwise highly overcomplicates fork() and
PTE-mapping a THP.
III. Swapout and GUP-fast
Whenever replacing a present page table entry that maps an exclusive
anonymous page by a swap entry, we have to mark the page possibly shared
and synchronize against GUP-fast by a proper clear/invalidate+flush to
make the following scenario impossible:
1. try_to_unmap() places a swap entry after checking for GUP pins and
clears exclusivity information on the page.
2. GUP-fast pins the page due to lack of synchronization.
-> We have a pinned page that is not marked exclusive anymore.
If we'd ever store information about exclusivity in the swap entry,
similar to migration handling, the same considerations as in II would
apply. This is future work.
Link: https://lkml.kernel.org/r/20220428083441.37290-13-david@redhat.com
Signed-off-by: David Hildenbrand <david@redhat.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Christoph Hellwig <hch@lst.de>
Cc: David Rientjes <rientjes@google.com>
Cc: Don Dutile <ddutile@redhat.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Jan Kara <jack@suse.cz>
Cc: Jann Horn <jannh@google.com>
Cc: Jason Gunthorpe <jgg@nvidia.com>
Cc: John Hubbard <jhubbard@nvidia.com>
Cc: Khalid Aziz <khalid.aziz@oracle.com>
Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Liang Zhang <zhangliang5@huawei.com>
Cc: "Matthew Wilcox (Oracle)" <willy@infradead.org>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Mike Rapoport <rppt@linux.ibm.com>
Cc: Nadav Amit <namit@vmware.com>
Cc: Oded Gabbay <oded.gabbay@gmail.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Pedro Demarchi Gomes <pedrodemargomes@gmail.com>
Cc: Peter Xu <peterx@redhat.com>
Cc: Rik van Riel <riel@surriel.com>
Cc: Roman Gushchin <guro@fb.com>
Cc: Shakeel Butt <shakeelb@google.com>
Cc: Yang Shi <shy828301@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2022-05-09 18:20:44 -07:00
|
|
|
|
2017-02-24 14:58:01 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
mm: support madvise(MADV_FREE)
Linux doesn't have an ability to free pages lazy while other OS already
have been supported that named by madvise(MADV_FREE).
The gain is clear that kernel can discard freed pages rather than
swapping out or OOM if memory pressure happens.
Without memory pressure, freed pages would be reused by userspace
without another additional overhead(ex, page fault + allocation +
zeroing).
Jason Evans said:
: Facebook has been using MAP_UNINITIALIZED
: (https://lkml.org/lkml/2012/1/18/308) in some of its applications for
: several years, but there are operational costs to maintaining this
: out-of-tree in our kernel and in jemalloc, and we are anxious to retire it
: in favor of MADV_FREE. When we first enabled MAP_UNINITIALIZED it
: increased throughput for much of our workload by ~5%, and although the
: benefit has decreased using newer hardware and kernels, there is still
: enough benefit that we cannot reasonably retire it without a replacement.
:
: Aside from Facebook operations, there are numerous broadly used
: applications that would benefit from MADV_FREE. The ones that immediately
: come to mind are redis, varnish, and MariaDB. I don't have much insight
: into Android internals and development process, but I would hope to see
: MADV_FREE support eventually end up there as well to benefit applications
: linked with the integrated jemalloc.
:
: jemalloc will use MADV_FREE once it becomes available in the Linux kernel.
: In fact, jemalloc already uses MADV_FREE or equivalent everywhere it's
: available: *BSD, OS X, Windows, and Solaris -- every platform except Linux
: (and AIX, but I'm not sure it even compiles on AIX). The lack of
: MADV_FREE on Linux forced me down a long series of increasingly
: sophisticated heuristics for madvise() volume reduction, and even so this
: remains a common performance issue for people using jemalloc on Linux.
: Please integrate MADV_FREE; many people will benefit substantially.
How it works:
When madvise syscall is called, VM clears dirty bit of ptes of the
range. If memory pressure happens, VM checks dirty bit of page table
and if it found still "clean", it means it's a "lazyfree pages" so VM
could discard the page instead of swapping out. Once there was store
operation for the page before VM peek a page to reclaim, dirty bit is
set so VM can swap out the page instead of discarding.
One thing we should notice is that basically, MADV_FREE relies on dirty
bit in page table entry to decide whether VM allows to discard the page
or not. IOW, if page table entry includes marked dirty bit, VM
shouldn't discard the page.
However, as a example, if swap-in by read fault happens, page table
entry doesn't have dirty bit so MADV_FREE could discard the page
wrongly.
For avoiding the problem, MADV_FREE did more checks with PageDirty and
PageSwapCache. It worked out because swapped-in page lives on swap
cache and since it is evicted from the swap cache, the page has PG_dirty
flag. So both page flags check effectively prevent wrong discarding by
MADV_FREE.
However, a problem in above logic is that swapped-in page has PG_dirty
still after they are removed from swap cache so VM cannot consider the
page as freeable any more even if madvise_free is called in future.
Look at below example for detail.
ptr = malloc();
memset(ptr);
..
..
.. heavy memory pressure so all of pages are swapped out
..
..
var = *ptr; -> a page swapped-in and could be removed from
swapcache. Then, page table doesn't mark
dirty bit and page descriptor includes PG_dirty
..
..
madvise_free(ptr); -> It doesn't clear PG_dirty of the page.
..
..
..
.. heavy memory pressure again.
.. In this time, VM cannot discard the page because the page
.. has *PG_dirty*
To solve the problem, this patch clears PG_dirty if only the page is
owned exclusively by current process when madvise is called because
PG_dirty represents ptes's dirtiness in several processes so we could
clear it only if we own it exclusively.
Firstly, heavy users would be general allocators(ex, jemalloc, tcmalloc
and hope glibc supports it) and jemalloc/tcmalloc already have supported
the feature for other OS(ex, FreeBSD)
barrios@blaptop:~/benchmark/ebizzy$ lscpu
Architecture: x86_64
CPU op-mode(s): 32-bit, 64-bit
Byte Order: Little Endian
CPU(s): 12
On-line CPU(s) list: 0-11
Thread(s) per core: 1
Core(s) per socket: 1
Socket(s): 12
NUMA node(s): 1
Vendor ID: GenuineIntel
CPU family: 6
Model: 2
Stepping: 3
CPU MHz: 3200.185
BogoMIPS: 6400.53
Virtualization: VT-x
Hypervisor vendor: KVM
Virtualization type: full
L1d cache: 32K
L1i cache: 32K
L2 cache: 4096K
NUMA node0 CPU(s): 0-11
ebizzy benchmark(./ebizzy -S 10 -n 512)
Higher avg is better.
vanilla-jemalloc MADV_free-jemalloc
1 thread
records: 10 records: 10
avg: 2961.90 avg: 12069.70
std: 71.96(2.43%) std: 186.68(1.55%)
max: 3070.00 max: 12385.00
min: 2796.00 min: 11746.00
2 thread
records: 10 records: 10
avg: 5020.00 avg: 17827.00
std: 264.87(5.28%) std: 358.52(2.01%)
max: 5244.00 max: 18760.00
min: 4251.00 min: 17382.00
4 thread
records: 10 records: 10
avg: 8988.80 avg: 27930.80
std: 1175.33(13.08%) std: 3317.33(11.88%)
max: 9508.00 max: 30879.00
min: 5477.00 min: 21024.00
8 thread
records: 10 records: 10
avg: 13036.50 avg: 33739.40
std: 170.67(1.31%) std: 5146.22(15.25%)
max: 13371.00 max: 40572.00
min: 12785.00 min: 24088.00
16 thread
records: 10 records: 10
avg: 11092.40 avg: 31424.20
std: 710.60(6.41%) std: 3763.89(11.98%)
max: 12446.00 max: 36635.00
min: 9949.00 min: 25669.00
32 thread
records: 10 records: 10
avg: 11067.00 avg: 34495.80
std: 971.06(8.77%) std: 2721.36(7.89%)
max: 12010.00 max: 38598.00
min: 9002.00 min: 30636.00
In summary, MADV_FREE is about much faster than MADV_DONTNEED.
This patch (of 12):
Add core MADV_FREE implementation.
[akpm@linux-foundation.org: small cleanups]
Signed-off-by: Minchan Kim <minchan@kernel.org>
Acked-by: Michal Hocko <mhocko@suse.com>
Acked-by: Hugh Dickins <hughd@google.com>
Cc: Mika Penttil <mika.penttila@nextfour.com>
Cc: Michael Kerrisk <mtk.manpages@gmail.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Rik van Riel <riel@redhat.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Jason Evans <je@fb.com>
Cc: Daniel Micay <danielmicay@gmail.com>
Cc: "Kirill A. Shutemov" <kirill@shutemov.name>
Cc: Shaohua Li <shli@kernel.org>
Cc: <yalin.wang2010@gmail.com>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: "James E.J. Bottomley" <jejb@parisc-linux.org>
Cc: "Kirill A. Shutemov" <kirill@shutemov.name>
Cc: "Shaohua Li" <shli@kernel.org>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Chen Gang <gang.chen.5i5j@gmail.com>
Cc: Chris Zankel <chris@zankel.net>
Cc: Darrick J. Wong <darrick.wong@oracle.com>
Cc: David S. Miller <davem@davemloft.net>
Cc: Helge Deller <deller@gmx.de>
Cc: Ivan Kokshaysky <ink@jurassic.park.msu.ru>
Cc: Matt Turner <mattst88@gmail.com>
Cc: Max Filippov <jcmvbkbc@gmail.com>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: Richard Henderson <rth@twiddle.net>
Cc: Roland Dreier <roland@kernel.org>
Cc: Russell King <rmk@arm.linux.org.uk>
Cc: Shaohua Li <shli@kernel.org>
Cc: Will Deacon <will.deacon@arm.com>
Cc: Wu Fengguang <fengguang.wu@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-01-15 16:54:53 -08:00
|
|
|
|
2017-02-24 14:58:01 -08:00
|
|
|
|
|
|
|
|
|
mm/swap: remember PG_anon_exclusive via a swp pte bit
Patch series "mm: COW fixes part 3: reliable GUP R/W FOLL_GET of anonymous pages", v2.
This series fixes memory corruptions when a GUP R/W reference (FOLL_WRITE
| FOLL_GET) was taken on an anonymous page and COW logic fails to detect
exclusivity of the page to then replacing the anonymous page by a copy in
the page table: The GUP reference lost synchronicity with the pages mapped
into the page tables. This series focuses on x86, arm64, s390x and
ppc64/book3s -- other architectures are fairly easy to support by
implementing __HAVE_ARCH_PTE_SWP_EXCLUSIVE.
This primarily fixes the O_DIRECT memory corruptions that can happen on
concurrent swapout, whereby we lose DMA reads to a page (modifying the
user page by writing to it).
O_DIRECT currently uses FOLL_GET for short-term (!FOLL_LONGTERM) DMA
from/to a user page. In the long run, we want to convert it to properly
use FOLL_PIN, and John is working on it, but that might take a while and
might not be easy to backport. In the meantime, let's restore what used
to work before we started modifying our COW logic: make R/W FOLL_GET
references reliable as long as there is no fork() after GUP involved.
This is just the natural follow-up of part 2, that will also further
reduce "wrong COW" on the swapin path, for example, when we cannot remove
a page from the swapcache due to concurrent writeback, or if we have two
threads faulting on the same swapped-out page. Fixing O_DIRECT is just a
nice side-product
This issue, including other related COW issues, has been summarized in [3]
under 2):
"
2. Intra Process Memory Corruptions due to Wrong COW (FOLL_GET)
It was discovered that we can create a memory corruption by reading a
file via O_DIRECT to a part (e.g., first 512 bytes) of a page,
concurrently writing to an unrelated part (e.g., last byte) of the same
page, and concurrently write-protecting the page via clear_refs
SOFTDIRTY tracking [6].
For the reproducer, the issue is that O_DIRECT grabs a reference of the
target page (via FOLL_GET) and clear_refs write-protects the relevant
page table entry. On successive write access to the page from the
process itself, we wrongly COW the page when resolving the write fault,
resulting in a loss of synchronicity and consequently a memory corruption.
While some people might think that using clear_refs in this combination
is a corner cases, it turns out to be a more generic problem unfortunately.
For example, it was just recently discovered that we can similarly
create a memory corruption without clear_refs, simply by concurrently
swapping out the buffer pages [7]. Note that we nowadays even use the
swap infrastructure in Linux without an actual swap disk/partition: the
prime example is zram which is enabled as default under Fedora [10].
The root issue is that a write-fault on a page that has additional
references results in a COW and thereby a loss of synchronicity
and consequently a memory corruption if two parties believe they are
referencing the same page.
"
We don't particularly care about R/O FOLL_GET references: they were never
reliable and O_DIRECT doesn't expect to observe modifications from a page
after DMA was started.
Note that:
* this only fixes the issue on x86, arm64, s390x and ppc64/book3s
("enterprise architectures"). Other architectures have to implement
__HAVE_ARCH_PTE_SWP_EXCLUSIVE to achieve the same.
* this does *not * consider any kind of fork() after taking the reference:
fork() after GUP never worked reliably with FOLL_GET.
* Not losing PG_anon_exclusive during swapout was the last remaining
piece. KSM already makes sure that there are no other references on
a page before considering it for sharing. Page migration maintains
PG_anon_exclusive and simply fails when there are additional references
(freezing the refcount fails). Only swapout code dropped the
PG_anon_exclusive flag because it requires more work to remember +
restore it.
With this series in place, most COW issues of [3] are fixed on said
architectures. Other architectures can implement
__HAVE_ARCH_PTE_SWP_EXCLUSIVE fairly easily.
[1] https://lkml.kernel.org/r/20220329160440.193848-1-david@redhat.com
[2] https://lkml.kernel.org/r/20211217113049.23850-1-david@redhat.com
[3] https://lore.kernel.org/r/3ae33b08-d9ef-f846-56fb-645e3b9b4c66@redhat.com
This patch (of 8):
Currently, we clear PG_anon_exclusive in try_to_unmap() and forget about
it. We do this, to keep fork() logic on swap entries easy and efficient:
for example, if we wouldn't clear it when unmapping, we'd have to lookup
the page in the swapcache for each and every swap entry during fork() and
clear PG_anon_exclusive if set.
Instead, we want to store that information directly in the swap pte,
protected by the page table lock, similarly to how we handle
SWP_MIGRATION_READ_EXCLUSIVE for migration entries. However, for actual
swap entries, we don't want to mess with the swap type (e.g., still one
bit) because it overcomplicates swap code.
In try_to_unmap(), we already reject to unmap in case the page might be
pinned, because we must not lose PG_anon_exclusive on pinned pages ever.
Checking if there are other unexpected references reliably *before*
completely unmapping a page is unfortunately not really possible: THP
heavily overcomplicate the situation. Once fully unmapped it's easier --
we, for example, make sure that there are no unexpected references *after*
unmapping a page before starting writeback on that page.
So, we currently might end up unmapping a page and clearing
PG_anon_exclusive if that page has additional references, for example, due
to a FOLL_GET.
do_swap_page() has to re-determine if a page is exclusive, which will
easily fail if there are other references on a page, most prominently GUP
references via FOLL_GET. This can currently result in memory corruptions
when taking a FOLL_GET | FOLL_WRITE reference on a page even when fork()
is never involved: try_to_unmap() will succeed, and when refaulting the
page, it cannot be marked exclusive and will get replaced by a copy in the
page tables on the next write access, resulting in writes via the GUP
reference to the page being lost.
In an ideal world, everybody that uses GUP and wants to modify page
content, such as O_DIRECT, would properly use FOLL_PIN. However, that
conversion will take a while. It's easier to fix what used to work in the
past (FOLL_GET | FOLL_WRITE) remembering PG_anon_exclusive. In addition,
by remembering PG_anon_exclusive we can further reduce unnecessary COW in
some cases, so it's the natural thing to do.
So let's transfer the PG_anon_exclusive information to the swap pte and
store it via an architecture-dependant pte bit; use that information when
restoring the swap pte in do_swap_page() and unuse_pte(). During fork(),
we simply have to clear the pte bit and are done.
Of course, there is one corner case to handle: swap backends that don't
support concurrent page modifications while the page is under writeback.
Special case these, and drop the exclusive marker. Add a comment why that
is just fine (also, reuse_swap_page() would have done the same in the
past).
In the future, we'll hopefully have all architectures support
__HAVE_ARCH_PTE_SWP_EXCLUSIVE, such that we can get rid of the empty stubs
and the define completely. Then, we can also convert
SWP_MIGRATION_READ_EXCLUSIVE. For architectures it's fairly easy to
support: either simply use a yet unused pte bit that can be used for swap
entries, steal one from the arch type bits if they exceed 5, or steal one
from the offset bits.
Note: R/O FOLL_GET references were never really reliable, especially when
taking one on a shared page and then writing to the page (e.g., GUP after
fork()). FOLL_GET, including R/W references, were never really reliable
once fork was involved (e.g., GUP before fork(), GUP during fork()). KSM
steps back in case it stumbles over unexpected references and is,
therefore, fine.
[david@redhat.com: fix SWP_STABLE_WRITES test]
Link: https://lkml.kernel.org/r/ac725bcb-313a-4fff-250a-68ba9a8f85fb@redhat.comLink: https://lkml.kernel.org/r/20220329164329.208407-1-david@redhat.com
Link: https://lkml.kernel.org/r/20220329164329.208407-2-david@redhat.com
Signed-off-by: David Hildenbrand <david@redhat.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Hugh Dickins <hughd@google.com>
Cc: Shakeel Butt <shakeelb@google.com>
Cc: John Hubbard <jhubbard@nvidia.com>
Cc: Jason Gunthorpe <jgg@nvidia.com>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Mike Rapoport <rppt@linux.ibm.com>
Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Matthew Wilcox (Oracle) <willy@infradead.org>
Cc: Jann Horn <jannh@google.com>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Nadav Amit <namit@vmware.com>
Cc: Rik van Riel <riel@surriel.com>
Cc: Roman Gushchin <guro@fb.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Peter Xu <peterx@redhat.com>
Cc: Don Dutile <ddutile@redhat.com>
Cc: Christoph Hellwig <hch@lst.de>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Jan Kara <jack@suse.cz>
Cc: Liang Zhang <zhangliang5@huawei.com>
Cc: Pedro Demarchi Gomes <pedrodemargomes@gmail.com>
Cc: Oded Gabbay <oded.gabbay@gmail.com>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Will Deacon <will@kernel.org>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Heiko Carstens <hca@linux.ibm.com>
Cc: Vasily Gorbik <gor@linux.ibm.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Gerald Schaefer <gerald.schaefer@linux.ibm.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2022-05-09 18:20:45 -07:00
|
|
|
|
|
|
|
|
|
2025-02-10 20:37:52 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
Revert "rmap: do not call mmu_notifier_invalidate_page() under ptl"
This reverts commit aac2fea94f7a3df8ad1eeb477eb2643f81fd5393.
It turns out that that patch was complete and utter garbage, and broke
KVM, resulting in odd oopses.
Quoting Andrea Arcangeli:
"The aforementioned commit has 3 bugs.
1) mmu_notifier_invalidate_range cannot be used in replacement of
mmu_notifier_invalidate_range_start/end.
For KVM mmu_notifier_invalidate_range is a noop and rightfully so.
A MMU notifier implementation has to implement either
->invalidate_range method or the invalidate_range_start/end
methods, not both. And if you implement invalidate_range_start/end
like KVM is forced to do, calling mmu_notifier_invalidate_range in
common code is a noop for KVM.
For those MMU notifiers that can get away only implementing
->invalidate_range, the ->invalidate_range is implicitly called by
mmu_notifier_invalidate_range_end(). And only those secondary MMUs
that share the same pagetable with the primary MMU (like AMD
iommuv2) can get away only implementing ->invalidate_range.
So all cases (THP on/off) are broken right now.
To fix this is enough to replace mmu_notifier_invalidate_range with
mmu_notifier_invalidate_range_start;mmu_notifier_invalidate_range_end.
Either that or call multiple mmu_notifier_invalidate_page like
before.
2) address + (1UL << compound_order(page) is buggy, it should be
PAGE_SIZE << compound_order(page), it's bytes not pages, 2M not
512.
3) The whole invalidate_range thing was an attempt to call a single
invalidate while walking multiple 4k ptes that maps the same THP
(after a pmd virtual split without physical compound page THP
split).
It's unclear if the rmap_walk will always provide an address that
is 2M aligned as parameter to try_to_unmap_one, in presence of THP.
I think it needs also an address &= (PAGE_SIZE <<
compound_order(page)) - 1 to be safe"
In general, we should stop making excuses for horrible MMU notifier
users. It's much more important that the core VM is sane and safe, than
letting MMU notifiers sleep.
So if some MMU notifier is sleeping under a spinlock, we need to fix the
notifier, not try to make excuses for that garbage in the core VM.
Reported-and-tested-by: Bernhard Held <berny156@gmx.de>
Reported-and-tested-by: Adam Borowski <kilobyte@angband.pl>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Radim Krčmář <rkrcmar@redhat.com>
Cc: Wanpeng Li <kernellwp@gmail.com>
Cc: Paolo Bonzini <pbonzini@redhat.com>
Cc: Takashi Iwai <tiwai@suse.de>
Cc: Nadav Amit <nadav.amit@gmail.com>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Jérôme Glisse <jglisse@redhat.com>
Cc: axie <axie@amd.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2017-08-29 09:11:06 -07:00
|
|
|
|
2017-11-15 17:34:07 -08:00
|
|
|
|
|
|
|
|
|
2022-02-15 09:28:49 -05:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-11-15 17:34:07 -08:00
|
|
|
|
2022-06-27 09:00:26 +03:00
|
|
|
|
2017-11-15 17:34:07 -08:00
|
|
|
|
2026-02-09 22:07:28 +08:00
|
|
|
|
2017-11-15 17:34:07 -08:00
|
|
|
|
mm: support madvise(MADV_FREE)
Linux doesn't have an ability to free pages lazy while other OS already
have been supported that named by madvise(MADV_FREE).
The gain is clear that kernel can discard freed pages rather than
swapping out or OOM if memory pressure happens.
Without memory pressure, freed pages would be reused by userspace
without another additional overhead(ex, page fault + allocation +
zeroing).
Jason Evans said:
: Facebook has been using MAP_UNINITIALIZED
: (https://lkml.org/lkml/2012/1/18/308) in some of its applications for
: several years, but there are operational costs to maintaining this
: out-of-tree in our kernel and in jemalloc, and we are anxious to retire it
: in favor of MADV_FREE. When we first enabled MAP_UNINITIALIZED it
: increased throughput for much of our workload by ~5%, and although the
: benefit has decreased using newer hardware and kernels, there is still
: enough benefit that we cannot reasonably retire it without a replacement.
:
: Aside from Facebook operations, there are numerous broadly used
: applications that would benefit from MADV_FREE. The ones that immediately
: come to mind are redis, varnish, and MariaDB. I don't have much insight
: into Android internals and development process, but I would hope to see
: MADV_FREE support eventually end up there as well to benefit applications
: linked with the integrated jemalloc.
:
: jemalloc will use MADV_FREE once it becomes available in the Linux kernel.
: In fact, jemalloc already uses MADV_FREE or equivalent everywhere it's
: available: *BSD, OS X, Windows, and Solaris -- every platform except Linux
: (and AIX, but I'm not sure it even compiles on AIX). The lack of
: MADV_FREE on Linux forced me down a long series of increasingly
: sophisticated heuristics for madvise() volume reduction, and even so this
: remains a common performance issue for people using jemalloc on Linux.
: Please integrate MADV_FREE; many people will benefit substantially.
How it works:
When madvise syscall is called, VM clears dirty bit of ptes of the
range. If memory pressure happens, VM checks dirty bit of page table
and if it found still "clean", it means it's a "lazyfree pages" so VM
could discard the page instead of swapping out. Once there was store
operation for the page before VM peek a page to reclaim, dirty bit is
set so VM can swap out the page instead of discarding.
One thing we should notice is that basically, MADV_FREE relies on dirty
bit in page table entry to decide whether VM allows to discard the page
or not. IOW, if page table entry includes marked dirty bit, VM
shouldn't discard the page.
However, as a example, if swap-in by read fault happens, page table
entry doesn't have dirty bit so MADV_FREE could discard the page
wrongly.
For avoiding the problem, MADV_FREE did more checks with PageDirty and
PageSwapCache. It worked out because swapped-in page lives on swap
cache and since it is evicted from the swap cache, the page has PG_dirty
flag. So both page flags check effectively prevent wrong discarding by
MADV_FREE.
However, a problem in above logic is that swapped-in page has PG_dirty
still after they are removed from swap cache so VM cannot consider the
page as freeable any more even if madvise_free is called in future.
Look at below example for detail.
ptr = malloc();
memset(ptr);
..
..
.. heavy memory pressure so all of pages are swapped out
..
..
var = *ptr; -> a page swapped-in and could be removed from
swapcache. Then, page table doesn't mark
dirty bit and page descriptor includes PG_dirty
..
..
madvise_free(ptr); -> It doesn't clear PG_dirty of the page.
..
..
..
.. heavy memory pressure again.
.. In this time, VM cannot discard the page because the page
.. has *PG_dirty*
To solve the problem, this patch clears PG_dirty if only the page is
owned exclusively by current process when madvise is called because
PG_dirty represents ptes's dirtiness in several processes so we could
clear it only if we own it exclusively.
Firstly, heavy users would be general allocators(ex, jemalloc, tcmalloc
and hope glibc supports it) and jemalloc/tcmalloc already have supported
the feature for other OS(ex, FreeBSD)
barrios@blaptop:~/benchmark/ebizzy$ lscpu
Architecture: x86_64
CPU op-mode(s): 32-bit, 64-bit
Byte Order: Little Endian
CPU(s): 12
On-line CPU(s) list: 0-11
Thread(s) per core: 1
Core(s) per socket: 1
Socket(s): 12
NUMA node(s): 1
Vendor ID: GenuineIntel
CPU family: 6
Model: 2
Stepping: 3
CPU MHz: 3200.185
BogoMIPS: 6400.53
Virtualization: VT-x
Hypervisor vendor: KVM
Virtualization type: full
L1d cache: 32K
L1i cache: 32K
L2 cache: 4096K
NUMA node0 CPU(s): 0-11
ebizzy benchmark(./ebizzy -S 10 -n 512)
Higher avg is better.
vanilla-jemalloc MADV_free-jemalloc
1 thread
records: 10 records: 10
avg: 2961.90 avg: 12069.70
std: 71.96(2.43%) std: 186.68(1.55%)
max: 3070.00 max: 12385.00
min: 2796.00 min: 11746.00
2 thread
records: 10 records: 10
avg: 5020.00 avg: 17827.00
std: 264.87(5.28%) std: 358.52(2.01%)
max: 5244.00 max: 18760.00
min: 4251.00 min: 17382.00
4 thread
records: 10 records: 10
avg: 8988.80 avg: 27930.80
std: 1175.33(13.08%) std: 3317.33(11.88%)
max: 9508.00 max: 30879.00
min: 5477.00 min: 21024.00
8 thread
records: 10 records: 10
avg: 13036.50 avg: 33739.40
std: 170.67(1.31%) std: 5146.22(15.25%)
max: 13371.00 max: 40572.00
min: 12785.00 min: 24088.00
16 thread
records: 10 records: 10
avg: 11092.40 avg: 31424.20
std: 710.60(6.41%) std: 3763.89(11.98%)
max: 12446.00 max: 36635.00
min: 9949.00 min: 25669.00
32 thread
records: 10 records: 10
avg: 11067.00 avg: 34495.80
std: 971.06(8.77%) std: 2721.36(7.89%)
max: 12010.00 max: 38598.00
min: 9002.00 min: 30636.00
In summary, MADV_FREE is about much faster than MADV_DONTNEED.
This patch (of 12):
Add core MADV_FREE implementation.
[akpm@linux-foundation.org: small cleanups]
Signed-off-by: Minchan Kim <minchan@kernel.org>
Acked-by: Michal Hocko <mhocko@suse.com>
Acked-by: Hugh Dickins <hughd@google.com>
Cc: Mika Penttil <mika.penttila@nextfour.com>
Cc: Michael Kerrisk <mtk.manpages@gmail.com>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Rik van Riel <riel@redhat.com>
Cc: Mel Gorman <mgorman@suse.de>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Jason Evans <je@fb.com>
Cc: Daniel Micay <danielmicay@gmail.com>
Cc: "Kirill A. Shutemov" <kirill@shutemov.name>
Cc: Shaohua Li <shli@kernel.org>
Cc: <yalin.wang2010@gmail.com>
Cc: Andy Lutomirski <luto@amacapital.net>
Cc: "James E.J. Bottomley" <jejb@parisc-linux.org>
Cc: "Kirill A. Shutemov" <kirill@shutemov.name>
Cc: "Shaohua Li" <shli@kernel.org>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Chen Gang <gang.chen.5i5j@gmail.com>
Cc: Chris Zankel <chris@zankel.net>
Cc: Darrick J. Wong <darrick.wong@oracle.com>
Cc: David S. Miller <davem@davemloft.net>
Cc: Helge Deller <deller@gmx.de>
Cc: Ivan Kokshaysky <ink@jurassic.park.msu.ru>
Cc: Matt Turner <mattst88@gmail.com>
Cc: Max Filippov <jcmvbkbc@gmail.com>
Cc: Ralf Baechle <ralf@linux-mips.org>
Cc: Richard Henderson <rth@twiddle.net>
Cc: Roland Dreier <roland@kernel.org>
Cc: Russell King <rmk@arm.linux.org.uk>
Cc: Shaohua Li <shli@kernel.org>
Cc: Will Deacon <will.deacon@arm.com>
Cc: Wu Fengguang <fengguang.wu@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2016-01-15 16:54:53 -08:00
|
|
|
|
mm: support batched unmap for lazyfree large folios during reclamation
Currently, the PTEs and rmap of a large folio are removed one at a time.
This is not only slow but also causes the large folio to be unnecessarily
added to deferred_split, which can lead to races between the
deferred_split shrinker callback and memory reclamation. This patch
releases all PTEs and rmap entries in a batch. Currently, it only handles
lazyfree large folios.
The below microbench tries to reclaim 128MB lazyfree large folios
whose sizes are 64KiB:
#include <stdio.h>
#include <sys/mman.h>
#include <string.h>
#include <time.h>
#define SIZE 128*1024*1024 // 128 MB
unsigned long read_split_deferred()
{
FILE *file = fopen("/sys/kernel/mm/transparent_hugepage"
"/hugepages-64kB/stats/split_deferred", "r");
if (!file) {
perror("Error opening file");
return 0;
}
unsigned long value;
if (fscanf(file, "%lu", &value) != 1) {
perror("Error reading value");
fclose(file);
return 0;
}
fclose(file);
return value;
}
int main(int argc, char *argv[])
{
while(1) {
volatile int *p = mmap(0, SIZE, PROT_READ | PROT_WRITE,
MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
memset((void *)p, 1, SIZE);
madvise((void *)p, SIZE, MADV_FREE);
clock_t start_time = clock();
unsigned long start_split = read_split_deferred();
madvise((void *)p, SIZE, MADV_PAGEOUT);
clock_t end_time = clock();
unsigned long end_split = read_split_deferred();
double elapsed_time = (double)(end_time - start_time) / CLOCKS_PER_SEC;
printf("Time taken by reclamation: %f seconds, split_deferred: %ld\n",
elapsed_time, end_split - start_split);
munmap((void *)p, SIZE);
}
return 0;
}
w/o patch:
~ # ./a.out
Time taken by reclamation: 0.177418 seconds, split_deferred: 2048
Time taken by reclamation: 0.178348 seconds, split_deferred: 2048
Time taken by reclamation: 0.174525 seconds, split_deferred: 2048
Time taken by reclamation: 0.171620 seconds, split_deferred: 2048
Time taken by reclamation: 0.172241 seconds, split_deferred: 2048
Time taken by reclamation: 0.174003 seconds, split_deferred: 2048
Time taken by reclamation: 0.171058 seconds, split_deferred: 2048
Time taken by reclamation: 0.171993 seconds, split_deferred: 2048
Time taken by reclamation: 0.169829 seconds, split_deferred: 2048
Time taken by reclamation: 0.172895 seconds, split_deferred: 2048
Time taken by reclamation: 0.176063 seconds, split_deferred: 2048
Time taken by reclamation: 0.172568 seconds, split_deferred: 2048
Time taken by reclamation: 0.171185 seconds, split_deferred: 2048
Time taken by reclamation: 0.170632 seconds, split_deferred: 2048
Time taken by reclamation: 0.170208 seconds, split_deferred: 2048
Time taken by reclamation: 0.174192 seconds, split_deferred: 2048
...
w/ patch:
~ # ./a.out
Time taken by reclamation: 0.074231 seconds, split_deferred: 0
Time taken by reclamation: 0.071026 seconds, split_deferred: 0
Time taken by reclamation: 0.072029 seconds, split_deferred: 0
Time taken by reclamation: 0.071873 seconds, split_deferred: 0
Time taken by reclamation: 0.073573 seconds, split_deferred: 0
Time taken by reclamation: 0.071906 seconds, split_deferred: 0
Time taken by reclamation: 0.073604 seconds, split_deferred: 0
Time taken by reclamation: 0.075903 seconds, split_deferred: 0
Time taken by reclamation: 0.073191 seconds, split_deferred: 0
Time taken by reclamation: 0.071228 seconds, split_deferred: 0
Time taken by reclamation: 0.071391 seconds, split_deferred: 0
Time taken by reclamation: 0.071468 seconds, split_deferred: 0
Time taken by reclamation: 0.071896 seconds, split_deferred: 0
Time taken by reclamation: 0.072508 seconds, split_deferred: 0
Time taken by reclamation: 0.071884 seconds, split_deferred: 0
Time taken by reclamation: 0.072433 seconds, split_deferred: 0
Time taken by reclamation: 0.071939 seconds, split_deferred: 0
...
Link: https://lkml.kernel.org/r/20250214093015.51024-4-21cnbao@gmail.com
Signed-off-by: Barry Song <v-songbaohua@oppo.com>
Cc: Albert Ou <aou@eecs.berkeley.edu>
Cc: Anshuman Khandual <anshuman.khandual@arm.com>
Cc: Baolin Wang <baolin.wang@linux.alibaba.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Chis Li <chrisl@kernel.org>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Gavin Shan <gshan@redhat.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: "Huang, Ying" <ying.huang@intel.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Kairui Song <kasong@tencent.com>
Cc: Kefeng Wang <wangkefeng.wang@huawei.com>
Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Lance Yang <ioworker0@gmail.com>
Cc: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Mauricio Faria de Oliveira <mfo@canonical.com>
Cc: Palmer Dabbelt <palmer@dabbelt.com>
Cc: Paul Walmsley <paul.walmsley@sifive.com>
Cc: Ryan Roberts <ryan.roberts@arm.com>
Cc: Shaoqin Huang <shahuang@redhat.com>
Cc: Tangquan Zheng <zhengtangquan@oppo.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Will Deacon <will@kernel.org>
Cc: Yicong Yang <yangyicong@hisilicon.com>
Cc: Yosry Ahmed <yosryahmed@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2025-02-14 22:30:14 +13:00
|
|
|
|
2023-12-20 23:44:26 +01:00
|
|
|
|
mm: support batched unmap for lazyfree large folios during reclamation
Currently, the PTEs and rmap of a large folio are removed one at a time.
This is not only slow but also causes the large folio to be unnecessarily
added to deferred_split, which can lead to races between the
deferred_split shrinker callback and memory reclamation. This patch
releases all PTEs and rmap entries in a batch. Currently, it only handles
lazyfree large folios.
The below microbench tries to reclaim 128MB lazyfree large folios
whose sizes are 64KiB:
#include <stdio.h>
#include <sys/mman.h>
#include <string.h>
#include <time.h>
#define SIZE 128*1024*1024 // 128 MB
unsigned long read_split_deferred()
{
FILE *file = fopen("/sys/kernel/mm/transparent_hugepage"
"/hugepages-64kB/stats/split_deferred", "r");
if (!file) {
perror("Error opening file");
return 0;
}
unsigned long value;
if (fscanf(file, "%lu", &value) != 1) {
perror("Error reading value");
fclose(file);
return 0;
}
fclose(file);
return value;
}
int main(int argc, char *argv[])
{
while(1) {
volatile int *p = mmap(0, SIZE, PROT_READ | PROT_WRITE,
MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
memset((void *)p, 1, SIZE);
madvise((void *)p, SIZE, MADV_FREE);
clock_t start_time = clock();
unsigned long start_split = read_split_deferred();
madvise((void *)p, SIZE, MADV_PAGEOUT);
clock_t end_time = clock();
unsigned long end_split = read_split_deferred();
double elapsed_time = (double)(end_time - start_time) / CLOCKS_PER_SEC;
printf("Time taken by reclamation: %f seconds, split_deferred: %ld\n",
elapsed_time, end_split - start_split);
munmap((void *)p, SIZE);
}
return 0;
}
w/o patch:
~ # ./a.out
Time taken by reclamation: 0.177418 seconds, split_deferred: 2048
Time taken by reclamation: 0.178348 seconds, split_deferred: 2048
Time taken by reclamation: 0.174525 seconds, split_deferred: 2048
Time taken by reclamation: 0.171620 seconds, split_deferred: 2048
Time taken by reclamation: 0.172241 seconds, split_deferred: 2048
Time taken by reclamation: 0.174003 seconds, split_deferred: 2048
Time taken by reclamation: 0.171058 seconds, split_deferred: 2048
Time taken by reclamation: 0.171993 seconds, split_deferred: 2048
Time taken by reclamation: 0.169829 seconds, split_deferred: 2048
Time taken by reclamation: 0.172895 seconds, split_deferred: 2048
Time taken by reclamation: 0.176063 seconds, split_deferred: 2048
Time taken by reclamation: 0.172568 seconds, split_deferred: 2048
Time taken by reclamation: 0.171185 seconds, split_deferred: 2048
Time taken by reclamation: 0.170632 seconds, split_deferred: 2048
Time taken by reclamation: 0.170208 seconds, split_deferred: 2048
Time taken by reclamation: 0.174192 seconds, split_deferred: 2048
...
w/ patch:
~ # ./a.out
Time taken by reclamation: 0.074231 seconds, split_deferred: 0
Time taken by reclamation: 0.071026 seconds, split_deferred: 0
Time taken by reclamation: 0.072029 seconds, split_deferred: 0
Time taken by reclamation: 0.071873 seconds, split_deferred: 0
Time taken by reclamation: 0.073573 seconds, split_deferred: 0
Time taken by reclamation: 0.071906 seconds, split_deferred: 0
Time taken by reclamation: 0.073604 seconds, split_deferred: 0
Time taken by reclamation: 0.075903 seconds, split_deferred: 0
Time taken by reclamation: 0.073191 seconds, split_deferred: 0
Time taken by reclamation: 0.071228 seconds, split_deferred: 0
Time taken by reclamation: 0.071391 seconds, split_deferred: 0
Time taken by reclamation: 0.071468 seconds, split_deferred: 0
Time taken by reclamation: 0.071896 seconds, split_deferred: 0
Time taken by reclamation: 0.072508 seconds, split_deferred: 0
Time taken by reclamation: 0.071884 seconds, split_deferred: 0
Time taken by reclamation: 0.072433 seconds, split_deferred: 0
Time taken by reclamation: 0.071939 seconds, split_deferred: 0
...
Link: https://lkml.kernel.org/r/20250214093015.51024-4-21cnbao@gmail.com
Signed-off-by: Barry Song <v-songbaohua@oppo.com>
Cc: Albert Ou <aou@eecs.berkeley.edu>
Cc: Anshuman Khandual <anshuman.khandual@arm.com>
Cc: Baolin Wang <baolin.wang@linux.alibaba.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Chis Li <chrisl@kernel.org>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Gavin Shan <gshan@redhat.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: "Huang, Ying" <ying.huang@intel.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Kairui Song <kasong@tencent.com>
Cc: Kefeng Wang <wangkefeng.wang@huawei.com>
Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Lance Yang <ioworker0@gmail.com>
Cc: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Mauricio Faria de Oliveira <mfo@canonical.com>
Cc: Palmer Dabbelt <palmer@dabbelt.com>
Cc: Paul Walmsley <paul.walmsley@sifive.com>
Cc: Ryan Roberts <ryan.roberts@arm.com>
Cc: Shaoqin Huang <shahuang@redhat.com>
Cc: Tangquan Zheng <zhengtangquan@oppo.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Will Deacon <will@kernel.org>
Cc: Yicong Yang <yangyicong@hisilicon.com>
Cc: Yosry Ahmed <yosryahmed@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2025-02-14 22:30:14 +13:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-02-14 18:38:47 -08:00
|
|
|
|
2023-01-12 12:39:31 +00:00
|
|
|
|
2025-06-27 14:23:19 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
mm: support batched unmap for lazyfree large folios during reclamation
Currently, the PTEs and rmap of a large folio are removed one at a time.
This is not only slow but also causes the large folio to be unnecessarily
added to deferred_split, which can lead to races between the
deferred_split shrinker callback and memory reclamation. This patch
releases all PTEs and rmap entries in a batch. Currently, it only handles
lazyfree large folios.
The below microbench tries to reclaim 128MB lazyfree large folios
whose sizes are 64KiB:
#include <stdio.h>
#include <sys/mman.h>
#include <string.h>
#include <time.h>
#define SIZE 128*1024*1024 // 128 MB
unsigned long read_split_deferred()
{
FILE *file = fopen("/sys/kernel/mm/transparent_hugepage"
"/hugepages-64kB/stats/split_deferred", "r");
if (!file) {
perror("Error opening file");
return 0;
}
unsigned long value;
if (fscanf(file, "%lu", &value) != 1) {
perror("Error reading value");
fclose(file);
return 0;
}
fclose(file);
return value;
}
int main(int argc, char *argv[])
{
while(1) {
volatile int *p = mmap(0, SIZE, PROT_READ | PROT_WRITE,
MAP_PRIVATE | MAP_ANONYMOUS, -1, 0);
memset((void *)p, 1, SIZE);
madvise((void *)p, SIZE, MADV_FREE);
clock_t start_time = clock();
unsigned long start_split = read_split_deferred();
madvise((void *)p, SIZE, MADV_PAGEOUT);
clock_t end_time = clock();
unsigned long end_split = read_split_deferred();
double elapsed_time = (double)(end_time - start_time) / CLOCKS_PER_SEC;
printf("Time taken by reclamation: %f seconds, split_deferred: %ld\n",
elapsed_time, end_split - start_split);
munmap((void *)p, SIZE);
}
return 0;
}
w/o patch:
~ # ./a.out
Time taken by reclamation: 0.177418 seconds, split_deferred: 2048
Time taken by reclamation: 0.178348 seconds, split_deferred: 2048
Time taken by reclamation: 0.174525 seconds, split_deferred: 2048
Time taken by reclamation: 0.171620 seconds, split_deferred: 2048
Time taken by reclamation: 0.172241 seconds, split_deferred: 2048
Time taken by reclamation: 0.174003 seconds, split_deferred: 2048
Time taken by reclamation: 0.171058 seconds, split_deferred: 2048
Time taken by reclamation: 0.171993 seconds, split_deferred: 2048
Time taken by reclamation: 0.169829 seconds, split_deferred: 2048
Time taken by reclamation: 0.172895 seconds, split_deferred: 2048
Time taken by reclamation: 0.176063 seconds, split_deferred: 2048
Time taken by reclamation: 0.172568 seconds, split_deferred: 2048
Time taken by reclamation: 0.171185 seconds, split_deferred: 2048
Time taken by reclamation: 0.170632 seconds, split_deferred: 2048
Time taken by reclamation: 0.170208 seconds, split_deferred: 2048
Time taken by reclamation: 0.174192 seconds, split_deferred: 2048
...
w/ patch:
~ # ./a.out
Time taken by reclamation: 0.074231 seconds, split_deferred: 0
Time taken by reclamation: 0.071026 seconds, split_deferred: 0
Time taken by reclamation: 0.072029 seconds, split_deferred: 0
Time taken by reclamation: 0.071873 seconds, split_deferred: 0
Time taken by reclamation: 0.073573 seconds, split_deferred: 0
Time taken by reclamation: 0.071906 seconds, split_deferred: 0
Time taken by reclamation: 0.073604 seconds, split_deferred: 0
Time taken by reclamation: 0.075903 seconds, split_deferred: 0
Time taken by reclamation: 0.073191 seconds, split_deferred: 0
Time taken by reclamation: 0.071228 seconds, split_deferred: 0
Time taken by reclamation: 0.071391 seconds, split_deferred: 0
Time taken by reclamation: 0.071468 seconds, split_deferred: 0
Time taken by reclamation: 0.071896 seconds, split_deferred: 0
Time taken by reclamation: 0.072508 seconds, split_deferred: 0
Time taken by reclamation: 0.071884 seconds, split_deferred: 0
Time taken by reclamation: 0.072433 seconds, split_deferred: 0
Time taken by reclamation: 0.071939 seconds, split_deferred: 0
...
Link: https://lkml.kernel.org/r/20250214093015.51024-4-21cnbao@gmail.com
Signed-off-by: Barry Song <v-songbaohua@oppo.com>
Cc: Albert Ou <aou@eecs.berkeley.edu>
Cc: Anshuman Khandual <anshuman.khandual@arm.com>
Cc: Baolin Wang <baolin.wang@linux.alibaba.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Chis Li <chrisl@kernel.org>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Gavin Shan <gshan@redhat.com>
Cc: "H. Peter Anvin" <hpa@zytor.com>
Cc: "Huang, Ying" <ying.huang@intel.com>
Cc: Ingo Molnar <mingo@redhat.com>
Cc: Kairui Song <kasong@tencent.com>
Cc: Kefeng Wang <wangkefeng.wang@huawei.com>
Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Lance Yang <ioworker0@gmail.com>
Cc: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Mauricio Faria de Oliveira <mfo@canonical.com>
Cc: Palmer Dabbelt <palmer@dabbelt.com>
Cc: Paul Walmsley <paul.walmsley@sifive.com>
Cc: Ryan Roberts <ryan.roberts@arm.com>
Cc: Shaoqin Huang <shahuang@redhat.com>
Cc: Tangquan Zheng <zhengtangquan@oppo.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Will Deacon <will@kernel.org>
Cc: Yicong Yang <yangyicong@hisilicon.com>
Cc: Yosry Ahmed <yosryahmed@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2025-02-14 22:30:14 +13:00
|
|
|
|
2024-06-14 09:51:36 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2017-02-24 14:58:01 -08:00
|
|
|
|
2017-08-31 17:17:27 -04:00
|
|
|
|
2018-12-28 00:38:09 -08:00
|
|
|
|
2017-08-31 17:17:27 -04:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2014-01-21 15:49:50 -08:00
|
|
|
|
|
|
|
|
|
2020-04-01 21:07:52 -07:00
|
|
|
|
2014-01-21 15:49:50 -08:00
|
|
|
|
|
|
|
|
|
2022-09-27 14:38:26 +08:00
|
|
|
|
2014-01-21 15:49:50 -08:00
|
|
|
|
2022-01-29 16:06:53 -05:00
|
|
|
|
2016-03-17 14:20:04 -07:00
|
|
|
|
2014-01-21 15:49:50 -08:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2022-02-15 09:28:49 -05:00
|
|
|
|
|
|
|
|
|
2009-09-16 11:50:10 +02:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
|
|
|
|
|
2022-02-15 09:28:49 -05:00
|
|
|
|
|
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2022-02-15 09:28:49 -05:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2022-02-15 09:28:49 -05:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2014-01-21 15:49:50 -08:00
|
|
|
|
|
|
|
|
|
2017-05-03 14:52:32 -07:00
|
|
|
|
2022-09-27 14:38:26 +08:00
|
|
|
|
2022-01-29 16:06:53 -05:00
|
|
|
|
2014-01-21 15:49:50 -08:00
|
|
|
|
2005-04-16 15:20:36 -07:00
|
|
|
|
2021-06-30 18:54:16 -07:00
|
|
|
|
2022-01-29 16:06:53 -05:00
|
|
|
|
2021-06-30 18:54:16 -07:00
|
|
|
|
2022-01-29 16:06:53 -05:00
|
|
|
|
2021-06-30 18:54:16 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-07-07 13:06:17 -07:00
|
|
|
|
2021-06-30 18:54:16 -07:00
|
|
|
|
2022-01-29 16:06:53 -05:00
|
|
|
|
2021-06-30 18:54:16 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-01-28 14:29:43 -05:00
|
|
|
|
2025-02-10 20:37:53 +01:00
|
|
|
|
2021-06-30 18:54:16 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
mm: ptep_get() conversion
Convert all instances of direct pte_t* dereferencing to instead use
ptep_get() helper. This means that by default, the accesses change from a
C dereference to a READ_ONCE(). This is technically the correct thing to
do since where pgtables are modified by HW (for access/dirty) they are
volatile and therefore we should always ensure READ_ONCE() semantics.
But more importantly, by always using the helper, it can be overridden by
the architecture to fully encapsulate the contents of the pte. Arch code
is deliberately not converted, as the arch code knows best. It is
intended that arch code (arm64) will override the default with its own
implementation that can (e.g.) hide certain bits from the core code, or
determine young/dirty status by mixing in state from another source.
Conversion was done using Coccinelle:
----
// $ make coccicheck \
// COCCI=ptepget.cocci \
// SPFLAGS="--include-headers" \
// MODE=patch
virtual patch
@ depends on patch @
pte_t *v;
@@
- *v
+ ptep_get(v)
----
Then reviewed and hand-edited to avoid multiple unnecessary calls to
ptep_get(), instead opting to store the result of a single call in a
variable, where it is correct to do so. This aims to negate any cost of
READ_ONCE() and will benefit arch-overrides that may be more complex.
Included is a fix for an issue in an earlier version of this patch that
was pointed out by kernel test robot. The issue arose because config
MMU=n elides definition of the ptep helper functions, including
ptep_get(). HUGETLB_PAGE=n configs still define a simple
huge_ptep_clear_flush() for linking purposes, which dereferences the ptep.
So when both configs are disabled, this caused a build error because
ptep_get() is not defined. Fix by continuing to do a direct dereference
when MMU=n. This is safe because for this config the arch code cannot be
trying to virtualize the ptes because none of the ptep helpers are
defined.
Link: https://lkml.kernel.org/r/20230612151545.3317766-4-ryan.roberts@arm.com
Reported-by: kernel test robot <lkp@intel.com>
Link: https://lore.kernel.org/oe-kbuild-all/202305120142.yXsNEo6H-lkp@intel.com/
Signed-off-by: Ryan Roberts <ryan.roberts@arm.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Alexander Potapenko <glider@google.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Alex Williamson <alex.williamson@redhat.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Andrey Konovalov <andreyknvl@gmail.com>
Cc: Andrey Ryabinin <ryabinin.a.a@gmail.com>
Cc: Christian Brauner <brauner@kernel.org>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: Daniel Vetter <daniel@ffwll.ch>
Cc: Dave Airlie <airlied@gmail.com>
Cc: Dimitri Sivanich <dimitri.sivanich@hpe.com>
Cc: Dmitry Vyukov <dvyukov@google.com>
Cc: Ian Rogers <irogers@google.com>
Cc: Jason Gunthorpe <jgg@ziepe.ca>
Cc: Jérôme Glisse <jglisse@redhat.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Lorenzo Stoakes <lstoakes@gmail.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Miaohe Lin <linmiaohe@huawei.com>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Mike Rapoport (IBM) <rppt@kernel.org>
Cc: Muchun Song <muchun.song@linux.dev>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Naoya Horiguchi <naoya.horiguchi@nec.com>
Cc: Oleksandr Tyshchenko <oleksandr_tyshchenko@epam.com>
Cc: Pavel Tatashin <pasha.tatashin@soleen.com>
Cc: Roman Gushchin <roman.gushchin@linux.dev>
Cc: SeongJae Park <sj@kernel.org>
Cc: Shakeel Butt <shakeelb@google.com>
Cc: Uladzislau Rezki (Sony) <urezki@gmail.com>
Cc: Vincenzo Frascino <vincenzo.frascino@arm.com>
Cc: Yu Zhao <yuzhao@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2023-06-12 16:15:45 +01:00
|
|
|
|
mm: hugetlb: add huge page size param to set_huge_pte_at()
Patch series "Fix set_huge_pte_at() panic on arm64", v2.
This series fixes a bug in arm64's implementation of set_huge_pte_at(),
which can result in an unprivileged user causing a kernel panic. The
problem was triggered when running the new uffd poison mm selftest for
HUGETLB memory. This test (and the uffd poison feature) was merged for
v6.5-rc7.
Ideally, I'd like to get this fix in for v6.6 and I've cc'ed stable
(correctly this time) to get it backported to v6.5, where the issue first
showed up.
Description of Bug
==================
arm64's huge pte implementation supports multiple huge page sizes, some of
which are implemented in the page table with multiple contiguous entries.
So set_huge_pte_at() needs to work out how big the logical pte is, so that
it can also work out how many physical ptes (or pmds) need to be written.
It previously did this by grabbing the folio out of the pte and querying
its size.
However, there are cases when the pte being set is actually a swap entry.
But this also used to work fine, because for huge ptes, we only ever saw
migration entries and hwpoison entries. And both of these types of swap
entries have a PFN embedded, so the code would grab that and everything
still worked out.
But over time, more calls to set_huge_pte_at() have been added that set
swap entry types that do not embed a PFN. And this causes the code to go
bang. The triggering case is for the uffd poison test, commit
99aa77215ad0 ("selftests/mm: add uffd unit test for UFFDIO_POISON"), which
causes a PTE_MARKER_POISONED swap entry to be set, coutesey of commit
8a13897fb0da ("mm: userfaultfd: support UFFDIO_POISON for hugetlbfs") -
added in v6.5-rc7. Although review shows that there are other call sites
that set PTE_MARKER_UFFD_WP (which also has no PFN), these don't trigger
on arm64 because arm64 doesn't support UFFD WP.
If CONFIG_DEBUG_VM is enabled, we do at least get a BUG(), but otherwise,
it will dereference a bad pointer in page_folio():
static inline struct folio *hugetlb_swap_entry_to_folio(swp_entry_t entry)
{
VM_BUG_ON(!is_migration_entry(entry) && !is_hwpoison_entry(entry));
return page_folio(pfn_to_page(swp_offset_pfn(entry)));
}
Fix
===
The simplest fix would have been to revert the dodgy cleanup commit
18f3962953e4 ("mm: hugetlb: kill set_huge_swap_pte_at()"), but since
things have moved on, this would have required an audit of all the new
set_huge_pte_at() call sites to see if they should be converted to
set_huge_swap_pte_at(). As per the original intent of the change, it
would also leave us open to future bugs when people invariably get it
wrong and call the wrong helper.
So instead, I've added a huge page size parameter to set_huge_pte_at().
This means that the arm64 code has the size in all cases. It's a bigger
change, due to needing to touch the arches that implement the function,
but it is entirely mechanical, so in my view, low risk.
I've compile-tested all touched arches; arm64, parisc, powerpc, riscv,
s390, sparc (and additionally x86_64). I've additionally booted and run
mm selftests against arm64, where I observe the uffd poison test is fixed,
and there are no other regressions.
This patch (of 2):
In order to fix a bug, arm64 needs to be told the size of the huge page
for which the pte is being set in set_huge_pte_at(). Provide for this by
adding an `unsigned long sz` parameter to the function. This follows the
same pattern as huge_pte_clear().
This commit makes the required interface modifications to the core mm as
well as all arches that implement this function (arm64, parisc, powerpc,
riscv, s390, sparc). The actual arm64 bug will be fixed in a separate
commit.
No behavioral changes intended.
Link: https://lkml.kernel.org/r/20230922115804.2043771-1-ryan.roberts@arm.com
Link: https://lkml.kernel.org/r/20230922115804.2043771-2-ryan.roberts@arm.com
Fixes: 8a13897fb0da ("mm: userfaultfd: support UFFDIO_POISON for hugetlbfs")
Signed-off-by: Ryan Roberts <ryan.roberts@arm.com>
Reviewed-by: Christophe Leroy <christophe.leroy@csgroup.eu> [powerpc 8xx]
Reviewed-by: Lorenzo Stoakes <lstoakes@gmail.com> [vmalloc change]
Cc: Alexandre Ghiti <alex@ghiti.fr>
Cc: Albert Ou <aou@eecs.berkeley.edu>
Cc: Alexander Gordeev <agordeev@linux.ibm.com>
Cc: Anshuman Khandual <anshuman.khandual@arm.com>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Axel Rasmussen <axelrasmussen@google.com>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Christian Borntraeger <borntraeger@linux.ibm.com>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: David S. Miller <davem@davemloft.net>
Cc: Gerald Schaefer <gerald.schaefer@linux.ibm.com>
Cc: Heiko Carstens <hca@linux.ibm.com>
Cc: Helge Deller <deller@gmx.de>
Cc: "James E.J. Bottomley" <James.Bottomley@HansenPartnership.com>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Muchun Song <muchun.song@linux.dev>
Cc: Nicholas Piggin <npiggin@gmail.com>
Cc: Palmer Dabbelt <palmer@dabbelt.com>
Cc: Paul Walmsley <paul.walmsley@sifive.com>
Cc: Peter Xu <peterx@redhat.com>
Cc: Qi Zheng <zhengqi.arch@bytedance.com>
Cc: Ryan Roberts <ryan.roberts@arm.com>
Cc: SeongJae Park <sj@kernel.org>
Cc: Sven Schnelle <svens@linux.ibm.com>
Cc: Uladzislau Rezki (Sony) <urezki@gmail.com>
Cc: Vasily Gorbik <gor@linux.ibm.com>
Cc: Will Deacon <will@kernel.org>
Cc: <stable@vger.kernel.org> [6.5+]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2023-09-22 12:58:03 +01:00
|
|
|
|
2021-06-30 18:54:16 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-12-20 23:44:54 +01:00
|
|
|
|
2026-04-27 13:43:16 +02:00
|
|
|
|
2021-06-30 18:54:16 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-02-03 11:40:17 -05:00
|
|
|
|
2023-01-10 13:57:22 +11:00
|
|
|
|
2021-06-30 18:54:16 -07:00
|
|
|
|
2022-01-28 14:29:43 -05:00
|
|
|
|
2021-06-30 18:54:16 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
mm: hugetlb: add huge page size param to set_huge_pte_at()
Patch series "Fix set_huge_pte_at() panic on arm64", v2.
This series fixes a bug in arm64's implementation of set_huge_pte_at(),
which can result in an unprivileged user causing a kernel panic. The
problem was triggered when running the new uffd poison mm selftest for
HUGETLB memory. This test (and the uffd poison feature) was merged for
v6.5-rc7.
Ideally, I'd like to get this fix in for v6.6 and I've cc'ed stable
(correctly this time) to get it backported to v6.5, where the issue first
showed up.
Description of Bug
==================
arm64's huge pte implementation supports multiple huge page sizes, some of
which are implemented in the page table with multiple contiguous entries.
So set_huge_pte_at() needs to work out how big the logical pte is, so that
it can also work out how many physical ptes (or pmds) need to be written.
It previously did this by grabbing the folio out of the pte and querying
its size.
However, there are cases when the pte being set is actually a swap entry.
But this also used to work fine, because for huge ptes, we only ever saw
migration entries and hwpoison entries. And both of these types of swap
entries have a PFN embedded, so the code would grab that and everything
still worked out.
But over time, more calls to set_huge_pte_at() have been added that set
swap entry types that do not embed a PFN. And this causes the code to go
bang. The triggering case is for the uffd poison test, commit
99aa77215ad0 ("selftests/mm: add uffd unit test for UFFDIO_POISON"), which
causes a PTE_MARKER_POISONED swap entry to be set, coutesey of commit
8a13897fb0da ("mm: userfaultfd: support UFFDIO_POISON for hugetlbfs") -
added in v6.5-rc7. Although review shows that there are other call sites
that set PTE_MARKER_UFFD_WP (which also has no PFN), these don't trigger
on arm64 because arm64 doesn't support UFFD WP.
If CONFIG_DEBUG_VM is enabled, we do at least get a BUG(), but otherwise,
it will dereference a bad pointer in page_folio():
static inline struct folio *hugetlb_swap_entry_to_folio(swp_entry_t entry)
{
VM_BUG_ON(!is_migration_entry(entry) && !is_hwpoison_entry(entry));
return page_folio(pfn_to_page(swp_offset_pfn(entry)));
}
Fix
===
The simplest fix would have been to revert the dodgy cleanup commit
18f3962953e4 ("mm: hugetlb: kill set_huge_swap_pte_at()"), but since
things have moved on, this would have required an audit of all the new
set_huge_pte_at() call sites to see if they should be converted to
set_huge_swap_pte_at(). As per the original intent of the change, it
would also leave us open to future bugs when people invariably get it
wrong and call the wrong helper.
So instead, I've added a huge page size parameter to set_huge_pte_at().
This means that the arm64 code has the size in all cases. It's a bigger
change, due to needing to touch the arches that implement the function,
but it is entirely mechanical, so in my view, low risk.
I've compile-tested all touched arches; arm64, parisc, powerpc, riscv,
s390, sparc (and additionally x86_64). I've additionally booted and run
mm selftests against arm64, where I observe the uffd poison test is fixed,
and there are no other regressions.
This patch (of 2):
In order to fix a bug, arm64 needs to be told the size of the huge page
for which the pte is being set in set_huge_pte_at(). Provide for this by
adding an `unsigned long sz` parameter to the function. This follows the
same pattern as huge_pte_clear().
This commit makes the required interface modifications to the core mm as
well as all arches that implement this function (arm64, parisc, powerpc,
riscv, s390, sparc). The actual arm64 bug will be fixed in a separate
commit.
No behavioral changes intended.
Link: https://lkml.kernel.org/r/20230922115804.2043771-1-ryan.roberts@arm.com
Link: https://lkml.kernel.org/r/20230922115804.2043771-2-ryan.roberts@arm.com
Fixes: 8a13897fb0da ("mm: userfaultfd: support UFFDIO_POISON for hugetlbfs")
Signed-off-by: Ryan Roberts <ryan.roberts@arm.com>
Reviewed-by: Christophe Leroy <christophe.leroy@csgroup.eu> [powerpc 8xx]
Reviewed-by: Lorenzo Stoakes <lstoakes@gmail.com> [vmalloc change]
Cc: Alexandre Ghiti <alex@ghiti.fr>
Cc: Albert Ou <aou@eecs.berkeley.edu>
Cc: Alexander Gordeev <agordeev@linux.ibm.com>
Cc: Anshuman Khandual <anshuman.khandual@arm.com>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Axel Rasmussen <axelrasmussen@google.com>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Christian Borntraeger <borntraeger@linux.ibm.com>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: David S. Miller <davem@davemloft.net>
Cc: Gerald Schaefer <gerald.schaefer@linux.ibm.com>
Cc: Heiko Carstens <hca@linux.ibm.com>
Cc: Helge Deller <deller@gmx.de>
Cc: "James E.J. Bottomley" <James.Bottomley@HansenPartnership.com>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Muchun Song <muchun.song@linux.dev>
Cc: Nicholas Piggin <npiggin@gmail.com>
Cc: Palmer Dabbelt <palmer@dabbelt.com>
Cc: Paul Walmsley <paul.walmsley@sifive.com>
Cc: Peter Xu <peterx@redhat.com>
Cc: Qi Zheng <zhengqi.arch@bytedance.com>
Cc: Ryan Roberts <ryan.roberts@arm.com>
Cc: SeongJae Park <sj@kernel.org>
Cc: Sven Schnelle <svens@linux.ibm.com>
Cc: Uladzislau Rezki (Sony) <urezki@gmail.com>
Cc: Vasily Gorbik <gor@linux.ibm.com>
Cc: Will Deacon <will@kernel.org>
Cc: <stable@vger.kernel.org> [6.5+]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2023-09-22 12:58:03 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-06-30 18:54:16 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-10-01 16:56:55 +10:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-04-25 18:38:58 +08:00
|
|
|
|
mm/huge_memory: fix early failure try_to_migrate() when split huge pmd for shared THP
Commit 60fbb14396d5 ("mm/huge_memory: adjust try_to_migrate_one() and
split_huge_pmd_locked()") return false unconditionally after
split_huge_pmd_locked(). This may fail try_to_migrate() early when
TTU_SPLIT_HUGE_PMD is specified.
The reason is the above commit adjusted try_to_migrate_one() to, when a
PMD-mapped THP entry is found, and TTU_SPLIT_HUGE_PMD is specified (for
example, via unmap_folio()), return false unconditionally. This breaks
the rmap walk and fail try_to_migrate() early, if this PMD-mapped THP is
mapped in multiple processes.
The user sensible impact of this bug could be:
* On memory pressure, shrink_folio_list() may split partially mapped
folio with split_folio_to_list(). Then free unmapped pages without IO.
If failed, it may not be reclaimed.
* On memory failure, memory_failure() would call try_to_split_thp_page()
to split folio contains the bad page. If succeed, the PG_has_hwpoisoned
bit is only set in the after-split folio contains @split_at. By doing
so, we limit bad memory. If failed to split, the whole folios is not
usable.
One way to reproduce:
Create an anonymous THP range and fork 512 children, so we have a
THP shared mapped in 513 processes. Then trigger folio split with
/sys/kernel/debug/split_huge_pages debugfs to split the THP folio to
order 0.
Without the above commit, we can successfully split to order 0. With the
above commit, the folio is still a large folio.
And currently there are two core users of TTU_SPLIT_HUGE_PMD:
* try_to_unmap_one()
* try_to_migrate_one()
try_to_unmap_one() would restart the rmap walk, so only
try_to_migrate_one() is affected.
We can't simply revert commit 60fbb14396d5 ("mm/huge_memory: adjust
try_to_migrate_one() and split_huge_pmd_locked()"), since it removed some
duplicated check covered by page_vma_mapped_walk().
This patch fixes this by restart page_vma_mapped_walk() after
split_huge_pmd_locked(). Since we cannot simply return "true" to fix the
problem, as that would affect another case:
When invoking folio_try_share_anon_rmap_pmd() from
split_huge_pmd_locked(), the latter can fail and leave a large folio
mapped through PTEs, in which case we ought to return true from
try_to_migrate_one(). This might result in unnecessary walking of the
rmap but is relatively harmless.
Link: https://lkml.kernel.org/r/20260305015006.27343-1-richard.weiyang@gmail.com
Fixes: 60fbb14396d5 ("mm/huge_memory: adjust try_to_migrate_one() and split_huge_pmd_locked()")
Signed-off-by: Wei Yang <richard.weiyang@gmail.com>
Reviewed-by: Baolin Wang <baolin.wang@linux.alibaba.com>
Reviewed-by: Zi Yan <ziy@nvidia.com>
Tested-by: Lance Yang <lance.yang@linux.dev>
Reviewed-by: Lance Yang <lance.yang@linux.dev>
Reviewed-by: Gavin Guo <gavinguo@igalia.com>
Acked-by: David Hildenbrand (arm) <david@kernel.org>
Reviewed-by: Lorenzo Stoakes (Oracle) <ljs@kernel.org>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2026-03-05 01:50:06 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-04-25 18:38:58 +08:00
|
|
|
|
2025-04-25 18:38:59 +08:00
|
|
|
|
mm/huge_memory: fix early failure try_to_migrate() when split huge pmd for shared THP
Commit 60fbb14396d5 ("mm/huge_memory: adjust try_to_migrate_one() and
split_huge_pmd_locked()") return false unconditionally after
split_huge_pmd_locked(). This may fail try_to_migrate() early when
TTU_SPLIT_HUGE_PMD is specified.
The reason is the above commit adjusted try_to_migrate_one() to, when a
PMD-mapped THP entry is found, and TTU_SPLIT_HUGE_PMD is specified (for
example, via unmap_folio()), return false unconditionally. This breaks
the rmap walk and fail try_to_migrate() early, if this PMD-mapped THP is
mapped in multiple processes.
The user sensible impact of this bug could be:
* On memory pressure, shrink_folio_list() may split partially mapped
folio with split_folio_to_list(). Then free unmapped pages without IO.
If failed, it may not be reclaimed.
* On memory failure, memory_failure() would call try_to_split_thp_page()
to split folio contains the bad page. If succeed, the PG_has_hwpoisoned
bit is only set in the after-split folio contains @split_at. By doing
so, we limit bad memory. If failed to split, the whole folios is not
usable.
One way to reproduce:
Create an anonymous THP range and fork 512 children, so we have a
THP shared mapped in 513 processes. Then trigger folio split with
/sys/kernel/debug/split_huge_pages debugfs to split the THP folio to
order 0.
Without the above commit, we can successfully split to order 0. With the
above commit, the folio is still a large folio.
And currently there are two core users of TTU_SPLIT_HUGE_PMD:
* try_to_unmap_one()
* try_to_migrate_one()
try_to_unmap_one() would restart the rmap walk, so only
try_to_migrate_one() is affected.
We can't simply revert commit 60fbb14396d5 ("mm/huge_memory: adjust
try_to_migrate_one() and split_huge_pmd_locked()"), since it removed some
duplicated check covered by page_vma_mapped_walk().
This patch fixes this by restart page_vma_mapped_walk() after
split_huge_pmd_locked(). Since we cannot simply return "true" to fix the
problem, as that would affect another case:
When invoking folio_try_share_anon_rmap_pmd() from
split_huge_pmd_locked(), the latter can fail and leave a large folio
mapped through PTEs, in which case we ought to return true from
try_to_migrate_one(). This might result in unnecessary walking of the
rmap but is relatively harmless.
Link: https://lkml.kernel.org/r/20260305015006.27343-1-richard.weiyang@gmail.com
Fixes: 60fbb14396d5 ("mm/huge_memory: adjust try_to_migrate_one() and split_huge_pmd_locked()")
Signed-off-by: Wei Yang <richard.weiyang@gmail.com>
Reviewed-by: Baolin Wang <baolin.wang@linux.alibaba.com>
Reviewed-by: Zi Yan <ziy@nvidia.com>
Tested-by: Lance Yang <lance.yang@linux.dev>
Reviewed-by: Lance Yang <lance.yang@linux.dev>
Reviewed-by: Gavin Guo <gavinguo@igalia.com>
Acked-by: David Hildenbrand (arm) <david@kernel.org>
Reviewed-by: Lorenzo Stoakes (Oracle) <ljs@kernel.org>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2026-03-05 01:50:06 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-04-25 18:38:58 +08:00
|
|
|
|
|
|
|
|
|
2025-10-01 16:56:55 +10:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-11-10 22:21:28 +00:00
|
|
|
|
2025-10-01 16:56:55 +10:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-01-28 14:29:43 -05:00
|
|
|
|
|
|
|
|
|
2021-06-30 18:54:16 -07:00
|
|
|
|
2022-05-09 18:20:44 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-06-30 18:54:16 -07:00
|
|
|
|
|
|
|
|
|
2025-04-25 18:38:58 +08:00
|
|
|
|
2021-06-30 18:54:16 -07:00
|
|
|
|
|
|
|
|
|
2022-01-28 14:29:43 -05:00
|
|
|
|
2021-06-30 18:54:16 -07:00
|
|
|
|
2025-02-10 20:37:53 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-06-23 22:53:32 +02:00
|
|
|
|
2025-11-10 22:21:34 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-02-10 20:37:53 +01:00
|
|
|
|
2022-06-23 22:53:32 +02:00
|
|
|
|
2025-02-10 20:37:53 +01:00
|
|
|
|
|
|
|
|
|
2021-06-30 18:54:16 -07:00
|
|
|
|
mm: remember exclusively mapped anonymous pages with PG_anon_exclusive
Let's mark exclusively mapped anonymous pages with PG_anon_exclusive as
exclusive, and use that information to make GUP pins reliable and stay
consistent with the page mapped into the page table even if the page table
entry gets write-protected.
With that information at hand, we can extend our COW logic to always reuse
anonymous pages that are exclusive. For anonymous pages that might be
shared, the existing logic applies.
As already documented, PG_anon_exclusive is usually only expressive in
combination with a page table entry. Especially PTE vs. PMD-mapped
anonymous pages require more thought, some examples: due to mremap() we
can easily have a single compound page PTE-mapped into multiple page
tables exclusively in a single process -- multiple page table locks apply.
Further, due to MADV_WIPEONFORK we might not necessarily write-protect
all PTEs, and only some subpages might be pinned. Long story short: once
PTE-mapped, we have to track information about exclusivity per sub-page,
but until then, we can just track it for the compound page in the head
page and not having to update a whole bunch of subpages all of the time
for a simple PMD mapping of a THP.
For simplicity, this commit mostly talks about "anonymous pages", while
it's for THP actually "the part of an anonymous folio referenced via a
page table entry".
To not spill PG_anon_exclusive code all over the mm code-base, we let the
anon rmap code to handle all PG_anon_exclusive logic it can easily handle.
If a writable, present page table entry points at an anonymous (sub)page,
that (sub)page must be PG_anon_exclusive. If GUP wants to take a reliably
pin (FOLL_PIN) on an anonymous page references via a present page table
entry, it must only pin if PG_anon_exclusive is set for the mapped
(sub)page.
This commit doesn't adjust GUP, so this is only implicitly handled for
FOLL_WRITE, follow-up commits will teach GUP to also respect it for
FOLL_PIN without FOLL_WRITE, to make all GUP pins of anonymous pages fully
reliable.
Whenever an anonymous page is to be shared (fork(), KSM), or when
temporarily unmapping an anonymous page (swap, migration), the relevant
PG_anon_exclusive bit has to be cleared to mark the anonymous page
possibly shared. Clearing will fail if there are GUP pins on the page:
* For fork(), this means having to copy the page and not being able to
share it. fork() protects against concurrent GUP using the PT lock and
the src_mm->write_protect_seq.
* For KSM, this means sharing will fail. For swap this means, unmapping
will fail, For migration this means, migration will fail early. All
three cases protect against concurrent GUP using the PT lock and a
proper clear/invalidate+flush of the relevant page table entry.
This fixes memory corruptions reported for FOLL_PIN | FOLL_WRITE, when a
pinned page gets mapped R/O and the successive write fault ends up
replacing the page instead of reusing it. It improves the situation for
O_DIRECT/vmsplice/... that still use FOLL_GET instead of FOLL_PIN, if
fork() is *not* involved, however swapout and fork() are still
problematic. Properly using FOLL_PIN instead of FOLL_GET for these GUP
users will fix the issue for them.
I. Details about basic handling
I.1. Fresh anonymous pages
page_add_new_anon_rmap() and hugepage_add_new_anon_rmap() will mark the
given page exclusive via __page_set_anon_rmap(exclusive=1). As that is
the mechanism fresh anonymous pages come into life (besides migration code
where we copy the page->mapping), all fresh anonymous pages will start out
as exclusive.
I.2. COW reuse handling of anonymous pages
When a COW handler stumbles over a (sub)page that's marked exclusive, it
simply reuses it. Otherwise, the handler tries harder under page lock to
detect if the (sub)page is exclusive and can be reused. If exclusive,
page_move_anon_rmap() will mark the given (sub)page exclusive.
Note that hugetlb code does not yet check for PageAnonExclusive(), as it
still uses the old COW logic that is prone to the COW security issue
because hugetlb code cannot really tolerate unnecessary/wrong COW as huge
pages are a scarce resource.
I.3. Migration handling
try_to_migrate() has to try marking an exclusive anonymous page shared via
page_try_share_anon_rmap(). If it fails because there are GUP pins on the
page, unmap fails. migrate_vma_collect_pmd() and
__split_huge_pmd_locked() are handled similarly.
Writable migration entries implicitly point at shared anonymous pages.
For readable migration entries that information is stored via a new
"readable-exclusive" migration entry, specific to anonymous pages.
When restoring a migration entry in remove_migration_pte(), information
about exlusivity is detected via the migration entry type, and
RMAP_EXCLUSIVE is set accordingly for
page_add_anon_rmap()/hugepage_add_anon_rmap() to restore that information.
I.4. Swapout handling
try_to_unmap() has to try marking the mapped page possibly shared via
page_try_share_anon_rmap(). If it fails because there are GUP pins on the
page, unmap fails. For now, information about exclusivity is lost. In
the future, we might want to remember that information in the swap entry
in some cases, however, it requires more thought, care, and a way to store
that information in swap entries.
I.5. Swapin handling
do_swap_page() will never stumble over exclusive anonymous pages in the
swap cache, as try_to_migrate() prohibits that. do_swap_page() always has
to detect manually if an anonymous page is exclusive and has to set
RMAP_EXCLUSIVE for page_add_anon_rmap() accordingly.
I.6. THP handling
__split_huge_pmd_locked() has to move the information about exclusivity
from the PMD to the PTEs.
a) In case we have a readable-exclusive PMD migration entry, simply
insert readable-exclusive PTE migration entries.
b) In case we have a present PMD entry and we don't want to freeze
("convert to migration entries"), simply forward PG_anon_exclusive to
all sub-pages, no need to temporarily clear the bit.
c) In case we have a present PMD entry and want to freeze, handle it
similar to try_to_migrate(): try marking the page shared first. In
case we fail, we ignore the "freeze" instruction and simply split
ordinarily. try_to_migrate() will properly fail because the THP is
still mapped via PTEs.
When splitting a compound anonymous folio (THP), the information about
exclusivity is implicitly handled via the migration entries: no need to
replicate PG_anon_exclusive manually.
I.7. fork() handling fork() handling is relatively easy, because
PG_anon_exclusive is only expressive for some page table entry types.
a) Present anonymous pages
page_try_dup_anon_rmap() will mark the given subpage shared -- which will
fail if the page is pinned. If it failed, we have to copy (or PTE-map a
PMD to handle it on the PTE level).
Note that device exclusive entries are just a pointer at a PageAnon()
page. fork() will first convert a device exclusive entry to a present
page table and handle it just like present anonymous pages.
b) Device private entry
Device private entries point at PageAnon() pages that cannot be mapped
directly and, therefore, cannot get pinned.
page_try_dup_anon_rmap() will mark the given subpage shared, which cannot
fail because they cannot get pinned.
c) HW poison entries
PG_anon_exclusive will remain untouched and is stale -- the page table
entry is just a placeholder after all.
d) Migration entries
Writable and readable-exclusive entries are converted to readable entries:
possibly shared.
I.8. mprotect() handling
mprotect() only has to properly handle the new readable-exclusive
migration entry:
When write-protecting a migration entry that points at an anonymous page,
remember the information about exclusivity via the "readable-exclusive"
migration entry type.
II. Migration and GUP-fast
Whenever replacing a present page table entry that maps an exclusive
anonymous page by a migration entry, we have to mark the page possibly
shared and synchronize against GUP-fast by a proper clear/invalidate+flush
to make the following scenario impossible:
1. try_to_migrate() places a migration entry after checking for GUP pins
and marks the page possibly shared.
2. GUP-fast pins the page due to lack of synchronization
3. fork() converts the "writable/readable-exclusive" migration entry into a
readable migration entry
4. Migration fails due to the GUP pin (failing to freeze the refcount)
5. Migration entries are restored. PG_anon_exclusive is lost
-> We have a pinned page that is not marked exclusive anymore.
Note that we move information about exclusivity from the page to the
migration entry as it otherwise highly overcomplicates fork() and
PTE-mapping a THP.
III. Swapout and GUP-fast
Whenever replacing a present page table entry that maps an exclusive
anonymous page by a swap entry, we have to mark the page possibly shared
and synchronize against GUP-fast by a proper clear/invalidate+flush to
make the following scenario impossible:
1. try_to_unmap() places a swap entry after checking for GUP pins and
clears exclusivity information on the page.
2. GUP-fast pins the page due to lack of synchronization.
-> We have a pinned page that is not marked exclusive anymore.
If we'd ever store information about exclusivity in the swap entry,
similar to migration handling, the same considerations as in II would
apply. This is future work.
Link: https://lkml.kernel.org/r/20220428083441.37290-13-david@redhat.com
Signed-off-by: David Hildenbrand <david@redhat.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Christoph Hellwig <hch@lst.de>
Cc: David Rientjes <rientjes@google.com>
Cc: Don Dutile <ddutile@redhat.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Jan Kara <jack@suse.cz>
Cc: Jann Horn <jannh@google.com>
Cc: Jason Gunthorpe <jgg@nvidia.com>
Cc: John Hubbard <jhubbard@nvidia.com>
Cc: Khalid Aziz <khalid.aziz@oracle.com>
Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Liang Zhang <zhangliang5@huawei.com>
Cc: "Matthew Wilcox (Oracle)" <willy@infradead.org>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Mike Rapoport <rppt@linux.ibm.com>
Cc: Nadav Amit <namit@vmware.com>
Cc: Oded Gabbay <oded.gabbay@gmail.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Pedro Demarchi Gomes <pedrodemargomes@gmail.com>
Cc: Peter Xu <peterx@redhat.com>
Cc: Rik van Riel <riel@surriel.com>
Cc: Roman Gushchin <guro@fb.com>
Cc: Shakeel Butt <shakeelb@google.com>
Cc: Yang Shi <shy828301@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2022-05-09 18:20:44 -07:00
|
|
|
|
|
|
|
|
|
2021-06-30 18:54:16 -07:00
|
|
|
|
2022-05-09 18:20:53 -07:00
|
|
|
|
2022-06-20 19:47:15 +08:00
|
|
|
|
|
|
|
|
|
2022-05-09 18:20:53 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-06-20 19:47:15 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-09-14 15:18:09 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-06-20 19:47:15 +08:00
|
|
|
|
2022-09-14 15:18:09 -07:00
|
|
|
|
mm/hugetlb: fix excessive IPI broadcasts when unsharing PMD tables using mmu_gather
As reported, ever since commit 1013af4f585f ("mm/hugetlb: fix
huge_pmd_unshare() vs GUP-fast race") we can end up in some situations
where we perform so many IPI broadcasts when unsharing hugetlb PMD page
tables that it severely regresses some workloads.
In particular, when we fork()+exit(), or when we munmap() a large
area backed by many shared PMD tables, we perform one IPI broadcast per
unshared PMD table.
There are two optimizations to be had:
(1) When we process (unshare) multiple such PMD tables, such as during
exit(), it is sufficient to send a single IPI broadcast (as long as
we respect locking rules) instead of one per PMD table.
Locking prevents that any of these PMD tables could get reused before
we drop the lock.
(2) When we are not the last sharer (> 2 users including us), there is
no need to send the IPI broadcast. The shared PMD tables cannot
become exclusive (fully unshared) before an IPI will be broadcasted
by the last sharer.
Concurrent GUP-fast could walk into a PMD table just before we
unshared it. It could then succeed in grabbing a page from the
shared page table even after munmap() etc succeeded (and supressed
an IPI). But there is not difference compared to GUP-fast just
sleeping for a while after grabbing the page and re-enabling IRQs.
Most importantly, GUP-fast will never walk into page tables that are
no-longer shared, because the last sharer will issue an IPI
broadcast.
(if ever required, checking whether the PUD changed in GUP-fast
after grabbing the page like we do in the PTE case could handle
this)
So let's rework PMD sharing TLB flushing + IPI sync to use the mmu_gather
infrastructure so we can implement these optimizations and demystify the
code at least a bit. Extend the mmu_gather infrastructure to be able to
deal with our special hugetlb PMD table sharing implementation.
To make initialization of the mmu_gather easier when working on a single
VMA (in particular, when dealing with hugetlb), provide
tlb_gather_mmu_vma().
We'll consolidate the handling for (full) unsharing of PMD tables in
tlb_unshare_pmd_ptdesc() and tlb_flush_unshared_tables(), and track
in "struct mmu_gather" whether we had (full) unsharing of PMD tables.
Because locking is very special (concurrent unsharing+reuse must be
prevented), we disallow deferring flushing to tlb_finish_mmu() and instead
require an explicit earlier call to tlb_flush_unshared_tables().
From hugetlb code, we call huge_pmd_unshare_flush() where we make sure
that the expected lock protecting us from concurrent unsharing+reuse is
still held.
Check with a VM_WARN_ON_ONCE() in tlb_finish_mmu() that
tlb_flush_unshared_tables() was properly called earlier.
Document it all properly.
Notes about tlb_remove_table_sync_one() interaction with unsharing:
There are two fairly tricky things:
(1) tlb_remove_table_sync_one() is a NOP on architectures without
CONFIG_MMU_GATHER_RCU_TABLE_FREE.
Here, the assumption is that the previous TLB flush would send an
IPI to all relevant CPUs. Careful: some architectures like x86 only
send IPIs to all relevant CPUs when tlb->freed_tables is set.
The relevant architectures should be selecting
MMU_GATHER_RCU_TABLE_FREE, but x86 might not do that in stable
kernels and it might have been problematic before this patch.
Also, the arch flushing behavior (independent of IPIs) is different
when tlb->freed_tables is set. Do we have to enlighten them to also
take care of tlb->unshared_tables? So far we didn't care, so
hopefully we are fine. Of course, we could be setting
tlb->freed_tables as well, but that might then unnecessarily flush
too much, because the semantics of tlb->freed_tables are a bit
fuzzy.
This patch changes nothing in this regard.
(2) tlb_remove_table_sync_one() is not a NOP on architectures with
CONFIG_MMU_GATHER_RCU_TABLE_FREE that actually don't need a sync.
Take x86 as an example: in the common case (!pv, !X86_FEATURE_INVLPGB)
we still issue IPIs during TLB flushes and don't actually need the
second tlb_remove_table_sync_one().
This optimized can be implemented on top of this, by checking e.g., in
tlb_remove_table_sync_one() whether we really need IPIs. But as
described in (1), it really must honor tlb->freed_tables then to
send IPIs to all relevant CPUs.
Notes on TLB flushing changes:
(1) Flushing for non-shared PMD tables
We're converting from flush_hugetlb_tlb_range() to
tlb_remove_huge_tlb_entry(). Given that we properly initialize the
MMU gather in tlb_gather_mmu_vma() to be hugetlb aware, similar to
__unmap_hugepage_range(), that should be fine.
(2) Flushing for shared PMD tables
We're converting from various things (flush_hugetlb_tlb_range(),
tlb_flush_pmd_range(), flush_tlb_range()) to tlb_flush_pmd_range().
tlb_flush_pmd_range() achieves the same that
tlb_remove_huge_tlb_entry() would achieve in these scenarios.
Note that tlb_remove_huge_tlb_entry() also calls
__tlb_remove_tlb_entry(), however that is only implemented on
powerpc, which does not support PMD table sharing.
Similar to (1), tlb_gather_mmu_vma() should make sure that TLB
flushing keeps on working as expected.
Further, note that the ptdesc_pmd_pts_dec() in huge_pmd_share() is not a
concern, as we are holding the i_mmap_lock the whole time, preventing
concurrent unsharing. That ptdesc_pmd_pts_dec() usage will be removed
separately as a cleanup later.
There are plenty more cleanups to be had, but they have to wait until
this is fixed.
[david@kernel.org: fix kerneldoc]
Link: https://lkml.kernel.org/r/f223dd74-331c-412d-93fc-69e360a5006c@kernel.org
Link: https://lkml.kernel.org/r/20251223214037.580860-5-david@kernel.org
Fixes: 1013af4f585f ("mm/hugetlb: fix huge_pmd_unshare() vs GUP-fast race")
Signed-off-by: David Hildenbrand (Red Hat) <david@kernel.org>
Reported-by: Uschakow, Stanislav" <suschako@amazon.de>
Closes: https://lore.kernel.org/all/4d3878531c76479d9f8ca9789dc6485d@amazon.de/
Tested-by: Laurence Oberman <loberman@redhat.com>
Acked-by: Harry Yoo <harry.yoo@oracle.com>
Reviewed-by: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: Lance Yang <lance.yang@linux.dev>
Cc: Liu Shixin <liushixin2@huawei.com>
Cc: Oscar Salvador <osalvador@suse.de>
Cc: Rik van Riel <riel@surriel.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2025-12-23 22:40:37 +01:00
|
|
|
|
|
|
|
|
|
2022-09-14 15:18:09 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-06-20 19:47:15 +08:00
|
|
|
|
mm/hugetlb: fix excessive IPI broadcasts when unsharing PMD tables using mmu_gather
As reported, ever since commit 1013af4f585f ("mm/hugetlb: fix
huge_pmd_unshare() vs GUP-fast race") we can end up in some situations
where we perform so many IPI broadcasts when unsharing hugetlb PMD page
tables that it severely regresses some workloads.
In particular, when we fork()+exit(), or when we munmap() a large
area backed by many shared PMD tables, we perform one IPI broadcast per
unshared PMD table.
There are two optimizations to be had:
(1) When we process (unshare) multiple such PMD tables, such as during
exit(), it is sufficient to send a single IPI broadcast (as long as
we respect locking rules) instead of one per PMD table.
Locking prevents that any of these PMD tables could get reused before
we drop the lock.
(2) When we are not the last sharer (> 2 users including us), there is
no need to send the IPI broadcast. The shared PMD tables cannot
become exclusive (fully unshared) before an IPI will be broadcasted
by the last sharer.
Concurrent GUP-fast could walk into a PMD table just before we
unshared it. It could then succeed in grabbing a page from the
shared page table even after munmap() etc succeeded (and supressed
an IPI). But there is not difference compared to GUP-fast just
sleeping for a while after grabbing the page and re-enabling IRQs.
Most importantly, GUP-fast will never walk into page tables that are
no-longer shared, because the last sharer will issue an IPI
broadcast.
(if ever required, checking whether the PUD changed in GUP-fast
after grabbing the page like we do in the PTE case could handle
this)
So let's rework PMD sharing TLB flushing + IPI sync to use the mmu_gather
infrastructure so we can implement these optimizations and demystify the
code at least a bit. Extend the mmu_gather infrastructure to be able to
deal with our special hugetlb PMD table sharing implementation.
To make initialization of the mmu_gather easier when working on a single
VMA (in particular, when dealing with hugetlb), provide
tlb_gather_mmu_vma().
We'll consolidate the handling for (full) unsharing of PMD tables in
tlb_unshare_pmd_ptdesc() and tlb_flush_unshared_tables(), and track
in "struct mmu_gather" whether we had (full) unsharing of PMD tables.
Because locking is very special (concurrent unsharing+reuse must be
prevented), we disallow deferring flushing to tlb_finish_mmu() and instead
require an explicit earlier call to tlb_flush_unshared_tables().
From hugetlb code, we call huge_pmd_unshare_flush() where we make sure
that the expected lock protecting us from concurrent unsharing+reuse is
still held.
Check with a VM_WARN_ON_ONCE() in tlb_finish_mmu() that
tlb_flush_unshared_tables() was properly called earlier.
Document it all properly.
Notes about tlb_remove_table_sync_one() interaction with unsharing:
There are two fairly tricky things:
(1) tlb_remove_table_sync_one() is a NOP on architectures without
CONFIG_MMU_GATHER_RCU_TABLE_FREE.
Here, the assumption is that the previous TLB flush would send an
IPI to all relevant CPUs. Careful: some architectures like x86 only
send IPIs to all relevant CPUs when tlb->freed_tables is set.
The relevant architectures should be selecting
MMU_GATHER_RCU_TABLE_FREE, but x86 might not do that in stable
kernels and it might have been problematic before this patch.
Also, the arch flushing behavior (independent of IPIs) is different
when tlb->freed_tables is set. Do we have to enlighten them to also
take care of tlb->unshared_tables? So far we didn't care, so
hopefully we are fine. Of course, we could be setting
tlb->freed_tables as well, but that might then unnecessarily flush
too much, because the semantics of tlb->freed_tables are a bit
fuzzy.
This patch changes nothing in this regard.
(2) tlb_remove_table_sync_one() is not a NOP on architectures with
CONFIG_MMU_GATHER_RCU_TABLE_FREE that actually don't need a sync.
Take x86 as an example: in the common case (!pv, !X86_FEATURE_INVLPGB)
we still issue IPIs during TLB flushes and don't actually need the
second tlb_remove_table_sync_one().
This optimized can be implemented on top of this, by checking e.g., in
tlb_remove_table_sync_one() whether we really need IPIs. But as
described in (1), it really must honor tlb->freed_tables then to
send IPIs to all relevant CPUs.
Notes on TLB flushing changes:
(1) Flushing for non-shared PMD tables
We're converting from flush_hugetlb_tlb_range() to
tlb_remove_huge_tlb_entry(). Given that we properly initialize the
MMU gather in tlb_gather_mmu_vma() to be hugetlb aware, similar to
__unmap_hugepage_range(), that should be fine.
(2) Flushing for shared PMD tables
We're converting from various things (flush_hugetlb_tlb_range(),
tlb_flush_pmd_range(), flush_tlb_range()) to tlb_flush_pmd_range().
tlb_flush_pmd_range() achieves the same that
tlb_remove_huge_tlb_entry() would achieve in these scenarios.
Note that tlb_remove_huge_tlb_entry() also calls
__tlb_remove_tlb_entry(), however that is only implemented on
powerpc, which does not support PMD table sharing.
Similar to (1), tlb_gather_mmu_vma() should make sure that TLB
flushing keeps on working as expected.
Further, note that the ptdesc_pmd_pts_dec() in huge_pmd_share() is not a
concern, as we are holding the i_mmap_lock the whole time, preventing
concurrent unsharing. That ptdesc_pmd_pts_dec() usage will be removed
separately as a cleanup later.
There are plenty more cleanups to be had, but they have to wait until
this is fixed.
[david@kernel.org: fix kerneldoc]
Link: https://lkml.kernel.org/r/f223dd74-331c-412d-93fc-69e360a5006c@kernel.org
Link: https://lkml.kernel.org/r/20251223214037.580860-5-david@kernel.org
Fixes: 1013af4f585f ("mm/hugetlb: fix huge_pmd_unshare() vs GUP-fast race")
Signed-off-by: David Hildenbrand (Red Hat) <david@kernel.org>
Reported-by: Uschakow, Stanislav" <suschako@amazon.de>
Closes: https://lore.kernel.org/all/4d3878531c76479d9f8ca9789dc6485d@amazon.de/
Tested-by: Laurence Oberman <loberman@redhat.com>
Acked-by: Harry Yoo <harry.yoo@oracle.com>
Reviewed-by: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: Lance Yang <lance.yang@linux.dev>
Cc: Liu Shixin <liushixin2@huawei.com>
Cc: Oscar Salvador <osalvador@suse.de>
Cc: Rik van Riel <riel@surriel.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2025-12-23 22:40:37 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-09-14 15:18:09 -07:00
|
|
|
|
2025-12-23 22:40:36 +01:00
|
|
|
|
|
|
|
|
|
2022-09-14 15:18:09 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
mm/hugetlb: fix excessive IPI broadcasts when unsharing PMD tables using mmu_gather
As reported, ever since commit 1013af4f585f ("mm/hugetlb: fix
huge_pmd_unshare() vs GUP-fast race") we can end up in some situations
where we perform so many IPI broadcasts when unsharing hugetlb PMD page
tables that it severely regresses some workloads.
In particular, when we fork()+exit(), or when we munmap() a large
area backed by many shared PMD tables, we perform one IPI broadcast per
unshared PMD table.
There are two optimizations to be had:
(1) When we process (unshare) multiple such PMD tables, such as during
exit(), it is sufficient to send a single IPI broadcast (as long as
we respect locking rules) instead of one per PMD table.
Locking prevents that any of these PMD tables could get reused before
we drop the lock.
(2) When we are not the last sharer (> 2 users including us), there is
no need to send the IPI broadcast. The shared PMD tables cannot
become exclusive (fully unshared) before an IPI will be broadcasted
by the last sharer.
Concurrent GUP-fast could walk into a PMD table just before we
unshared it. It could then succeed in grabbing a page from the
shared page table even after munmap() etc succeeded (and supressed
an IPI). But there is not difference compared to GUP-fast just
sleeping for a while after grabbing the page and re-enabling IRQs.
Most importantly, GUP-fast will never walk into page tables that are
no-longer shared, because the last sharer will issue an IPI
broadcast.
(if ever required, checking whether the PUD changed in GUP-fast
after grabbing the page like we do in the PTE case could handle
this)
So let's rework PMD sharing TLB flushing + IPI sync to use the mmu_gather
infrastructure so we can implement these optimizations and demystify the
code at least a bit. Extend the mmu_gather infrastructure to be able to
deal with our special hugetlb PMD table sharing implementation.
To make initialization of the mmu_gather easier when working on a single
VMA (in particular, when dealing with hugetlb), provide
tlb_gather_mmu_vma().
We'll consolidate the handling for (full) unsharing of PMD tables in
tlb_unshare_pmd_ptdesc() and tlb_flush_unshared_tables(), and track
in "struct mmu_gather" whether we had (full) unsharing of PMD tables.
Because locking is very special (concurrent unsharing+reuse must be
prevented), we disallow deferring flushing to tlb_finish_mmu() and instead
require an explicit earlier call to tlb_flush_unshared_tables().
From hugetlb code, we call huge_pmd_unshare_flush() where we make sure
that the expected lock protecting us from concurrent unsharing+reuse is
still held.
Check with a VM_WARN_ON_ONCE() in tlb_finish_mmu() that
tlb_flush_unshared_tables() was properly called earlier.
Document it all properly.
Notes about tlb_remove_table_sync_one() interaction with unsharing:
There are two fairly tricky things:
(1) tlb_remove_table_sync_one() is a NOP on architectures without
CONFIG_MMU_GATHER_RCU_TABLE_FREE.
Here, the assumption is that the previous TLB flush would send an
IPI to all relevant CPUs. Careful: some architectures like x86 only
send IPIs to all relevant CPUs when tlb->freed_tables is set.
The relevant architectures should be selecting
MMU_GATHER_RCU_TABLE_FREE, but x86 might not do that in stable
kernels and it might have been problematic before this patch.
Also, the arch flushing behavior (independent of IPIs) is different
when tlb->freed_tables is set. Do we have to enlighten them to also
take care of tlb->unshared_tables? So far we didn't care, so
hopefully we are fine. Of course, we could be setting
tlb->freed_tables as well, but that might then unnecessarily flush
too much, because the semantics of tlb->freed_tables are a bit
fuzzy.
This patch changes nothing in this regard.
(2) tlb_remove_table_sync_one() is not a NOP on architectures with
CONFIG_MMU_GATHER_RCU_TABLE_FREE that actually don't need a sync.
Take x86 as an example: in the common case (!pv, !X86_FEATURE_INVLPGB)
we still issue IPIs during TLB flushes and don't actually need the
second tlb_remove_table_sync_one().
This optimized can be implemented on top of this, by checking e.g., in
tlb_remove_table_sync_one() whether we really need IPIs. But as
described in (1), it really must honor tlb->freed_tables then to
send IPIs to all relevant CPUs.
Notes on TLB flushing changes:
(1) Flushing for non-shared PMD tables
We're converting from flush_hugetlb_tlb_range() to
tlb_remove_huge_tlb_entry(). Given that we properly initialize the
MMU gather in tlb_gather_mmu_vma() to be hugetlb aware, similar to
__unmap_hugepage_range(), that should be fine.
(2) Flushing for shared PMD tables
We're converting from various things (flush_hugetlb_tlb_range(),
tlb_flush_pmd_range(), flush_tlb_range()) to tlb_flush_pmd_range().
tlb_flush_pmd_range() achieves the same that
tlb_remove_huge_tlb_entry() would achieve in these scenarios.
Note that tlb_remove_huge_tlb_entry() also calls
__tlb_remove_tlb_entry(), however that is only implemented on
powerpc, which does not support PMD table sharing.
Similar to (1), tlb_gather_mmu_vma() should make sure that TLB
flushing keeps on working as expected.
Further, note that the ptdesc_pmd_pts_dec() in huge_pmd_share() is not a
concern, as we are holding the i_mmap_lock the whole time, preventing
concurrent unsharing. That ptdesc_pmd_pts_dec() usage will be removed
separately as a cleanup later.
There are plenty more cleanups to be had, but they have to wait until
this is fixed.
[david@kernel.org: fix kerneldoc]
Link: https://lkml.kernel.org/r/f223dd74-331c-412d-93fc-69e360a5006c@kernel.org
Link: https://lkml.kernel.org/r/20251223214037.580860-5-david@kernel.org
Fixes: 1013af4f585f ("mm/hugetlb: fix huge_pmd_unshare() vs GUP-fast race")
Signed-off-by: David Hildenbrand (Red Hat) <david@kernel.org>
Reported-by: Uschakow, Stanislav" <suschako@amazon.de>
Closes: https://lore.kernel.org/all/4d3878531c76479d9f8ca9789dc6485d@amazon.de/
Tested-by: Laurence Oberman <loberman@redhat.com>
Acked-by: Harry Yoo <harry.yoo@oracle.com>
Reviewed-by: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: Lance Yang <lance.yang@linux.dev>
Cc: Liu Shixin <liushixin2@huawei.com>
Cc: Oscar Salvador <osalvador@suse.de>
Cc: Rik van Riel <riel@surriel.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2025-12-23 22:40:37 +01:00
|
|
|
|
2021-06-30 18:54:16 -07:00
|
|
|
|
mm: rmap: fix CONT-PTE/PMD size hugetlb issue when migration
On some architectures (like ARM64), it can support CONT-PTE/PMD size
hugetlb, which means it can support not only PMD/PUD size hugetlb: 2M and
1G, but also CONT-PTE/PMD size: 64K and 32M if a 4K page size specified.
When migrating a hugetlb page, we will get the relevant page table entry
by huge_pte_offset() only once to nuke it and remap it with a migration
pte entry. This is correct for PMD or PUD size hugetlb, since they always
contain only one pmd entry or pud entry in the page table.
However this is incorrect for CONT-PTE and CONT-PMD size hugetlb, since
they can contain several continuous pte or pmd entry with same page table
attributes. So we will nuke or remap only one pte or pmd entry for this
CONT-PTE/PMD size hugetlb page, which is not expected for hugetlb
migration. The problem is we can still continue to modify the subpages'
data of a hugetlb page during migrating a hugetlb page, which can cause a
serious data consistent issue, since we did not nuke the page table entry
and set a migration pte for the subpages of a hugetlb page.
To fix this issue, we should change to use huge_ptep_clear_flush() to nuke
a hugetlb page table, and remap it with set_huge_pte_at() and
set_huge_swap_pte_at() when migrating a hugetlb page, which already
considered the CONT-PTE or CONT-PMD size hugetlb.
[akpm@linux-foundation.org: fix nommu build]
[baolin.wang@linux.alibaba.com: fix build errors for !CONFIG_MMU]
Link: https://lkml.kernel.org/r/a4baca670aca637e7198d9ae4543b8873cb224dc.1652270205.git.baolin.wang@linux.alibaba.com
Link: https://lkml.kernel.org/r/ea5abf529f0997b5430961012bfda6166c1efc8c.1652147571.git.baolin.wang@linux.alibaba.com
Signed-off-by: Baolin Wang <baolin.wang@linux.alibaba.com>
Reviewed-by: Muchun Song <songmuchun@bytedance.com>
Reviewed-by: Mike Kravetz <mike.kravetz@oracle.com>
Acked-by: David Hildenbrand <david@redhat.com>
Cc: Alexander Gordeev <agordeev@linux.ibm.com>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Christian Borntraeger <borntraeger@linux.ibm.com>
Cc: David S. Miller <davem@davemloft.net>
Cc: Gerald Schaefer <gerald.schaefer@linux.ibm.com>
Cc: Heiko Carstens <hca@linux.ibm.com>
Cc: Helge Deller <deller@gmx.de>
Cc: James Bottomley <James.Bottomley@HansenPartnership.com>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Rich Felker <dalias@libc.org>
Cc: Sven Schnelle <svens@linux.ibm.com>
Cc: Thomas Bogendoerfer <tsbogend@alpha.franken.de>
Cc: Vasily Gorbik <gor@linux.ibm.com>
Cc: Will Deacon <will@kernel.org>
Cc: Yoshinori Sato <ysato@users.osdn.me>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2022-05-13 16:48:55 -07:00
|
|
|
|
|
|
|
|
|
2025-02-10 20:37:53 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
mm: ptep_get() conversion
Convert all instances of direct pte_t* dereferencing to instead use
ptep_get() helper. This means that by default, the accesses change from a
C dereference to a READ_ONCE(). This is technically the correct thing to
do since where pgtables are modified by HW (for access/dirty) they are
volatile and therefore we should always ensure READ_ONCE() semantics.
But more importantly, by always using the helper, it can be overridden by
the architecture to fully encapsulate the contents of the pte. Arch code
is deliberately not converted, as the arch code knows best. It is
intended that arch code (arm64) will override the default with its own
implementation that can (e.g.) hide certain bits from the core code, or
determine young/dirty status by mixing in state from another source.
Conversion was done using Coccinelle:
----
// $ make coccicheck \
// COCCI=ptepget.cocci \
// SPFLAGS="--include-headers" \
// MODE=patch
virtual patch
@ depends on patch @
pte_t *v;
@@
- *v
+ ptep_get(v)
----
Then reviewed and hand-edited to avoid multiple unnecessary calls to
ptep_get(), instead opting to store the result of a single call in a
variable, where it is correct to do so. This aims to negate any cost of
READ_ONCE() and will benefit arch-overrides that may be more complex.
Included is a fix for an issue in an earlier version of this patch that
was pointed out by kernel test robot. The issue arose because config
MMU=n elides definition of the ptep helper functions, including
ptep_get(). HUGETLB_PAGE=n configs still define a simple
huge_ptep_clear_flush() for linking purposes, which dereferences the ptep.
So when both configs are disabled, this caused a build error because
ptep_get() is not defined. Fix by continuing to do a direct dereference
when MMU=n. This is safe because for this config the arch code cannot be
trying to virtualize the ptes because none of the ptep helpers are
defined.
Link: https://lkml.kernel.org/r/20230612151545.3317766-4-ryan.roberts@arm.com
Reported-by: kernel test robot <lkp@intel.com>
Link: https://lore.kernel.org/oe-kbuild-all/202305120142.yXsNEo6H-lkp@intel.com/
Signed-off-by: Ryan Roberts <ryan.roberts@arm.com>
Cc: Adrian Hunter <adrian.hunter@intel.com>
Cc: Alexander Potapenko <glider@google.com>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Alex Williamson <alex.williamson@redhat.com>
Cc: Al Viro <viro@zeniv.linux.org.uk>
Cc: Andrey Konovalov <andreyknvl@gmail.com>
Cc: Andrey Ryabinin <ryabinin.a.a@gmail.com>
Cc: Christian Brauner <brauner@kernel.org>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: Daniel Vetter <daniel@ffwll.ch>
Cc: Dave Airlie <airlied@gmail.com>
Cc: Dimitri Sivanich <dimitri.sivanich@hpe.com>
Cc: Dmitry Vyukov <dvyukov@google.com>
Cc: Ian Rogers <irogers@google.com>
Cc: Jason Gunthorpe <jgg@ziepe.ca>
Cc: Jérôme Glisse <jglisse@redhat.com>
Cc: Jiri Olsa <jolsa@kernel.org>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Lorenzo Stoakes <lstoakes@gmail.com>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Miaohe Lin <linmiaohe@huawei.com>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Mike Rapoport (IBM) <rppt@kernel.org>
Cc: Muchun Song <muchun.song@linux.dev>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Naoya Horiguchi <naoya.horiguchi@nec.com>
Cc: Oleksandr Tyshchenko <oleksandr_tyshchenko@epam.com>
Cc: Pavel Tatashin <pasha.tatashin@soleen.com>
Cc: Roman Gushchin <roman.gushchin@linux.dev>
Cc: SeongJae Park <sj@kernel.org>
Cc: Shakeel Butt <shakeelb@google.com>
Cc: Uladzislau Rezki (Sony) <urezki@gmail.com>
Cc: Vincenzo Frascino <vincenzo.frascino@arm.com>
Cc: Yu Zhao <yuzhao@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2023-06-12 16:15:45 +01:00
|
|
|
|
mm: rmap: fix CONT-PTE/PMD size hugetlb issue when migration
On some architectures (like ARM64), it can support CONT-PTE/PMD size
hugetlb, which means it can support not only PMD/PUD size hugetlb: 2M and
1G, but also CONT-PTE/PMD size: 64K and 32M if a 4K page size specified.
When migrating a hugetlb page, we will get the relevant page table entry
by huge_pte_offset() only once to nuke it and remap it with a migration
pte entry. This is correct for PMD or PUD size hugetlb, since they always
contain only one pmd entry or pud entry in the page table.
However this is incorrect for CONT-PTE and CONT-PMD size hugetlb, since
they can contain several continuous pte or pmd entry with same page table
attributes. So we will nuke or remap only one pte or pmd entry for this
CONT-PTE/PMD size hugetlb page, which is not expected for hugetlb
migration. The problem is we can still continue to modify the subpages'
data of a hugetlb page during migrating a hugetlb page, which can cause a
serious data consistent issue, since we did not nuke the page table entry
and set a migration pte for the subpages of a hugetlb page.
To fix this issue, we should change to use huge_ptep_clear_flush() to nuke
a hugetlb page table, and remap it with set_huge_pte_at() and
set_huge_swap_pte_at() when migrating a hugetlb page, which already
considered the CONT-PTE or CONT-PMD size hugetlb.
[akpm@linux-foundation.org: fix nommu build]
[baolin.wang@linux.alibaba.com: fix build errors for !CONFIG_MMU]
Link: https://lkml.kernel.org/r/a4baca670aca637e7198d9ae4543b8873cb224dc.1652270205.git.baolin.wang@linux.alibaba.com
Link: https://lkml.kernel.org/r/ea5abf529f0997b5430961012bfda6166c1efc8c.1652147571.git.baolin.wang@linux.alibaba.com
Signed-off-by: Baolin Wang <baolin.wang@linux.alibaba.com>
Reviewed-by: Muchun Song <songmuchun@bytedance.com>
Reviewed-by: Mike Kravetz <mike.kravetz@oracle.com>
Acked-by: David Hildenbrand <david@redhat.com>
Cc: Alexander Gordeev <agordeev@linux.ibm.com>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Christian Borntraeger <borntraeger@linux.ibm.com>
Cc: David S. Miller <davem@davemloft.net>
Cc: Gerald Schaefer <gerald.schaefer@linux.ibm.com>
Cc: Heiko Carstens <hca@linux.ibm.com>
Cc: Helge Deller <deller@gmx.de>
Cc: James Bottomley <James.Bottomley@HansenPartnership.com>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Rich Felker <dalias@libc.org>
Cc: Sven Schnelle <svens@linux.ibm.com>
Cc: Thomas Bogendoerfer <tsbogend@alpha.franken.de>
Cc: Vasily Gorbik <gor@linux.ibm.com>
Cc: Will Deacon <will@kernel.org>
Cc: Yoshinori Sato <ysato@users.osdn.me>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2022-05-13 16:48:55 -07:00
|
|
|
|
2023-02-13 20:34:43 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-02-14 22:30:13 +13:00
|
|
|
|
2023-02-13 20:34:43 +08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-02-10 20:37:53 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-11-10 22:21:33 +00:00
|
|
|
|
|
|
|
|
|
2025-02-10 20:37:53 +01:00
|
|
|
|
2025-11-10 22:21:33 +00:00
|
|
|
|
|
|
|
|
|
2021-06-30 18:54:16 -07:00
|
|
|
|
|
|
|
|
|
2025-02-10 20:37:53 +01:00
|
|
|
|
|
|
|
|
|
2021-06-30 18:54:16 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-02-10 20:37:53 +01:00
|
|
|
|
|
|
|
|
|
2021-06-30 18:54:16 -07:00
|
|
|
|
|
|
|
|
|
2022-01-28 14:29:43 -05:00
|
|
|
|
|
|
|
|
|
mm: hugetlb: add huge page size param to set_huge_pte_at()
Patch series "Fix set_huge_pte_at() panic on arm64", v2.
This series fixes a bug in arm64's implementation of set_huge_pte_at(),
which can result in an unprivileged user causing a kernel panic. The
problem was triggered when running the new uffd poison mm selftest for
HUGETLB memory. This test (and the uffd poison feature) was merged for
v6.5-rc7.
Ideally, I'd like to get this fix in for v6.6 and I've cc'ed stable
(correctly this time) to get it backported to v6.5, where the issue first
showed up.
Description of Bug
==================
arm64's huge pte implementation supports multiple huge page sizes, some of
which are implemented in the page table with multiple contiguous entries.
So set_huge_pte_at() needs to work out how big the logical pte is, so that
it can also work out how many physical ptes (or pmds) need to be written.
It previously did this by grabbing the folio out of the pte and querying
its size.
However, there are cases when the pte being set is actually a swap entry.
But this also used to work fine, because for huge ptes, we only ever saw
migration entries and hwpoison entries. And both of these types of swap
entries have a PFN embedded, so the code would grab that and everything
still worked out.
But over time, more calls to set_huge_pte_at() have been added that set
swap entry types that do not embed a PFN. And this causes the code to go
bang. The triggering case is for the uffd poison test, commit
99aa77215ad0 ("selftests/mm: add uffd unit test for UFFDIO_POISON"), which
causes a PTE_MARKER_POISONED swap entry to be set, coutesey of commit
8a13897fb0da ("mm: userfaultfd: support UFFDIO_POISON for hugetlbfs") -
added in v6.5-rc7. Although review shows that there are other call sites
that set PTE_MARKER_UFFD_WP (which also has no PFN), these don't trigger
on arm64 because arm64 doesn't support UFFD WP.
If CONFIG_DEBUG_VM is enabled, we do at least get a BUG(), but otherwise,
it will dereference a bad pointer in page_folio():
static inline struct folio *hugetlb_swap_entry_to_folio(swp_entry_t entry)
{
VM_BUG_ON(!is_migration_entry(entry) && !is_hwpoison_entry(entry));
return page_folio(pfn_to_page(swp_offset_pfn(entry)));
}
Fix
===
The simplest fix would have been to revert the dodgy cleanup commit
18f3962953e4 ("mm: hugetlb: kill set_huge_swap_pte_at()"), but since
things have moved on, this would have required an audit of all the new
set_huge_pte_at() call sites to see if they should be converted to
set_huge_swap_pte_at(). As per the original intent of the change, it
would also leave us open to future bugs when people invariably get it
wrong and call the wrong helper.
So instead, I've added a huge page size parameter to set_huge_pte_at().
This means that the arm64 code has the size in all cases. It's a bigger
change, due to needing to touch the arches that implement the function,
but it is entirely mechanical, so in my view, low risk.
I've compile-tested all touched arches; arm64, parisc, powerpc, riscv,
s390, sparc (and additionally x86_64). I've additionally booted and run
mm selftests against arm64, where I observe the uffd poison test is fixed,
and there are no other regressions.
This patch (of 2):
In order to fix a bug, arm64 needs to be told the size of the huge page
for which the pte is being set in set_huge_pte_at(). Provide for this by
adding an `unsigned long sz` parameter to the function. This follows the
same pattern as huge_pte_clear().
This commit makes the required interface modifications to the core mm as
well as all arches that implement this function (arm64, parisc, powerpc,
riscv, s390, sparc). The actual arm64 bug will be fixed in a separate
commit.
No behavioral changes intended.
Link: https://lkml.kernel.org/r/20230922115804.2043771-1-ryan.roberts@arm.com
Link: https://lkml.kernel.org/r/20230922115804.2043771-2-ryan.roberts@arm.com
Fixes: 8a13897fb0da ("mm: userfaultfd: support UFFDIO_POISON for hugetlbfs")
Signed-off-by: Ryan Roberts <ryan.roberts@arm.com>
Reviewed-by: Christophe Leroy <christophe.leroy@csgroup.eu> [powerpc 8xx]
Reviewed-by: Lorenzo Stoakes <lstoakes@gmail.com> [vmalloc change]
Cc: Alexandre Ghiti <alex@ghiti.fr>
Cc: Albert Ou <aou@eecs.berkeley.edu>
Cc: Alexander Gordeev <agordeev@linux.ibm.com>
Cc: Anshuman Khandual <anshuman.khandual@arm.com>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Axel Rasmussen <axelrasmussen@google.com>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Christian Borntraeger <borntraeger@linux.ibm.com>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: David S. Miller <davem@davemloft.net>
Cc: Gerald Schaefer <gerald.schaefer@linux.ibm.com>
Cc: Heiko Carstens <hca@linux.ibm.com>
Cc: Helge Deller <deller@gmx.de>
Cc: "James E.J. Bottomley" <James.Bottomley@HansenPartnership.com>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Muchun Song <muchun.song@linux.dev>
Cc: Nicholas Piggin <npiggin@gmail.com>
Cc: Palmer Dabbelt <palmer@dabbelt.com>
Cc: Paul Walmsley <paul.walmsley@sifive.com>
Cc: Peter Xu <peterx@redhat.com>
Cc: Qi Zheng <zhengqi.arch@bytedance.com>
Cc: Ryan Roberts <ryan.roberts@arm.com>
Cc: SeongJae Park <sj@kernel.org>
Cc: Sven Schnelle <svens@linux.ibm.com>
Cc: Uladzislau Rezki (Sony) <urezki@gmail.com>
Cc: Vasily Gorbik <gor@linux.ibm.com>
Cc: Will Deacon <will@kernel.org>
Cc: <stable@vger.kernel.org> [6.5+]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2023-09-22 12:58:03 +01:00
|
|
|
|
|
|
|
|
|
2021-06-30 18:54:16 -07:00
|
|
|
|
2024-01-11 15:24:28 +00:00
|
|
|
|
2021-06-30 18:54:16 -07:00
|
|
|
|
|
|
|
|
|
2025-02-10 20:37:53 +01:00
|
|
|
|
|
|
|
|
|
2021-06-30 18:54:16 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2024-01-11 15:24:28 +00:00
|
|
|
|
2021-06-30 18:54:16 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-02-10 20:37:53 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-06-30 18:54:16 -07:00
|
|
|
|
mm: rmap: fix CONT-PTE/PMD size hugetlb issue when migration
On some architectures (like ARM64), it can support CONT-PTE/PMD size
hugetlb, which means it can support not only PMD/PUD size hugetlb: 2M and
1G, but also CONT-PTE/PMD size: 64K and 32M if a 4K page size specified.
When migrating a hugetlb page, we will get the relevant page table entry
by huge_pte_offset() only once to nuke it and remap it with a migration
pte entry. This is correct for PMD or PUD size hugetlb, since they always
contain only one pmd entry or pud entry in the page table.
However this is incorrect for CONT-PTE and CONT-PMD size hugetlb, since
they can contain several continuous pte or pmd entry with same page table
attributes. So we will nuke or remap only one pte or pmd entry for this
CONT-PTE/PMD size hugetlb page, which is not expected for hugetlb
migration. The problem is we can still continue to modify the subpages'
data of a hugetlb page during migrating a hugetlb page, which can cause a
serious data consistent issue, since we did not nuke the page table entry
and set a migration pte for the subpages of a hugetlb page.
To fix this issue, we should change to use huge_ptep_clear_flush() to nuke
a hugetlb page table, and remap it with set_huge_pte_at() and
set_huge_swap_pte_at() when migrating a hugetlb page, which already
considered the CONT-PTE or CONT-PMD size hugetlb.
[akpm@linux-foundation.org: fix nommu build]
[baolin.wang@linux.alibaba.com: fix build errors for !CONFIG_MMU]
Link: https://lkml.kernel.org/r/a4baca670aca637e7198d9ae4543b8873cb224dc.1652270205.git.baolin.wang@linux.alibaba.com
Link: https://lkml.kernel.org/r/ea5abf529f0997b5430961012bfda6166c1efc8c.1652147571.git.baolin.wang@linux.alibaba.com
Signed-off-by: Baolin Wang <baolin.wang@linux.alibaba.com>
Reviewed-by: Muchun Song <songmuchun@bytedance.com>
Reviewed-by: Mike Kravetz <mike.kravetz@oracle.com>
Acked-by: David Hildenbrand <david@redhat.com>
Cc: Alexander Gordeev <agordeev@linux.ibm.com>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Christian Borntraeger <borntraeger@linux.ibm.com>
Cc: David S. Miller <davem@davemloft.net>
Cc: Gerald Schaefer <gerald.schaefer@linux.ibm.com>
Cc: Heiko Carstens <hca@linux.ibm.com>
Cc: Helge Deller <deller@gmx.de>
Cc: James Bottomley <James.Bottomley@HansenPartnership.com>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Rich Felker <dalias@libc.org>
Cc: Sven Schnelle <svens@linux.ibm.com>
Cc: Thomas Bogendoerfer <tsbogend@alpha.franken.de>
Cc: Vasily Gorbik <gor@linux.ibm.com>
Cc: Will Deacon <will@kernel.org>
Cc: Yoshinori Sato <ysato@users.osdn.me>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2022-05-13 16:48:55 -07:00
|
|
|
|
mm: hugetlb: add huge page size param to set_huge_pte_at()
Patch series "Fix set_huge_pte_at() panic on arm64", v2.
This series fixes a bug in arm64's implementation of set_huge_pte_at(),
which can result in an unprivileged user causing a kernel panic. The
problem was triggered when running the new uffd poison mm selftest for
HUGETLB memory. This test (and the uffd poison feature) was merged for
v6.5-rc7.
Ideally, I'd like to get this fix in for v6.6 and I've cc'ed stable
(correctly this time) to get it backported to v6.5, where the issue first
showed up.
Description of Bug
==================
arm64's huge pte implementation supports multiple huge page sizes, some of
which are implemented in the page table with multiple contiguous entries.
So set_huge_pte_at() needs to work out how big the logical pte is, so that
it can also work out how many physical ptes (or pmds) need to be written.
It previously did this by grabbing the folio out of the pte and querying
its size.
However, there are cases when the pte being set is actually a swap entry.
But this also used to work fine, because for huge ptes, we only ever saw
migration entries and hwpoison entries. And both of these types of swap
entries have a PFN embedded, so the code would grab that and everything
still worked out.
But over time, more calls to set_huge_pte_at() have been added that set
swap entry types that do not embed a PFN. And this causes the code to go
bang. The triggering case is for the uffd poison test, commit
99aa77215ad0 ("selftests/mm: add uffd unit test for UFFDIO_POISON"), which
causes a PTE_MARKER_POISONED swap entry to be set, coutesey of commit
8a13897fb0da ("mm: userfaultfd: support UFFDIO_POISON for hugetlbfs") -
added in v6.5-rc7. Although review shows that there are other call sites
that set PTE_MARKER_UFFD_WP (which also has no PFN), these don't trigger
on arm64 because arm64 doesn't support UFFD WP.
If CONFIG_DEBUG_VM is enabled, we do at least get a BUG(), but otherwise,
it will dereference a bad pointer in page_folio():
static inline struct folio *hugetlb_swap_entry_to_folio(swp_entry_t entry)
{
VM_BUG_ON(!is_migration_entry(entry) && !is_hwpoison_entry(entry));
return page_folio(pfn_to_page(swp_offset_pfn(entry)));
}
Fix
===
The simplest fix would have been to revert the dodgy cleanup commit
18f3962953e4 ("mm: hugetlb: kill set_huge_swap_pte_at()"), but since
things have moved on, this would have required an audit of all the new
set_huge_pte_at() call sites to see if they should be converted to
set_huge_swap_pte_at(). As per the original intent of the change, it
would also leave us open to future bugs when people invariably get it
wrong and call the wrong helper.
So instead, I've added a huge page size parameter to set_huge_pte_at().
This means that the arm64 code has the size in all cases. It's a bigger
change, due to needing to touch the arches that implement the function,
but it is entirely mechanical, so in my view, low risk.
I've compile-tested all touched arches; arm64, parisc, powerpc, riscv,
s390, sparc (and additionally x86_64). I've additionally booted and run
mm selftests against arm64, where I observe the uffd poison test is fixed,
and there are no other regressions.
This patch (of 2):
In order to fix a bug, arm64 needs to be told the size of the huge page
for which the pte is being set in set_huge_pte_at(). Provide for this by
adding an `unsigned long sz` parameter to the function. This follows the
same pattern as huge_pte_clear().
This commit makes the required interface modifications to the core mm as
well as all arches that implement this function (arm64, parisc, powerpc,
riscv, s390, sparc). The actual arm64 bug will be fixed in a separate
commit.
No behavioral changes intended.
Link: https://lkml.kernel.org/r/20230922115804.2043771-1-ryan.roberts@arm.com
Link: https://lkml.kernel.org/r/20230922115804.2043771-2-ryan.roberts@arm.com
Fixes: 8a13897fb0da ("mm: userfaultfd: support UFFDIO_POISON for hugetlbfs")
Signed-off-by: Ryan Roberts <ryan.roberts@arm.com>
Reviewed-by: Christophe Leroy <christophe.leroy@csgroup.eu> [powerpc 8xx]
Reviewed-by: Lorenzo Stoakes <lstoakes@gmail.com> [vmalloc change]
Cc: Alexandre Ghiti <alex@ghiti.fr>
Cc: Albert Ou <aou@eecs.berkeley.edu>
Cc: Alexander Gordeev <agordeev@linux.ibm.com>
Cc: Anshuman Khandual <anshuman.khandual@arm.com>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Axel Rasmussen <axelrasmussen@google.com>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Christian Borntraeger <borntraeger@linux.ibm.com>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: David S. Miller <davem@davemloft.net>
Cc: Gerald Schaefer <gerald.schaefer@linux.ibm.com>
Cc: Heiko Carstens <hca@linux.ibm.com>
Cc: Helge Deller <deller@gmx.de>
Cc: "James E.J. Bottomley" <James.Bottomley@HansenPartnership.com>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Muchun Song <muchun.song@linux.dev>
Cc: Nicholas Piggin <npiggin@gmail.com>
Cc: Palmer Dabbelt <palmer@dabbelt.com>
Cc: Paul Walmsley <paul.walmsley@sifive.com>
Cc: Peter Xu <peterx@redhat.com>
Cc: Qi Zheng <zhengqi.arch@bytedance.com>
Cc: Ryan Roberts <ryan.roberts@arm.com>
Cc: SeongJae Park <sj@kernel.org>
Cc: Sven Schnelle <svens@linux.ibm.com>
Cc: Uladzislau Rezki (Sony) <urezki@gmail.com>
Cc: Vasily Gorbik <gor@linux.ibm.com>
Cc: Will Deacon <will@kernel.org>
Cc: <stable@vger.kernel.org> [6.5+]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2023-09-22 12:58:03 +01:00
|
|
|
|
|
|
|
|
|
mm: rmap: fix CONT-PTE/PMD size hugetlb issue when migration
On some architectures (like ARM64), it can support CONT-PTE/PMD size
hugetlb, which means it can support not only PMD/PUD size hugetlb: 2M and
1G, but also CONT-PTE/PMD size: 64K and 32M if a 4K page size specified.
When migrating a hugetlb page, we will get the relevant page table entry
by huge_pte_offset() only once to nuke it and remap it with a migration
pte entry. This is correct for PMD or PUD size hugetlb, since they always
contain only one pmd entry or pud entry in the page table.
However this is incorrect for CONT-PTE and CONT-PMD size hugetlb, since
they can contain several continuous pte or pmd entry with same page table
attributes. So we will nuke or remap only one pte or pmd entry for this
CONT-PTE/PMD size hugetlb page, which is not expected for hugetlb
migration. The problem is we can still continue to modify the subpages'
data of a hugetlb page during migrating a hugetlb page, which can cause a
serious data consistent issue, since we did not nuke the page table entry
and set a migration pte for the subpages of a hugetlb page.
To fix this issue, we should change to use huge_ptep_clear_flush() to nuke
a hugetlb page table, and remap it with set_huge_pte_at() and
set_huge_swap_pte_at() when migrating a hugetlb page, which already
considered the CONT-PTE or CONT-PMD size hugetlb.
[akpm@linux-foundation.org: fix nommu build]
[baolin.wang@linux.alibaba.com: fix build errors for !CONFIG_MMU]
Link: https://lkml.kernel.org/r/a4baca670aca637e7198d9ae4543b8873cb224dc.1652270205.git.baolin.wang@linux.alibaba.com
Link: https://lkml.kernel.org/r/ea5abf529f0997b5430961012bfda6166c1efc8c.1652147571.git.baolin.wang@linux.alibaba.com
Signed-off-by: Baolin Wang <baolin.wang@linux.alibaba.com>
Reviewed-by: Muchun Song <songmuchun@bytedance.com>
Reviewed-by: Mike Kravetz <mike.kravetz@oracle.com>
Acked-by: David Hildenbrand <david@redhat.com>
Cc: Alexander Gordeev <agordeev@linux.ibm.com>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Christian Borntraeger <borntraeger@linux.ibm.com>
Cc: David S. Miller <davem@davemloft.net>
Cc: Gerald Schaefer <gerald.schaefer@linux.ibm.com>
Cc: Heiko Carstens <hca@linux.ibm.com>
Cc: Helge Deller <deller@gmx.de>
Cc: James Bottomley <James.Bottomley@HansenPartnership.com>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Rich Felker <dalias@libc.org>
Cc: Sven Schnelle <svens@linux.ibm.com>
Cc: Thomas Bogendoerfer <tsbogend@alpha.franken.de>
Cc: Vasily Gorbik <gor@linux.ibm.com>
Cc: Will Deacon <will@kernel.org>
Cc: Yoshinori Sato <ysato@users.osdn.me>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2022-05-13 16:48:55 -07:00
|
|
|
|
|
|
|
|
|
2021-06-30 18:54:16 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
mm: fix PageAnonExclusive clearing racing with concurrent RCU GUP-fast
commit 6c287605fd56 ("mm: remember exclusively mapped anonymous pages with
PG_anon_exclusive") made sure that when PageAnonExclusive() has to be
cleared during temporary unmapping of a page, that the PTE is
cleared/invalidated and that the TLB is flushed.
What we want to achieve in all cases is that we cannot end up with a pin on
an anonymous page that may be shared, because such pins would be
unreliable and could result in memory corruptions when the mapped page
and the pin go out of sync due to a write fault.
That TLB flush handling was inspired by an outdated comment in
mm/ksm.c:write_protect_page(), which similarly required the TLB flush in
the past to synchronize with GUP-fast. However, ever since general RCU GUP
fast was introduced in commit 2667f50e8b81 ("mm: introduce a general RCU
get_user_pages_fast()"), a TLB flush is no longer sufficient to handle
concurrent GUP-fast in all cases -- it only handles traditional IPI-based
GUP-fast correctly.
Peter Xu (thankfully) questioned whether that TLB flush is really
required. On architectures that send an IPI broadcast on TLB flush,
it works as expected. To synchronize with RCU GUP-fast properly, we're
conceptually fine, however, we have to enforce a certain memory order and
are missing memory barriers.
Let's document that, avoid the TLB flush where possible and use proper
explicit memory barriers where required. We shouldn't really care about the
additional memory barriers here, as we're not on extremely hot paths --
and we're getting rid of some TLB flushes.
We use a smp_mb() pair for handling concurrent pinning and a
smp_rmb()/smp_wmb() pair for handling the corner case of only temporary
PTE changes but permanent PageAnonExclusive changes.
One extreme example, whereby GUP-fast takes a R/O pin and KSM wants to
convert an exclusive anonymous page to a KSM page, and that page is already
mapped write-protected (-> no PTE change) would be:
Thread 0 (KSM) Thread 1 (GUP-fast)
(B1) Read the PTE
# (B2) skipped without FOLL_WRITE
(A1) Clear PTE
smp_mb()
(A2) Check pinned
(B3) Pin the mapped page
smp_mb()
(A3) Clear PageAnonExclusive
smp_wmb()
(A4) Restore PTE
(B4) Check if the PTE changed
smp_rmb()
(B5) Check PageAnonExclusive
Thread 1 will properly detect that PageAnonExclusive was cleared and
back off.
Note that we don't need a memory barrier between checking if the page is
pinned and clearing PageAnonExclusive, because stores are not
speculated.
The possible issues due to reordering are of theoretical nature so far
and attempts to reproduce the race failed.
Especially the "no PTE change" case isn't the common case, because we'd
need an exclusive anonymous page that's mapped R/O and the PTE is clean
in KSM code -- and using KSM with page pinning isn't extremely common.
Further, the clear+TLB flush we used for now implies a memory barrier.
So the problematic missing part should be the missing memory barrier
after pinning but before checking if the PTE changed.
Link: https://lkml.kernel.org/r/20220901083559.67446-1-david@redhat.com
Fixes: 6c287605fd56 ("mm: remember exclusively mapped anonymous pages with PG_anon_exclusive")
Signed-off-by: David Hildenbrand <david@redhat.com>
Cc: Jason Gunthorpe <jgg@nvidia.com>
Cc: John Hubbard <jhubbard@nvidia.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Peter Xu <peterx@redhat.com>
Cc: Alistair Popple <apopple@nvidia.com>
Cc: Nadav Amit <namit@vmware.com>
Cc: Yang Shi <shy828301@gmail.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Andrea Parri <parri.andrea@gmail.com>
Cc: Will Deacon <will@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: "Paul E. McKenney" <paulmck@kernel.org>
Cc: Christoph von Recklinghausen <crecklin@redhat.com>
Cc: Don Dutile <ddutile@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2022-09-01 10:35:59 +02:00
|
|
|
|
2023-12-20 23:45:02 +01:00
|
|
|
|
2023-12-20 23:44:29 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
mm: hugetlb: add huge page size param to set_huge_pte_at()
Patch series "Fix set_huge_pte_at() panic on arm64", v2.
This series fixes a bug in arm64's implementation of set_huge_pte_at(),
which can result in an unprivileged user causing a kernel panic. The
problem was triggered when running the new uffd poison mm selftest for
HUGETLB memory. This test (and the uffd poison feature) was merged for
v6.5-rc7.
Ideally, I'd like to get this fix in for v6.6 and I've cc'ed stable
(correctly this time) to get it backported to v6.5, where the issue first
showed up.
Description of Bug
==================
arm64's huge pte implementation supports multiple huge page sizes, some of
which are implemented in the page table with multiple contiguous entries.
So set_huge_pte_at() needs to work out how big the logical pte is, so that
it can also work out how many physical ptes (or pmds) need to be written.
It previously did this by grabbing the folio out of the pte and querying
its size.
However, there are cases when the pte being set is actually a swap entry.
But this also used to work fine, because for huge ptes, we only ever saw
migration entries and hwpoison entries. And both of these types of swap
entries have a PFN embedded, so the code would grab that and everything
still worked out.
But over time, more calls to set_huge_pte_at() have been added that set
swap entry types that do not embed a PFN. And this causes the code to go
bang. The triggering case is for the uffd poison test, commit
99aa77215ad0 ("selftests/mm: add uffd unit test for UFFDIO_POISON"), which
causes a PTE_MARKER_POISONED swap entry to be set, coutesey of commit
8a13897fb0da ("mm: userfaultfd: support UFFDIO_POISON for hugetlbfs") -
added in v6.5-rc7. Although review shows that there are other call sites
that set PTE_MARKER_UFFD_WP (which also has no PFN), these don't trigger
on arm64 because arm64 doesn't support UFFD WP.
If CONFIG_DEBUG_VM is enabled, we do at least get a BUG(), but otherwise,
it will dereference a bad pointer in page_folio():
static inline struct folio *hugetlb_swap_entry_to_folio(swp_entry_t entry)
{
VM_BUG_ON(!is_migration_entry(entry) && !is_hwpoison_entry(entry));
return page_folio(pfn_to_page(swp_offset_pfn(entry)));
}
Fix
===
The simplest fix would have been to revert the dodgy cleanup commit
18f3962953e4 ("mm: hugetlb: kill set_huge_swap_pte_at()"), but since
things have moved on, this would have required an audit of all the new
set_huge_pte_at() call sites to see if they should be converted to
set_huge_swap_pte_at(). As per the original intent of the change, it
would also leave us open to future bugs when people invariably get it
wrong and call the wrong helper.
So instead, I've added a huge page size parameter to set_huge_pte_at().
This means that the arm64 code has the size in all cases. It's a bigger
change, due to needing to touch the arches that implement the function,
but it is entirely mechanical, so in my view, low risk.
I've compile-tested all touched arches; arm64, parisc, powerpc, riscv,
s390, sparc (and additionally x86_64). I've additionally booted and run
mm selftests against arm64, where I observe the uffd poison test is fixed,
and there are no other regressions.
This patch (of 2):
In order to fix a bug, arm64 needs to be told the size of the huge page
for which the pte is being set in set_huge_pte_at(). Provide for this by
adding an `unsigned long sz` parameter to the function. This follows the
same pattern as huge_pte_clear().
This commit makes the required interface modifications to the core mm as
well as all arches that implement this function (arm64, parisc, powerpc,
riscv, s390, sparc). The actual arm64 bug will be fixed in a separate
commit.
No behavioral changes intended.
Link: https://lkml.kernel.org/r/20230922115804.2043771-1-ryan.roberts@arm.com
Link: https://lkml.kernel.org/r/20230922115804.2043771-2-ryan.roberts@arm.com
Fixes: 8a13897fb0da ("mm: userfaultfd: support UFFDIO_POISON for hugetlbfs")
Signed-off-by: Ryan Roberts <ryan.roberts@arm.com>
Reviewed-by: Christophe Leroy <christophe.leroy@csgroup.eu> [powerpc 8xx]
Reviewed-by: Lorenzo Stoakes <lstoakes@gmail.com> [vmalloc change]
Cc: Alexandre Ghiti <alex@ghiti.fr>
Cc: Albert Ou <aou@eecs.berkeley.edu>
Cc: Alexander Gordeev <agordeev@linux.ibm.com>
Cc: Anshuman Khandual <anshuman.khandual@arm.com>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Axel Rasmussen <axelrasmussen@google.com>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Christian Borntraeger <borntraeger@linux.ibm.com>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: David S. Miller <davem@davemloft.net>
Cc: Gerald Schaefer <gerald.schaefer@linux.ibm.com>
Cc: Heiko Carstens <hca@linux.ibm.com>
Cc: Helge Deller <deller@gmx.de>
Cc: "James E.J. Bottomley" <James.Bottomley@HansenPartnership.com>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Muchun Song <muchun.song@linux.dev>
Cc: Nicholas Piggin <npiggin@gmail.com>
Cc: Palmer Dabbelt <palmer@dabbelt.com>
Cc: Paul Walmsley <paul.walmsley@sifive.com>
Cc: Peter Xu <peterx@redhat.com>
Cc: Qi Zheng <zhengqi.arch@bytedance.com>
Cc: Ryan Roberts <ryan.roberts@arm.com>
Cc: SeongJae Park <sj@kernel.org>
Cc: Sven Schnelle <svens@linux.ibm.com>
Cc: Uladzislau Rezki (Sony) <urezki@gmail.com>
Cc: Vasily Gorbik <gor@linux.ibm.com>
Cc: Will Deacon <will@kernel.org>
Cc: <stable@vger.kernel.org> [6.5+]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2023-09-22 12:58:03 +01:00
|
|
|
|
|
|
|
|
|
2023-12-20 23:44:29 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-12-20 23:45:02 +01:00
|
|
|
|
2023-12-20 23:44:29 +01:00
|
|
|
|
mm: remember exclusively mapped anonymous pages with PG_anon_exclusive
Let's mark exclusively mapped anonymous pages with PG_anon_exclusive as
exclusive, and use that information to make GUP pins reliable and stay
consistent with the page mapped into the page table even if the page table
entry gets write-protected.
With that information at hand, we can extend our COW logic to always reuse
anonymous pages that are exclusive. For anonymous pages that might be
shared, the existing logic applies.
As already documented, PG_anon_exclusive is usually only expressive in
combination with a page table entry. Especially PTE vs. PMD-mapped
anonymous pages require more thought, some examples: due to mremap() we
can easily have a single compound page PTE-mapped into multiple page
tables exclusively in a single process -- multiple page table locks apply.
Further, due to MADV_WIPEONFORK we might not necessarily write-protect
all PTEs, and only some subpages might be pinned. Long story short: once
PTE-mapped, we have to track information about exclusivity per sub-page,
but until then, we can just track it for the compound page in the head
page and not having to update a whole bunch of subpages all of the time
for a simple PMD mapping of a THP.
For simplicity, this commit mostly talks about "anonymous pages", while
it's for THP actually "the part of an anonymous folio referenced via a
page table entry".
To not spill PG_anon_exclusive code all over the mm code-base, we let the
anon rmap code to handle all PG_anon_exclusive logic it can easily handle.
If a writable, present page table entry points at an anonymous (sub)page,
that (sub)page must be PG_anon_exclusive. If GUP wants to take a reliably
pin (FOLL_PIN) on an anonymous page references via a present page table
entry, it must only pin if PG_anon_exclusive is set for the mapped
(sub)page.
This commit doesn't adjust GUP, so this is only implicitly handled for
FOLL_WRITE, follow-up commits will teach GUP to also respect it for
FOLL_PIN without FOLL_WRITE, to make all GUP pins of anonymous pages fully
reliable.
Whenever an anonymous page is to be shared (fork(), KSM), or when
temporarily unmapping an anonymous page (swap, migration), the relevant
PG_anon_exclusive bit has to be cleared to mark the anonymous page
possibly shared. Clearing will fail if there are GUP pins on the page:
* For fork(), this means having to copy the page and not being able to
share it. fork() protects against concurrent GUP using the PT lock and
the src_mm->write_protect_seq.
* For KSM, this means sharing will fail. For swap this means, unmapping
will fail, For migration this means, migration will fail early. All
three cases protect against concurrent GUP using the PT lock and a
proper clear/invalidate+flush of the relevant page table entry.
This fixes memory corruptions reported for FOLL_PIN | FOLL_WRITE, when a
pinned page gets mapped R/O and the successive write fault ends up
replacing the page instead of reusing it. It improves the situation for
O_DIRECT/vmsplice/... that still use FOLL_GET instead of FOLL_PIN, if
fork() is *not* involved, however swapout and fork() are still
problematic. Properly using FOLL_PIN instead of FOLL_GET for these GUP
users will fix the issue for them.
I. Details about basic handling
I.1. Fresh anonymous pages
page_add_new_anon_rmap() and hugepage_add_new_anon_rmap() will mark the
given page exclusive via __page_set_anon_rmap(exclusive=1). As that is
the mechanism fresh anonymous pages come into life (besides migration code
where we copy the page->mapping), all fresh anonymous pages will start out
as exclusive.
I.2. COW reuse handling of anonymous pages
When a COW handler stumbles over a (sub)page that's marked exclusive, it
simply reuses it. Otherwise, the handler tries harder under page lock to
detect if the (sub)page is exclusive and can be reused. If exclusive,
page_move_anon_rmap() will mark the given (sub)page exclusive.
Note that hugetlb code does not yet check for PageAnonExclusive(), as it
still uses the old COW logic that is prone to the COW security issue
because hugetlb code cannot really tolerate unnecessary/wrong COW as huge
pages are a scarce resource.
I.3. Migration handling
try_to_migrate() has to try marking an exclusive anonymous page shared via
page_try_share_anon_rmap(). If it fails because there are GUP pins on the
page, unmap fails. migrate_vma_collect_pmd() and
__split_huge_pmd_locked() are handled similarly.
Writable migration entries implicitly point at shared anonymous pages.
For readable migration entries that information is stored via a new
"readable-exclusive" migration entry, specific to anonymous pages.
When restoring a migration entry in remove_migration_pte(), information
about exlusivity is detected via the migration entry type, and
RMAP_EXCLUSIVE is set accordingly for
page_add_anon_rmap()/hugepage_add_anon_rmap() to restore that information.
I.4. Swapout handling
try_to_unmap() has to try marking the mapped page possibly shared via
page_try_share_anon_rmap(). If it fails because there are GUP pins on the
page, unmap fails. For now, information about exclusivity is lost. In
the future, we might want to remember that information in the swap entry
in some cases, however, it requires more thought, care, and a way to store
that information in swap entries.
I.5. Swapin handling
do_swap_page() will never stumble over exclusive anonymous pages in the
swap cache, as try_to_migrate() prohibits that. do_swap_page() always has
to detect manually if an anonymous page is exclusive and has to set
RMAP_EXCLUSIVE for page_add_anon_rmap() accordingly.
I.6. THP handling
__split_huge_pmd_locked() has to move the information about exclusivity
from the PMD to the PTEs.
a) In case we have a readable-exclusive PMD migration entry, simply
insert readable-exclusive PTE migration entries.
b) In case we have a present PMD entry and we don't want to freeze
("convert to migration entries"), simply forward PG_anon_exclusive to
all sub-pages, no need to temporarily clear the bit.
c) In case we have a present PMD entry and want to freeze, handle it
similar to try_to_migrate(): try marking the page shared first. In
case we fail, we ignore the "freeze" instruction and simply split
ordinarily. try_to_migrate() will properly fail because the THP is
still mapped via PTEs.
When splitting a compound anonymous folio (THP), the information about
exclusivity is implicitly handled via the migration entries: no need to
replicate PG_anon_exclusive manually.
I.7. fork() handling fork() handling is relatively easy, because
PG_anon_exclusive is only expressive for some page table entry types.
a) Present anonymous pages
page_try_dup_anon_rmap() will mark the given subpage shared -- which will
fail if the page is pinned. If it failed, we have to copy (or PTE-map a
PMD to handle it on the PTE level).
Note that device exclusive entries are just a pointer at a PageAnon()
page. fork() will first convert a device exclusive entry to a present
page table and handle it just like present anonymous pages.
b) Device private entry
Device private entries point at PageAnon() pages that cannot be mapped
directly and, therefore, cannot get pinned.
page_try_dup_anon_rmap() will mark the given subpage shared, which cannot
fail because they cannot get pinned.
c) HW poison entries
PG_anon_exclusive will remain untouched and is stale -- the page table
entry is just a placeholder after all.
d) Migration entries
Writable and readable-exclusive entries are converted to readable entries:
possibly shared.
I.8. mprotect() handling
mprotect() only has to properly handle the new readable-exclusive
migration entry:
When write-protecting a migration entry that points at an anonymous page,
remember the information about exclusivity via the "readable-exclusive"
migration entry type.
II. Migration and GUP-fast
Whenever replacing a present page table entry that maps an exclusive
anonymous page by a migration entry, we have to mark the page possibly
shared and synchronize against GUP-fast by a proper clear/invalidate+flush
to make the following scenario impossible:
1. try_to_migrate() places a migration entry after checking for GUP pins
and marks the page possibly shared.
2. GUP-fast pins the page due to lack of synchronization
3. fork() converts the "writable/readable-exclusive" migration entry into a
readable migration entry
4. Migration fails due to the GUP pin (failing to freeze the refcount)
5. Migration entries are restored. PG_anon_exclusive is lost
-> We have a pinned page that is not marked exclusive anymore.
Note that we move information about exclusivity from the page to the
migration entry as it otherwise highly overcomplicates fork() and
PTE-mapping a THP.
III. Swapout and GUP-fast
Whenever replacing a present page table entry that maps an exclusive
anonymous page by a swap entry, we have to mark the page possibly shared
and synchronize against GUP-fast by a proper clear/invalidate+flush to
make the following scenario impossible:
1. try_to_unmap() places a swap entry after checking for GUP pins and
clears exclusivity information on the page.
2. GUP-fast pins the page due to lack of synchronization.
-> We have a pinned page that is not marked exclusive anymore.
If we'd ever store information about exclusivity in the swap entry,
similar to migration handling, the same considerations as in II would
apply. This is future work.
Link: https://lkml.kernel.org/r/20220428083441.37290-13-david@redhat.com
Signed-off-by: David Hildenbrand <david@redhat.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Christoph Hellwig <hch@lst.de>
Cc: David Rientjes <rientjes@google.com>
Cc: Don Dutile <ddutile@redhat.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Jan Kara <jack@suse.cz>
Cc: Jann Horn <jannh@google.com>
Cc: Jason Gunthorpe <jgg@nvidia.com>
Cc: John Hubbard <jhubbard@nvidia.com>
Cc: Khalid Aziz <khalid.aziz@oracle.com>
Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Liang Zhang <zhangliang5@huawei.com>
Cc: "Matthew Wilcox (Oracle)" <willy@infradead.org>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Mike Rapoport <rppt@linux.ibm.com>
Cc: Nadav Amit <namit@vmware.com>
Cc: Oded Gabbay <oded.gabbay@gmail.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Pedro Demarchi Gomes <pedrodemargomes@gmail.com>
Cc: Peter Xu <peterx@redhat.com>
Cc: Rik van Riel <riel@surriel.com>
Cc: Roman Gushchin <guro@fb.com>
Cc: Shakeel Butt <shakeelb@google.com>
Cc: Yang Shi <shy828301@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2022-05-09 18:20:44 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-06-30 18:54:16 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-02-10 20:37:53 +01:00
|
|
|
|
2021-06-30 18:54:16 -07:00
|
|
|
|
|
|
|
|
|
mm: remember exclusively mapped anonymous pages with PG_anon_exclusive
Let's mark exclusively mapped anonymous pages with PG_anon_exclusive as
exclusive, and use that information to make GUP pins reliable and stay
consistent with the page mapped into the page table even if the page table
entry gets write-protected.
With that information at hand, we can extend our COW logic to always reuse
anonymous pages that are exclusive. For anonymous pages that might be
shared, the existing logic applies.
As already documented, PG_anon_exclusive is usually only expressive in
combination with a page table entry. Especially PTE vs. PMD-mapped
anonymous pages require more thought, some examples: due to mremap() we
can easily have a single compound page PTE-mapped into multiple page
tables exclusively in a single process -- multiple page table locks apply.
Further, due to MADV_WIPEONFORK we might not necessarily write-protect
all PTEs, and only some subpages might be pinned. Long story short: once
PTE-mapped, we have to track information about exclusivity per sub-page,
but until then, we can just track it for the compound page in the head
page and not having to update a whole bunch of subpages all of the time
for a simple PMD mapping of a THP.
For simplicity, this commit mostly talks about "anonymous pages", while
it's for THP actually "the part of an anonymous folio referenced via a
page table entry".
To not spill PG_anon_exclusive code all over the mm code-base, we let the
anon rmap code to handle all PG_anon_exclusive logic it can easily handle.
If a writable, present page table entry points at an anonymous (sub)page,
that (sub)page must be PG_anon_exclusive. If GUP wants to take a reliably
pin (FOLL_PIN) on an anonymous page references via a present page table
entry, it must only pin if PG_anon_exclusive is set for the mapped
(sub)page.
This commit doesn't adjust GUP, so this is only implicitly handled for
FOLL_WRITE, follow-up commits will teach GUP to also respect it for
FOLL_PIN without FOLL_WRITE, to make all GUP pins of anonymous pages fully
reliable.
Whenever an anonymous page is to be shared (fork(), KSM), or when
temporarily unmapping an anonymous page (swap, migration), the relevant
PG_anon_exclusive bit has to be cleared to mark the anonymous page
possibly shared. Clearing will fail if there are GUP pins on the page:
* For fork(), this means having to copy the page and not being able to
share it. fork() protects against concurrent GUP using the PT lock and
the src_mm->write_protect_seq.
* For KSM, this means sharing will fail. For swap this means, unmapping
will fail, For migration this means, migration will fail early. All
three cases protect against concurrent GUP using the PT lock and a
proper clear/invalidate+flush of the relevant page table entry.
This fixes memory corruptions reported for FOLL_PIN | FOLL_WRITE, when a
pinned page gets mapped R/O and the successive write fault ends up
replacing the page instead of reusing it. It improves the situation for
O_DIRECT/vmsplice/... that still use FOLL_GET instead of FOLL_PIN, if
fork() is *not* involved, however swapout and fork() are still
problematic. Properly using FOLL_PIN instead of FOLL_GET for these GUP
users will fix the issue for them.
I. Details about basic handling
I.1. Fresh anonymous pages
page_add_new_anon_rmap() and hugepage_add_new_anon_rmap() will mark the
given page exclusive via __page_set_anon_rmap(exclusive=1). As that is
the mechanism fresh anonymous pages come into life (besides migration code
where we copy the page->mapping), all fresh anonymous pages will start out
as exclusive.
I.2. COW reuse handling of anonymous pages
When a COW handler stumbles over a (sub)page that's marked exclusive, it
simply reuses it. Otherwise, the handler tries harder under page lock to
detect if the (sub)page is exclusive and can be reused. If exclusive,
page_move_anon_rmap() will mark the given (sub)page exclusive.
Note that hugetlb code does not yet check for PageAnonExclusive(), as it
still uses the old COW logic that is prone to the COW security issue
because hugetlb code cannot really tolerate unnecessary/wrong COW as huge
pages are a scarce resource.
I.3. Migration handling
try_to_migrate() has to try marking an exclusive anonymous page shared via
page_try_share_anon_rmap(). If it fails because there are GUP pins on the
page, unmap fails. migrate_vma_collect_pmd() and
__split_huge_pmd_locked() are handled similarly.
Writable migration entries implicitly point at shared anonymous pages.
For readable migration entries that information is stored via a new
"readable-exclusive" migration entry, specific to anonymous pages.
When restoring a migration entry in remove_migration_pte(), information
about exlusivity is detected via the migration entry type, and
RMAP_EXCLUSIVE is set accordingly for
page_add_anon_rmap()/hugepage_add_anon_rmap() to restore that information.
I.4. Swapout handling
try_to_unmap() has to try marking the mapped page possibly shared via
page_try_share_anon_rmap(). If it fails because there are GUP pins on the
page, unmap fails. For now, information about exclusivity is lost. In
the future, we might want to remember that information in the swap entry
in some cases, however, it requires more thought, care, and a way to store
that information in swap entries.
I.5. Swapin handling
do_swap_page() will never stumble over exclusive anonymous pages in the
swap cache, as try_to_migrate() prohibits that. do_swap_page() always has
to detect manually if an anonymous page is exclusive and has to set
RMAP_EXCLUSIVE for page_add_anon_rmap() accordingly.
I.6. THP handling
__split_huge_pmd_locked() has to move the information about exclusivity
from the PMD to the PTEs.
a) In case we have a readable-exclusive PMD migration entry, simply
insert readable-exclusive PTE migration entries.
b) In case we have a present PMD entry and we don't want to freeze
("convert to migration entries"), simply forward PG_anon_exclusive to
all sub-pages, no need to temporarily clear the bit.
c) In case we have a present PMD entry and want to freeze, handle it
similar to try_to_migrate(): try marking the page shared first. In
case we fail, we ignore the "freeze" instruction and simply split
ordinarily. try_to_migrate() will properly fail because the THP is
still mapped via PTEs.
When splitting a compound anonymous folio (THP), the information about
exclusivity is implicitly handled via the migration entries: no need to
replicate PG_anon_exclusive manually.
I.7. fork() handling fork() handling is relatively easy, because
PG_anon_exclusive is only expressive for some page table entry types.
a) Present anonymous pages
page_try_dup_anon_rmap() will mark the given subpage shared -- which will
fail if the page is pinned. If it failed, we have to copy (or PTE-map a
PMD to handle it on the PTE level).
Note that device exclusive entries are just a pointer at a PageAnon()
page. fork() will first convert a device exclusive entry to a present
page table and handle it just like present anonymous pages.
b) Device private entry
Device private entries point at PageAnon() pages that cannot be mapped
directly and, therefore, cannot get pinned.
page_try_dup_anon_rmap() will mark the given subpage shared, which cannot
fail because they cannot get pinned.
c) HW poison entries
PG_anon_exclusive will remain untouched and is stale -- the page table
entry is just a placeholder after all.
d) Migration entries
Writable and readable-exclusive entries are converted to readable entries:
possibly shared.
I.8. mprotect() handling
mprotect() only has to properly handle the new readable-exclusive
migration entry:
When write-protecting a migration entry that points at an anonymous page,
remember the information about exclusivity via the "readable-exclusive"
migration entry type.
II. Migration and GUP-fast
Whenever replacing a present page table entry that maps an exclusive
anonymous page by a migration entry, we have to mark the page possibly
shared and synchronize against GUP-fast by a proper clear/invalidate+flush
to make the following scenario impossible:
1. try_to_migrate() places a migration entry after checking for GUP pins
and marks the page possibly shared.
2. GUP-fast pins the page due to lack of synchronization
3. fork() converts the "writable/readable-exclusive" migration entry into a
readable migration entry
4. Migration fails due to the GUP pin (failing to freeze the refcount)
5. Migration entries are restored. PG_anon_exclusive is lost
-> We have a pinned page that is not marked exclusive anymore.
Note that we move information about exclusivity from the page to the
migration entry as it otherwise highly overcomplicates fork() and
PTE-mapping a THP.
III. Swapout and GUP-fast
Whenever replacing a present page table entry that maps an exclusive
anonymous page by a swap entry, we have to mark the page possibly shared
and synchronize against GUP-fast by a proper clear/invalidate+flush to
make the following scenario impossible:
1. try_to_unmap() places a swap entry after checking for GUP pins and
clears exclusivity information on the page.
2. GUP-fast pins the page due to lack of synchronization.
-> We have a pinned page that is not marked exclusive anymore.
If we'd ever store information about exclusivity in the swap entry,
similar to migration handling, the same considerations as in II would
apply. This is future work.
Link: https://lkml.kernel.org/r/20220428083441.37290-13-david@redhat.com
Signed-off-by: David Hildenbrand <david@redhat.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Christoph Hellwig <hch@lst.de>
Cc: David Rientjes <rientjes@google.com>
Cc: Don Dutile <ddutile@redhat.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Jan Kara <jack@suse.cz>
Cc: Jann Horn <jannh@google.com>
Cc: Jason Gunthorpe <jgg@nvidia.com>
Cc: John Hubbard <jhubbard@nvidia.com>
Cc: Khalid Aziz <khalid.aziz@oracle.com>
Cc: "Kirill A. Shutemov" <kirill.shutemov@linux.intel.com>
Cc: Liang Zhang <zhangliang5@huawei.com>
Cc: "Matthew Wilcox (Oracle)" <willy@infradead.org>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Mike Rapoport <rppt@linux.ibm.com>
Cc: Nadav Amit <namit@vmware.com>
Cc: Oded Gabbay <oded.gabbay@gmail.com>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Pedro Demarchi Gomes <pedrodemargomes@gmail.com>
Cc: Peter Xu <peterx@redhat.com>
Cc: Rik van Riel <riel@surriel.com>
Cc: Roman Gushchin <guro@fb.com>
Cc: Shakeel Butt <shakeelb@google.com>
Cc: Yang Shi <shy828301@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2022-05-09 18:20:44 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-06-30 18:54:16 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-02-10 20:37:53 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
mm: rmap: fix CONT-PTE/PMD size hugetlb issue when migration
On some architectures (like ARM64), it can support CONT-PTE/PMD size
hugetlb, which means it can support not only PMD/PUD size hugetlb: 2M and
1G, but also CONT-PTE/PMD size: 64K and 32M if a 4K page size specified.
When migrating a hugetlb page, we will get the relevant page table entry
by huge_pte_offset() only once to nuke it and remap it with a migration
pte entry. This is correct for PMD or PUD size hugetlb, since they always
contain only one pmd entry or pud entry in the page table.
However this is incorrect for CONT-PTE and CONT-PMD size hugetlb, since
they can contain several continuous pte or pmd entry with same page table
attributes. So we will nuke or remap only one pte or pmd entry for this
CONT-PTE/PMD size hugetlb page, which is not expected for hugetlb
migration. The problem is we can still continue to modify the subpages'
data of a hugetlb page during migrating a hugetlb page, which can cause a
serious data consistent issue, since we did not nuke the page table entry
and set a migration pte for the subpages of a hugetlb page.
To fix this issue, we should change to use huge_ptep_clear_flush() to nuke
a hugetlb page table, and remap it with set_huge_pte_at() and
set_huge_swap_pte_at() when migrating a hugetlb page, which already
considered the CONT-PTE or CONT-PMD size hugetlb.
[akpm@linux-foundation.org: fix nommu build]
[baolin.wang@linux.alibaba.com: fix build errors for !CONFIG_MMU]
Link: https://lkml.kernel.org/r/a4baca670aca637e7198d9ae4543b8873cb224dc.1652270205.git.baolin.wang@linux.alibaba.com
Link: https://lkml.kernel.org/r/ea5abf529f0997b5430961012bfda6166c1efc8c.1652147571.git.baolin.wang@linux.alibaba.com
Signed-off-by: Baolin Wang <baolin.wang@linux.alibaba.com>
Reviewed-by: Muchun Song <songmuchun@bytedance.com>
Reviewed-by: Mike Kravetz <mike.kravetz@oracle.com>
Acked-by: David Hildenbrand <david@redhat.com>
Cc: Alexander Gordeev <agordeev@linux.ibm.com>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Christian Borntraeger <borntraeger@linux.ibm.com>
Cc: David S. Miller <davem@davemloft.net>
Cc: Gerald Schaefer <gerald.schaefer@linux.ibm.com>
Cc: Heiko Carstens <hca@linux.ibm.com>
Cc: Helge Deller <deller@gmx.de>
Cc: James Bottomley <James.Bottomley@HansenPartnership.com>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Rich Felker <dalias@libc.org>
Cc: Sven Schnelle <svens@linux.ibm.com>
Cc: Thomas Bogendoerfer <tsbogend@alpha.franken.de>
Cc: Vasily Gorbik <gor@linux.ibm.com>
Cc: Will Deacon <will@kernel.org>
Cc: Yoshinori Sato <ysato@users.osdn.me>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2022-05-13 16:48:55 -07:00
|
|
|
|
mm: hugetlb: add huge page size param to set_huge_pte_at()
Patch series "Fix set_huge_pte_at() panic on arm64", v2.
This series fixes a bug in arm64's implementation of set_huge_pte_at(),
which can result in an unprivileged user causing a kernel panic. The
problem was triggered when running the new uffd poison mm selftest for
HUGETLB memory. This test (and the uffd poison feature) was merged for
v6.5-rc7.
Ideally, I'd like to get this fix in for v6.6 and I've cc'ed stable
(correctly this time) to get it backported to v6.5, where the issue first
showed up.
Description of Bug
==================
arm64's huge pte implementation supports multiple huge page sizes, some of
which are implemented in the page table with multiple contiguous entries.
So set_huge_pte_at() needs to work out how big the logical pte is, so that
it can also work out how many physical ptes (or pmds) need to be written.
It previously did this by grabbing the folio out of the pte and querying
its size.
However, there are cases when the pte being set is actually a swap entry.
But this also used to work fine, because for huge ptes, we only ever saw
migration entries and hwpoison entries. And both of these types of swap
entries have a PFN embedded, so the code would grab that and everything
still worked out.
But over time, more calls to set_huge_pte_at() have been added that set
swap entry types that do not embed a PFN. And this causes the code to go
bang. The triggering case is for the uffd poison test, commit
99aa77215ad0 ("selftests/mm: add uffd unit test for UFFDIO_POISON"), which
causes a PTE_MARKER_POISONED swap entry to be set, coutesey of commit
8a13897fb0da ("mm: userfaultfd: support UFFDIO_POISON for hugetlbfs") -
added in v6.5-rc7. Although review shows that there are other call sites
that set PTE_MARKER_UFFD_WP (which also has no PFN), these don't trigger
on arm64 because arm64 doesn't support UFFD WP.
If CONFIG_DEBUG_VM is enabled, we do at least get a BUG(), but otherwise,
it will dereference a bad pointer in page_folio():
static inline struct folio *hugetlb_swap_entry_to_folio(swp_entry_t entry)
{
VM_BUG_ON(!is_migration_entry(entry) && !is_hwpoison_entry(entry));
return page_folio(pfn_to_page(swp_offset_pfn(entry)));
}
Fix
===
The simplest fix would have been to revert the dodgy cleanup commit
18f3962953e4 ("mm: hugetlb: kill set_huge_swap_pte_at()"), but since
things have moved on, this would have required an audit of all the new
set_huge_pte_at() call sites to see if they should be converted to
set_huge_swap_pte_at(). As per the original intent of the change, it
would also leave us open to future bugs when people invariably get it
wrong and call the wrong helper.
So instead, I've added a huge page size parameter to set_huge_pte_at().
This means that the arm64 code has the size in all cases. It's a bigger
change, due to needing to touch the arches that implement the function,
but it is entirely mechanical, so in my view, low risk.
I've compile-tested all touched arches; arm64, parisc, powerpc, riscv,
s390, sparc (and additionally x86_64). I've additionally booted and run
mm selftests against arm64, where I observe the uffd poison test is fixed,
and there are no other regressions.
This patch (of 2):
In order to fix a bug, arm64 needs to be told the size of the huge page
for which the pte is being set in set_huge_pte_at(). Provide for this by
adding an `unsigned long sz` parameter to the function. This follows the
same pattern as huge_pte_clear().
This commit makes the required interface modifications to the core mm as
well as all arches that implement this function (arm64, parisc, powerpc,
riscv, s390, sparc). The actual arm64 bug will be fixed in a separate
commit.
No behavioral changes intended.
Link: https://lkml.kernel.org/r/20230922115804.2043771-1-ryan.roberts@arm.com
Link: https://lkml.kernel.org/r/20230922115804.2043771-2-ryan.roberts@arm.com
Fixes: 8a13897fb0da ("mm: userfaultfd: support UFFDIO_POISON for hugetlbfs")
Signed-off-by: Ryan Roberts <ryan.roberts@arm.com>
Reviewed-by: Christophe Leroy <christophe.leroy@csgroup.eu> [powerpc 8xx]
Reviewed-by: Lorenzo Stoakes <lstoakes@gmail.com> [vmalloc change]
Cc: Alexandre Ghiti <alex@ghiti.fr>
Cc: Albert Ou <aou@eecs.berkeley.edu>
Cc: Alexander Gordeev <agordeev@linux.ibm.com>
Cc: Anshuman Khandual <anshuman.khandual@arm.com>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Axel Rasmussen <axelrasmussen@google.com>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Christian Borntraeger <borntraeger@linux.ibm.com>
Cc: Christoph Hellwig <hch@infradead.org>
Cc: David S. Miller <davem@davemloft.net>
Cc: Gerald Schaefer <gerald.schaefer@linux.ibm.com>
Cc: Heiko Carstens <hca@linux.ibm.com>
Cc: Helge Deller <deller@gmx.de>
Cc: "James E.J. Bottomley" <James.Bottomley@HansenPartnership.com>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Muchun Song <muchun.song@linux.dev>
Cc: Nicholas Piggin <npiggin@gmail.com>
Cc: Palmer Dabbelt <palmer@dabbelt.com>
Cc: Paul Walmsley <paul.walmsley@sifive.com>
Cc: Peter Xu <peterx@redhat.com>
Cc: Qi Zheng <zhengqi.arch@bytedance.com>
Cc: Ryan Roberts <ryan.roberts@arm.com>
Cc: SeongJae Park <sj@kernel.org>
Cc: Sven Schnelle <svens@linux.ibm.com>
Cc: Uladzislau Rezki (Sony) <urezki@gmail.com>
Cc: Vasily Gorbik <gor@linux.ibm.com>
Cc: Will Deacon <will@kernel.org>
Cc: <stable@vger.kernel.org> [6.5+]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2023-09-22 12:58:03 +01:00
|
|
|
|
|
|
|
|
|
mm: rmap: fix CONT-PTE/PMD size hugetlb issue when migration
On some architectures (like ARM64), it can support CONT-PTE/PMD size
hugetlb, which means it can support not only PMD/PUD size hugetlb: 2M and
1G, but also CONT-PTE/PMD size: 64K and 32M if a 4K page size specified.
When migrating a hugetlb page, we will get the relevant page table entry
by huge_pte_offset() only once to nuke it and remap it with a migration
pte entry. This is correct for PMD or PUD size hugetlb, since they always
contain only one pmd entry or pud entry in the page table.
However this is incorrect for CONT-PTE and CONT-PMD size hugetlb, since
they can contain several continuous pte or pmd entry with same page table
attributes. So we will nuke or remap only one pte or pmd entry for this
CONT-PTE/PMD size hugetlb page, which is not expected for hugetlb
migration. The problem is we can still continue to modify the subpages'
data of a hugetlb page during migrating a hugetlb page, which can cause a
serious data consistent issue, since we did not nuke the page table entry
and set a migration pte for the subpages of a hugetlb page.
To fix this issue, we should change to use huge_ptep_clear_flush() to nuke
a hugetlb page table, and remap it with set_huge_pte_at() and
set_huge_swap_pte_at() when migrating a hugetlb page, which already
considered the CONT-PTE or CONT-PMD size hugetlb.
[akpm@linux-foundation.org: fix nommu build]
[baolin.wang@linux.alibaba.com: fix build errors for !CONFIG_MMU]
Link: https://lkml.kernel.org/r/a4baca670aca637e7198d9ae4543b8873cb224dc.1652270205.git.baolin.wang@linux.alibaba.com
Link: https://lkml.kernel.org/r/ea5abf529f0997b5430961012bfda6166c1efc8c.1652147571.git.baolin.wang@linux.alibaba.com
Signed-off-by: Baolin Wang <baolin.wang@linux.alibaba.com>
Reviewed-by: Muchun Song <songmuchun@bytedance.com>
Reviewed-by: Mike Kravetz <mike.kravetz@oracle.com>
Acked-by: David Hildenbrand <david@redhat.com>
Cc: Alexander Gordeev <agordeev@linux.ibm.com>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Christian Borntraeger <borntraeger@linux.ibm.com>
Cc: David S. Miller <davem@davemloft.net>
Cc: Gerald Schaefer <gerald.schaefer@linux.ibm.com>
Cc: Heiko Carstens <hca@linux.ibm.com>
Cc: Helge Deller <deller@gmx.de>
Cc: James Bottomley <James.Bottomley@HansenPartnership.com>
Cc: Michael Ellerman <mpe@ellerman.id.au>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Rich Felker <dalias@libc.org>
Cc: Sven Schnelle <svens@linux.ibm.com>
Cc: Thomas Bogendoerfer <tsbogend@alpha.franken.de>
Cc: Vasily Gorbik <gor@linux.ibm.com>
Cc: Will Deacon <will@kernel.org>
Cc: Yoshinori Sato <ysato@users.osdn.me>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2022-05-13 16:48:55 -07:00
|
|
|
|
|
|
|
|
|
2022-03-24 18:10:01 -07:00
|
|
|
|
2024-02-15 20:53:05 +00:00
|
|
|
|
2021-06-30 18:54:16 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-12-20 23:44:26 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-12-20 23:44:54 +01:00
|
|
|
|
2022-02-14 18:38:47 -08:00
|
|
|
|
2023-01-12 12:39:31 +00:00
|
|
|
|
2022-01-28 14:29:43 -05:00
|
|
|
|
2021-06-30 18:54:16 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-01-28 14:29:43 -05:00
|
|
|
|
2021-06-30 18:54:16 -07:00
|
|
|
|
|
|
|
|
|
2022-01-28 14:29:43 -05:00
|
|
|
|
|
|
|
|
|
2021-06-30 18:54:16 -07:00
|
|
|
|
2022-01-28 14:29:43 -05:00
|
|
|
|
2021-06-30 18:54:16 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-09-27 14:38:26 +08:00
|
|
|
|
2022-01-29 16:06:53 -05:00
|
|
|
|
2021-06-30 18:54:16 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2023-02-13 20:34:43 +08:00
|
|
|
|
2021-06-30 18:54:16 -07:00
|
|
|
|
|
|
|
|
|
2023-02-13 20:34:43 +08:00
|
|
|
|
2021-06-30 18:54:16 -07:00
|
|
|
|
|
|
|
|
|
2022-07-15 10:05:10 -05:00
|
|
|
|
|
|
|
|
|
2021-07-07 13:13:33 -07:00
|
|
|
|
|
|
|
|
|
2014-01-21 15:49:50 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-01-28 14:29:43 -05:00
|
|
|
|
2014-01-21 15:49:50 -08:00
|
|
|
|
|
|
|
|
|
2016-03-17 14:20:04 -07:00
|
|
|
|
2022-01-29 16:06:53 -05:00
|
|
|
|
2016-03-17 14:20:04 -07:00
|
|
|
|
2022-01-29 16:06:53 -05:00
|
|
|
|
mlock: mlocked pages are unevictable
Make sure that mlocked pages also live on the unevictable LRU, so kswapd
will not scan them over and over again.
This is achieved through various strategies:
1) add yet another page flag--PG_mlocked--to indicate that
the page is locked for efficient testing in vmscan and,
optionally, fault path. This allows early culling of
unevictable pages, preventing them from getting to
page_referenced()/try_to_unmap(). Also allows separate
accounting of mlock'd pages, as Nick's original patch
did.
Note: Nick's original mlock patch used a PG_mlocked
flag. I had removed this in favor of the PG_unevictable
flag + an mlock_count [new page struct member]. I
restored the PG_mlocked flag to eliminate the new
count field.
2) add the mlock/unevictable infrastructure to mm/mlock.c,
with internal APIs in mm/internal.h. This is a rework
of Nick's original patch to these files, taking into
account that mlocked pages are now kept on unevictable
LRU list.
3) update vmscan.c:page_evictable() to check PageMlocked()
and, if vma passed in, the vm_flags. Note that the vma
will only be passed in for new pages in the fault path;
and then only if the "cull unevictable pages in fault
path" patch is included.
4) add try_to_unlock() to rmap.c to walk a page's rmap and
ClearPageMlocked() if no other vmas have it mlocked.
Reuses as much of try_to_unmap() as possible. This
effectively replaces the use of one of the lru list links
as an mlock count. If this mechanism let's pages in mlocked
vmas leak through w/o PG_mlocked set [I don't know that it
does], we should catch them later in try_to_unmap(). One
hopes this will be rare, as it will be relatively expensive.
Original mm/internal.h, mm/rmap.c and mm/mlock.c changes:
Signed-off-by: Nick Piggin <npiggin@suse.de>
splitlru: introduce __get_user_pages():
New munlock processing need to GUP_FLAGS_IGNORE_VMA_PERMISSIONS.
because current get_user_pages() can't grab PROT_NONE pages theresore it
cause PROT_NONE pages can't munlock.
[akpm@linux-foundation.org: fix this for pagemap-pass-mm-into-pagewalkers.patch]
[akpm@linux-foundation.org: untangle patch interdependencies]
[akpm@linux-foundation.org: fix things after out-of-order merging]
[hugh@veritas.com: fix page-flags mess]
[lee.schermerhorn@hp.com: fix munlock page table walk - now requires 'mm']
[kosaki.motohiro@jp.fujitsu.com: build fix]
[kosaki.motohiro@jp.fujitsu.com: fix truncate race and sevaral comments]
[kosaki.motohiro@jp.fujitsu.com: splitlru: introduce __get_user_pages()]
Signed-off-by: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Signed-off-by: Rik van Riel <riel@redhat.com>
Signed-off-by: Lee Schermerhorn <lee.schermerhorn@hp.com>
Cc: Nick Piggin <npiggin@suse.de>
Cc: Dave Hansen <dave@linux.vnet.ibm.com>
Cc: Matt Mackall <mpm@selenic.com>
Signed-off-by: Hugh Dickins <hugh@veritas.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2008-10-18 20:26:44 -07:00
|
|
|
|
ksm: rmap_walk to remove_migation_ptes
A side-effect of making ksm pages swappable is that they have to be placed
on the LRUs: which then exposes them to isolate_lru_page() and hence to
page migration.
Add rmap_walk() for remove_migration_ptes() to use: rmap_walk_anon() and
rmap_walk_file() in rmap.c, but rmap_walk_ksm() in ksm.c. Perhaps some
consolidation with existing code is possible, but don't attempt that yet
(try_to_unmap needs to handle nonlinears, but migration pte removal does
not).
rmap_walk() is sadly less general than it appears: rmap_walk_anon(), like
remove_anon_migration_ptes() which it replaces, avoids calling
page_lock_anon_vma(), because that includes a page_mapped() test which
fails when all migration ptes are in place. That was valid when NUMA page
migration was introduced (holding mmap_sem provided the missing guarantee
that anon_vma's slab had not already been destroyed), but I believe not
valid in the memory hotremove case added since.
For now do the same as before, and consider the best way to fix that
unlikely race later on. When fixed, we can probably use rmap_walk() on
hwpoisoned ksm pages too: for now, they remain among hwpoison's various
exceptions (its PageKsm test comes before the page is locked, but its
page_lock_anon_vma fails safely if an anon gets upgraded).
Signed-off-by: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Cc: Izik Eidus <ieidus@redhat.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Chris Wright <chrisw@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-12-14 17:59:31 -08:00
|
|
|
|
2021-06-30 18:54:25 -07:00
|
|
|
|
|
|
|
|
|
2025-02-10 20:37:45 +01:00
|
|
|
|
2022-05-09 18:20:54 -07:00
|
|
|
|
2025-02-10 20:37:45 +01:00
|
|
|
|
2021-06-30 18:54:25 -07:00
|
|
|
|
2025-02-10 20:37:45 +01:00
|
|
|
|
2021-06-30 18:54:25 -07:00
|
|
|
|
2025-02-10 20:37:45 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-06-30 18:54:25 -07:00
|
|
|
|
2025-02-10 20:37:45 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-06-30 18:54:25 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-02-10 20:37:45 +01:00
|
|
|
|
2021-06-30 18:54:25 -07:00
|
|
|
|
2025-02-10 20:37:45 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-06-30 18:54:25 -07:00
|
|
|
|
2025-02-10 20:37:45 +01:00
|
|
|
|
|
|
|
|
|
2021-06-30 18:54:25 -07:00
|
|
|
|
2025-02-10 20:37:46 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-02-10 20:37:45 +01:00
|
|
|
|
2025-02-10 20:37:46 +01:00
|
|
|
|
|
|
|
|
|
mm/rmap: avoid -EBUSY from make_device_exclusive()
Failing to obtain the folio lock, for example because the folio is
concurrently getting migrated or swapped out, can easily make the callers
fail: for example, the hmm selftest can sometimes be observed to fail
because of this. Instead of forcing the caller to retry, let's simply
retry in this to-be-expected case.
Similarly, avoid spurious failures simply because we raced with someone
(e.g., swapout) modifying the page table such that our folio_walk fails.
Simply unconditionally lock the folio, and retry GUP if our folio_walk
fails. Note that the folio_walk repeatedly failing is not something we
expect.
Note that we might want to avoid grabbing the folio lock at some point;
for now, keep that as is and only unconditionally lock the folio.
With this change, the hmm selftests don't fail simply because the folio is
already locked. While this fixes the selftests in some cases, it's likely
not something that deserves a "Fixes:".
Link: https://lkml.kernel.org/r/20250210193801.781278-18-david@redhat.com
Signed-off-by: David Hildenbrand <david@redhat.com>
Tested-by: Alistair Popple <apopple@nvidia.com>
Cc: Alex Shi <alexs@kernel.org>
Cc: Danilo Krummrich <dakr@kernel.org>
Cc: Dave Airlie <airlied@gmail.com>
Cc: Jann Horn <jannh@google.com>
Cc: Jason Gunthorpe <jgg@nvidia.com>
Cc: Jerome Glisse <jglisse@redhat.com>
Cc: John Hubbard <jhubbard@nvidia.com>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Karol Herbst <kherbst@redhat.com>
Cc: Liam Howlett <liam.howlett@oracle.com>
Cc: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: Lyude <lyude@redhat.com>
Cc: "Masami Hiramatsu (Google)" <mhiramat@kernel.org>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Pasha Tatashin <pasha.tatashin@soleen.com>
Cc: Peter Xu <peterx@redhat.com>
Cc: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: SeongJae Park <sj@kernel.org>
Cc: Simona Vetter <simona.vetter@ffwll.ch>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Yanteng Si <si.yanteng@linux.dev>
Cc: Barry Song <v-songbaohua@oppo.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2025-02-10 20:37:59 +01:00
|
|
|
|
2021-06-30 18:54:25 -07:00
|
|
|
|
2025-02-10 20:37:45 +01:00
|
|
|
|
2025-02-10 20:37:46 +01:00
|
|
|
|
2025-02-10 20:37:45 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-02-10 20:37:46 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-02-10 20:37:45 +01:00
|
|
|
|
mm/rmap: avoid -EBUSY from make_device_exclusive()
Failing to obtain the folio lock, for example because the folio is
concurrently getting migrated or swapped out, can easily make the callers
fail: for example, the hmm selftest can sometimes be observed to fail
because of this. Instead of forcing the caller to retry, let's simply
retry in this to-be-expected case.
Similarly, avoid spurious failures simply because we raced with someone
(e.g., swapout) modifying the page table such that our folio_walk fails.
Simply unconditionally lock the folio, and retry GUP if our folio_walk
fails. Note that the folio_walk repeatedly failing is not something we
expect.
Note that we might want to avoid grabbing the folio lock at some point;
for now, keep that as is and only unconditionally lock the folio.
With this change, the hmm selftests don't fail simply because the folio is
already locked. While this fixes the selftests in some cases, it's likely
not something that deserves a "Fixes:".
Link: https://lkml.kernel.org/r/20250210193801.781278-18-david@redhat.com
Signed-off-by: David Hildenbrand <david@redhat.com>
Tested-by: Alistair Popple <apopple@nvidia.com>
Cc: Alex Shi <alexs@kernel.org>
Cc: Danilo Krummrich <dakr@kernel.org>
Cc: Dave Airlie <airlied@gmail.com>
Cc: Jann Horn <jannh@google.com>
Cc: Jason Gunthorpe <jgg@nvidia.com>
Cc: Jerome Glisse <jglisse@redhat.com>
Cc: John Hubbard <jhubbard@nvidia.com>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Karol Herbst <kherbst@redhat.com>
Cc: Liam Howlett <liam.howlett@oracle.com>
Cc: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: Lyude <lyude@redhat.com>
Cc: "Masami Hiramatsu (Google)" <mhiramat@kernel.org>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Pasha Tatashin <pasha.tatashin@soleen.com>
Cc: Peter Xu <peterx@redhat.com>
Cc: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: SeongJae Park <sj@kernel.org>
Cc: Simona Vetter <simona.vetter@ffwll.ch>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Yanteng Si <si.yanteng@linux.dev>
Cc: Barry Song <v-songbaohua@oppo.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2025-02-10 20:37:59 +01:00
|
|
|
|
2025-02-10 20:37:46 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-02-10 20:37:45 +01:00
|
|
|
|
2021-06-30 18:54:25 -07:00
|
|
|
|
2025-02-10 20:37:45 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2021-06-30 18:54:25 -07:00
|
|
|
|
|
|
|
|
|
mm/rmap: avoid -EBUSY from make_device_exclusive()
Failing to obtain the folio lock, for example because the folio is
concurrently getting migrated or swapped out, can easily make the callers
fail: for example, the hmm selftest can sometimes be observed to fail
because of this. Instead of forcing the caller to retry, let's simply
retry in this to-be-expected case.
Similarly, avoid spurious failures simply because we raced with someone
(e.g., swapout) modifying the page table such that our folio_walk fails.
Simply unconditionally lock the folio, and retry GUP if our folio_walk
fails. Note that the folio_walk repeatedly failing is not something we
expect.
Note that we might want to avoid grabbing the folio lock at some point;
for now, keep that as is and only unconditionally lock the folio.
With this change, the hmm selftests don't fail simply because the folio is
already locked. While this fixes the selftests in some cases, it's likely
not something that deserves a "Fixes:".
Link: https://lkml.kernel.org/r/20250210193801.781278-18-david@redhat.com
Signed-off-by: David Hildenbrand <david@redhat.com>
Tested-by: Alistair Popple <apopple@nvidia.com>
Cc: Alex Shi <alexs@kernel.org>
Cc: Danilo Krummrich <dakr@kernel.org>
Cc: Dave Airlie <airlied@gmail.com>
Cc: Jann Horn <jannh@google.com>
Cc: Jason Gunthorpe <jgg@nvidia.com>
Cc: Jerome Glisse <jglisse@redhat.com>
Cc: John Hubbard <jhubbard@nvidia.com>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Karol Herbst <kherbst@redhat.com>
Cc: Liam Howlett <liam.howlett@oracle.com>
Cc: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: Lyude <lyude@redhat.com>
Cc: "Masami Hiramatsu (Google)" <mhiramat@kernel.org>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Pasha Tatashin <pasha.tatashin@soleen.com>
Cc: Peter Xu <peterx@redhat.com>
Cc: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: SeongJae Park <sj@kernel.org>
Cc: Simona Vetter <simona.vetter@ffwll.ch>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Yanteng Si <si.yanteng@linux.dev>
Cc: Barry Song <v-songbaohua@oppo.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2025-02-10 20:37:59 +01:00
|
|
|
|
|
|
|
|
|
2025-02-10 20:37:45 +01:00
|
|
|
|
mm/rmap: avoid -EBUSY from make_device_exclusive()
Failing to obtain the folio lock, for example because the folio is
concurrently getting migrated or swapped out, can easily make the callers
fail: for example, the hmm selftest can sometimes be observed to fail
because of this. Instead of forcing the caller to retry, let's simply
retry in this to-be-expected case.
Similarly, avoid spurious failures simply because we raced with someone
(e.g., swapout) modifying the page table such that our folio_walk fails.
Simply unconditionally lock the folio, and retry GUP if our folio_walk
fails. Note that the folio_walk repeatedly failing is not something we
expect.
Note that we might want to avoid grabbing the folio lock at some point;
for now, keep that as is and only unconditionally lock the folio.
With this change, the hmm selftests don't fail simply because the folio is
already locked. While this fixes the selftests in some cases, it's likely
not something that deserves a "Fixes:".
Link: https://lkml.kernel.org/r/20250210193801.781278-18-david@redhat.com
Signed-off-by: David Hildenbrand <david@redhat.com>
Tested-by: Alistair Popple <apopple@nvidia.com>
Cc: Alex Shi <alexs@kernel.org>
Cc: Danilo Krummrich <dakr@kernel.org>
Cc: Dave Airlie <airlied@gmail.com>
Cc: Jann Horn <jannh@google.com>
Cc: Jason Gunthorpe <jgg@nvidia.com>
Cc: Jerome Glisse <jglisse@redhat.com>
Cc: John Hubbard <jhubbard@nvidia.com>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Karol Herbst <kherbst@redhat.com>
Cc: Liam Howlett <liam.howlett@oracle.com>
Cc: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: Lyude <lyude@redhat.com>
Cc: "Masami Hiramatsu (Google)" <mhiramat@kernel.org>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Pasha Tatashin <pasha.tatashin@soleen.com>
Cc: Peter Xu <peterx@redhat.com>
Cc: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: SeongJae Park <sj@kernel.org>
Cc: Simona Vetter <simona.vetter@ffwll.ch>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Yanteng Si <si.yanteng@linux.dev>
Cc: Barry Song <v-songbaohua@oppo.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2025-02-10 20:37:59 +01:00
|
|
|
|
2025-02-10 20:37:45 +01:00
|
|
|
|
|
|
|
|
|
2025-02-10 20:37:46 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-02-10 20:37:45 +01:00
|
|
|
|
|
|
|
|
|
mm/rmap: avoid -EBUSY from make_device_exclusive()
Failing to obtain the folio lock, for example because the folio is
concurrently getting migrated or swapped out, can easily make the callers
fail: for example, the hmm selftest can sometimes be observed to fail
because of this. Instead of forcing the caller to retry, let's simply
retry in this to-be-expected case.
Similarly, avoid spurious failures simply because we raced with someone
(e.g., swapout) modifying the page table such that our folio_walk fails.
Simply unconditionally lock the folio, and retry GUP if our folio_walk
fails. Note that the folio_walk repeatedly failing is not something we
expect.
Note that we might want to avoid grabbing the folio lock at some point;
for now, keep that as is and only unconditionally lock the folio.
With this change, the hmm selftests don't fail simply because the folio is
already locked. While this fixes the selftests in some cases, it's likely
not something that deserves a "Fixes:".
Link: https://lkml.kernel.org/r/20250210193801.781278-18-david@redhat.com
Signed-off-by: David Hildenbrand <david@redhat.com>
Tested-by: Alistair Popple <apopple@nvidia.com>
Cc: Alex Shi <alexs@kernel.org>
Cc: Danilo Krummrich <dakr@kernel.org>
Cc: Dave Airlie <airlied@gmail.com>
Cc: Jann Horn <jannh@google.com>
Cc: Jason Gunthorpe <jgg@nvidia.com>
Cc: Jerome Glisse <jglisse@redhat.com>
Cc: John Hubbard <jhubbard@nvidia.com>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Karol Herbst <kherbst@redhat.com>
Cc: Liam Howlett <liam.howlett@oracle.com>
Cc: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: Lyude <lyude@redhat.com>
Cc: "Masami Hiramatsu (Google)" <mhiramat@kernel.org>
Cc: Oleg Nesterov <oleg@redhat.com>
Cc: Pasha Tatashin <pasha.tatashin@soleen.com>
Cc: Peter Xu <peterx@redhat.com>
Cc: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: SeongJae Park <sj@kernel.org>
Cc: Simona Vetter <simona.vetter@ffwll.ch>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Yanteng Si <si.yanteng@linux.dev>
Cc: Barry Song <v-songbaohua@oppo.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2025-02-10 20:37:59 +01:00
|
|
|
|
2025-02-10 20:37:45 +01:00
|
|
|
|
2025-02-10 20:37:46 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-02-10 20:37:48 +01:00
|
|
|
|
2025-02-10 20:37:46 +01:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2025-02-10 20:37:45 +01:00
|
|
|
|
|
|
|
|
|
2021-06-30 18:54:25 -07:00
|
|
|
|
2025-02-10 20:37:45 +01:00
|
|
|
|
2021-06-30 18:54:25 -07:00
|
|
|
|
|
|
|
|
|
2011-03-22 16:32:49 -07:00
|
|
|
|
2010-08-09 17:18:41 -07:00
|
|
|
|
2011-03-22 16:32:49 -07:00
|
|
|
|
2010-08-09 17:18:41 -07:00
|
|
|
|
2014-06-06 19:09:30 +04:00
|
|
|
|
2011-03-22 16:32:49 -07:00
|
|
|
|
|
|
|
|
|
2010-08-09 17:18:41 -07:00
|
|
|
|
|
|
|
|
|
2024-10-05 21:01:15 +01:00
|
|
|
|
mm: don't be stuck to rmap lock on reclaim path
The rmap locks(i_mmap_rwsem and anon_vma->root->rwsem) could be contended
under memory pressure if processes keep working on their vmas(e.g., fork,
mmap, munmap). It makes reclaim path stuck. In our real workload traces,
we see kswapd is waiting the lock for 300ms+(worst case, a sec) and it
makes other processes entering direct reclaim, which were also stuck on
the lock.
This patch makes lru aging path try_lock mode like shink_page_list so the
reclaim context will keep working with next lru pages without being stuck.
if it found the rmap lock contended, it rotates the page back to head of
lru in both active/inactive lrus to make them consistent behavior, which
is basic starting point rather than adding more heristic.
Since this patch introduces a new "contended" field as out-param along
with try_lock in-param in rmap_walk_control, it's not immutable any longer
if the try_lock is set so remove const keywords on rmap related functions.
Since rmap walking is already expensive operation, I doubt the const
would help sizable benefit( And we didn't have it until 5.17).
In a heavy app workload in Android, trace shows following statistics. It
almost removes rmap lock contention from reclaim path.
Martin Liu reported:
Before:
max_dur(ms) min_dur(ms) max-min(dur)ms avg_dur(ms) sum_dur(ms) count blocked_function
1632 0 1631 151.542173 31672 209 page_lock_anon_vma_read
601 0 601 145.544681 28817 198 rmap_walk_file
After:
max_dur(ms) min_dur(ms) max-min(dur)ms avg_dur(ms) sum_dur(ms) count blocked_function
NaN NaN NaN NaN NaN 0.0 NaN
0 0 0 0.127645 1 12 rmap_walk_file
[minchan@kernel.org: add comment, per Matthew]
Link: https://lkml.kernel.org/r/YnNqeB5tUf6LZ57b@google.com
Link: https://lkml.kernel.org/r/20220510215423.164547-1-minchan@kernel.org
Signed-off-by: Minchan Kim <minchan@kernel.org>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: Suren Baghdasaryan <surenb@google.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: John Dias <joaodias@google.com>
Cc: Tim Murray <timmurray@google.com>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Vladimir Davydov <vdavydov.dev@gmail.com>
Cc: Martin Liu <liumartin@google.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Matthew Wilcox <willy@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2022-05-19 14:08:54 -07:00
|
|
|
|
2014-01-21 15:49:46 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2014-01-21 15:49:49 -08:00
|
|
|
|
mm: don't be stuck to rmap lock on reclaim path
The rmap locks(i_mmap_rwsem and anon_vma->root->rwsem) could be contended
under memory pressure if processes keep working on their vmas(e.g., fork,
mmap, munmap). It makes reclaim path stuck. In our real workload traces,
we see kswapd is waiting the lock for 300ms+(worst case, a sec) and it
makes other processes entering direct reclaim, which were also stuck on
the lock.
This patch makes lru aging path try_lock mode like shink_page_list so the
reclaim context will keep working with next lru pages without being stuck.
if it found the rmap lock contended, it rotates the page back to head of
lru in both active/inactive lrus to make them consistent behavior, which
is basic starting point rather than adding more heristic.
Since this patch introduces a new "contended" field as out-param along
with try_lock in-param in rmap_walk_control, it's not immutable any longer
if the try_lock is set so remove const keywords on rmap related functions.
Since rmap walking is already expensive operation, I doubt the const
would help sizable benefit( And we didn't have it until 5.17).
In a heavy app workload in Android, trace shows following statistics. It
almost removes rmap lock contention from reclaim path.
Martin Liu reported:
Before:
max_dur(ms) min_dur(ms) max-min(dur)ms avg_dur(ms) sum_dur(ms) count blocked_function
1632 0 1631 151.542173 31672 209 page_lock_anon_vma_read
601 0 601 145.544681 28817 198 rmap_walk_file
After:
max_dur(ms) min_dur(ms) max-min(dur)ms avg_dur(ms) sum_dur(ms) count blocked_function
NaN NaN NaN NaN NaN 0.0 NaN
0 0 0 0.127645 1 12 rmap_walk_file
[minchan@kernel.org: add comment, per Matthew]
Link: https://lkml.kernel.org/r/YnNqeB5tUf6LZ57b@google.com
Link: https://lkml.kernel.org/r/20220510215423.164547-1-minchan@kernel.org
Signed-off-by: Minchan Kim <minchan@kernel.org>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: Suren Baghdasaryan <surenb@google.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: John Dias <joaodias@google.com>
Cc: Tim Murray <timmurray@google.com>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Vladimir Davydov <vdavydov.dev@gmail.com>
Cc: Martin Liu <liumartin@google.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Matthew Wilcox <willy@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2022-05-19 14:08:54 -07:00
|
|
|
|
2014-01-21 15:49:49 -08:00
|
|
|
|
2014-01-21 15:49:46 -08:00
|
|
|
|
2022-01-29 16:06:53 -05:00
|
|
|
|
2026-04-27 13:43:16 +02:00
|
|
|
|
2020-06-08 21:33:54 -07:00
|
|
|
|
2014-01-21 15:49:46 -08:00
|
|
|
|
|
|
|
|
|
2022-01-29 11:52:52 -05:00
|
|
|
|
2014-01-21 15:49:46 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
mm: don't be stuck to rmap lock on reclaim path
The rmap locks(i_mmap_rwsem and anon_vma->root->rwsem) could be contended
under memory pressure if processes keep working on their vmas(e.g., fork,
mmap, munmap). It makes reclaim path stuck. In our real workload traces,
we see kswapd is waiting the lock for 300ms+(worst case, a sec) and it
makes other processes entering direct reclaim, which were also stuck on
the lock.
This patch makes lru aging path try_lock mode like shink_page_list so the
reclaim context will keep working with next lru pages without being stuck.
if it found the rmap lock contended, it rotates the page back to head of
lru in both active/inactive lrus to make them consistent behavior, which
is basic starting point rather than adding more heristic.
Since this patch introduces a new "contended" field as out-param along
with try_lock in-param in rmap_walk_control, it's not immutable any longer
if the try_lock is set so remove const keywords on rmap related functions.
Since rmap walking is already expensive operation, I doubt the const
would help sizable benefit( And we didn't have it until 5.17).
In a heavy app workload in Android, trace shows following statistics. It
almost removes rmap lock contention from reclaim path.
Martin Liu reported:
Before:
max_dur(ms) min_dur(ms) max-min(dur)ms avg_dur(ms) sum_dur(ms) count blocked_function
1632 0 1631 151.542173 31672 209 page_lock_anon_vma_read
601 0 601 145.544681 28817 198 rmap_walk_file
After:
max_dur(ms) min_dur(ms) max-min(dur)ms avg_dur(ms) sum_dur(ms) count blocked_function
NaN NaN NaN NaN NaN 0.0 NaN
0 0 0 0.127645 1 12 rmap_walk_file
[minchan@kernel.org: add comment, per Matthew]
Link: https://lkml.kernel.org/r/YnNqeB5tUf6LZ57b@google.com
Link: https://lkml.kernel.org/r/20220510215423.164547-1-minchan@kernel.org
Signed-off-by: Minchan Kim <minchan@kernel.org>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: Suren Baghdasaryan <surenb@google.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: John Dias <joaodias@google.com>
Cc: Tim Murray <timmurray@google.com>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Vladimir Davydov <vdavydov.dev@gmail.com>
Cc: Martin Liu <liumartin@google.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Matthew Wilcox <willy@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2022-05-19 14:08:54 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2014-01-21 15:49:46 -08:00
|
|
|
|
mm: don't be stuck to rmap lock on reclaim path
The rmap locks(i_mmap_rwsem and anon_vma->root->rwsem) could be contended
under memory pressure if processes keep working on their vmas(e.g., fork,
mmap, munmap). It makes reclaim path stuck. In our real workload traces,
we see kswapd is waiting the lock for 300ms+(worst case, a sec) and it
makes other processes entering direct reclaim, which were also stuck on
the lock.
This patch makes lru aging path try_lock mode like shink_page_list so the
reclaim context will keep working with next lru pages without being stuck.
if it found the rmap lock contended, it rotates the page back to head of
lru in both active/inactive lrus to make them consistent behavior, which
is basic starting point rather than adding more heristic.
Since this patch introduces a new "contended" field as out-param along
with try_lock in-param in rmap_walk_control, it's not immutable any longer
if the try_lock is set so remove const keywords on rmap related functions.
Since rmap walking is already expensive operation, I doubt the const
would help sizable benefit( And we didn't have it until 5.17).
In a heavy app workload in Android, trace shows following statistics. It
almost removes rmap lock contention from reclaim path.
Martin Liu reported:
Before:
max_dur(ms) min_dur(ms) max-min(dur)ms avg_dur(ms) sum_dur(ms) count blocked_function
1632 0 1631 151.542173 31672 209 page_lock_anon_vma_read
601 0 601 145.544681 28817 198 rmap_walk_file
After:
max_dur(ms) min_dur(ms) max-min(dur)ms avg_dur(ms) sum_dur(ms) count blocked_function
NaN NaN NaN NaN NaN 0.0 NaN
0 0 0 0.127645 1 12 rmap_walk_file
[minchan@kernel.org: add comment, per Matthew]
Link: https://lkml.kernel.org/r/YnNqeB5tUf6LZ57b@google.com
Link: https://lkml.kernel.org/r/20220510215423.164547-1-minchan@kernel.org
Signed-off-by: Minchan Kim <minchan@kernel.org>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: Suren Baghdasaryan <surenb@google.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: John Dias <joaodias@google.com>
Cc: Tim Murray <timmurray@google.com>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Vladimir Davydov <vdavydov.dev@gmail.com>
Cc: Martin Liu <liumartin@google.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Matthew Wilcox <willy@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2022-05-19 14:08:54 -07:00
|
|
|
|
2014-01-21 15:49:46 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
ksm: rmap_walk to remove_migation_ptes
A side-effect of making ksm pages swappable is that they have to be placed
on the LRUs: which then exposes them to isolate_lru_page() and hence to
page migration.
Add rmap_walk() for remove_migration_ptes() to use: rmap_walk_anon() and
rmap_walk_file() in rmap.c, but rmap_walk_ksm() in ksm.c. Perhaps some
consolidation with existing code is possible, but don't attempt that yet
(try_to_unmap needs to handle nonlinears, but migration pte removal does
not).
rmap_walk() is sadly less general than it appears: rmap_walk_anon(), like
remove_anon_migration_ptes() which it replaces, avoids calling
page_lock_anon_vma(), because that includes a page_mapped() test which
fails when all migration ptes are in place. That was valid when NUMA page
migration was introduced (holding mmap_sem provided the missing guarantee
that anon_vma's slab had not already been destroyed), but I believe not
valid in the memory hotremove case added since.
For now do the same as before, and consider the best way to fix that
unlikely race later on. When fixed, we can probably use rmap_walk() on
hwpoisoned ksm pages too: for now, they remain among hwpoison's various
exceptions (its PageKsm test comes before the page is locked, but its
page_lock_anon_vma fails safely if an anon gets upgraded).
Signed-off-by: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Cc: Izik Eidus <ieidus@redhat.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Chris Wright <chrisw@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-12-14 17:59:31 -08:00
|
|
|
|
2014-01-21 15:49:52 -08:00
|
|
|
|
|
|
|
|
|
2023-07-18 17:21:36 +08:00
|
|
|
|
2014-01-21 15:49:52 -08:00
|
|
|
|
2023-07-18 17:21:36 +08:00
|
|
|
|
2014-01-21 15:49:52 -08:00
|
|
|
|
2023-07-18 17:21:36 +08:00
|
|
|
|
|
|
|
|
|
ksm: rmap_walk to remove_migation_ptes
A side-effect of making ksm pages swappable is that they have to be placed
on the LRUs: which then exposes them to isolate_lru_page() and hence to
page migration.
Add rmap_walk() for remove_migration_ptes() to use: rmap_walk_anon() and
rmap_walk_file() in rmap.c, but rmap_walk_ksm() in ksm.c. Perhaps some
consolidation with existing code is possible, but don't attempt that yet
(try_to_unmap needs to handle nonlinears, but migration pte removal does
not).
rmap_walk() is sadly less general than it appears: rmap_walk_anon(), like
remove_anon_migration_ptes() which it replaces, avoids calling
page_lock_anon_vma(), because that includes a page_mapped() test which
fails when all migration ptes are in place. That was valid when NUMA page
migration was introduced (holding mmap_sem provided the missing guarantee
that anon_vma's slab had not already been destroyed), but I believe not
valid in the memory hotremove case added since.
For now do the same as before, and consider the best way to fix that
unlikely race later on. When fixed, we can probably use rmap_walk() on
hwpoisoned ksm pages too: for now, they remain among hwpoison's various
exceptions (its PageKsm test comes before the page is locked, but its
page_lock_anon_vma fails safely if an anon gets upgraded).
Signed-off-by: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Cc: Izik Eidus <ieidus@redhat.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Chris Wright <chrisw@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-12-14 17:59:31 -08:00
|
|
|
|
2022-01-29 16:16:54 -05:00
|
|
|
|
mm: don't be stuck to rmap lock on reclaim path
The rmap locks(i_mmap_rwsem and anon_vma->root->rwsem) could be contended
under memory pressure if processes keep working on their vmas(e.g., fork,
mmap, munmap). It makes reclaim path stuck. In our real workload traces,
we see kswapd is waiting the lock for 300ms+(worst case, a sec) and it
makes other processes entering direct reclaim, which were also stuck on
the lock.
This patch makes lru aging path try_lock mode like shink_page_list so the
reclaim context will keep working with next lru pages without being stuck.
if it found the rmap lock contended, it rotates the page back to head of
lru in both active/inactive lrus to make them consistent behavior, which
is basic starting point rather than adding more heristic.
Since this patch introduces a new "contended" field as out-param along
with try_lock in-param in rmap_walk_control, it's not immutable any longer
if the try_lock is set so remove const keywords on rmap related functions.
Since rmap walking is already expensive operation, I doubt the const
would help sizable benefit( And we didn't have it until 5.17).
In a heavy app workload in Android, trace shows following statistics. It
almost removes rmap lock contention from reclaim path.
Martin Liu reported:
Before:
max_dur(ms) min_dur(ms) max-min(dur)ms avg_dur(ms) sum_dur(ms) count blocked_function
1632 0 1631 151.542173 31672 209 page_lock_anon_vma_read
601 0 601 145.544681 28817 198 rmap_walk_file
After:
max_dur(ms) min_dur(ms) max-min(dur)ms avg_dur(ms) sum_dur(ms) count blocked_function
NaN NaN NaN NaN NaN 0.0 NaN
0 0 0 0.127645 1 12 rmap_walk_file
[minchan@kernel.org: add comment, per Matthew]
Link: https://lkml.kernel.org/r/YnNqeB5tUf6LZ57b@google.com
Link: https://lkml.kernel.org/r/20220510215423.164547-1-minchan@kernel.org
Signed-off-by: Minchan Kim <minchan@kernel.org>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: Suren Baghdasaryan <surenb@google.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: John Dias <joaodias@google.com>
Cc: Tim Murray <timmurray@google.com>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Vladimir Davydov <vdavydov.dev@gmail.com>
Cc: Martin Liu <liumartin@google.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Matthew Wilcox <willy@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2022-05-19 14:08:54 -07:00
|
|
|
|
ksm: rmap_walk to remove_migation_ptes
A side-effect of making ksm pages swappable is that they have to be placed
on the LRUs: which then exposes them to isolate_lru_page() and hence to
page migration.
Add rmap_walk() for remove_migration_ptes() to use: rmap_walk_anon() and
rmap_walk_file() in rmap.c, but rmap_walk_ksm() in ksm.c. Perhaps some
consolidation with existing code is possible, but don't attempt that yet
(try_to_unmap needs to handle nonlinears, but migration pte removal does
not).
rmap_walk() is sadly less general than it appears: rmap_walk_anon(), like
remove_anon_migration_ptes() which it replaces, avoids calling
page_lock_anon_vma(), because that includes a page_mapped() test which
fails when all migration ptes are in place. That was valid when NUMA page
migration was introduced (holding mmap_sem provided the missing guarantee
that anon_vma's slab had not already been destroyed), but I believe not
valid in the memory hotremove case added since.
For now do the same as before, and consider the best way to fix that
unlikely race later on. When fixed, we can probably use rmap_walk() on
hwpoisoned ksm pages too: for now, they remain among hwpoison's various
exceptions (its PageKsm test comes before the page is locked, but its
page_lock_anon_vma fails safely if an anon gets upgraded).
Signed-off-by: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Cc: Izik Eidus <ieidus@redhat.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Chris Wright <chrisw@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-12-14 17:59:31 -08:00
|
|
|
|
|
|
|
|
|
2017-02-24 14:57:54 -08:00
|
|
|
|
mm: change anon_vma linking to fix multi-process server scalability issue
The old anon_vma code can lead to scalability issues with heavily forking
workloads. Specifically, each anon_vma will be shared between the parent
process and all its child processes.
In a workload with 1000 child processes and a VMA with 1000 anonymous
pages per process that get COWed, this leads to a system with a million
anonymous pages in the same anon_vma, each of which is mapped in just one
of the 1000 processes. However, the current rmap code needs to walk them
all, leading to O(N) scanning complexity for each page.
This can result in systems where one CPU is walking the page tables of
1000 processes in page_referenced_one, while all other CPUs are stuck on
the anon_vma lock. This leads to catastrophic failure for a benchmark
like AIM7, where the total number of processes can reach in the tens of
thousands. Real workloads are still a factor 10 less process intensive
than AIM7, but they are catching up.
This patch changes the way anon_vmas and VMAs are linked, which allows us
to associate multiple anon_vmas with a VMA. At fork time, each child
process gets its own anon_vmas, in which its COWed pages will be
instantiated. The parents' anon_vma is also linked to the VMA, because
non-COWed pages could be present in any of the children.
This reduces rmap scanning complexity to O(1) for the pages of the 1000
child processes, with O(N) complexity for at most 1/N pages in the system.
This reduces the average scanning cost in heavily forking workloads from
O(N) to 2.
The only real complexity in this patch stems from the fact that linking a
VMA to anon_vmas now involves memory allocations. This means vma_adjust
can fail, if it needs to attach a VMA to anon_vma structures. This in
turn means error handling needs to be added to the calling functions.
A second source of complexity is that, because there can be multiple
anon_vmas, the anon_vma linking in vma_adjust can no longer be done under
"the" anon_vma lock. To prevent the rmap code from walking up an
incomplete VMA, this patch introduces the VM_LOCK_RMAP VMA flag. This bit
flag uses the same slot as the NOMMU VM_MAPPED_COPY, with an ifdef in mm.h
to make sure it is impossible to compile a kernel that needs both symbolic
values for the same bitflag.
Some test results:
Without the anon_vma changes, when AIM7 hits around 9.7k users (on a test
box with 16GB RAM and not quite enough IO), the system ends up running
>99% in system time, with every CPU on the same anon_vma lock in the
pageout code.
With these changes, AIM7 hits the cross-over point around 29.7k users.
This happens with ~99% IO wait time, there never seems to be any spike in
system time. The anon_vma lock contention appears to be resolved.
[akpm@linux-foundation.org: cleanups]
Signed-off-by: Rik van Riel <riel@redhat.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Larry Woodman <lwoodman@redhat.com>
Cc: Lee Schermerhorn <Lee.Schermerhorn@hp.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-03-05 13:42:07 -08:00
|
|
|
|
ksm: rmap_walk to remove_migation_ptes
A side-effect of making ksm pages swappable is that they have to be placed
on the LRUs: which then exposes them to isolate_lru_page() and hence to
page migration.
Add rmap_walk() for remove_migration_ptes() to use: rmap_walk_anon() and
rmap_walk_file() in rmap.c, but rmap_walk_ksm() in ksm.c. Perhaps some
consolidation with existing code is possible, but don't attempt that yet
(try_to_unmap needs to handle nonlinears, but migration pte removal does
not).
rmap_walk() is sadly less general than it appears: rmap_walk_anon(), like
remove_anon_migration_ptes() which it replaces, avoids calling
page_lock_anon_vma(), because that includes a page_mapped() test which
fails when all migration ptes are in place. That was valid when NUMA page
migration was introduced (holding mmap_sem provided the missing guarantee
that anon_vma's slab had not already been destroyed), but I believe not
valid in the memory hotremove case added since.
For now do the same as before, and consider the best way to fix that
unlikely race later on. When fixed, we can probably use rmap_walk() on
hwpoisoned ksm pages too: for now, they remain among hwpoison's various
exceptions (its PageKsm test comes before the page is locked, but its
page_lock_anon_vma fails safely if an anon gets upgraded).
Signed-off-by: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Cc: Izik Eidus <ieidus@redhat.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Chris Wright <chrisw@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-12-14 17:59:31 -08:00
|
|
|
|
mm: always call rmap_walk() on locked folios
Patch series "Improve UFFDIO_MOVE scalability by removing anon_vma lock", v2.
Userfaultfd has a scalability issue in its UFFDIO_MOVE ioctl, which is
heavily used in Android as its java garbage collector uses it for
concurrent heap compaction.
The issue arises because UFFDIO_MOVE updates folio->mapping to an anon_vma
with a different root, in order to move the folio from a src VMA to dst
VMA. It performs the operation with the folio locked, but this is
insufficient, because rmap_walk() can be performed on non-KSM anonymous
folios without folio lock.
This means that UFFDIO_MOVE has to acquire the anon_vma write lock of the
root anon_vma belonging to the folio it wishes to move.
This causes scalability bottleneck when multiple threads perform
UFFDIO_MOVE simultanously on distinct pages of the same src VMA. In field
traces of arm64 android devices, we have observed janky user interactions
due to long (sometimes over ~50ms) uninterruptible sleeps on main UI
thread caused by anon_vma lock contention in UFFDIO_MOVE. This is
particularly severe during the beginning of GC's compaction phase when it
is likely to have multiple threads involved.
This patch resolves the issue by removing the exception in rmap_walk() for
non-KSM anon folios by ensuring that all folios are locked during rmap
walk. This is less problematic than it might seem, as the only major
caller which utilises this mode is shrink_active_list(), which is covered
in detail in the first patch of this series.
As a result of changing our approach to locking, we can remove all the
code that took steps to acquire an anon_vma write lock instead of a folio
lock. This results in a significant simplification and scalability
improvement of the code (currently only in UFFDIO_MOVE). Furthermore, as
a side-effect, folio_lock_anon_vma_read() gets simpler as we don't need to
worry that folio->mapping may have changed under us.
This patch (of 2):
Guarantee that rmap_walk() is called on locked folios so that threads
changing folio->mapping and folio->index for non-KSM anon folios can
serialize on fine-grained folio lock rather than anon_vma lock. Other
folio types are already always locked before rmap_walk(). With this, we
are going from 'not necessarily' locking the non-KSM anon folio to
'definitely' locking it during rmap walks.
This patch is in preparation for removing anon_vma write-lock from
UFFDIO_MOVE.
With this patch, three functions are now expected to be called with a
locked folio. To be careful of not missing any case, here is the
exhaustive list of all their callers.
1) rmap_walk() is called from:
a) folio_referenced()
b) damon_folio_mkold()
c) damon_folio_young()
d) page_idle_clear_pte_refs()
e) try_to_unmap()
f) try_to_migrate()
g) folio_mkclean()
h) remove_migration_ptes()
In the above list, first 4 are changed in this patch to try-lock non-KSM
anon folios, similar to other types of folios. The remaining functions in
the list already hold folio lock when calling rmap_walk().
2) folio_lock_anon_vma_read() is called from following functions:
a) collect_procs_anon()
b) page_idle_clear_pte_refs()
c) damon_folio_mkold()
d) damon_folio_young()
e) folio_referenced()
f) try_to_unmap()
g) try_to_migrate()
All the functions in above list, except collect_procs_anon(), are covered
by the rmap_walk() list above. For collect_procs_anon(), with
kill_procs_now() changed to take folio lock in this patch ensures that all
callers of folio_lock_anon_vma_read() now hold the lock.
3) folio_get_anon_vma() is called from following functions, all of which
already hold the folio lock:
a) move_pages_huge_pmd()
b) __folio_split()
c) move_pages_ptes()
d) migrate_folio_unmap()
e) unmap_and_move_huge_page()
Functionally, this patch doesn't break the logic because rmap walkers
generally do some other check to see if what is expected to mapped did
happen so it's fine, or otherwise treat things as best-effort.
Among the 4 functions changed in this patch, folio_referenced() is the
only core-mm function, and is also frequently accessed. To assess the
impact of locking non-KSM anon folios in
shrink_active_list()->folio_referenced() path, we performed an app cycle
test on an arm64 android device. During the whole duration of the test
there were over 140k invocations of shrink_active_list(), out of which
over 29k had at least one non-KSM anon folio on which folio_referenced()
was called. In none of these invocations folio_trylock() failed.
Of course, we now take a lock where we wouldn't previously have. In the
past it would have had a major impact in causing a CoW write fault to copy
a page in do_wp_page(), as commit 09854ba94c6a ("mm: do_wp_page()
simplification") caused a failure to obtain folio lock to result in a page
copy even if one wasn't necessary.
However, since commit 6c287605fd56 ("mm: remember exclusively mapped
anonymous pages with PG_anon_exclusive"), and the introduction of the
folio anon exclusive flag, this issue is significantly mitigated.
The only case remaining that we might worry about from this perspective is
that of read-only folios immediately after fork where the anon exclusive
bit will not have been set yet.
We note however in the case of read-only just-forked folios that
wp_can_reuse_anon_folio() will notice the raised reference count
established by shrink_active_list() via isolate_lru_folios() and refuse to
reuse in any case, so this will in fact have no impact - the folio lock is
ultimately immaterial here.
All-in-all it appears that there is little opportunity for meaningful
negative impact from this change.
Link: https://lkml.kernel.org/r/20250923071019.775806-1-lokeshgidra@google.com
Link: https://lkml.kernel.org/r/20250923071019.775806-2-lokeshgidra@google.com
Signed-off-by: Lokesh Gidra <lokeshgidra@google.com>
Acked-by: David Hildenbrand <david@redhat.com>
Acked-by: Peter Xu <peterx@redhat.com>
Reviewed-by: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: Harry Yoo <harry.yoo@oracle.com>
Cc: Suren Baghdasaryan <surenb@google.com>
Cc: Barry Song <baohua@kernel.org>
Cc: SeongJae Park <sj@kernel.org>
Cc: Jann Horn <jannh@google.com>
Cc: Kalesh Singh <kaleshsingh@google.com>
Cc: Nicolas Geoffray <ngeoffray@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2025-09-23 00:10:18 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2016-03-17 14:20:01 -07:00
|
|
|
|
2022-01-29 11:52:52 -05:00
|
|
|
|
2016-03-17 14:20:01 -07:00
|
|
|
|
2022-01-29 11:52:52 -05:00
|
|
|
|
2016-03-17 14:20:01 -07:00
|
|
|
|
2022-01-29 16:06:53 -05:00
|
|
|
|
2016-03-17 14:20:01 -07:00
|
|
|
|
ksm: rmap_walk to remove_migation_ptes
A side-effect of making ksm pages swappable is that they have to be placed
on the LRUs: which then exposes them to isolate_lru_page() and hence to
page migration.
Add rmap_walk() for remove_migration_ptes() to use: rmap_walk_anon() and
rmap_walk_file() in rmap.c, but rmap_walk_ksm() in ksm.c. Perhaps some
consolidation with existing code is possible, but don't attempt that yet
(try_to_unmap needs to handle nonlinears, but migration pte removal does
not).
rmap_walk() is sadly less general than it appears: rmap_walk_anon(), like
remove_anon_migration_ptes() which it replaces, avoids calling
page_lock_anon_vma(), because that includes a page_mapped() test which
fails when all migration ptes are in place. That was valid when NUMA page
migration was introduced (holding mmap_sem provided the missing guarantee
that anon_vma's slab had not already been destroyed), but I believe not
valid in the memory hotremove case added since.
For now do the same as before, and consider the best way to fix that
unlikely race later on. When fixed, we can probably use rmap_walk() on
hwpoisoned ksm pages too: for now, they remain among hwpoison's various
exceptions (its PageKsm test comes before the page is locked, but its
page_lock_anon_vma fails safely if an anon gets upgraded).
Signed-off-by: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Cc: Izik Eidus <ieidus@redhat.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Chris Wright <chrisw@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-12-14 17:59:31 -08:00
|
|
|
|
2017-05-03 14:54:23 -07:00
|
|
|
|
2014-01-21 15:49:46 -08:00
|
|
|
|
2022-01-29 16:06:53 -05:00
|
|
|
|
|
|
|
|
|
2017-02-24 14:57:54 -08:00
|
|
|
|
|
|
|
|
|
mm: change anon_vma linking to fix multi-process server scalability issue
The old anon_vma code can lead to scalability issues with heavily forking
workloads. Specifically, each anon_vma will be shared between the parent
process and all its child processes.
In a workload with 1000 child processes and a VMA with 1000 anonymous
pages per process that get COWed, this leads to a system with a million
anonymous pages in the same anon_vma, each of which is mapped in just one
of the 1000 processes. However, the current rmap code needs to walk them
all, leading to O(N) scanning complexity for each page.
This can result in systems where one CPU is walking the page tables of
1000 processes in page_referenced_one, while all other CPUs are stuck on
the anon_vma lock. This leads to catastrophic failure for a benchmark
like AIM7, where the total number of processes can reach in the tens of
thousands. Real workloads are still a factor 10 less process intensive
than AIM7, but they are catching up.
This patch changes the way anon_vmas and VMAs are linked, which allows us
to associate multiple anon_vmas with a VMA. At fork time, each child
process gets its own anon_vmas, in which its COWed pages will be
instantiated. The parents' anon_vma is also linked to the VMA, because
non-COWed pages could be present in any of the children.
This reduces rmap scanning complexity to O(1) for the pages of the 1000
child processes, with O(N) complexity for at most 1/N pages in the system.
This reduces the average scanning cost in heavily forking workloads from
O(N) to 2.
The only real complexity in this patch stems from the fact that linking a
VMA to anon_vmas now involves memory allocations. This means vma_adjust
can fail, if it needs to attach a VMA to anon_vma structures. This in
turn means error handling needs to be added to the calling functions.
A second source of complexity is that, because there can be multiple
anon_vmas, the anon_vma linking in vma_adjust can no longer be done under
"the" anon_vma lock. To prevent the rmap code from walking up an
incomplete VMA, this patch introduces the VM_LOCK_RMAP VMA flag. This bit
flag uses the same slot as the NOMMU VM_MAPPED_COPY, with an ifdef in mm.h
to make sure it is impossible to compile a kernel that needs both symbolic
values for the same bitflag.
Some test results:
Without the anon_vma changes, when AIM7 hits around 9.7k users (on a test
box with 16GB RAM and not quite enough IO), the system ends up running
>99% in system time, with every CPU on the same anon_vma lock in the
pageout code.
With these changes, AIM7 hits the cross-over point around 29.7k users.
This happens with ~99% IO wait time, there never seems to be any spike in
system time. The anon_vma lock contention appears to be resolved.
[akpm@linux-foundation.org: cleanups]
Signed-off-by: Rik van Riel <riel@redhat.com>
Cc: KOSAKI Motohiro <kosaki.motohiro@jp.fujitsu.com>
Cc: Larry Woodman <lwoodman@redhat.com>
Cc: Lee Schermerhorn <Lee.Schermerhorn@hp.com>
Cc: Minchan Kim <minchan.kim@gmail.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2010-03-05 13:42:07 -08:00
|
|
|
|
2024-03-28 22:58:29 +00:00
|
|
|
|
|
|
|
|
|
2014-01-21 15:49:49 -08:00
|
|
|
|
mm/thp: fix vma_address() if virtual address below file offset
Running certain tests with a DEBUG_VM kernel would crash within hours,
on the total_mapcount BUG() in split_huge_page_to_list(), while trying
to free up some memory by punching a hole in a shmem huge page: split's
try_to_unmap() was unable to find all the mappings of the page (which,
on a !DEBUG_VM kernel, would then keep the huge page pinned in memory).
When that BUG() was changed to a WARN(), it would later crash on the
VM_BUG_ON_VMA(end < vma->vm_start || start >= vma->vm_end, vma) in
mm/internal.h:vma_address(), used by rmap_walk_file() for
try_to_unmap().
vma_address() is usually correct, but there's a wraparound case when the
vm_start address is unusually low, but vm_pgoff not so low:
vma_address() chooses max(start, vma->vm_start), but that decides on the
wrong address, because start has become almost ULONG_MAX.
Rewrite vma_address() to be more careful about vm_pgoff; move the
VM_BUG_ON_VMA() out of it, returning -EFAULT for errors, so that it can
be safely used from page_mapped_in_vma() and page_address_in_vma() too.
Add vma_address_end() to apply similar care to end address calculation,
in page_vma_mapped_walk() and page_mkclean_one() and try_to_unmap_one();
though it raises a question of whether callers would do better to supply
pvmw->end to page_vma_mapped_walk() - I chose not, for a smaller patch.
An irritation is that their apparent generality breaks down on KSM
pages, which cannot be located by the page->index that page_to_pgoff()
uses: as commit 4b0ece6fa016 ("mm: migrate: fix remove_migration_pte()
for ksm pages") once discovered. I dithered over the best thing to do
about that, and have ended up with a VM_BUG_ON_PAGE(PageKsm) in both
vma_address() and vma_address_end(); though the only place in danger of
using it on them was try_to_unmap_one().
Sidenote: vma_address() and vma_address_end() now use compound_nr() on a
head page, instead of thp_size(): to make the right calculation on a
hugetlbfs page, whether or not THPs are configured. try_to_unmap() is
used on hugetlbfs pages, but perhaps the wrong calculation never
mattered.
Link: https://lkml.kernel.org/r/caf1c1a3-7cfb-7f8f-1beb-ba816e932825@google.com
Fixes: a8fa41ad2f6f ("mm, rmap: check all VMAs that PTE-mapped THP can be part of")
Signed-off-by: Hugh Dickins <hughd@google.com>
Acked-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Alistair Popple <apopple@nvidia.com>
Cc: Jan Kara <jack@suse.cz>
Cc: Jue Wang <juew@google.com>
Cc: "Matthew Wilcox (Oracle)" <willy@infradead.org>
Cc: Miaohe Lin <linmiaohe@huawei.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Naoya Horiguchi <naoya.horiguchi@nec.com>
Cc: Oscar Salvador <osalvador@suse.de>
Cc: Peter Xu <peterx@redhat.com>
Cc: Ralph Campbell <rcampbell@nvidia.com>
Cc: Shakeel Butt <shakeelb@google.com>
Cc: Wang Yugui <wangyugui@e16-tech.com>
Cc: Yang Shi <shy828301@gmail.com>
Cc: Zi Yan <ziy@nvidia.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2021-06-15 18:23:56 -07:00
|
|
|
|
2015-11-05 18:49:07 -08:00
|
|
|
|
|
|
|
|
|
2014-01-21 15:49:49 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-01-29 16:06:53 -05:00
|
|
|
|
ksm: rmap_walk to remove_migation_ptes
A side-effect of making ksm pages swappable is that they have to be placed
on the LRUs: which then exposes them to isolate_lru_page() and hence to
page migration.
Add rmap_walk() for remove_migration_ptes() to use: rmap_walk_anon() and
rmap_walk_file() in rmap.c, but rmap_walk_ksm() in ksm.c. Perhaps some
consolidation with existing code is possible, but don't attempt that yet
(try_to_unmap needs to handle nonlinears, but migration pte removal does
not).
rmap_walk() is sadly less general than it appears: rmap_walk_anon(), like
remove_anon_migration_ptes() which it replaces, avoids calling
page_lock_anon_vma(), because that includes a page_mapped() test which
fails when all migration ptes are in place. That was valid when NUMA page
migration was introduced (holding mmap_sem provided the missing guarantee
that anon_vma's slab had not already been destroyed), but I believe not
valid in the memory hotremove case added since.
For now do the same as before, and consider the best way to fix that
unlikely race later on. When fixed, we can probably use rmap_walk() on
hwpoisoned ksm pages too: for now, they remain among hwpoison's various
exceptions (its PageKsm test comes before the page is locked, but its
page_lock_anon_vma fails safely if an anon gets upgraded).
Signed-off-by: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Cc: Izik Eidus <ieidus@redhat.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Chris Wright <chrisw@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-12-14 17:59:31 -08:00
|
|
|
|
2022-01-29 16:06:53 -05:00
|
|
|
|
2014-01-21 15:49:49 -08:00
|
|
|
|
ksm: rmap_walk to remove_migation_ptes
A side-effect of making ksm pages swappable is that they have to be placed
on the LRUs: which then exposes them to isolate_lru_page() and hence to
page migration.
Add rmap_walk() for remove_migration_ptes() to use: rmap_walk_anon() and
rmap_walk_file() in rmap.c, but rmap_walk_ksm() in ksm.c. Perhaps some
consolidation with existing code is possible, but don't attempt that yet
(try_to_unmap needs to handle nonlinears, but migration pte removal does
not).
rmap_walk() is sadly less general than it appears: rmap_walk_anon(), like
remove_anon_migration_ptes() which it replaces, avoids calling
page_lock_anon_vma(), because that includes a page_mapped() test which
fails when all migration ptes are in place. That was valid when NUMA page
migration was introduced (holding mmap_sem provided the missing guarantee
that anon_vma's slab had not already been destroyed), but I believe not
valid in the memory hotremove case added since.
For now do the same as before, and consider the best way to fix that
unlikely race later on. When fixed, we can probably use rmap_walk() on
hwpoisoned ksm pages too: for now, they remain among hwpoison's various
exceptions (its PageKsm test comes before the page is locked, but its
page_lock_anon_vma fails safely if an anon gets upgraded).
Signed-off-by: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Cc: Izik Eidus <ieidus@redhat.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Chris Wright <chrisw@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-12-14 17:59:31 -08:00
|
|
|
|
2016-03-17 14:20:01 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
ksm: rmap_walk to remove_migation_ptes
A side-effect of making ksm pages swappable is that they have to be placed
on the LRUs: which then exposes them to isolate_lru_page() and hence to
page migration.
Add rmap_walk() for remove_migration_ptes() to use: rmap_walk_anon() and
rmap_walk_file() in rmap.c, but rmap_walk_ksm() in ksm.c. Perhaps some
consolidation with existing code is possible, but don't attempt that yet
(try_to_unmap needs to handle nonlinears, but migration pte removal does
not).
rmap_walk() is sadly less general than it appears: rmap_walk_anon(), like
remove_anon_migration_ptes() which it replaces, avoids calling
page_lock_anon_vma(), because that includes a page_mapped() test which
fails when all migration ptes are in place. That was valid when NUMA page
migration was introduced (holding mmap_sem provided the missing guarantee
that anon_vma's slab had not already been destroyed), but I believe not
valid in the memory hotremove case added since.
For now do the same as before, and consider the best way to fix that
unlikely race later on. When fixed, we can probably use rmap_walk() on
hwpoisoned ksm pages too: for now, they remain among hwpoison's various
exceptions (its PageKsm test comes before the page is locked, but its
page_lock_anon_vma fails safely if an anon gets upgraded).
Signed-off-by: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Cc: Izik Eidus <ieidus@redhat.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Chris Wright <chrisw@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-12-14 17:59:31 -08:00
|
|
|
|
|
|
|
|
|
mm: refactor rmap_walk_file() to separate out traversal logic
Patch series "expose mapping wrprotect, fix fb_defio use", v3.
Right now the only means by which we can write-protect a range using the
reverse mapping is via folio_mkclean().
However this is not always the appropriate means of doing so, specifically
in the case of the framebuffer deferred I/O logic (fb_defio enabled by
CONFIG_FB_DEFERRED_IO). There, kernel pages are mapped read-only and
write-protect faults used to batch up I/O operations.
Each time the deferred work is done, folio_mkclean() is used to mark the
framebuffer page as having had I/O performed on it. However doing so
requires the kernel page (perhaps allocated via vmalloc()) to have its
page->mapping, index fields set so the rmap can find everything that maps
it in order to write-protect.
This is problematic as firstly, these fields should not be set for
kernel-allocated memory, and secondly these are not folios (it's not user
memory) and page->index, mapping fields are now deprecated and soon to be
removed.
The removal of these fields is imminent, rendering this series more urgent
than it might first appear.
The implementers cannot be blamed for having used this however, as there
is simply no other way of performing this operation correctly.
This series fixes this - we provide the mapping_wrprotect_range() function
to allow the reverse mapping to be used to look up mappings from the page
cache object (i.e. its address_space pointer) at a specific offset.
The fb_defio logic already stores this offset, and can simply be expanded
to keep track of the page cache object, so the change then becomes
straight-forward.
This series should have no functional change.
This patch (of 3):
In order to permit the traversal of the reverse mapping at a specified
mapping and offset rather than those specified by an input folio, we need
to separate out the portion of the rmap file logic which deals with this
traversal from those parts of the logic which interact with the folio.
This patch achieves this by adding a new static __rmap_walk_file()
function which rmap_walk_file() invokes.
This function permits the ability to pass NULL folio, on the assumption
that the caller has provided for this correctly in the callbacks specified
in the rmap_walk_control object.
Though it provides for this, and adds debug asserts to ensure that, should
a folio be specified, these are equal to the mapping and offset specified
in the folio, there should be no functional change as a result of this
patch.
The reason for adding this is to enable for future changes to permit users
to be able to traverse mappings of userland-mapped kernel memory,
write-protecting those mappings to enable page_mkwrite() or pfn_mkwrite()
fault handlers to be retriggered on subsequent dirty.
Link: https://lkml.kernel.org/r/cover.1739029358.git.lorenzo.stoakes@oracle.com
Link: https://lkml.kernel.org/r/0d1acec0cba1e5a12f9b53efcabc397541c90517.1739029358.git.lorenzo.stoakes@oracle.com
Signed-off-by: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Helge Deller <deller@gmx.de>
Cc: Jaya Kumar <jayakumar.lkml@gmail.com>
Cc: Kajtar Zsolt <soci@c64.rulez.org>
Cc: Maíra Canal <mcanal@igalia.com>
Cc: Matthew Wilcox <willy@infradead.org> [English fixes]
Cc: Simona Vetter <simona.vetter@ffwll.ch>
Cc: Thomas Zimemrmann <tzimmermann@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2025-02-08 15:52:54 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2014-01-21 15:49:52 -08:00
|
|
|
|
mm: refactor rmap_walk_file() to separate out traversal logic
Patch series "expose mapping wrprotect, fix fb_defio use", v3.
Right now the only means by which we can write-protect a range using the
reverse mapping is via folio_mkclean().
However this is not always the appropriate means of doing so, specifically
in the case of the framebuffer deferred I/O logic (fb_defio enabled by
CONFIG_FB_DEFERRED_IO). There, kernel pages are mapped read-only and
write-protect faults used to batch up I/O operations.
Each time the deferred work is done, folio_mkclean() is used to mark the
framebuffer page as having had I/O performed on it. However doing so
requires the kernel page (perhaps allocated via vmalloc()) to have its
page->mapping, index fields set so the rmap can find everything that maps
it in order to write-protect.
This is problematic as firstly, these fields should not be set for
kernel-allocated memory, and secondly these are not folios (it's not user
memory) and page->index, mapping fields are now deprecated and soon to be
removed.
The removal of these fields is imminent, rendering this series more urgent
than it might first appear.
The implementers cannot be blamed for having used this however, as there
is simply no other way of performing this operation correctly.
This series fixes this - we provide the mapping_wrprotect_range() function
to allow the reverse mapping to be used to look up mappings from the page
cache object (i.e. its address_space pointer) at a specific offset.
The fb_defio logic already stores this offset, and can simply be expanded
to keep track of the page cache object, so the change then becomes
straight-forward.
This series should have no functional change.
This patch (of 3):
In order to permit the traversal of the reverse mapping at a specified
mapping and offset rather than those specified by an input folio, we need
to separate out the portion of the rmap file logic which deals with this
traversal from those parts of the logic which interact with the folio.
This patch achieves this by adding a new static __rmap_walk_file()
function which rmap_walk_file() invokes.
This function permits the ability to pass NULL folio, on the assumption
that the caller has provided for this correctly in the callbacks specified
in the rmap_walk_control object.
Though it provides for this, and adds debug asserts to ensure that, should
a folio be specified, these are equal to the mapping and offset specified
in the folio, there should be no functional change as a result of this
patch.
The reason for adding this is to enable for future changes to permit users
to be able to traverse mappings of userland-mapped kernel memory,
write-protecting those mappings to enable page_mkwrite() or pfn_mkwrite()
fault handlers to be retriggered on subsequent dirty.
Link: https://lkml.kernel.org/r/cover.1739029358.git.lorenzo.stoakes@oracle.com
Link: https://lkml.kernel.org/r/0d1acec0cba1e5a12f9b53efcabc397541c90517.1739029358.git.lorenzo.stoakes@oracle.com
Signed-off-by: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Helge Deller <deller@gmx.de>
Cc: Jaya Kumar <jayakumar.lkml@gmail.com>
Cc: Kajtar Zsolt <soci@c64.rulez.org>
Cc: Maíra Canal <mcanal@igalia.com>
Cc: Matthew Wilcox <willy@infradead.org> [English fixes]
Cc: Simona Vetter <simona.vetter@ffwll.ch>
Cc: Thomas Zimemrmann <tzimmermann@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2025-02-08 15:52:54 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2014-01-21 15:49:52 -08:00
|
|
|
|
mm: refactor rmap_walk_file() to separate out traversal logic
Patch series "expose mapping wrprotect, fix fb_defio use", v3.
Right now the only means by which we can write-protect a range using the
reverse mapping is via folio_mkclean().
However this is not always the appropriate means of doing so, specifically
in the case of the framebuffer deferred I/O logic (fb_defio enabled by
CONFIG_FB_DEFERRED_IO). There, kernel pages are mapped read-only and
write-protect faults used to batch up I/O operations.
Each time the deferred work is done, folio_mkclean() is used to mark the
framebuffer page as having had I/O performed on it. However doing so
requires the kernel page (perhaps allocated via vmalloc()) to have its
page->mapping, index fields set so the rmap can find everything that maps
it in order to write-protect.
This is problematic as firstly, these fields should not be set for
kernel-allocated memory, and secondly these are not folios (it's not user
memory) and page->index, mapping fields are now deprecated and soon to be
removed.
The removal of these fields is imminent, rendering this series more urgent
than it might first appear.
The implementers cannot be blamed for having used this however, as there
is simply no other way of performing this operation correctly.
This series fixes this - we provide the mapping_wrprotect_range() function
to allow the reverse mapping to be used to look up mappings from the page
cache object (i.e. its address_space pointer) at a specific offset.
The fb_defio logic already stores this offset, and can simply be expanded
to keep track of the page cache object, so the change then becomes
straight-forward.
This series should have no functional change.
This patch (of 3):
In order to permit the traversal of the reverse mapping at a specified
mapping and offset rather than those specified by an input folio, we need
to separate out the portion of the rmap file logic which deals with this
traversal from those parts of the logic which interact with the folio.
This patch achieves this by adding a new static __rmap_walk_file()
function which rmap_walk_file() invokes.
This function permits the ability to pass NULL folio, on the assumption
that the caller has provided for this correctly in the callbacks specified
in the rmap_walk_control object.
Though it provides for this, and adds debug asserts to ensure that, should
a folio be specified, these are equal to the mapping and offset specified
in the folio, there should be no functional change as a result of this
patch.
The reason for adding this is to enable for future changes to permit users
to be able to traverse mappings of userland-mapped kernel memory,
write-protecting those mappings to enable page_mkwrite() or pfn_mkwrite()
fault handlers to be retriggered on subsequent dirty.
Link: https://lkml.kernel.org/r/cover.1739029358.git.lorenzo.stoakes@oracle.com
Link: https://lkml.kernel.org/r/0d1acec0cba1e5a12f9b53efcabc397541c90517.1739029358.git.lorenzo.stoakes@oracle.com
Signed-off-by: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Helge Deller <deller@gmx.de>
Cc: Jaya Kumar <jayakumar.lkml@gmail.com>
Cc: Kajtar Zsolt <soci@c64.rulez.org>
Cc: Maíra Canal <mcanal@igalia.com>
Cc: Matthew Wilcox <willy@infradead.org> [English fixes]
Cc: Simona Vetter <simona.vetter@ffwll.ch>
Cc: Thomas Zimemrmann <tzimmermann@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2025-02-08 15:52:54 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
ksm: rmap_walk to remove_migation_ptes
A side-effect of making ksm pages swappable is that they have to be placed
on the LRUs: which then exposes them to isolate_lru_page() and hence to
page migration.
Add rmap_walk() for remove_migration_ptes() to use: rmap_walk_anon() and
rmap_walk_file() in rmap.c, but rmap_walk_ksm() in ksm.c. Perhaps some
consolidation with existing code is possible, but don't attempt that yet
(try_to_unmap needs to handle nonlinears, but migration pte removal does
not).
rmap_walk() is sadly less general than it appears: rmap_walk_anon(), like
remove_anon_migration_ptes() which it replaces, avoids calling
page_lock_anon_vma(), because that includes a page_mapped() test which
fails when all migration ptes are in place. That was valid when NUMA page
migration was introduced (holding mmap_sem provided the missing guarantee
that anon_vma's slab had not already been destroyed), but I believe not
valid in the memory hotremove case added since.
For now do the same as before, and consider the best way to fix that
unlikely race later on. When fixed, we can probably use rmap_walk() on
hwpoisoned ksm pages too: for now, they remain among hwpoison's various
exceptions (its PageKsm test comes before the page is locked, but its
page_lock_anon_vma fails safely if an anon gets upgraded).
Signed-off-by: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Cc: Izik Eidus <ieidus@redhat.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Chris Wright <chrisw@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-12-14 17:59:31 -08:00
|
|
|
|
mm: refactor rmap_walk_file() to separate out traversal logic
Patch series "expose mapping wrprotect, fix fb_defio use", v3.
Right now the only means by which we can write-protect a range using the
reverse mapping is via folio_mkclean().
However this is not always the appropriate means of doing so, specifically
in the case of the framebuffer deferred I/O logic (fb_defio enabled by
CONFIG_FB_DEFERRED_IO). There, kernel pages are mapped read-only and
write-protect faults used to batch up I/O operations.
Each time the deferred work is done, folio_mkclean() is used to mark the
framebuffer page as having had I/O performed on it. However doing so
requires the kernel page (perhaps allocated via vmalloc()) to have its
page->mapping, index fields set so the rmap can find everything that maps
it in order to write-protect.
This is problematic as firstly, these fields should not be set for
kernel-allocated memory, and secondly these are not folios (it's not user
memory) and page->index, mapping fields are now deprecated and soon to be
removed.
The removal of these fields is imminent, rendering this series more urgent
than it might first appear.
The implementers cannot be blamed for having used this however, as there
is simply no other way of performing this operation correctly.
This series fixes this - we provide the mapping_wrprotect_range() function
to allow the reverse mapping to be used to look up mappings from the page
cache object (i.e. its address_space pointer) at a specific offset.
The fb_defio logic already stores this offset, and can simply be expanded
to keep track of the page cache object, so the change then becomes
straight-forward.
This series should have no functional change.
This patch (of 3):
In order to permit the traversal of the reverse mapping at a specified
mapping and offset rather than those specified by an input folio, we need
to separate out the portion of the rmap file logic which deals with this
traversal from those parts of the logic which interact with the folio.
This patch achieves this by adding a new static __rmap_walk_file()
function which rmap_walk_file() invokes.
This function permits the ability to pass NULL folio, on the assumption
that the caller has provided for this correctly in the callbacks specified
in the rmap_walk_control object.
Though it provides for this, and adds debug asserts to ensure that, should
a folio be specified, these are equal to the mapping and offset specified
in the folio, there should be no functional change as a result of this
patch.
The reason for adding this is to enable for future changes to permit users
to be able to traverse mappings of userland-mapped kernel memory,
write-protecting those mappings to enable page_mkwrite() or pfn_mkwrite()
fault handlers to be retriggered on subsequent dirty.
Link: https://lkml.kernel.org/r/cover.1739029358.git.lorenzo.stoakes@oracle.com
Link: https://lkml.kernel.org/r/0d1acec0cba1e5a12f9b53efcabc397541c90517.1739029358.git.lorenzo.stoakes@oracle.com
Signed-off-by: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Helge Deller <deller@gmx.de>
Cc: Jaya Kumar <jayakumar.lkml@gmail.com>
Cc: Kajtar Zsolt <soci@c64.rulez.org>
Cc: Maíra Canal <mcanal@igalia.com>
Cc: Matthew Wilcox <willy@infradead.org> [English fixes]
Cc: Simona Vetter <simona.vetter@ffwll.ch>
Cc: Thomas Zimemrmann <tzimmermann@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2025-02-08 15:52:54 +00:00
|
|
|
|
ksm: rmap_walk to remove_migation_ptes
A side-effect of making ksm pages swappable is that they have to be placed
on the LRUs: which then exposes them to isolate_lru_page() and hence to
page migration.
Add rmap_walk() for remove_migration_ptes() to use: rmap_walk_anon() and
rmap_walk_file() in rmap.c, but rmap_walk_ksm() in ksm.c. Perhaps some
consolidation with existing code is possible, but don't attempt that yet
(try_to_unmap needs to handle nonlinears, but migration pte removal does
not).
rmap_walk() is sadly less general than it appears: rmap_walk_anon(), like
remove_anon_migration_ptes() which it replaces, avoids calling
page_lock_anon_vma(), because that includes a page_mapped() test which
fails when all migration ptes are in place. That was valid when NUMA page
migration was introduced (holding mmap_sem provided the missing guarantee
that anon_vma's slab had not already been destroyed), but I believe not
valid in the memory hotremove case added since.
For now do the same as before, and consider the best way to fix that
unlikely race later on. When fixed, we can probably use rmap_walk() on
hwpoisoned ksm pages too: for now, they remain among hwpoison's various
exceptions (its PageKsm test comes before the page is locked, but its
page_lock_anon_vma fails safely if an anon gets upgraded).
Signed-off-by: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Cc: Izik Eidus <ieidus@redhat.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Chris Wright <chrisw@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-12-14 17:59:31 -08:00
|
|
|
|
|
|
|
|
|
mm: refactor rmap_walk_file() to separate out traversal logic
Patch series "expose mapping wrprotect, fix fb_defio use", v3.
Right now the only means by which we can write-protect a range using the
reverse mapping is via folio_mkclean().
However this is not always the appropriate means of doing so, specifically
in the case of the framebuffer deferred I/O logic (fb_defio enabled by
CONFIG_FB_DEFERRED_IO). There, kernel pages are mapped read-only and
write-protect faults used to batch up I/O operations.
Each time the deferred work is done, folio_mkclean() is used to mark the
framebuffer page as having had I/O performed on it. However doing so
requires the kernel page (perhaps allocated via vmalloc()) to have its
page->mapping, index fields set so the rmap can find everything that maps
it in order to write-protect.
This is problematic as firstly, these fields should not be set for
kernel-allocated memory, and secondly these are not folios (it's not user
memory) and page->index, mapping fields are now deprecated and soon to be
removed.
The removal of these fields is imminent, rendering this series more urgent
than it might first appear.
The implementers cannot be blamed for having used this however, as there
is simply no other way of performing this operation correctly.
This series fixes this - we provide the mapping_wrprotect_range() function
to allow the reverse mapping to be used to look up mappings from the page
cache object (i.e. its address_space pointer) at a specific offset.
The fb_defio logic already stores this offset, and can simply be expanded
to keep track of the page cache object, so the change then becomes
straight-forward.
This series should have no functional change.
This patch (of 3):
In order to permit the traversal of the reverse mapping at a specified
mapping and offset rather than those specified by an input folio, we need
to separate out the portion of the rmap file logic which deals with this
traversal from those parts of the logic which interact with the folio.
This patch achieves this by adding a new static __rmap_walk_file()
function which rmap_walk_file() invokes.
This function permits the ability to pass NULL folio, on the assumption
that the caller has provided for this correctly in the callbacks specified
in the rmap_walk_control object.
Though it provides for this, and adds debug asserts to ensure that, should
a folio be specified, these are equal to the mapping and offset specified
in the folio, there should be no functional change as a result of this
patch.
The reason for adding this is to enable for future changes to permit users
to be able to traverse mappings of userland-mapped kernel memory,
write-protecting those mappings to enable page_mkwrite() or pfn_mkwrite()
fault handlers to be retriggered on subsequent dirty.
Link: https://lkml.kernel.org/r/cover.1739029358.git.lorenzo.stoakes@oracle.com
Link: https://lkml.kernel.org/r/0d1acec0cba1e5a12f9b53efcabc397541c90517.1739029358.git.lorenzo.stoakes@oracle.com
Signed-off-by: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Helge Deller <deller@gmx.de>
Cc: Jaya Kumar <jayakumar.lkml@gmail.com>
Cc: Kajtar Zsolt <soci@c64.rulez.org>
Cc: Maíra Canal <mcanal@igalia.com>
Cc: Matthew Wilcox <willy@infradead.org> [English fixes]
Cc: Simona Vetter <simona.vetter@ffwll.ch>
Cc: Thomas Zimemrmann <tzimmermann@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2025-02-08 15:52:54 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2014-01-21 15:49:53 -08:00
|
|
|
|
mm: don't be stuck to rmap lock on reclaim path
The rmap locks(i_mmap_rwsem and anon_vma->root->rwsem) could be contended
under memory pressure if processes keep working on their vmas(e.g., fork,
mmap, munmap). It makes reclaim path stuck. In our real workload traces,
we see kswapd is waiting the lock for 300ms+(worst case, a sec) and it
makes other processes entering direct reclaim, which were also stuck on
the lock.
This patch makes lru aging path try_lock mode like shink_page_list so the
reclaim context will keep working with next lru pages without being stuck.
if it found the rmap lock contended, it rotates the page back to head of
lru in both active/inactive lrus to make them consistent behavior, which
is basic starting point rather than adding more heristic.
Since this patch introduces a new "contended" field as out-param along
with try_lock in-param in rmap_walk_control, it's not immutable any longer
if the try_lock is set so remove const keywords on rmap related functions.
Since rmap walking is already expensive operation, I doubt the const
would help sizable benefit( And we didn't have it until 5.17).
In a heavy app workload in Android, trace shows following statistics. It
almost removes rmap lock contention from reclaim path.
Martin Liu reported:
Before:
max_dur(ms) min_dur(ms) max-min(dur)ms avg_dur(ms) sum_dur(ms) count blocked_function
1632 0 1631 151.542173 31672 209 page_lock_anon_vma_read
601 0 601 145.544681 28817 198 rmap_walk_file
After:
max_dur(ms) min_dur(ms) max-min(dur)ms avg_dur(ms) sum_dur(ms) count blocked_function
NaN NaN NaN NaN NaN 0.0 NaN
0 0 0 0.127645 1 12 rmap_walk_file
[minchan@kernel.org: add comment, per Matthew]
Link: https://lkml.kernel.org/r/YnNqeB5tUf6LZ57b@google.com
Link: https://lkml.kernel.org/r/20220510215423.164547-1-minchan@kernel.org
Signed-off-by: Minchan Kim <minchan@kernel.org>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: Suren Baghdasaryan <surenb@google.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: John Dias <joaodias@google.com>
Cc: Tim Murray <timmurray@google.com>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Vladimir Davydov <vdavydov.dev@gmail.com>
Cc: Martin Liu <liumartin@google.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Matthew Wilcox <willy@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2022-05-19 14:08:54 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2016-03-17 14:20:01 -07:00
|
|
|
|
mm: don't be stuck to rmap lock on reclaim path
The rmap locks(i_mmap_rwsem and anon_vma->root->rwsem) could be contended
under memory pressure if processes keep working on their vmas(e.g., fork,
mmap, munmap). It makes reclaim path stuck. In our real workload traces,
we see kswapd is waiting the lock for 300ms+(worst case, a sec) and it
makes other processes entering direct reclaim, which were also stuck on
the lock.
This patch makes lru aging path try_lock mode like shink_page_list so the
reclaim context will keep working with next lru pages without being stuck.
if it found the rmap lock contended, it rotates the page back to head of
lru in both active/inactive lrus to make them consistent behavior, which
is basic starting point rather than adding more heristic.
Since this patch introduces a new "contended" field as out-param along
with try_lock in-param in rmap_walk_control, it's not immutable any longer
if the try_lock is set so remove const keywords on rmap related functions.
Since rmap walking is already expensive operation, I doubt the const
would help sizable benefit( And we didn't have it until 5.17).
In a heavy app workload in Android, trace shows following statistics. It
almost removes rmap lock contention from reclaim path.
Martin Liu reported:
Before:
max_dur(ms) min_dur(ms) max-min(dur)ms avg_dur(ms) sum_dur(ms) count blocked_function
1632 0 1631 151.542173 31672 209 page_lock_anon_vma_read
601 0 601 145.544681 28817 198 rmap_walk_file
After:
max_dur(ms) min_dur(ms) max-min(dur)ms avg_dur(ms) sum_dur(ms) count blocked_function
NaN NaN NaN NaN NaN 0.0 NaN
0 0 0 0.127645 1 12 rmap_walk_file
[minchan@kernel.org: add comment, per Matthew]
Link: https://lkml.kernel.org/r/YnNqeB5tUf6LZ57b@google.com
Link: https://lkml.kernel.org/r/20220510215423.164547-1-minchan@kernel.org
Signed-off-by: Minchan Kim <minchan@kernel.org>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: Suren Baghdasaryan <surenb@google.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: John Dias <joaodias@google.com>
Cc: Tim Murray <timmurray@google.com>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Vladimir Davydov <vdavydov.dev@gmail.com>
Cc: Martin Liu <liumartin@google.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Matthew Wilcox <willy@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2022-05-19 14:08:54 -07:00
|
|
|
|
|
|
|
|
|
2017-02-24 14:57:54 -08:00
|
|
|
|
|
|
|
|
|
mm: refactor rmap_walk_file() to separate out traversal logic
Patch series "expose mapping wrprotect, fix fb_defio use", v3.
Right now the only means by which we can write-protect a range using the
reverse mapping is via folio_mkclean().
However this is not always the appropriate means of doing so, specifically
in the case of the framebuffer deferred I/O logic (fb_defio enabled by
CONFIG_FB_DEFERRED_IO). There, kernel pages are mapped read-only and
write-protect faults used to batch up I/O operations.
Each time the deferred work is done, folio_mkclean() is used to mark the
framebuffer page as having had I/O performed on it. However doing so
requires the kernel page (perhaps allocated via vmalloc()) to have its
page->mapping, index fields set so the rmap can find everything that maps
it in order to write-protect.
This is problematic as firstly, these fields should not be set for
kernel-allocated memory, and secondly these are not folios (it's not user
memory) and page->index, mapping fields are now deprecated and soon to be
removed.
The removal of these fields is imminent, rendering this series more urgent
than it might first appear.
The implementers cannot be blamed for having used this however, as there
is simply no other way of performing this operation correctly.
This series fixes this - we provide the mapping_wrprotect_range() function
to allow the reverse mapping to be used to look up mappings from the page
cache object (i.e. its address_space pointer) at a specific offset.
The fb_defio logic already stores this offset, and can simply be expanded
to keep track of the page cache object, so the change then becomes
straight-forward.
This series should have no functional change.
This patch (of 3):
In order to permit the traversal of the reverse mapping at a specified
mapping and offset rather than those specified by an input folio, we need
to separate out the portion of the rmap file logic which deals with this
traversal from those parts of the logic which interact with the folio.
This patch achieves this by adding a new static __rmap_walk_file()
function which rmap_walk_file() invokes.
This function permits the ability to pass NULL folio, on the assumption
that the caller has provided for this correctly in the callbacks specified
in the rmap_walk_control object.
Though it provides for this, and adds debug asserts to ensure that, should
a folio be specified, these are equal to the mapping and offset specified
in the folio, there should be no functional change as a result of this
patch.
The reason for adding this is to enable for future changes to permit users
to be able to traverse mappings of userland-mapped kernel memory,
write-protecting those mappings to enable page_mkwrite() or pfn_mkwrite()
fault handlers to be retriggered on subsequent dirty.
Link: https://lkml.kernel.org/r/cover.1739029358.git.lorenzo.stoakes@oracle.com
Link: https://lkml.kernel.org/r/0d1acec0cba1e5a12f9b53efcabc397541c90517.1739029358.git.lorenzo.stoakes@oracle.com
Signed-off-by: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Helge Deller <deller@gmx.de>
Cc: Jaya Kumar <jayakumar.lkml@gmail.com>
Cc: Kajtar Zsolt <soci@c64.rulez.org>
Cc: Maíra Canal <mcanal@igalia.com>
Cc: Matthew Wilcox <willy@infradead.org> [English fixes]
Cc: Simona Vetter <simona.vetter@ffwll.ch>
Cc: Thomas Zimemrmann <tzimmermann@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2025-02-08 15:52:54 +00:00
|
|
|
|
2014-01-21 15:49:49 -08:00
|
|
|
|
mm/thp: fix vma_address() if virtual address below file offset
Running certain tests with a DEBUG_VM kernel would crash within hours,
on the total_mapcount BUG() in split_huge_page_to_list(), while trying
to free up some memory by punching a hole in a shmem huge page: split's
try_to_unmap() was unable to find all the mappings of the page (which,
on a !DEBUG_VM kernel, would then keep the huge page pinned in memory).
When that BUG() was changed to a WARN(), it would later crash on the
VM_BUG_ON_VMA(end < vma->vm_start || start >= vma->vm_end, vma) in
mm/internal.h:vma_address(), used by rmap_walk_file() for
try_to_unmap().
vma_address() is usually correct, but there's a wraparound case when the
vm_start address is unusually low, but vm_pgoff not so low:
vma_address() chooses max(start, vma->vm_start), but that decides on the
wrong address, because start has become almost ULONG_MAX.
Rewrite vma_address() to be more careful about vm_pgoff; move the
VM_BUG_ON_VMA() out of it, returning -EFAULT for errors, so that it can
be safely used from page_mapped_in_vma() and page_address_in_vma() too.
Add vma_address_end() to apply similar care to end address calculation,
in page_vma_mapped_walk() and page_mkclean_one() and try_to_unmap_one();
though it raises a question of whether callers would do better to supply
pvmw->end to page_vma_mapped_walk() - I chose not, for a smaller patch.
An irritation is that their apparent generality breaks down on KSM
pages, which cannot be located by the page->index that page_to_pgoff()
uses: as commit 4b0ece6fa016 ("mm: migrate: fix remove_migration_pte()
for ksm pages") once discovered. I dithered over the best thing to do
about that, and have ended up with a VM_BUG_ON_PAGE(PageKsm) in both
vma_address() and vma_address_end(); though the only place in danger of
using it on them was try_to_unmap_one().
Sidenote: vma_address() and vma_address_end() now use compound_nr() on a
head page, instead of thp_size(): to make the right calculation on a
hugetlbfs page, whether or not THPs are configured. try_to_unmap() is
used on hugetlbfs pages, but perhaps the wrong calculation never
mattered.
Link: https://lkml.kernel.org/r/caf1c1a3-7cfb-7f8f-1beb-ba816e932825@google.com
Fixes: a8fa41ad2f6f ("mm, rmap: check all VMAs that PTE-mapped THP can be part of")
Signed-off-by: Hugh Dickins <hughd@google.com>
Acked-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Alistair Popple <apopple@nvidia.com>
Cc: Jan Kara <jack@suse.cz>
Cc: Jue Wang <juew@google.com>
Cc: "Matthew Wilcox (Oracle)" <willy@infradead.org>
Cc: Miaohe Lin <linmiaohe@huawei.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Naoya Horiguchi <naoya.horiguchi@nec.com>
Cc: Oscar Salvador <osalvador@suse.de>
Cc: Peter Xu <peterx@redhat.com>
Cc: Ralph Campbell <rcampbell@nvidia.com>
Cc: Shakeel Butt <shakeelb@google.com>
Cc: Wang Yugui <wangyugui@e16-tech.com>
Cc: Yang Shi <shy828301@gmail.com>
Cc: Zi Yan <ziy@nvidia.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2021-06-15 18:23:56 -07:00
|
|
|
|
2015-11-05 18:49:07 -08:00
|
|
|
|
|
|
|
|
|
2014-01-21 15:49:49 -08:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2022-01-29 16:06:53 -05:00
|
|
|
|
2014-01-21 15:49:49 -08:00
|
|
|
|
2022-01-29 16:06:53 -05:00
|
|
|
|
2014-01-21 15:49:49 -08:00
|
|
|
|
ksm: rmap_walk to remove_migation_ptes
A side-effect of making ksm pages swappable is that they have to be placed
on the LRUs: which then exposes them to isolate_lru_page() and hence to
page migration.
Add rmap_walk() for remove_migration_ptes() to use: rmap_walk_anon() and
rmap_walk_file() in rmap.c, but rmap_walk_ksm() in ksm.c. Perhaps some
consolidation with existing code is possible, but don't attempt that yet
(try_to_unmap needs to handle nonlinears, but migration pte removal does
not).
rmap_walk() is sadly less general than it appears: rmap_walk_anon(), like
remove_anon_migration_ptes() which it replaces, avoids calling
page_lock_anon_vma(), because that includes a page_mapped() test which
fails when all migration ptes are in place. That was valid when NUMA page
migration was introduced (holding mmap_sem provided the missing guarantee
that anon_vma's slab had not already been destroyed), but I believe not
valid in the memory hotremove case added since.
For now do the same as before, and consider the best way to fix that
unlikely race later on. When fixed, we can probably use rmap_walk() on
hwpoisoned ksm pages too: for now, they remain among hwpoison's various
exceptions (its PageKsm test comes before the page is locked, but its
page_lock_anon_vma fails safely if an anon gets upgraded).
Signed-off-by: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Cc: Izik Eidus <ieidus@redhat.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Chris Wright <chrisw@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-12-14 17:59:31 -08:00
|
|
|
|
2014-01-21 15:49:49 -08:00
|
|
|
|
2016-03-17 14:20:01 -07:00
|
|
|
|
|
|
|
|
|
ksm: rmap_walk to remove_migation_ptes
A side-effect of making ksm pages swappable is that they have to be placed
on the LRUs: which then exposes them to isolate_lru_page() and hence to
page migration.
Add rmap_walk() for remove_migration_ptes() to use: rmap_walk_anon() and
rmap_walk_file() in rmap.c, but rmap_walk_ksm() in ksm.c. Perhaps some
consolidation with existing code is possible, but don't attempt that yet
(try_to_unmap needs to handle nonlinears, but migration pte removal does
not).
rmap_walk() is sadly less general than it appears: rmap_walk_anon(), like
remove_anon_migration_ptes() which it replaces, avoids calling
page_lock_anon_vma(), because that includes a page_mapped() test which
fails when all migration ptes are in place. That was valid when NUMA page
migration was introduced (holding mmap_sem provided the missing guarantee
that anon_vma's slab had not already been destroyed), but I believe not
valid in the memory hotremove case added since.
For now do the same as before, and consider the best way to fix that
unlikely race later on. When fixed, we can probably use rmap_walk() on
hwpoisoned ksm pages too: for now, they remain among hwpoison's various
exceptions (its PageKsm test comes before the page is locked, but its
page_lock_anon_vma fails safely if an anon gets upgraded).
Signed-off-by: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Cc: Izik Eidus <ieidus@redhat.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Chris Wright <chrisw@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-12-14 17:59:31 -08:00
|
|
|
|
|
|
|
|
|
mm: refactor rmap_walk_file() to separate out traversal logic
Patch series "expose mapping wrprotect, fix fb_defio use", v3.
Right now the only means by which we can write-protect a range using the
reverse mapping is via folio_mkclean().
However this is not always the appropriate means of doing so, specifically
in the case of the framebuffer deferred I/O logic (fb_defio enabled by
CONFIG_FB_DEFERRED_IO). There, kernel pages are mapped read-only and
write-protect faults used to batch up I/O operations.
Each time the deferred work is done, folio_mkclean() is used to mark the
framebuffer page as having had I/O performed on it. However doing so
requires the kernel page (perhaps allocated via vmalloc()) to have its
page->mapping, index fields set so the rmap can find everything that maps
it in order to write-protect.
This is problematic as firstly, these fields should not be set for
kernel-allocated memory, and secondly these are not folios (it's not user
memory) and page->index, mapping fields are now deprecated and soon to be
removed.
The removal of these fields is imminent, rendering this series more urgent
than it might first appear.
The implementers cannot be blamed for having used this however, as there
is simply no other way of performing this operation correctly.
This series fixes this - we provide the mapping_wrprotect_range() function
to allow the reverse mapping to be used to look up mappings from the page
cache object (i.e. its address_space pointer) at a specific offset.
The fb_defio logic already stores this offset, and can simply be expanded
to keep track of the page cache object, so the change then becomes
straight-forward.
This series should have no functional change.
This patch (of 3):
In order to permit the traversal of the reverse mapping at a specified
mapping and offset rather than those specified by an input folio, we need
to separate out the portion of the rmap file logic which deals with this
traversal from those parts of the logic which interact with the folio.
This patch achieves this by adding a new static __rmap_walk_file()
function which rmap_walk_file() invokes.
This function permits the ability to pass NULL folio, on the assumption
that the caller has provided for this correctly in the callbacks specified
in the rmap_walk_control object.
Though it provides for this, and adds debug asserts to ensure that, should
a folio be specified, these are equal to the mapping and offset specified
in the folio, there should be no functional change as a result of this
patch.
The reason for adding this is to enable for future changes to permit users
to be able to traverse mappings of userland-mapped kernel memory,
write-protecting those mappings to enable page_mkwrite() or pfn_mkwrite()
fault handlers to be retriggered on subsequent dirty.
Link: https://lkml.kernel.org/r/cover.1739029358.git.lorenzo.stoakes@oracle.com
Link: https://lkml.kernel.org/r/0d1acec0cba1e5a12f9b53efcabc397541c90517.1739029358.git.lorenzo.stoakes@oracle.com
Signed-off-by: Lorenzo Stoakes <lorenzo.stoakes@oracle.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: Helge Deller <deller@gmx.de>
Cc: Jaya Kumar <jayakumar.lkml@gmail.com>
Cc: Kajtar Zsolt <soci@c64.rulez.org>
Cc: Maíra Canal <mcanal@igalia.com>
Cc: Matthew Wilcox <willy@infradead.org> [English fixes]
Cc: Simona Vetter <simona.vetter@ffwll.ch>
Cc: Thomas Zimemrmann <tzimmermann@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2025-02-08 15:52:54 +00:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
mm: don't be stuck to rmap lock on reclaim path
The rmap locks(i_mmap_rwsem and anon_vma->root->rwsem) could be contended
under memory pressure if processes keep working on their vmas(e.g., fork,
mmap, munmap). It makes reclaim path stuck. In our real workload traces,
we see kswapd is waiting the lock for 300ms+(worst case, a sec) and it
makes other processes entering direct reclaim, which were also stuck on
the lock.
This patch makes lru aging path try_lock mode like shink_page_list so the
reclaim context will keep working with next lru pages without being stuck.
if it found the rmap lock contended, it rotates the page back to head of
lru in both active/inactive lrus to make them consistent behavior, which
is basic starting point rather than adding more heristic.
Since this patch introduces a new "contended" field as out-param along
with try_lock in-param in rmap_walk_control, it's not immutable any longer
if the try_lock is set so remove const keywords on rmap related functions.
Since rmap walking is already expensive operation, I doubt the const
would help sizable benefit( And we didn't have it until 5.17).
In a heavy app workload in Android, trace shows following statistics. It
almost removes rmap lock contention from reclaim path.
Martin Liu reported:
Before:
max_dur(ms) min_dur(ms) max-min(dur)ms avg_dur(ms) sum_dur(ms) count blocked_function
1632 0 1631 151.542173 31672 209 page_lock_anon_vma_read
601 0 601 145.544681 28817 198 rmap_walk_file
After:
max_dur(ms) min_dur(ms) max-min(dur)ms avg_dur(ms) sum_dur(ms) count blocked_function
NaN NaN NaN NaN NaN 0.0 NaN
0 0 0 0.127645 1 12 rmap_walk_file
[minchan@kernel.org: add comment, per Matthew]
Link: https://lkml.kernel.org/r/YnNqeB5tUf6LZ57b@google.com
Link: https://lkml.kernel.org/r/20220510215423.164547-1-minchan@kernel.org
Signed-off-by: Minchan Kim <minchan@kernel.org>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: Suren Baghdasaryan <surenb@google.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: John Dias <joaodias@google.com>
Cc: Tim Murray <timmurray@google.com>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Vladimir Davydov <vdavydov.dev@gmail.com>
Cc: Martin Liu <liumartin@google.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Matthew Wilcox <willy@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2022-05-19 14:08:54 -07:00
|
|
|
|
ksm: rmap_walk to remove_migation_ptes
A side-effect of making ksm pages swappable is that they have to be placed
on the LRUs: which then exposes them to isolate_lru_page() and hence to
page migration.
Add rmap_walk() for remove_migration_ptes() to use: rmap_walk_anon() and
rmap_walk_file() in rmap.c, but rmap_walk_ksm() in ksm.c. Perhaps some
consolidation with existing code is possible, but don't attempt that yet
(try_to_unmap needs to handle nonlinears, but migration pte removal does
not).
rmap_walk() is sadly less general than it appears: rmap_walk_anon(), like
remove_anon_migration_ptes() which it replaces, avoids calling
page_lock_anon_vma(), because that includes a page_mapped() test which
fails when all migration ptes are in place. That was valid when NUMA page
migration was introduced (holding mmap_sem provided the missing guarantee
that anon_vma's slab had not already been destroyed), but I believe not
valid in the memory hotremove case added since.
For now do the same as before, and consider the best way to fix that
unlikely race later on. When fixed, we can probably use rmap_walk() on
hwpoisoned ksm pages too: for now, they remain among hwpoison's various
exceptions (its PageKsm test comes before the page is locked, but its
page_lock_anon_vma fails safely if an anon gets upgraded).
Signed-off-by: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Cc: Izik Eidus <ieidus@redhat.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Chris Wright <chrisw@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-12-14 17:59:31 -08:00
|
|
|
|
2022-01-29 16:06:53 -05:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
ksm: rmap_walk to remove_migation_ptes
A side-effect of making ksm pages swappable is that they have to be placed
on the LRUs: which then exposes them to isolate_lru_page() and hence to
page migration.
Add rmap_walk() for remove_migration_ptes() to use: rmap_walk_anon() and
rmap_walk_file() in rmap.c, but rmap_walk_ksm() in ksm.c. Perhaps some
consolidation with existing code is possible, but don't attempt that yet
(try_to_unmap needs to handle nonlinears, but migration pte removal does
not).
rmap_walk() is sadly less general than it appears: rmap_walk_anon(), like
remove_anon_migration_ptes() which it replaces, avoids calling
page_lock_anon_vma(), because that includes a page_mapped() test which
fails when all migration ptes are in place. That was valid when NUMA page
migration was introduced (holding mmap_sem provided the missing guarantee
that anon_vma's slab had not already been destroyed), but I believe not
valid in the memory hotremove case added since.
For now do the same as before, and consider the best way to fix that
unlikely race later on. When fixed, we can probably use rmap_walk() on
hwpoisoned ksm pages too: for now, they remain among hwpoison's various
exceptions (its PageKsm test comes before the page is locked, but its
page_lock_anon_vma fails safely if an anon gets upgraded).
Signed-off-by: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Cc: Izik Eidus <ieidus@redhat.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Chris Wright <chrisw@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-12-14 17:59:31 -08:00
|
|
|
|
2022-01-29 16:06:53 -05:00
|
|
|
|
2016-03-17 14:20:01 -07:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
mm: don't be stuck to rmap lock on reclaim path
The rmap locks(i_mmap_rwsem and anon_vma->root->rwsem) could be contended
under memory pressure if processes keep working on their vmas(e.g., fork,
mmap, munmap). It makes reclaim path stuck. In our real workload traces,
we see kswapd is waiting the lock for 300ms+(worst case, a sec) and it
makes other processes entering direct reclaim, which were also stuck on
the lock.
This patch makes lru aging path try_lock mode like shink_page_list so the
reclaim context will keep working with next lru pages without being stuck.
if it found the rmap lock contended, it rotates the page back to head of
lru in both active/inactive lrus to make them consistent behavior, which
is basic starting point rather than adding more heristic.
Since this patch introduces a new "contended" field as out-param along
with try_lock in-param in rmap_walk_control, it's not immutable any longer
if the try_lock is set so remove const keywords on rmap related functions.
Since rmap walking is already expensive operation, I doubt the const
would help sizable benefit( And we didn't have it until 5.17).
In a heavy app workload in Android, trace shows following statistics. It
almost removes rmap lock contention from reclaim path.
Martin Liu reported:
Before:
max_dur(ms) min_dur(ms) max-min(dur)ms avg_dur(ms) sum_dur(ms) count blocked_function
1632 0 1631 151.542173 31672 209 page_lock_anon_vma_read
601 0 601 145.544681 28817 198 rmap_walk_file
After:
max_dur(ms) min_dur(ms) max-min(dur)ms avg_dur(ms) sum_dur(ms) count blocked_function
NaN NaN NaN NaN NaN 0.0 NaN
0 0 0 0.127645 1 12 rmap_walk_file
[minchan@kernel.org: add comment, per Matthew]
Link: https://lkml.kernel.org/r/YnNqeB5tUf6LZ57b@google.com
Link: https://lkml.kernel.org/r/20220510215423.164547-1-minchan@kernel.org
Signed-off-by: Minchan Kim <minchan@kernel.org>
Acked-by: Johannes Weiner <hannes@cmpxchg.org>
Cc: Suren Baghdasaryan <surenb@google.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: John Dias <joaodias@google.com>
Cc: Tim Murray <timmurray@google.com>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Vladimir Davydov <vdavydov.dev@gmail.com>
Cc: Martin Liu <liumartin@google.com>
Cc: Minchan Kim <minchan@kernel.org>
Cc: Matthew Wilcox <willy@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2022-05-19 14:08:54 -07:00
|
|
|
|
2016-03-17 14:20:01 -07:00
|
|
|
|
|
|
|
|
|
2022-01-29 16:06:53 -05:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
2016-03-17 14:20:01 -07:00
|
|
|
|
2022-01-29 16:06:53 -05:00
|
|
|
|
ksm: rmap_walk to remove_migation_ptes
A side-effect of making ksm pages swappable is that they have to be placed
on the LRUs: which then exposes them to isolate_lru_page() and hence to
page migration.
Add rmap_walk() for remove_migration_ptes() to use: rmap_walk_anon() and
rmap_walk_file() in rmap.c, but rmap_walk_ksm() in ksm.c. Perhaps some
consolidation with existing code is possible, but don't attempt that yet
(try_to_unmap needs to handle nonlinears, but migration pte removal does
not).
rmap_walk() is sadly less general than it appears: rmap_walk_anon(), like
remove_anon_migration_ptes() which it replaces, avoids calling
page_lock_anon_vma(), because that includes a page_mapped() test which
fails when all migration ptes are in place. That was valid when NUMA page
migration was introduced (holding mmap_sem provided the missing guarantee
that anon_vma's slab had not already been destroyed), but I believe not
valid in the memory hotremove case added since.
For now do the same as before, and consider the best way to fix that
unlikely race later on. When fixed, we can probably use rmap_walk() on
hwpoisoned ksm pages too: for now, they remain among hwpoison's various
exceptions (its PageKsm test comes before the page is locked, but its
page_lock_anon_vma fails safely if an anon gets upgraded).
Signed-off-by: Hugh Dickins <hugh.dickins@tiscali.co.uk>
Cc: Izik Eidus <ieidus@redhat.com>
Cc: Andrea Arcangeli <aarcange@redhat.com>
Cc: Chris Wright <chrisw@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2009-12-14 17:59:31 -08:00
|
|
|
|
2010-05-28 09:29:16 +09:00
|
|
|
|
2010-06-15 13:18:13 +09:00
|
|
|
|
2010-05-28 09:29:16 +09:00
|
|
|
|
2018-12-28 00:39:31 -08:00
|
|
|
|
2010-05-28 09:29:16 +09:00
|
|
|
|
|
|
|
|
|
|
|
|
|
|
mm/rmap: rename hugepage_add* to hugetlb_add*
Patch series "mm/rmap: interface overhaul", v2.
This series overhauls the rmap interface, to get rid of the "bool
compound" / RMAP_COMPOUND parameter with the goal of making the interface
less error prone, more future proof, and more natural to extend to
"batching". Also, this converts the interface to always consume
folio+subpage, which speeds up operations on large folios.
Further, this series adds PTE-batching variants for 4 rmap functions,
whereby only folio_add_anon_rmap_ptes() is used for batching in this
series when PTE-remapping a PMD-mapped THP. folio_remove_rmap_ptes(),
folio_try_dup_anon_rmap_ptes() and folio_dup_file_rmap_ptes() will soon
come in handy[1,2].
This series performs a lot of folio conversion along the way. Most of the
added LOC in the diff are only due to documentation.
As we're moving to a pte/pmd interface where we clearly express the
mapping granularity we are dealing with, we first get the remainder of
hugetlb out of the way, as it is special and expected to remain special:
it treats everything as a "single logical PTE" and only currently allows
entire mappings.
Even if we'd ever support partial mappings, I strongly assume the
interface and implementation will still differ heavily: hopefull we can
avoid working on subpages/subpage mapcounts completely and only add a
"count" parameter for them to enable batching.
New (extended) hugetlb interface that operates on entire folio:
* hugetlb_add_new_anon_rmap() -> Already existed
* hugetlb_add_anon_rmap() -> Already existed
* hugetlb_try_dup_anon_rmap()
* hugetlb_try_share_anon_rmap()
* hugetlb_add_file_rmap()
* hugetlb_remove_rmap()
New "ordinary" interface for small folios / THP::
* folio_add_new_anon_rmap() -> Already existed
* folio_add_anon_rmap_[pte|ptes|pmd]()
* folio_try_dup_anon_rmap_[pte|ptes|pmd]()
* folio_try_share_anon_rmap_[pte|pmd]()
* folio_add_file_rmap_[pte|ptes|pmd]()
* folio_dup_file_rmap_[pte|ptes|pmd]()
* folio_remove_rmap_[pte|ptes|pmd]()
folio_add_new_anon_rmap() will always map at the largest granularity
possible (currently, a single PMD to cover a PMD-sized THP). Could be
extended if ever required.
In the future, we might want "_pud" variants and eventually "_pmds"
variants for batching.
I ran some simple microbenchmarks on an Intel(R) Xeon(R) Silver 4210R:
measuring munmap(), fork(), cow, MADV_DONTNEED on each PTE ... and PTE
remapping PMD-mapped THPs on 1 GiB of memory.
For small folios, there is barely a change (< 1% improvement for me).
For PTE-mapped THP:
* PTE-remapping a PMD-mapped THP is more than 10% faster.
* fork() is more than 4% faster.
* MADV_DONTNEED is 2% faster
* COW when writing only a single byte on a COW-shared PTE is 1% faster
* munmap() barely changes (< 1%).
[1] https://lkml.kernel.org/r/20230810103332.3062143-1-ryan.roberts@arm.com
[2] https://lkml.kernel.org/r/20231204105440.61448-1-ryan.roberts@arm.com
This patch (of 40):
Let's just call it "hugetlb_".
Yes, it's all already inconsistent and confusing because we have a lot of
"hugepage_" functions for legacy reasons. But "hugetlb" cannot possibly
be confused with transparent huge pages, and it matches "hugetlb.c" and
"folio_test_hugetlb()". So let's minimize confusion in rmap code.
Link: https://lkml.kernel.org/r/20231220224504.646757-1-david@redhat.com
Link: https://lkml.kernel.org/r/20231220224504.646757-2-david@redhat.com
Signed-off-by: David Hildenbrand <david@redhat.com>
Reviewed-by: Muchun Song <songmuchun@bytedance.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Matthew Wilcox (Oracle) <willy@infradead.org>
Cc: Muchun Song <muchun.song@linux.dev>
Cc: Peter Xu <peterx@redhat.com>
Cc: Ryan Roberts <ryan.roberts@arm.com>
Cc: Yin Fengwei <fengwei.yin@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2023-12-20 23:44:25 +01:00
|
|
|
|
|
|
|
|
|
2010-05-28 09:29:16 +09:00
|
|
|
|
2023-12-20 23:44:30 +01:00
|
|
|
|
2023-09-13 14:51:10 +02:00
|
|
|
|
|
|
|
|
|
2023-09-13 14:51:12 +02:00
|
|
|
|
mm: track mapcount of large folios in single value
Let's track the mapcount of large folios in a single value. The mapcount
of a large folio currently corresponds to the sum of the entire mapcount
and all page mapcounts.
This sum is what we actually want to know in folio_mapcount() and it is
also sufficient for implementing folio_mapped().
With PTE-mapped THP becoming more important and more widely used, we want
to avoid looping over all pages of a folio just to obtain the mapcount of
large folios. The comment "In the common case, avoid the loop when no
pages mapped by PTE" in folio_total_mapcount() does no longer hold for
mTHP that are always mapped by PTE.
Further, we are planning on using folio_mapcount() more frequently, and
might even want to remove page mapcounts for large folios in some kernel
configs. Therefore, allow for reading the mapcount of large folios
efficiently and atomically without looping over any pages.
Maintain the mapcount also for hugetlb pages for simplicity. Use the new
mapcount to implement folio_mapcount() and folio_mapped(). Make
page_mapped() simply call folio_mapped(). We can now get rid of
folio_large_is_mapped().
_nr_pages_mapped is now only used in rmap code and for debugging purposes.
Keep folio_nr_pages_mapped() around, but document that its use should be
limited to rmap internals and debugging purposes.
This change implies one additional atomic add/sub whenever
mapping/unmapping (parts of) a large folio.
As we now batch RMAP operations for PTE-mapped THP during fork(), during
unmap/zap, and when PTE-remapping a PMD-mapped THP, and we adjust the
large mapcount for a PTE batch only once, the added overhead in the common
case is small. Only when unmapping individual pages of a large folio
(e.g., during COW), the overhead might be bigger in comparison, but it's
essentially one additional atomic operation.
Note that before the new mapcount would overflow, already our refcount
would overflow: each mapping requires a folio reference. Extend the
focumentation of folio_mapcount().
Link: https://lkml.kernel.org/r/20240409192301.907377-5-david@redhat.com
Signed-off-by: David Hildenbrand <david@redhat.com>
Reviewed-by: Yin Fengwei <fengwei.yin@intel.com>
Cc: Chris Zankel <chris@zankel.net>
Cc: Hugh Dickins <hughd@google.com>
Cc: John Paul Adrian Glaubitz <glaubitz@physik.fu-berlin.de>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Matthew Wilcox (Oracle) <willy@infradead.org>
Cc: Max Filippov <jcmvbkbc@gmail.com>
Cc: Miaohe Lin <linmiaohe@huawei.com>
Cc: Muchun Song <muchun.song@linux.dev>
Cc: Naoya Horiguchi <nao.horiguchi@gmail.com>
Cc: Peter Xu <peterx@redhat.com>
Cc: Richard Chang <richardycc@google.com>
Cc: Rich Felker <dalias@libc.org>
Cc: Ryan Roberts <ryan.roberts@arm.com>
Cc: Yang Shi <shy828301@gmail.com>
Cc: Yoshinori Sato <ysato@users.sourceforge.jp>
Cc: Zi Yan <ziy@nvidia.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2024-04-09 21:22:47 +02:00
|
|
|
|
2023-09-13 14:51:09 +02:00
|
|
|
|
2023-09-13 14:51:13 +02:00
|
|
|
|
2023-09-13 14:51:12 +02:00
|
|
|
|
2023-09-13 14:51:13 +02:00
|
|
|
|
2010-05-28 09:29:16 +09:00
|
|
|
|
|
|
|
|
|
mm/rmap: rename hugepage_add* to hugetlb_add*
Patch series "mm/rmap: interface overhaul", v2.
This series overhauls the rmap interface, to get rid of the "bool
compound" / RMAP_COMPOUND parameter with the goal of making the interface
less error prone, more future proof, and more natural to extend to
"batching". Also, this converts the interface to always consume
folio+subpage, which speeds up operations on large folios.
Further, this series adds PTE-batching variants for 4 rmap functions,
whereby only folio_add_anon_rmap_ptes() is used for batching in this
series when PTE-remapping a PMD-mapped THP. folio_remove_rmap_ptes(),
folio_try_dup_anon_rmap_ptes() and folio_dup_file_rmap_ptes() will soon
come in handy[1,2].
This series performs a lot of folio conversion along the way. Most of the
added LOC in the diff are only due to documentation.
As we're moving to a pte/pmd interface where we clearly express the
mapping granularity we are dealing with, we first get the remainder of
hugetlb out of the way, as it is special and expected to remain special:
it treats everything as a "single logical PTE" and only currently allows
entire mappings.
Even if we'd ever support partial mappings, I strongly assume the
interface and implementation will still differ heavily: hopefull we can
avoid working on subpages/subpage mapcounts completely and only add a
"count" parameter for them to enable batching.
New (extended) hugetlb interface that operates on entire folio:
* hugetlb_add_new_anon_rmap() -> Already existed
* hugetlb_add_anon_rmap() -> Already existed
* hugetlb_try_dup_anon_rmap()
* hugetlb_try_share_anon_rmap()
* hugetlb_add_file_rmap()
* hugetlb_remove_rmap()
New "ordinary" interface for small folios / THP::
* folio_add_new_anon_rmap() -> Already existed
* folio_add_anon_rmap_[pte|ptes|pmd]()
* folio_try_dup_anon_rmap_[pte|ptes|pmd]()
* folio_try_share_anon_rmap_[pte|pmd]()
* folio_add_file_rmap_[pte|ptes|pmd]()
* folio_dup_file_rmap_[pte|ptes|pmd]()
* folio_remove_rmap_[pte|ptes|pmd]()
folio_add_new_anon_rmap() will always map at the largest granularity
possible (currently, a single PMD to cover a PMD-sized THP). Could be
extended if ever required.
In the future, we might want "_pud" variants and eventually "_pmds"
variants for batching.
I ran some simple microbenchmarks on an Intel(R) Xeon(R) Silver 4210R:
measuring munmap(), fork(), cow, MADV_DONTNEED on each PTE ... and PTE
remapping PMD-mapped THPs on 1 GiB of memory.
For small folios, there is barely a change (< 1% improvement for me).
For PTE-mapped THP:
* PTE-remapping a PMD-mapped THP is more than 10% faster.
* fork() is more than 4% faster.
* MADV_DONTNEED is 2% faster
* COW when writing only a single byte on a COW-shared PTE is 1% faster
* munmap() barely changes (< 1%).
[1] https://lkml.kernel.org/r/20230810103332.3062143-1-ryan.roberts@arm.com
[2] https://lkml.kernel.org/r/20231204105440.61448-1-ryan.roberts@arm.com
This patch (of 40):
Let's just call it "hugetlb_".
Yes, it's all already inconsistent and confusing because we have a lot of
"hugepage_" functions for legacy reasons. But "hugetlb" cannot possibly
be confused with transparent huge pages, and it matches "hugetlb.c" and
"folio_test_hugetlb()". So let's minimize confusion in rmap code.
Link: https://lkml.kernel.org/r/20231220224504.646757-1-david@redhat.com
Link: https://lkml.kernel.org/r/20231220224504.646757-2-david@redhat.com
Signed-off-by: David Hildenbrand <david@redhat.com>
Reviewed-by: Muchun Song <songmuchun@bytedance.com>
Cc: Hugh Dickins <hughd@google.com>
Cc: Matthew Wilcox (Oracle) <willy@infradead.org>
Cc: Muchun Song <muchun.song@linux.dev>
Cc: Peter Xu <peterx@redhat.com>
Cc: Ryan Roberts <ryan.roberts@arm.com>
Cc: Yin Fengwei <fengwei.yin@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2023-12-20 23:44:25 +01:00
|
|
|
|
|
|
|
|
|
2010-05-28 09:29:16 +09:00
|
|
|
|
2023-12-20 23:44:30 +01:00
|
|
|
|
|
|
|
|
|
2010-05-28 09:29:16 +09:00
|
|
|
|
mm,thp,rmap: simplify compound page mapcount handling
Compound page (folio) mapcount calculations have been different for anon
and file (or shmem) THPs, and involved the obscure PageDoubleMap flag.
And each huge mapping and unmapping of a file (or shmem) THP involved
atomically incrementing and decrementing the mapcount of every subpage of
that huge page, dirtying many struct page cachelines.
Add subpages_mapcount field to the struct folio and first tail page, so
that the total of subpage mapcounts is available in one place near the
head: then page_mapcount() and total_mapcount() and page_mapped(), and
their folio equivalents, are so quick that anon and file and hugetlb don't
need to be optimized differently. Delete the unloved PageDoubleMap.
page_add and page_remove rmap functions must now maintain the
subpages_mapcount as well as the subpage _mapcount, when dealing with pte
mappings of huge pages; and correct maintenance of NR_ANON_MAPPED and
NR_FILE_MAPPED statistics still needs reading through the subpages, using
nr_subpages_unmapped() - but only when first or last pmd mapping finds
subpages_mapcount raised (double-map case, not the common case).
But are those counts (used to decide when to split an anon THP, and in
vmscan's pagecache_reclaimable heuristic) correctly maintained? Not
quite: since page_remove_rmap() (and also split_huge_pmd()) is often
called without page lock, there can be races when a subpage pte mapcount
0<->1 while compound pmd mapcount 0<->1 is scanning - races which the
previous implementation had prevented. The statistics might become
inaccurate, and even drift down until they underflow through 0. That is
not good enough, but is better dealt with in a followup patch.
Update a few comments on first and second tail page overlaid fields.
hugepage_add_new_anon_rmap() has to "increment" compound_mapcount, but
subpages_mapcount and compound_pincount are already correctly at 0, so
delete its reinitialization of compound_pincount.
A simple 100 X munmap(mmap(2GB, MAP_SHARED|MAP_POPULATE, tmpfs), 2GB) took
18 seconds on small pages, and used to take 1 second on huge pages, but
now takes 119 milliseconds on huge pages. Mapping by pmds a second time
used to take 860ms and now takes 92ms; mapping by pmds after mapping by
ptes (when the scan is needed) used to take 870ms and now takes 495ms.
But there might be some benchmarks which would show a slowdown, because
tail struct pages now fall out of cache until final freeing checks them.
Link: https://lkml.kernel.org/r/47ad693-717-79c8-e1ba-46c3a6602e48@google.com
Signed-off-by: Hugh Dickins <hughd@google.com>
Acked-by: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: David Hildenbrand <david@redhat.com>
Cc: James Houghton <jthoughton@google.com>
Cc: John Hubbard <jhubbard@nvidia.com>
Cc: Matthew Wilcox (Oracle) <willy@infradead.org>
Cc: Miaohe Lin <linmiaohe@huawei.com>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Mina Almasry <almasrymina@google.com>
Cc: Muchun Song <songmuchun@bytedance.com>
Cc: Naoya Horiguchi <naoya.horiguchi@linux.dev>
Cc: Peter Xu <peterx@redhat.com>
Cc: Sidhartha Kumar <sidhartha.kumar@oracle.com>
Cc: Vlastimil Babka <vbabka@suse.cz>
Cc: Yang Shi <shy828301@gmail.com>
Cc: Zach O'Keefe <zokeefe@google.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2022-11-02 18:51:38 -07:00
|
|
|
|
2023-01-11 14:28:56 +00:00
|
|
|
|
mm: track mapcount of large folios in single value
Let's track the mapcount of large folios in a single value. The mapcount
of a large folio currently corresponds to the sum of the entire mapcount
and all page mapcounts.
This sum is what we actually want to know in folio_mapcount() and it is
also sufficient for implementing folio_mapped().
With PTE-mapped THP becoming more important and more widely used, we want
to avoid looping over all pages of a folio just to obtain the mapcount of
large folios. The comment "In the common case, avoid the loop when no
pages mapped by PTE" in folio_total_mapcount() does no longer hold for
mTHP that are always mapped by PTE.
Further, we are planning on using folio_mapcount() more frequently, and
might even want to remove page mapcounts for large folios in some kernel
configs. Therefore, allow for reading the mapcount of large folios
efficiently and atomically without looping over any pages.
Maintain the mapcount also for hugetlb pages for simplicity. Use the new
mapcount to implement folio_mapcount() and folio_mapped(). Make
page_mapped() simply call folio_mapped(). We can now get rid of
folio_large_is_mapped().
_nr_pages_mapped is now only used in rmap code and for debugging purposes.
Keep folio_nr_pages_mapped() around, but document that its use should be
limited to rmap internals and debugging purposes.
This change implies one additional atomic add/sub whenever
mapping/unmapping (parts of) a large folio.
As we now batch RMAP operations for PTE-mapped THP during fork(), during
unmap/zap, and when PTE-remapping a PMD-mapped THP, and we adjust the
large mapcount for a PTE batch only once, the added overhead in the common
case is small. Only when unmapping individual pages of a large folio
(e.g., during COW), the overhead might be bigger in comparison, but it's
essentially one additional atomic operation.
Note that before the new mapcount would overflow, already our refcount
would overflow: each mapping requires a folio reference. Extend the
focumentation of folio_mapcount().
Link: https://lkml.kernel.org/r/20240409192301.907377-5-david@redhat.com
Signed-off-by: David Hildenbrand <david@redhat.com>
Reviewed-by: Yin Fengwei <fengwei.yin@intel.com>
Cc: Chris Zankel <chris@zankel.net>
Cc: Hugh Dickins <hughd@google.com>
Cc: John Paul Adrian Glaubitz <glaubitz@physik.fu-berlin.de>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Matthew Wilcox (Oracle) <willy@infradead.org>
Cc: Max Filippov <jcmvbkbc@gmail.com>
Cc: Miaohe Lin <linmiaohe@huawei.com>
Cc: Muchun Song <muchun.song@linux.dev>
Cc: Naoya Horiguchi <nao.horiguchi@gmail.com>
Cc: Peter Xu <peterx@redhat.com>
Cc: Richard Chang <richardycc@google.com>
Cc: Rich Felker <dalias@libc.org>
Cc: Ryan Roberts <ryan.roberts@arm.com>
Cc: Yang Shi <shy828301@gmail.com>
Cc: Yoshinori Sato <ysato@users.sourceforge.jp>
Cc: Zi Yan <ziy@nvidia.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
2024-04-09 21:22:47 +02:00
|
|
|
|
2023-01-11 14:28:56 +00:00
|
|
|
|
2023-09-13 14:51:09 +02:00
|
|
|
|
|
|
|
|
|
2010-05-28 09:29:16 +09:00
|
|
|
|
2010-06-15 13:18:13 +09:00
|
|
|
|
