Patch series "userfaultfd shmem updates".
Jann found two bugs in the userfaultfd shmem MAP_SHARED backend: the
lack of the VM_MAYWRITE check and the lack of i_size checks.
Then looking into the above we also fixed the MAP_PRIVATE case.
Hugh by source review also found a data loss source if UFFDIO_COPY is
used on shmem MAP_SHARED PROT_READ mappings (the production usages
incidentally run with PROT_READ|PROT_WRITE, so the data loss couldn't
happen in those production usages like with QEMU).
The whole patchset is marked for stable.
We verified QEMU postcopy live migration with guest running on shmem
MAP_PRIVATE run as well as before after the fix of shmem MAP_PRIVATE.
Regardless if it's shmem or hugetlbfs or MAP_PRIVATE or MAP_SHARED, QEMU
unconditionally invokes a punch hole if the guest mapping is filebacked
and a MADV_DONTNEED too (needed to get rid of the MAP_PRIVATE COWs and
for the anon backend).
This patch (of 5):
We internally used EFAULT to communicate with the caller, switch to
ENOENT, so EFAULT can be used as a non internal retval.
Link: http://lkml.kernel.org/r/20181126173452.26955-2-aarcange@redhat.com
Fixes: 4c27fe4c4c ("userfaultfd: shmem: add shmem_mcopy_atomic_pte for userfaultfd support")
Signed-off-by: Andrea Arcangeli <aarcange@redhat.com>
Reviewed-by: Mike Rapoport <rppt@linux.ibm.com>
Reviewed-by: Hugh Dickins <hughd@google.com>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Jann Horn <jannh@google.com>
Cc: Peter Xu <peterx@redhat.com>
Cc: "Dr. David Alan Gilbert" <dgilbert@redhat.com>
Cc: <stable@vger.kernel.org>
Cc: stable@vger.kernel.org
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
We free the misc device string twice on rmmod; fix this. Without this
we cannot remove the module without crashing.
Link: http://lkml.kernel.org/r/20181124050500.5257-1-mcgrof@kernel.org
Signed-off-by: Luis Chamberlain <mcgrof@kernel.org>
Reported-by: Randy Dunlap <rdunlap@infradead.org>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Cc: <stable@vger.kernel.org> [4.12+]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
hfs_bmap_free() frees node via hfs_bnode_put(node). However it then
reads node->this when dumping error message on an error path, which may
result in a use-after-free bug. This patch frees node only when it is
never used.
Link: http://lkml.kernel.org/r/1543053441-66942-1-git-send-email-bianpan2016@163.com
Signed-off-by: Pan Bian <bianpan2016@163.com>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Ernesto A. Fernandez <ernesto.mnd.fernandez@gmail.com>
Cc: Joe Perches <joe@perches.com>
Cc: Viacheslav Dubeyko <slava@dubeyko.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
hfs_bmap_free() frees the node via hfs_bnode_put(node). However, it
then reads node->this when dumping error message on an error path, which
may result in a use-after-free bug. This patch frees the node only when
it is never again used.
Link: http://lkml.kernel.org/r/1542963889-128825-1-git-send-email-bianpan2016@163.com
Fixes: a1185ffa2fc ("HFS rewrite")
Signed-off-by: Pan Bian <bianpan2016@163.com>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Cc: Joe Perches <joe@perches.com>
Cc: Ernesto A. Fernandez <ernesto.mnd.fernandez@gmail.com>
Cc: Viacheslav Dubeyko <slava@dubeyko.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Turns out that /proc has official documentation and people even trying
to keep it uptodate.
Link: http://lkml.kernel.org/r/20181116134630.GA8004@avx2
Signed-off-by: Alexey Dobriyan <adobriyan@gmail.com>
Reviewed-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
init_currently_empty_zone() will adjust pgdat->nr_zones and set it to
'zone_idx(zone) + 1' unconditionally. This is correct in the normal
case, while not exact in hot-plug situation.
This function is used in two places:
* free_area_init_core()
* move_pfn_range_to_zone()
In the first case, we are sure zone index increase monotonically. While
in the second one, this is under users control.
One way to reproduce this is:
----------------------------
1. create a virtual machine with empty node1
-m 4G,slots=32,maxmem=32G \
-smp 4,maxcpus=8 \
-numa node,nodeid=0,mem=4G,cpus=0-3 \
-numa node,nodeid=1,mem=0G,cpus=4-7
2. hot-add cpu 3-7
cpu-add [3-7]
2. hot-add memory to nod1
object_add memory-backend-ram,id=ram0,size=1G
device_add pc-dimm,id=dimm0,memdev=ram0,node=1
3. online memory with following order
echo online_movable > memory47/state
echo online > memory40/state
After this, node1 will have its nr_zones equals to (ZONE_NORMAL + 1)
instead of (ZONE_MOVABLE + 1).
Michal said:
"Having an incorrect nr_zones might result in all sorts of problems
which would be quite hard to debug (e.g. reclaim not considering the
movable zone). I do not expect many users would suffer from this it
but still this is trivial and obviously right thing to do so
backporting to the stable tree shouldn't be harmful (last famous
words)"
Link: http://lkml.kernel.org/r/20181117022022.9956-1-richard.weiyang@gmail.com
Fixes: f1dd2cd13c ("mm, memory_hotplug: do not associate hotadded memory to zones until online")
Signed-off-by: Wei Yang <richard.weiyang@gmail.com>
Acked-by: Michal Hocko <mhocko@suse.com>
Reviewed-by: Oscar Salvador <osalvador@suse.de>
Cc: Anshuman Khandual <anshuman.khandual@arm.com>
Cc: Dave Hansen <dave.hansen@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
We changed the key of swap cache tree from swp_entry_t.val to
swp_offset. We need to do so in shmem_replace_page() as well.
Hugh said:
"shmem_replace_page() has been wrong since the day I wrote it: good
enough to work on swap "type" 0, which is all most people ever use
(especially those few who need shmem_replace_page() at all), but
broken once there are any non-0 swp_type bits set in the higher order
bits"
Link: http://lkml.kernel.org/r/20181121215442.138545-1-yuzhao@google.com
Fixes: f6ab1f7f6b ("mm, swap: use offset of swap entry as key of swap cache")
Signed-off-by: Yu Zhao <yuzhao@google.com>
Reviewed-by: Matthew Wilcox <willy@infradead.org>
Acked-by: Hugh Dickins <hughd@google.com>
Cc: <stable@vger.kernel.org> [4.9+]
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
If all pages are deleted from the mapping by memory reclaim and also
moved to the cleancache:
__delete_from_page_cache
(no shadow case)
unaccount_page_cache_page
cleancache_put_page
page_cache_delete
mapping->nrpages -= nr
(nrpages becomes 0)
We don't clean the cleancache for an inode after final file truncation
(removal).
truncate_inode_pages_final
check (nrpages || nrexceptional) is false
no truncate_inode_pages
no cleancache_invalidate_inode(mapping)
These way when reading the new file created with same inode we may get
these trash leftover pages from cleancache and see wrong data instead of
the contents of the new file.
Fix it by always doing truncate_inode_pages which is already ready for
nrpages == 0 && nrexceptional == 0 case and just invalidates inode.
[akpm@linux-foundation.org: add comment, per Jan]
Link: http://lkml.kernel.org/r/20181112095734.17979-1-ptikhomirov@virtuozzo.com
Fixes: commit 91b0abe36a ("mm + fs: store shadow entries in page cache")
Signed-off-by: Pavel Tikhomirov <ptikhomirov@virtuozzo.com>
Reviewed-by: Vasily Averin <vvs@virtuozzo.com>
Reviewed-by: Andrey Ryabinin <aryabinin@virtuozzo.com>
Reviewed-by: Jan Kara <jack@suse.cz>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: Matthew Wilcox <willy@infradead.org>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
ocfs2_defrag_extent may fall into deadlock.
ocfs2_ioctl_move_extents
ocfs2_ioctl_move_extents
ocfs2_move_extents
ocfs2_defrag_extent
ocfs2_lock_allocators_move_extents
ocfs2_reserve_clusters
inode_lock GLOBAL_BITMAP_SYSTEM_INODE
__ocfs2_flush_truncate_log
inode_lock GLOBAL_BITMAP_SYSTEM_INODE
As backtrace shows above, ocfs2_reserve_clusters() will call inode_lock
against the global bitmap if local allocator has not sufficient cluters.
Once global bitmap could meet the demand, ocfs2_reserve_cluster will
return success with global bitmap locked.
After ocfs2_reserve_cluster(), if truncate log is full,
__ocfs2_flush_truncate_log() will definitely fall into deadlock because
it needs to inode_lock global bitmap, which has already been locked.
To fix this bug, we could remove from
ocfs2_lock_allocators_move_extents() the code which intends to lock
global allocator, and put the removed code after
__ocfs2_flush_truncate_log().
ocfs2_lock_allocators_move_extents() is referred by 2 places, one is
here, the other does not need the data allocator context, which means
this patch does not affect the caller so far.
Link: http://lkml.kernel.org/r/20181101071422.14470-1-lchen@suse.com
Signed-off-by: Larry Chen <lchen@suse.com>
Reviewed-by: Changwei Ge <ge.changwei@h3c.com>
Cc: Mark Fasheh <mark@fasheh.com>
Cc: Joel Becker <jlbec@evilplan.org>
Cc: Junxiao Bi <junxiao.bi@oracle.com>
Cc: Joseph Qi <jiangqi903@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Commit df06b37ffe ("mm/gup: cache dev_pagemap while pinning pages")
attempted to operate on each page that get_user_pages had retrieved. In
order to do that, it created a common exit point from the routine.
However, one case was missed, which this patch fixes up.
Also, there was still an unnecessary shadow declaration (with a
different type) of the "ret" variable, which this patch removes.
Keith's description of the situation is:
This also fixes a potentially leaked dev_pagemap reference count if a
failure occurs when an iteration crosses a vma boundary. I don't think
it's normal to have different vma's on a users mapped zone device
memory, but good to fix anyway.
I actually thought that this code:
/* first iteration or cross vma bound */
if (!vma || start >= vma->vm_end) {
vma = find_extend_vma(mm, start);
if (!vma && in_gate_area(mm, start)) {
ret = get_gate_page(mm, start & PAGE_MASK,
gup_flags, &vma,
pages ? &pages[i] : NULL);
if (ret)
goto out;
dealt with the "you're trying to pin the gate page, as part of this
call", rather than the generic case of crossing a vma boundary. (I
think there's a fine point that I must be overlooking.) But it's still a
valid case, either way.
Link: http://lkml.kernel.org/r/20181121081402.29641-2-jhubbard@nvidia.com
Fixes: df06b37ffe ("mm/gup: cache dev_pagemap while pinning pages")
Signed-off-by: John Hubbard <jhubbard@nvidia.com>
Reviewed-by: Keith Busch <keith.busch@intel.com>
Cc: Dan Williams <dan.j.williams@intel.com>
Cc: Kirill A. Shutemov <kirill.shutemov@linux.intel.com>
Cc: Dave Hansen <dave.hansen@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
My name has changed, works better than Global Entry I tell ya.
Link: http://lkml.kernel.org/r/20181122003138.7752-1-mcgrof@kernel.org
Signed-off-by: Luis Chamberlain <mcgrof@kernel.org>
Cc: Kees Cook <keescook@chromium.org>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
New versions of gcc reasonably warn about the odd pattern of
strncpy(p, q, strlen(q));
which really doesn't make sense: the strncpy() ends up being just a slow
and odd way to write memcpy() in this case.
There was a comment about _why_ the code used strncpy - to avoid the
terminating NUL byte, but memcpy does the same and avoids the warning.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
The message got changed a lot time ago.
This was responsible for 36 test case failures on sparc64.
Fixes: f1174f77b5 ("bpf/verifier: rework value tracking")
Signed-off-by: David S. Miller <davem@davemloft.net>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Quentin Monnet says:
====================
Hi,
Several items for bpftool are included in this set: the first three patches
are fixes for bpftool itself and bash completion, while the last two
slightly improve the information obtained when dumping programs or maps, on
Daniel's suggestion. Please refer to individual commit logs for more
details.
====================
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
For prog array maps, the type of the owner program, and the JIT-ed state
of that program, are available from the file descriptor information
under /proc. Add them to "bpftool map show" output. Example output:
# bpftool map show
158225: prog_array name jmp_table flags 0x0
key 4B value 4B max_entries 8 memlock 4096B
owner_prog_type flow_dissector owner jited
# bpftool --json --pretty map show
[{
"id": 1337,
"type": "prog_array",
"name": "jmp_table",
"flags": 0,
"bytes_key": 4,
"bytes_value": 4,
"max_entries": 8,
"bytes_memlock": 4096,
"owner_prog_type": "flow_dissector",
"owner_jited": true
}
]
As we move the table used for associating names to program types,
complete it with the missing types (lwt_seg6local and sk_reuseport).
Also add missing types to the help message for "bpftool prog"
(sk_reuseport and flow_dissector).
Suggested-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: Quentin Monnet <quentin.monnet@netronome.com>
Acked-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
In bpftool (plain) output for "bpftool prog show" or "bpftool map show",
an offloaded BPF object is simply denoted with "dev ifname", which is
not really explicit. Change it with something that clearly shows the
program is offloaded.
While at it also add an additional space, as done between other
information fields.
Example output, before:
# bpftool prog show
1337: xdp tag a04f5eef06a7f555 dev foo
loaded_at 2018-10-19T16:40:36+0100 uid 0
xlated 16B not jited memlock 4096B
After:
# bpftool prog show
1337: xdp tag a04f5eef06a7f555 offloaded_to foo
loaded_at 2018-10-19T16:40:36+0100 uid 0
xlated 16B not jited memlock 4096B
Suggested-by: Daniel Borkmann <daniel@iogearbox.net>
Signed-off-by: Quentin Monnet <quentin.monnet@netronome.com>
Acked-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Commit 197c2dac74 ("bpf: Add BPF_MAP_TYPE_QUEUE and BPF_MAP_TYPE_STACK
to bpftool-map") added support for queue and stack eBPF map types in
bpftool map handling. Let's update the bash completion accordingly.
Fixes: 197c2dac74 ("bpf: Add BPF_MAP_TYPE_QUEUE and BPF_MAP_TYPE_STACK to bpftool-map")
Signed-off-by: Quentin Monnet <quentin.monnet@netronome.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Fix bash completion for "bpftool prog (attach|detach) PROG TYPE MAP" so
that the list of indices proposed for MAP are map indices, and not PROG
indices. Also use variables for map and prog reference types ("id",
"pinned", and "tag" for programs).
Fixes: b7d3826c2e ("bpf: bpftool, add support for attaching programs to maps")
Signed-off-by: Quentin Monnet <quentin.monnet@netronome.com>
Reviewed-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
The getpid() function is called in a couple of places in bpftool to
craft links of the shape "/proc/<pid>/...". Instead, it is possible to
use the "/proc/self/" shortcut, which makes things a bit easier, in
particular in jit_disasm.c.
Do the replacement, and remove the includes of <sys/types.h> from the
relevant files, now we do not use getpid() anymore.
Signed-off-by: Quentin Monnet <quentin.monnet@netronome.com>
Reviewed-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Signed-off-by: Alexei Starovoitov <ast@kernel.org>
Even if we have no waiters on any of the sbitmap_queue wait states, we
still have to loop every entry to check. We do this for every IO, so
the cost adds up.
Shift a bit of the cost to the slow path, when we actually have waiters.
Wrap prepare_to_wait_exclusive() and finish_wait(), so we can maintain
an internal count of how many are currently active. Then we can simply
check this count in sbq_wake_ptr() and not have to loop if we don't
have any sleepers.
Convert the two users of sbitmap with waiting, blk-mq-tag and iSCSI.
Reviewed-by: Omar Sandoval <osandov@fb.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
sbitmap maintains a set of words that we use to set and clear bits, with
each bit representing a tag for blk-mq. Even though we spread the bits
out and maintain a hint cache, one particular bit allocated will end up
being cleared in the exact same spot.
This introduces batched clearing of bits. Instead of clearing a given
bit, the same bit is set in a cleared/free mask instead. If we fail
allocating a bit from a given word, then we check the free mask, and
batch move those cleared bits at that time. This trades 64 atomic bitops
for 2 cmpxchg().
In a threaded poll test case, half the overhead of getting and clearing
tags is removed with this change. On another poll test case with a
single thread, performance is unchanged.
Reviewed-by: Omar Sandoval <osandov@fb.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
v3d_bo_get_pages() checks this to decide to map the imported buffer
instead of the backing shmem file. The caller was about to set this
value anyway, and there's no error path in between. Ideally we
wouldn't even allocate the shmem file for our imports, but that's a
more invasive fix.
Signed-off-by: Eric Anholt <eric@anholt.net>
Fixes: 57692c94dc ("drm/v3d: Introduce a new DRM driver for Broadcom V3D V3.x+")
Link: https://patchwork.freedesktop.org/patch/msgid/20181128230927.10951-3-eric@anholt.net
Acked-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Reviewed-by: Dave Emett <david.emett@broadcom.com>
Ariel Elior says:
====================
qed*: Doorbell overflow recovery
Doorbell Overflow
If sufficient CPU cores will send doorbells at a sufficiently high rate, they
can cause an overflow in the doorbell queue block message fifo. When fill level
reaches maximum, the device stops accepting all doorbells from that PF until a
recovery procedure has taken place.
Doorbell Overflow Recovery
The recovery procedure basically means resending the last doorbell for every
doorbelling entity. A doorbelling entity is anything which may send doorbells:
L2 tx ring, rdma sq/rq/cq, light l2, vf l2 tx ring, spq, etc. This relies on
the design assumption that all doorbells are aggregative, so last doorbell
carries the information of all previous doorbells.
APIs
All doorbelling entities need to register with the mechanism before sending
doorbells. The registration entails providing the doorbell address the entity
would be using, and a virtual address where last doorbell data can be found.
Typically fastpath structures already have this construct.
Executing the recovery procedure
Handling the attentions, iterating over all the registered entities and
resending their doorbells, is all handled within qed core module.
Relevance
All doorbelling entities in all protocols need to register with the mechanism,
via the new APIs. Technically this is quite simple (just call the API). Some
protocol fastpath implementation may not have the doorbell data stored anywhere
(compute it from scratch every time) and will have to add such a place.
This is rare and is also better practice (save some cycles on the fastpath).
Performance Penalty
No performance penalty should incur as a result of this feature. If anything
performance can improve by avoiding recalcualtion of doorbell data everytime
doorbell is sent (in some flows).
Add the database used to register doorbelling entities, and APIs for adding
and deleting entries, and logic for traversing the database and doorbelling
once on behalf of all entities.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
All L2 queues funnel through this flow, so this would cover the
regular RSS queues, as well queues created for VFs, mqos queues,
xdp queues, etc.
Signed-off-by: Ariel Elior <Ariel.Elior@cavium.com>
Signed-off-by: Michal Kalderon <Michal.Kalderon@cavium.com>
Signed-off-by: Tomer Tayar <Tomer.Tayar@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Most of the doorbelling entities are outside of the core module.
L2 queues, Roce queues, iscsi and fcoe all need to register.
Make the APIs available for these drivers.
Signed-off-by: Ariel Elior <Ariel.Elior@cavium.com>
Signed-off-by: Michal Kalderon <Michal.Kalderon@cavium.com>
Signed-off-by: Tomer Tayar <Tomer.Tayar@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Light L2 queues are doorbelling entities. Modify the implementation
to keep the doorbell data necessary for doorbelling in well known
location instead of recomputing every time. Register the LL2 queue
with doorbell recovery mechanism.
Signed-off-by: Ariel Elior <Ariel.Elior@cavium.com>
Signed-off-by: Michal Kalderon <Michal.Kalderon@cavium.com>
Signed-off-by: Tomer Tayar <Tomer.Tayar@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Slow path queue is a doorbelling entity. Register it with the overflow mechanism.
Signed-off-by: Ariel Elior <Ariel.Elior@cavium.com>
Signed-off-by: Michal Kalderon <Michal.Kalderon@cavium.com>
Signed-off-by: Tomer Tayar <Tomer.Tayar@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
In case of an attention from the doorbell queue block, analyze the HW
indications. In case of a doorbell overflow, execute a doorbell recovery.
Since there can be spurious indications (race conditions between multiple PFs),
schedule a periodic task for checking whether a doorbell overflow may have been
missed. After a set time with no indications, terminate the periodic task.
Signed-off-by: Ariel Elior <Ariel.Elior@cavium.com>
Signed-off-by: Michal Kalderon <Michal.Kalderon@cavium.com>
Signed-off-by: Tomer Tayar <Tomer.Tayar@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Add the database used to register doorbelling entities, and APIs for adding
and deleting entries, and logic for traversing the database and doorbelling
once on behalf of all entities.
Signed-off-by: Ariel Elior <Ariel.Elior@cavium.com>
Signed-off-by: Michal Kalderon <Michal.Kalderon@cavium.com>
Signed-off-by: Tomer Tayar <Tomer.Tayar@cavium.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Jakub Kicinski says:
====================
rtnetlink: avoid a warning in rtnl_newlink()
I've been hoping for some time that someone more competent would fix
the stack frame size warning in rtnl_newlink(), but looks like I'll
have to take a stab at it myself :) That's the only warning I see
in most of my builds.
First patch refactors away a somewhat surprising if (1) code block.
Reindentation will most likely cause cherry-pick problems but OTOH
rtnl_newlink() doesn't seem to be changed often, so perhaps we can
risk it in the name of cleaner code?
Second patch fixes the warning in simplest possible way. I was
pondering if there is any more clever solution, but I can't see it..
rtnl_newlink() is quite long with a lot of possible execution paths
so doing memory allocations half way through leads to very ugly
results.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
Standard kernel compilation produces the following warning:
net/core/rtnetlink.c: In function ‘rtnl_newlink’:
net/core/rtnetlink.c:3232:1: warning: the frame size of 1288 bytes is larger than 1024 bytes [-Wframe-larger-than=]
}
^
This should not really be an issue, as rtnl_newlink() stack is
generally quite shallow.
Fix the warning by allocating attributes with kmalloc() in a wrapper
and passing it down to rtnl_newlink(), avoiding complexities on error
paths.
Alternatively we could kmalloc() some structure within rtnl_newlink(),
slave attributes look like a good candidate. In practice it adds to
already rather high complexity and length of the function.
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
rtnl_newlink() used to create VLAs based on link kind. Since
commit ccf8dbcd06 ("rtnetlink: Remove VLA usage") statically
sized array is created on the stack, so there is no more use
for a separate code block that used to be the VLA's live range.
While at it christmas tree the variables. Note that there is
a goto-based retry so to be on the safe side the variables can
no longer be initialized in place. It doesn't seem to matter,
logically, but why make the code harder to read..
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: David Ahern <dsahern@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Jakub Kicinski says:
====================
nfp: update TX path to enable repr offloads
This set starts with three micro optimizations to the TX path.
The improvement is measurable, but below 1% of CPU utilization.
Patches 4 - 9 add basic TX offloads to representor devices, like
checksum offload or TSO, and remove the unnecessary TX lock and
Qdisc (our representors are software constructs on top of the PF).
The last 2 patches add more info to error messages - id of command
which failed and exact location of incorrect TLVs, very useful for
debugging.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
FW reconfiguration timeouts are a common indicator of FW trouble.
To make debugging easier print requested update and control word
when reconfiguration fails.
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: Dirk van der Merwe <dirk.vandermerwe@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
When troubleshooting incorrect FW capabilities it's useful to know
where the faulty TLV is located. Add offset to all errors messages.
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: Dirk van der Merwe <dirk.vandermerwe@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
FW/HW can generally support the standard networking offloads
on representors without any trouble. Add the ability for FW
to advertise which features should be available on representors.
Because representors are muxed on top of the vNIC we need to listen
on feature changes of their lower devices, and update their features
appropriately.
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: John Hurley <john.hurley@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Up until now we never needed to keep a networking locks around
representors accesses, we only accessed them when device was
reconfigured (under nfp pf->lock) or on fast path (under RCU).
Now we want to be able to iterate over all representors during
notifications, so make sure representor assignment is done
under RTNL lock.
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: John Hurley <john.hurley@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Our representors are software devices built on top of the PF
vNIC, the queuing should only happen at the vNIC netdevice.
Allow representors to run qdisc-less.
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: John Hurley <john.hurley@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Our representors are software devices built on top of the PF
vNIC, the only state they have are per-cpu stats, so make
the TX run locklessly.
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: John Hurley <john.hurley@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
In preparation for TSO over representors make sure the port id
prepend will always fit in the frame. The current max header
length is 255, which is ample, so assume worst case scenario
of 8 byte prepend and save ourselves the conditionals.
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: Dirk van der Merwe <dirk.vandermerwe@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
The TSO-related offsets in the descriptor should not include
the length of the prepended metadata. Adjust them. Note that
this could not have caused issues in the past as we don't
support TSO with metadata prepend as of this patch.
Signed-off-by: Michael Rapson <michael.rapson@netronome.com>
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: Dirk van der Merwe <dirk.vandermerwe@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
nd_q is only used at the very end of nfp_net_tx(), there is no need
to initialize it early.
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: Dirk van der Merwe <dirk.vandermerwe@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Move temporary variables in scope of the loop in nfp_net_tx_complete(),
and add a temp for txbuf software structure. This saves us 0.2% of CPU.
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: Dirk van der Merwe <dirk.vandermerwe@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Chained descriptors for fragments need to duplicate all the descriptor
fields of the skb head, so we copy the descriptor and then modify the
relevant fields. This is wasteful, because the top half of the descriptor
will get overwritten entirely while the bottom half is not modified at all.
Copy only the bottom half. This saves us 0.3% of CPU in a GSO test.
Signed-off-by: Jakub Kicinski <jakub.kicinski@netronome.com>
Reviewed-by: Dirk van der Merwe <dirk.vandermerwe@netronome.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Most linux hosts never setup TCP MD5 keys. We can avoid a
cache line miss (accessing tp->md5ig_info) on RX and TX
using a jump label.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Eric Dumazet says:
====================
tcp: take a bit more care of backlog stress
While working on the SACK compression issue Jean-Louis Dupond
reported, we found that his linux box was suffering very hard
from tail drops on the socket backlog queue.
First patch hints the compiler about sack flows being the norm.
Second patch changes non-sack code in preparation of the ack
compression.
Third patch fixes tcp_space() to take backlog into account.
Fourth patch is attempting coalescing when a new packet must
be added to the backlog queue. Cooking bigger skbs helps
to keep backlog list smaller and speeds its handling when
user thread finally releases the socket lock.
v3: Neal/Yuchung feedback addressed :
Do not aggregate if any skb has URG bit set.
Do not aggregate if the skbs have different ECE/CWR bits
v2: added feedback from Neal : tcp: take care of compressed acks in tcp_add_reno_sack()
added : tcp: hint compiler about sack flows
added : tcp: make tcp_space() aware of socket backlog
====================
Signed-off-by: David S. Miller <davem@davemloft.net>
In case GRO is not as efficient as it should be or disabled,
we might have a user thread trapped in __release_sock() while
softirq handler flood packets up to the point we have to drop.
This patch balances work done from user thread and softirq,
to give more chances to __release_sock() to complete its work
before new packets are added the the backlog.
This also helps if we receive many ACK packets, since GRO
does not aggregate them.
This patch brings ~60% throughput increase on a receiver
without GRO, but the spectacular gain is really on
1000x release_sock() latency reduction I have measured.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Cc: Neal Cardwell <ncardwell@google.com>
Cc: Yuchung Cheng <ycheng@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Jean-Louis Dupond reported poor iscsi TCP receive performance
that we tracked to backlog drops.
Apparently we fail to send window updates reflecting the
fact that we are under stress.
Note that we might lack a proper window increase when
backlog is fully processed, since __release_sock() clears
sk->sk_backlog.len _after_ all skbs have been processed.
This should not matter in practice. If we had a significant
load through socket backlog, we are in a dangerous
situation.
Reported-by: Jean-Louis Dupond <jean-louis@dupond.be>
Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Acked-by: Yuchung Cheng <ycheng@google.com>
Tested-by: Jean-Louis Dupond<jean-louis@dupond.be>
Signed-off-by: David S. Miller <davem@davemloft.net>
Neal pointed out that non sack flows might suffer from ACK compression
added in the following patch ("tcp: implement coalescing on backlog queue")
Instead of tweaking tcp_add_backlog() we can take into
account how many ACK were coalesced, this information
will be available in skb_shinfo(skb)->gso_segs
Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
Tell the compiler that most TCP flows are using SACK these days.
There is no need to add the unlikely() clause in tcp_is_reno(),
the compiler is able to infer it.
Signed-off-by: Eric Dumazet <edumazet@google.com>
Acked-by: Neal Cardwell <ncardwell@google.com>
Acked-by: Yuchung Cheng <ycheng@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>
