This may run from a kernel thread via device_add_disk(). So this could
also use __fput_sync() if we were worried about EBUSY. But when it is
called from a kernel thread it's always BLK_OPEN_READ so EBUSY can't
really happen even if we do BLK_OPEN_RESTRICT_WRITES or BLK_OPEN_EXCL.
Otherwise it's called from an ioctl on the block device which is only
called from userspace and can rely on task work.
Link: https://lore.kernel.org/r/20240123-vfs-bdev-file-v2-3-adbd023e19cc@kernel.org
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Jan Kara <jack@suse.cz>
Signed-off-by: Christian Brauner <brauner@kernel.org>
Add two new helpers to allow opening block devices as files.
This is not the final infrastructure. This still opens the block device
before opening a struct a file. Until we have removed all references to
struct bdev_handle we can't switch the order:
* Introduce blk_to_file_flags() to translate from block specific to
flags usable to pen a new file.
* Introduce bdev_file_open_by_{dev,path}().
* Introduce temporary sb_bdev_handle() helper to retrieve a struct
bdev_handle from a block device file and update places that directly
reference struct bdev_handle to rely on it.
* Don't count block device openes against the number of open files. A
bdev_file_open_by_{dev,path}() file is never installed into any
file descriptor table.
One idea that came to mind was to use kernel_tmpfile_open() which
would require us to pass a path and it would then call do_dentry_open()
going through the regular fops->open::blkdev_open() path. But then we're
back to the problem of routing block specific flags such as
BLK_OPEN_RESTRICT_WRITES through the open path and would have to waste
FMODE_* flags every time we add a new one. With this we can avoid using
a flag bit and we have more leeway in how we open block devices from
bdev_open_by_{dev,path}().
Link: https://lore.kernel.org/r/20240123-vfs-bdev-file-v2-1-adbd023e19cc@kernel.org
Signed-off-by: Christian Brauner <brauner@kernel.org>
In order to add a helper to open files that aren't accounted split
alloc_file() and parts of alloc_file_pseudo() into helpers. One to
prepare a path, another one to setup the file.
Suggested-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20240129160241.GA2793@lst.de
Signed-off-by: Christian Brauner <brauner@kernel.org>
Obtain the PMU regmap using the new API added to exynos-pmu driver rather
than syscon_regmap_lookup_by_phandle(). As this driver no longer depends
on mfd syscon remove that header and Kconfig dependency.
Tested-by: Alexey Klimov <alexey.klimov@linaro.org>
Tested-by: Sam Protsenko <semen.protsenko@linaro.org>
Acked-by: Guenter Roeck <linux@roeck-us.net>
Reviewed-by: Sam Protsenko <semen.protsenko@linaro.org>
Signed-off-by: Peter Griffin <peter.griffin@linaro.org>
Link: https://lore.kernel.org/r/20240220220613.797068-3-peter.griffin@linaro.org
Signed-off-by: Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org>
Some Exynos based SoCs like Tensor gs101 protect the PMU registers for
security hardening reasons so that they are only write accessible in el3
via an SMC call.
As most Exynos drivers that need to write PMU registers currently obtain a
regmap via syscon (phys, pinctrl, watchdog). Support for the above usecase
is implemented in this driver using a custom regmap similar to syscon to
handle the SMC call. Platforms that don't secure PMU registers, get a mmio
regmap like before. As regmaps abstract out the underlying register access
changes to the leaf drivers are minimal.
A new API exynos_get_pmu_regmap_by_phandle() is provided for leaf drivers
that currently use syscon_regmap_lookup_by_phandle(). This also handles
deferred probing.
Tested-by: Sam Protsenko <semen.protsenko@linaro.org>
Tested-by: Alexey Klimov <alexey.klimov@linaro.org>
Reviewed-by: Sam Protsenko <semen.protsenko@linaro.org>
Signed-off-by: Peter Griffin <peter.griffin@linaro.org>
Link: https://lore.kernel.org/r/20240220220613.797068-2-peter.griffin@linaro.org
Signed-off-by: Krzysztof Kozlowski <krzysztof.kozlowski@linaro.org>
GCC recently added option -fmin-function-alignment, which should appear
in GCC 14. Unlike -falign-functions, this option causes all functions to
be aligned at the specified value, including the cold ones.
In particular, when an arm64 kernel is built with
DYNAMIC_FTRACE_WITH_CALL_OPS=y, the 8-byte function alignment is
required for correct functionality. This was done by -falign-functions=8
and having workarounds in the kernel to force the compiler to follow
this alignment. The new -fmin-function-alignment option directly
guarantees it.
Detect availability of -fmin-function-alignment and use it instead of
-falign-functions when present. Introduce CC_HAS_SANE_FUNCTION_ALIGNMENT
and enable __cold to work as expected when it is set.
Signed-off-by: Petr Pavlu <petr.pavlu@suse.com>
Reviewed-by: Nathan Chancellor <nathan@kernel.org>
Acked-by: Mark Rutland <mark.rutland@arm.com>
Signed-off-by: Masahiro Yamada <masahiroy@kernel.org>
The 'duplicates' bool argument is always true when efivar_init() is
called from its only caller so let's just drop it instead.
Signed-off-by: Ard Biesheuvel <ardb@kernel.org>
Al points out that kill_sb() will be called if efivarfs_fill_super()
fails and so there is no point in cleaning up the efivar entry list.
Reported-by: Alexander Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Ard Biesheuvel <ardb@kernel.org>
Work around a quirk in a few old (2011-ish) UEFI implementations, where
a call to `GetNextVariableName` with a buffer size larger than 512 bytes
will always return EFI_INVALID_PARAMETER.
There is some lore around EFI variable names being up to 1024 bytes in
size, but this has no basis in the UEFI specification, and the upper
bounds are typically platform specific, and apply to the entire variable
(name plus payload).
Given that Linux does not permit creating files with names longer than
NAME_MAX (255) bytes, 512 bytes (== 256 UTF-16 characters) is a
reasonable limit.
Cc: <stable@vger.kernel.org> # 6.1+
Signed-off-by: Tim Schumacher <timschumi@gmx.de>
Signed-off-by: Ard Biesheuvel <ardb@kernel.org>
1) errors from ext4_getblk() should not be propagated to caller
unless we are really sure that we would've gotten the same error
in non-RCU pathwalk.
2) we leak buffer_heads if ext4_getblk() is successful, but bh is
not uptodate.
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
->d_revalidate() bails out there, anyway. It's not enough
to prevent getting into ->get_link() in RCU mode, but that
could happen only in a very contrieved setup. Not worth
trying to do anything fancy here unless ->d_revalidate()
stops kicking out of RCU mode at least in some cases.
Reviewed-by: Christian Brauner <brauner@kernel.org>
Acked-by: Miklos Szeredi <mszeredi@redhat.com>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
->permission(), ->get_link() and ->inode_get_acl() might dereference
->s_fs_info (and, in case of ->permission(), ->s_fs_info->fc->user_ns
as well) when called from rcu pathwalk.
Freeing ->s_fs_info->fc is rcu-delayed; we need to make freeing ->s_fs_info
and dropping ->user_ns rcu-delayed too.
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
makes proc_pid_ns() safe from rcu pathwalk (put_pid_ns()
is still synchronous, but that's not a problem - it does
rcu-delay everything that needs to be)
Reviewed-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
that keeps both around until struct inode is freed, making access
to them safe from rcu-pathwalk
Acked-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
NFS ->d_revalidate(), ->permission() and ->get_link() need to access
some parts of nfs_server when called in RCU mode:
server->flags
server->caps
*(server->io_stats)
and, worst of all, call
server->nfs_client->rpc_ops->have_delegation
(the last one - as NFS_PROTO(inode)->have_delegation()). We really
don't want to RCU-delay the entire nfs_free_server() (it would have
to be done with schedule_work() from RCU callback, since it can't
be made to run from interrupt context), but actual freeing of
nfs_server and ->io_stats can be done via call_rcu() just fine.
nfs_client part is handled simply by making nfs_free_client() use
kfree_rcu().
Acked-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
nfs_set_verifier() relies upon dentry being pinned; if that's
the case, grabbing ->d_lock stabilizes ->d_parent and guarantees
that ->d_parent points to a positive dentry. For something
we'd run into in RCU mode that is *not* true - dentry might've
been through dentry_kill() just as we grabbed ->d_lock, with
its parent going through the same just as we get to into
nfs_set_verifier_locked(). It might get to detaching inode
(and zeroing ->d_inode) before nfs_set_verifier_locked() gets
to fetching that; we get an oops as the result.
That can happen in nfs{,4} ->d_revalidate(); the call chain in
question is nfs_set_verifier_locked() <- nfs_set_verifier() <-
nfs_lookup_revalidate_delegated() <- nfs{,4}_do_lookup_revalidate().
We have checked that the parent had been positive, but that's
done before we get to nfs_set_verifier() and it's possible for
memory pressure to pick our dentry as eviction candidate by that
time. If that happens, back-to-back attempts to kill dentry and
its parent are quite normal. Sure, in case of eviction we'll
fail the ->d_seq check in the caller, but we need to survive
until we return there...
Acked-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
In __afs_break_callback() we might check ->cb_nr_mmap and if it's non-zero
do queue_work(&vnode->cb_work). In afs_drop_open_mmap() we decrement
->cb_nr_mmap and do flush_work(&vnode->cb_work) if it reaches zero.
The trouble is, there's nothing to prevent __afs_break_callback() from
seeing ->cb_nr_mmap before the decrement and do queue_work() after both
the decrement and flush_work(). If that happens, we might be in trouble -
vnode might get freed before the queued work runs.
__afs_break_callback() is always done under ->cb_lock, so let's make
sure that ->cb_nr_mmap can change from non-zero to zero while holding
->cb_lock (the spinlock component of it - it's a seqlock and we don't
need to mess with the counter).
Acked-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
->d_hash() and ->d_compare() use those, so we need to delay freeing
them.
Reviewed-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
That stuff can be accessed by ->d_hash()/->d_compare(); as it is, we have
a hard-to-hit UAF if rcu pathwalk manages to get into ->d_hash() on a filesystem
that is in process of getting shut down.
Besides, having nls and upcase table cleanup moved from ->put_super() towards
the place where sbi is freed makes for simpler failure exits.
Acked-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
one of the flags in it is used by ->d_hash()/->d_compare()
Reviewed-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
If lazy call of ->permission() returns a hard error, check that
try_to_unlazy() succeeds before returning it. That both makes
life easier for ->permission() instances and closes the race
in ENOTDIR handling - it is possible that positive d_can_lookup()
seen in link_path_walk() applies to the state *after* unlink() +
mkdir(), while nd->inode matches the state prior to that.
Normally seeing e.g. EACCES from permission check in rcu pathwalk
means that with some timings non-rcu pathwalk would've run into
the same; however, running into a non-executable regular file
in the middle of a pathname would not get to permission check -
it would fail with ENOTDIR instead.
Reviewed-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Avoids fun races in RCU pathwalk... Same goes for freeing LSM shite
hanging off super_block's arse.
Reviewed-by: Christian Brauner <brauner@kernel.org>
Signed-off-by: Al Viro <viro@zeniv.linux.org.uk>
Since commit 8208181fe5 ("clk: imx: composite-8m:
Add imx8m_divider_determine_rate") the lcdif controller has
had the ability to set the disp_pixel_clk rate which propagates
up the tree and sets the video_pll rate automatically.
As such, there is no need to define it in the board file.
Signed-off-by: Adam Ford <aford173@gmail.com>
Signed-off-by: Shawn Guo <shawnguo@kernel.org>
Since commit 8208181fe5 ("clk: imx: composite-8m:
Add imx8m_divider_determine_rate") the lcdif controller has
had the ability to set the disp_pixel_clk rate which propagates
up the tree and sets the video_pll rate automatically.
By setting this value low, it will force the recalculation of
video_pll to the lowest rate needed by lcdif instead of
dividing a larger clock down to the desired clock speed. This
has the advantage of being able to lower the video_pll rate
from 594MHz to 148.5MHz when operating at 1080p. It can go even
lower when operating at lower resolutions and refresh rates.
Signed-off-by: Adam Ford <aford173@gmail.com>
Signed-off-by: Shawn Guo <shawnguo@kernel.org>
The imx8mp-beacon SOM has an integrated PHY connected to
the EQOS ethernet controller which can support up to five
queues. Configure these queues in the same manor as done
on the imx8mp-evk.
Signed-off-by: Adam Ford <aford173@gmail.com>
Signed-off-by: Shawn Guo <shawnguo@kernel.org>
The imx8mp-beacon SOM has wireless chip supporting Wi-Fi and
Bluetooth shared. The Wi-Fi is already enabled via the SDIO
interface, so enable the Bluetooth via UART1.
Signed-off-by: Adam Ford <aford173@gmail.com>
Signed-off-by: Shawn Guo <shawnguo@kernel.org>
check_snapshot() copies the bch_snapshot to a temporary to easily handle
older versions that don't have all the fields of the current version,
but it lacked a min() to correctly handle keys newer and larger than the
current version.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
If a journal write errored, the list of devices it was written to could
be empty - we're not supposed to mark an empty replicas list.
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
bch2_direct_IO_read() checks the request offset and size for sector
alignment and then falls through to a couple calculations to shrink
the size of the request based on the inode size. The problem is that
these checks round up to the fs block size, which runs the risk of
underflowing iter->count if the block size happens to be large
enough. This is triggered by fstest generic/361 with a 4k block
size, which subsequently leads to a crash. To avoid this crash,
check that the shorten length doesn't exceed the overall length of
the iter.
Fixes:
Signed-off-by: Brian Foster <bfoster@redhat.com>
Reviewed-by: Su Yue <glass.su@suse.com>
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
If we're in FILTER_SNAPSHOTS mode and we start scanning a range of the
keyspace where no keys are visible in the current snapshot, we have a
problem - we'll scan for a very long time before scanning terminates.
Awhile back, this was fixed for most cases with peek_upto() (and
assertions that enforce that it's being used).
But the fix missed the fact that the inodes btree is different - every
key offset is in a different snapshot tree, not just the inode field.
Fixes:
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
Recently, we fixed our __GFP_NOFAIL usage in the readahead path, but the
easy one in read_single_folio() (where wa can return an error) was
missed - oops.
Fixes:
Signed-off-by: Kent Overstreet <kent.overstreet@linux.dev>
- Fix a crash when hot adding a PCI device to an LPAR since recent changes.
- Fix nested KVM level-2 guest reboot failure due to empty 'arch_compat'.
Thanks to: Amit Machhiwal, Aneesh Kumar K.V (IBM), Brian King, Gaurav Batra,
Vaibhav Jain.
-----BEGIN PGP SIGNATURE-----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=+tcX
-----END PGP SIGNATURE-----
Merge tag 'powerpc-6.8-4' of git://git.kernel.org/pub/scm/linux/kernel/git/powerpc/linux
Pull powerpc fixes from Michael Ellerman:
- Fix a crash when hot adding a PCI device to an LPAR since
recent changes
- Fix nested KVM level-2 guest reboot failure due to empty
'arch_compat'
Thanks to Amit Machhiwal, Aneesh Kumar K.V (IBM), Brian King, Gaurav
Batra, and Vaibhav Jain.
* tag 'powerpc-6.8-4' of git://git.kernel.org/pub/scm/linux/kernel/git/powerpc/linux:
KVM: PPC: Book3S HV: Fix L2 guest reboot failure due to empty 'arch_compat'
powerpc/pseries/iommu: DLPAR add doesn't completely initialize pci_controller
Including:
- Intel VT-d fixes for nested domain handling:
- Cache invalidation for changes in a parent domain
- Dirty tracking setting for parent and nested domains
- Fix a constant-out-of-range warning
- ARM SMMU fixes:
- Fix CD allocation from atomic context when using SVA with SMMUv3
- Revert the conversion of SMMUv2 to domain_alloc_paging(), as it
breaks the boot for Qualcomm MSM8996 devices
- Restore SVA handle sharing in core code as it turned out there are
still drivers relying on it
-----BEGIN PGP SIGNATURE-----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=aVOB
-----END PGP SIGNATURE-----
Merge tag 'iommu-fixes-v6.8-rc5' of git://git.kernel.org/pub/scm/linux/kernel/git/joro/iommu
Pull iommu fixes from Joerg Roedel:
- Intel VT-d fixes for nested domain handling:
- Cache invalidation for changes in a parent domain
- Dirty tracking setting for parent and nested domains
- Fix a constant-out-of-range warning
- ARM SMMU fixes:
- Fix CD allocation from atomic context when using SVA with SMMUv3
- Revert the conversion of SMMUv2 to domain_alloc_paging(), as it
breaks the boot for Qualcomm MSM8996 devices
- Restore SVA handle sharing in core code as it turned out there are
still drivers relying on it
* tag 'iommu-fixes-v6.8-rc5' of git://git.kernel.org/pub/scm/linux/kernel/git/joro/iommu:
iommu/sva: Restore SVA handle sharing
iommu/arm-smmu-v3: Do not use GFP_KERNEL under as spinlock
iommu/vt-d: Fix constant-out-of-range warning
iommu/vt-d: Set SSADE when attaching to a parent with dirty tracking
iommu/vt-d: Add missing dirty tracking set for parent domain
iommu/vt-d: Wrap the dirty tracking loop to be a helper
iommu/vt-d: Remove domain parameter for intel_pasid_setup_dirty_tracking()
iommu/vt-d: Add missing device iotlb flush for parent domain
iommu/vt-d: Update iotlb in nested domain attach
iommu/vt-d: Add missing iotlb flush for parent domain
iommu/vt-d: Add __iommu_flush_iotlb_psi()
iommu/vt-d: Track nested domains in parent
Revert "iommu/arm-smmu: Convert to domain_alloc_paging()"
- Fix NUMA initialization from ACPI CEDT.CFMWS
- Fix region assembly failures due to async init order
- Fix / simplify export of qos_class information
- Fix cxl_acpi initialization vs single-window-init failures
- Fix handling of repeated 'pci_channel_io_frozen' notifications
- Workaround platforms that violate host-physical-address ==
system-physical address assumptions
- Defer CXL CPER notification handling to v6.9
-----BEGIN PGP SIGNATURE-----
iHUEABYKAB0WIQSbo+XnGs+rwLz9XGXfioYZHlFsZwUCZdpH9gAKCRDfioYZHlFs
ZwZlAQDE+PxTJnjCXDVnDylVF4yeJF2G/wSkH1CFVFVxa0OjhAD/ZFScS/nz/76l
1IYYiiLqmVO5DdmJtfKtq16m7e1cZwc=
=PuPF
-----END PGP SIGNATURE-----
Merge tag 'cxl-fixes-6.8-rc6' of git://git.kernel.org/pub/scm/linux/kernel/git/cxl/cxl
Pull cxl fixes from Dan Williams:
"A collection of significant fixes for the CXL subsystem.
The largest change in this set, that bordered on "new development", is
the fix for the fact that the location of the new qos_class attribute
did not match the Documentation. The fix ends up deleting more code
than it added, and it has a new unit test to backstop basic errors in
this interface going forward. So the "red-diff" and unit test saved
the "rip it out and try again" response.
In contrast, the new notification path for firmware reported CXL
errors (CXL CPER notifications) has a locking context bug that can not
be fixed with a red-diff. Given where the release cycle stands, it is
not comfortable to squeeze in that fix in these waning days. So, that
receives the "back it out and try again later" treatment.
There is a regression fix in the code that establishes memory NUMA
nodes for platform CXL regions. That has an ack from x86 folks. There
are a couple more fixups for Linux to understand (reassemble) CXL
regions instantiated by platform firmware. The policy around platforms
that do not match host-physical-address with system-physical-address
(i.e. systems that have an address translation mechanism between the
address range reported in the ACPI CEDT.CFMWS and endpoint decoders)
has been softened to abort driver load rather than teardown the memory
range (can cause system hangs). Lastly, there is a robustness /
regression fix for cases where the driver would previously continue in
the face of error, and a fixup for PCI error notification handling.
Summary:
- Fix NUMA initialization from ACPI CEDT.CFMWS
- Fix region assembly failures due to async init order
- Fix / simplify export of qos_class information
- Fix cxl_acpi initialization vs single-window-init failures
- Fix handling of repeated 'pci_channel_io_frozen' notifications
- Workaround platforms that violate host-physical-address ==
system-physical address assumptions
- Defer CXL CPER notification handling to v6.9"
* tag 'cxl-fixes-6.8-rc6' of git://git.kernel.org/pub/scm/linux/kernel/git/cxl/cxl:
cxl/acpi: Fix load failures due to single window creation failure
acpi/ghes: Remove CXL CPER notifications
cxl/pci: Fix disabling memory if DVSEC CXL Range does not match a CFMWS window
cxl/test: Add support for qos_class checking
cxl: Fix sysfs export of qos_class for memdev
cxl: Remove unnecessary type cast in cxl_qos_class_verify()
cxl: Change 'struct cxl_memdev_state' *_perf_list to single 'struct cxl_dpa_perf'
cxl/region: Allow out of order assembly of autodiscovered regions
cxl/region: Handle endpoint decoders in cxl_region_find_decoder()
x86/numa: Fix the sort compare func used in numa_fill_memblks()
x86/numa: Fix the address overlap check in numa_fill_memblks()
cxl/pci: Skip to handle RAS errors if CXL.mem device is detached
The SLAB_MEM_SPREAD flag is already a no-op as of 6.8-rc1, remove
its usage so we can delete it from slab. No functional change.
Signed-off-by: Chengming Zhou <zhouchengming@bytedance.com>
Link: https://lore.kernel.org/r/20240224134646.829105-1-chengming.zhou@linux.dev
Signed-off-by: Jens Axboe <axboe@kernel.dk>
The logic in blk_mq_complete_need_ipi() assumes SMP systems where all
CPUs have equal compute capacities and only LLC cache can make
a different on perceived performance. But this assumption falls apart on
HMP systems where LLC is shared, but the CPUs have different capacities.
Staying local then can have a big performance impact if the IO request
was done from a CPU with higher capacity but the interrupt is serviced
on a lower capacity CPU.
Use the new cpus_equal_capacity() function to check if we need to send
an IPI.
Without the patch I see the BLOCK softirq always running on little cores
(where the hardirq is serviced). With it I can see it running on all
cores.
This was noticed after the topology change [1] where now on a big.LITTLE
we truly get that the LLC is shared between all cores where as in the
past it was being misrepresented for historical reasons. The logic
exposed a missing dependency on capacities for such systems where there
can be a big performance difference between the CPUs.
This of course introduced a noticeable change in behavior depending on
how the topology is presented. Leading to regressions in some workloads
as the performance of the BLOCK softirq on littles can be noticeably
worse on some platforms.
Worth noting that we could have checked for capacities being greater
than or equal instead for equality. This will lead to favouring higher
performance always. But opted for equality instead to match the
performance of the requester without making an assumption that can lead
to power trade-offs which these systems tend to be sensitive about. If
the requester would like to run faster, it's better to rely on the
scheduler to give the IO requester via some facility to run on a faster
core; and then if the interrupt triggered on a CPU with different
capacity we'll make sure to match the performance the requester is
supposed to run at.
[1] https://lpc.events/event/16/contributions/1342/attachments/962/1883/LPC-2022-Android-MC-Phantom-Domains.pdf
Signed-off-by: Qais Yousef <qyousef@layalina.io>
Reviewed-by: Bart Van Assche <bvanassche@acm.org>
Link: https://lore.kernel.org/r/20240223155749.2958009-3-qyousef@layalina.io
Signed-off-by: Jens Axboe <axboe@kernel.dk>
