Merge mmu_notifier fixes from Jérôme Glisse:
"The invalidate_page callback suffered from 2 pitfalls. First it used
to happen after page table lock was release and thus a new page might
have been setup for the virtual address before the call to
invalidate_page().
This is in a weird way fixed by commit c7ab0d2fdc ("mm: convert
try_to_unmap_one() to use page_vma_mapped_walk()") which moved the
callback under the page table lock. Which also broke several existing
user of the mmu_notifier API that assumed they could sleep inside this
callback.
The second pitfall was invalidate_page being the only callback not
taking a range of address in respect to invalidation but was giving an
address and a page. Lot of the callback implementer assumed this could
never be THP and thus failed to invalidate the appropriate range for
THP pages.
By killing this callback we unify the mmu_notifier callback API to
always take a virtual address range as input.
There is now two clear API (I am not mentioning the youngess API which
is seldomly used):
- invalidate_range_start()/end() callback (which allow you to sleep)
- invalidate_range() where you can not sleep but happen right after
page table update under page table lock
Note that a lot of existing user feels broken in respect to
range_start/ range_end. Many user only have range_start() callback but
there is nothing preventing them to undo what was invalidated in their
range_start() callback after it returns but before any CPU page table
update take place.
The code pattern use in kvm or umem odp is an example on how to
properly avoid such race. In a nutshell use some kind of sequence
number and active range invalidation counter to block anything that
might undo what the range_start() callback did.
If you do not care about keeping fully in sync with CPU page table (ie
you can live with CPU page table pointing to new different page for a
given virtual address) then you can take a reference on the pages
inside the range_start callback and drop it in range_end or when your
driver is done with those pages.
Last alternative is to use invalidate_range() if you can do
invalidation without sleeping as invalidate_range() callback happens
under the CPU page table spinlock right after the page table is
updated.
The first two patches convert existing mmu_notifier_invalidate_page()
calls to mmu_notifier_invalidate_range() and bracket those call with
call to mmu_notifier_invalidate_range_start()/end().
The next ten patches remove existing invalidate_page() callback as it
can no longer happen.
Finally the last page remove the invalidate_page() callback completely
so it can RIP.
Changes since v1:
- remove more dead code in kvm (no testing impact)
- more accurate end address computation (patch 2) in page_mkclean_one
and try_to_unmap_one
- added tested-by/reviewed-by gotten so far"
* emailed patches from Jérôme Glisse <jglisse@redhat.com>:
mm/mmu_notifier: kill invalidate_page
KVM: update to new mmu_notifier semantic v2
xen/gntdev: update to new mmu_notifier semantic
sgi-gru: update to new mmu_notifier semantic
misc/mic/scif: update to new mmu_notifier semantic
iommu/intel: update to new mmu_notifier semantic
iommu/amd: update to new mmu_notifier semantic
IB/hfi1: update to new mmu_notifier semantic
IB/umem: update to new mmu_notifier semantic
drm/amdgpu: update to new mmu_notifier semantic
powerpc/powernv: update to new mmu_notifier semantic
mm/rmap: update to new mmu_notifier semantic v2
dax: update to new mmu_notifier semantic
jfs had previously avoided the use of MAX_LFS_FILESIZE because it hadn't
accounted for the whole 32-bit index range on 32-bit systems. That has
been fixed by commit 0cc3b0ec23 ("Clarify (and fix) MAX_LFS_FILESIZE
macros"), so we can simplify the code now.
Suggested by Andreas Dilger.
Signed-off-by: Dave Kleikamp <dave.kleikamp@oracle.com>
Reviewed-by: Andreas Dilger <adilger@dilger.ca>
Cc: jfs-discussion@lists.sourceforge.net
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
The invalidate_page callback suffered from two pitfalls. First it used
to happen after the page table lock was release and thus a new page
might have setup before the call to invalidate_page() happened.
This is in a weird way fixed by commit c7ab0d2fdc ("mm: convert
try_to_unmap_one() to use page_vma_mapped_walk()") that moved the
callback under the page table lock but this also broke several existing
users of the mmu_notifier API that assumed they could sleep inside this
callback.
The second pitfall was invalidate_page() being the only callback not
taking a range of address in respect to invalidation but was giving an
address and a page. Lots of the callback implementers assumed this
could never be THP and thus failed to invalidate the appropriate range
for THP.
By killing this callback we unify the mmu_notifier callback API to
always take a virtual address range as input.
Finally this also simplifies the end user life as there is now two clear
choices:
- invalidate_range_start()/end() callback (which allow you to sleep)
- invalidate_range() where you can not sleep but happen right after
page table update under page table lock
Signed-off-by: Jérôme Glisse <jglisse@redhat.com>
Cc: Bernhard Held <berny156@gmx.de>
Cc: 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: axie <axie@amd.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Pull libnvdimm fix from Dan Williams:
"A single patch removing some structure definitions from a uapi header
file. These payloads are never processed directly by the kernel they
are simply passed through an ioctl as opaque blobs to the ACPI _DSM
(Device Specific Method) interface.
Userspace should not be depending on the kernel to define these
payloads. We will instead provide these definitions via the existing
libndctl (https://github.com/pmem/ndctl) project that has NVDIMM
command helpers and other definitions"
* 'libnvdimm-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/nvdimm/nvdimm:
libnvdimm: clean up command definitions
Pull drm fixes from Dave Airlie:
"Two fixes (a vmwgfx and core drm fix) in the queue for 4.13 final,
hopefully that is it"
* tag 'drm-fixes-for-v4.13-rc8' of git://people.freedesktop.org/~airlied/linux:
drm/vmwgfx: Fix F26 Wayland screen update issue
drm/bridge/sii8620: Fix memory corruption
Pull SCSI fixes from James Bottomley:
"Three minor fixes: a NULL deref in qedf, an off by one in sg and a fix
to IPR to prevent an error on initialisation"
* tag 'scsi-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi:
scsi: qedf: Fix a potential NULL pointer dereference
scsi: sg: off by one in sg_ioctl()
scsi: ipr: Set no_report_opcodes for RAID arrays
Pull UML fix from Richard Weinberger:
"This contains a single fix for a regression which was introduced while
the merge window"
* 'for-linus-4.13-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/rw/uml:
um: Fix check for _xstate for older hosts
Pull alpha update from Matt Turner:
"A few fixes and wires up some additional syscalls."
[ Some of this is technically not really rc7 material, but it's alpha,
and it all looks safe anyway. Matt explains: "My alpha has been
offline, hence the very late-in-cycle pull request" and hasn't caused
problems before, so he gets to slide. - Linus ]
* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mattst88/alpha:
alpha: uapi: Add support for __SANE_USERSPACE_TYPES__
alpha: Define ioremap_wc
alpha: Fix section mismatches
alpha: support R_ALPHA_REFLONG relocations for module loading
alpha: Fix typo in ev6-copy_user.S
alpha: Package string routines together
alpha: Update for new syscalls
alpha: Fix build error without CONFIG_VGA_HOSE.
vmwgfx currently cannot support non-blocking commit because when
vmw_*_crtc_page_flip is called, drm_atomic_nonblocking_commit()
schedules the update on a thread. This means vmw_*_crtc_page_flip
cannot rely on the new surface being bound before the subsequent
dirty and flush operations happen.
Cc: <stable@vger.kernel.org> # 4.12.x
Signed-off-by: Sinclair Yeh <syeh@vmware.com>
Reviewed-by: Thomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: Charmaine Lee <charmainel@vmware.com>
This fixes compiler errors in perf such as:
tests/attr.c: In function 'store_event':
tests/attr.c:66:27: error: format '%llu' expects argument of type 'long long unsigned int', but argument 6 has type '__u64 {aka long unsigned int}' [-Werror=format=]
snprintf(path, PATH_MAX, "%s/event-%d-%llu-%d", dir,
^
Signed-off-by: Ben Hutchings <ben@decadent.org.uk>
Tested-by: Michael Cree <mcree@orcon.net.nz>
Cc: stable@vger.kernel.org
Signed-off-by: Matt Turner <mattst88@gmail.com>
Commit 3cc2dac5be ("drivers/video/fbdev/atyfb: Replace MTRR UC hole
with strong UC") introduces calls to ioremap_wc and ioremap_uc. This
causes build failures with alpha:allmodconfig. Map the missing functions
to ioremap_nocache.
Fixes: 3cc2dac5be ("drivers/video/fbdev/atyfb:
Replace MTRR UC hole with strong UC")
Cc: Paul Gortmaker <paul.gortmaker@windriver.com>
Cc: Luis R. Rodriguez <mcgrof@suse.com>
Signed-off-by: Guenter Roeck <linux@roeck-us.net>
Signed-off-by: Matt Turner <mattst88@gmail.com>
Since commit 71810db27c (modversions: treat symbol CRCs
as 32 bit quantities) R_ALPHA_REFLONG relocations can be required
to load modules. This implements it.
Tested-by: Bob Tracy <rct@gherkin.frus.com>
Reviewed-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Michael Cree <mcree@orcon.net.nz>
Signed-off-by: Matt Turner <mattst88@gmail.com>
Patch 8525023121 introduced a typo.
That said, the identity AND insns added by that patch are more
clearly written as MOV. At the same time, re-schedule the ev6
version so that the first dispatch can execute in parallel.
Signed-off-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Matt Turner <mattst88@gmail.com>
There are direct branches between {str*cpy,str*cat} and stx*cpy.
Ensure the branches are within range by merging these objects.
Signed-off-by: Richard Henderson <rth@twiddle.net>
Signed-off-by: Matt Turner <mattst88@gmail.com>
Pull cgroup fix from Tejun Heo:
"A late but obvious fix for cgroup.
I broke the 'cpuset.memory_pressure' file a long time ago (v4.4) by
accidentally deleting its file index, which made it a duplicate of the
'cpuset.memory_migrate' file. Spotted and fixed by Waiman"
* 'for-4.13-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/tj/cgroup:
cpuset: Fix incorrect memory_pressure control file mapping
Pull libata fixes from Tejun Heo:
"Late fixes for libata. There's a minor platform driver fix but the
important one is READ LOG PAGE.
This is a new ATA command which is used to test some optional features
but it broke probing of some devices - they locked up instead of
failing the unknown command.
Christoph tried blacklisting, but, after finding out there are
multiple devices which fail this way, backed off to testing feature
bit in IDENTIFY data first, which is a bit lossy (we can miss features
on some devices) but should be a lot safer"
* 'for-4.13-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/tj/libata:
Revert "libata: quirk read log on no-name M.2 SSD"
libata: check for trusted computing in IDENTIFY DEVICE data
libata: quirk read log on no-name M.2 SSD
sata: ahci-da850: Fix some error handling paths in 'ahci_da850_probe()'
This reverts commit aac2fea94f.
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>
This reverts commit 35f0b6a779.
We now conditionalize issuing of READ LOG PAGE on the TRUSTED
COMPUTING SUPPORTED bit in the identity data and this shouldn't be
necessary.
Signed-off-by: Tejun Heo <tj@kernel.org>
ATA-8 and later mirrors the TRUSTED COMPUTING SUPPORTED bit in word 48 of
the IDENTIFY DEVICE data. Check this before issuing a READ LOG PAGE
command to avoid issues with buggy devices. The only downside is that
we can't support Security Send / Receive for a device with an older
revision due to the conflicting use of this field in earlier
specifications.
tj: The reason we need this is because some devices which don't
support READ LOG PAGE lock up after getting issued that command.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Tested-by: David Ahern <dsahern@gmail.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
Commit 3510ca20ec ("Minor page waitqueue cleanups") made the page
queue code always add new waiters to the back of the queue, which helps
upcoming patches to batch the wakeups for some horrid loads where the
wait queues grow to thousands of entries.
However, I forgot about the nasrt add_page_wait_queue() special case
code that is only used by the cachefiles code. That one still continued
to add the new wait queue entries at the beginning of the list.
Fix it, because any sane batched wakeup will require that we don't
suddenly start getting new entries at the beginning of the list that we
already handled in a previous batch.
[ The current code always does the whole list while holding the lock, so
wait queue ordering doesn't matter for correctness, but even then it's
better to add later entries at the end from a fairness standpoint ]
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
When !NUMA, cpumask_of_node(@node) equals cpu_online_mask regardless of
@node. The assumption seems that if !NUMA, there shouldn't be more than
one node and thus reporting cpu_online_mask regardless of @node is
correct. However, that assumption was broken years ago to support
DISCONTIGMEM and whether a system has multiple nodes or not is
separately controlled by NEED_MULTIPLE_NODES.
This means that, on a system with !NUMA && NEED_MULTIPLE_NODES,
cpumask_of_node() will report cpu_online_mask for all possible nodes,
indicating that the CPUs are associated with multiple nodes which is an
impossible configuration.
This bug has been around forever but doesn't look like it has caused any
noticeable symptoms. However, it triggers a WARN recently added to
workqueue to verify NUMA affinity configuration.
Fix it by reporting empty cpumask on non-zero nodes if !NUMA.
Signed-off-by: Tejun Heo <tj@kernel.org>
Reported-and-tested-by: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: stable@vger.kernel.org
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Recent commit a8ec3ee861 "arc: Mask individual IRQ lines during core
INTC init" breaks interrupt handling on ARCv2 SMP systems.
That commit masked all interrupts at onset, as some controllers on some
boards (customer as well as internal), would assert interrutps early
before any handlers were installed. For SMP systems, the masking was
done at each cpu's core-intc. Later, when the IRQ was actually
requested, it was unmasked, but only on the requesting cpu.
For "common" interrupts, which were wired up from the 2nd level IDU
intc, this was as issue as they needed to be enabled on ALL the cpus
(given that IDU IRQs are by default served Round Robin across cpus)
So fix that by NOT masking "common" interrupts at core-intc, but instead
at the 2nd level IDU intc (latter already being done in idu_of_init())
Fixes: a8ec3ee861 ("arc: Mask individual IRQ lines during core INTC init")
Signed-off-by: Alexey Brodkin <abrodkin@synopsys.com>
[vgupta: reworked changelog, removed the extraneous idu_irq_mask_raw()]
Signed-off-by: Vineet Gupta <vgupta@synopsys.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Commit 464d62421c ("select: switch compat_{get,put}_fd_set() to
compat_{get,put}_bitmap()") changed the calculation on how many bytes
need to be zeroed when userspace handed over a NULL pointer for a fdset
array in the select syscall.
The calculation was changed in compat_get_fd_set() wrongly from
memset(fdset, 0, ((nr + 1) & ~1)*sizeof(compat_ulong_t));
to
memset(fdset, 0, ALIGN(nr, BITS_PER_LONG));
The ALIGN(nr, BITS_PER_LONG) calculates the number of _bits_ which need
to be zeroed in the target fdset array (rounded up to the next full bits
for an unsigned long).
But the memset() call expects the number of _bytes_ to be zeroed.
This leads to clearing more memory than wanted (on the stack area or
even at kmalloc()ed memory areas) and to random kernel crashes as we
have seen them on the parisc platform.
The correct change should have been
memset(fdset, 0, (ALIGN(nr, BITS_PER_LONG) / BITS_PER_LONG) * BYTES_PER_LONG);
which is the same as can be archieved with a call to
zero_fd_set(nr, fdset).
Fixes: 464d62421c ("select: switch compat_{get,put}_fd_set() to compat_{get,put}_bitmap()"
Acked-by:: Al Viro <viro@zeniv.linux.org.uk>
Signed-off-by: Helge Deller <deller@gmx.de>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Pull c6x tweaks from Mark Salter.
* tag 'for-linus' of git://linux-c6x.org/git/projects/linux-c6x-upstreaming:
c6x: Convert to using %pOF instead of full_name
c6x: defconfig: Cleanup from old Kconfig options
Ido reported that reading the log page on his systems fails,
so quirk it as it won't support ZBC or security protocols.
Signed-off-by: Christoph Hellwig <hch@lst.de>
Reported-by: Ido Schimmel <idosch@mellanox.com>
Tested-by: Ido Schimmel <idosch@mellanox.com>
Signed-off-by: Tejun Heo <tj@kernel.org>
Remove the command payloads that do not have an associated libnvdimm
ioctl. I.e. remove the payloads that would only ever be carried in the
ND_CMD_CALL envelope. This prevents userspace from growing unnecessary
dependencies on this kernel header when userspace already has everything
it needs to craft and send these commands.
Cc: Jerry Hoemann <jerry.hoemann@hpe.com>
Reported-by: Yasunori Goto <y-goto@jp.fujitsu.com>
Signed-off-by: Dan Williams <dan.j.williams@intel.com>
Pull IOMMU fix from Joerg Roedel:
"Another fix, this time in common IOMMU sysfs code.
In the conversion from the old iommu sysfs-code to the
iommu_device_register interface, I missed to update the release path
for the struct device associated with an IOMMU. It freed the 'struct
device', which was a pointer before, but is now embedded in another
struct.
Freeing from the middle of allocated memory had all kinds of nasty
side effects when an IOMMU was unplugged. Unfortunatly nobody
unplugged and IOMMU until now, so this was not discovered earlier. The
fix is to make the 'struct device' a pointer again"
* tag 'iommu-fixes-v4.13-rc6' of git://git.kernel.org/pub/scm/linux/kernel/git/joro/iommu:
iommu: Fix wrong freeing of iommu_device->dev
Pull char/misc fix from Greg KH:
"Here is a single misc driver fix for 4.13-rc7. It resolves a reported
problem in the Android binder driver due to previous patches in
4.13-rc.
It's been in linux-next with no reported issues"
* tag 'char-misc-4.13-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/char-misc:
ANDROID: binder: fix proc->tsk check.
Pull staging/iio fixes from Greg KH:
"Here are few small staging driver fixes, and some more IIO driver
fixes for 4.13-rc7. Nothing major, just resolutions for some reported
problems.
All of these have been in linux-next with no reported problems"
* tag 'staging-4.13-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/gregkh/staging:
iio: magnetometer: st_magn: remove ihl property for LSM303AGR
iio: magnetometer: st_magn: fix status register address for LSM303AGR
iio: hid-sensor-trigger: Fix the race with user space powering up sensors
iio: trigger: stm32-timer: fix get trigger mode
iio: imu: adis16480: Fix acceleration scale factor for adis16480
PATCH] iio: Fix some documentation warnings
staging: rtl8188eu: add RNX-N150NUB support
Revert "staging: fsl-mc: be consistent when checking strcmp() return"
iio: adc: stm32: fix common clock rate
iio: adc: ina219: Avoid underflow for sleeping time
iio: trigger: stm32-timer: add enable attribute
iio: trigger: stm32-timer: fix get/set down count direction
iio: trigger: stm32-timer: fix write_raw return value
iio: trigger: stm32-timer: fix quadrature mode get routine
iio: bmp280: properly initialize device for humidity reading
Pull NTB fixes from Jon Mason:
"NTB bug fixes to address an incorrect ntb_mw_count reference in the
NTB transport, improperly bringing down the link if SPADs are
corrupted, and an out-of-order issue regarding link negotiation and
data passing"
* tag 'ntb-4.13-bugfixes' of git://github.com/jonmason/ntb:
ntb: ntb_test: ensure the link is up before trying to configure the mws
ntb: transport shouldn't disable link due to bogus values in SPADs
ntb: use correct mw_count function in ntb_tool and ntb_transport
The "lock_page_killable()" function waits for exclusive access to the
page lock bit using the WQ_FLAG_EXCLUSIVE bit in the waitqueue entry
set.
That means that if it gets woken up, other waiters may have been
skipped.
That, in turn, means that if it sees the page being unlocked, it *must*
take that lock and return success, even if a lethal signal is also
pending.
So instead of checking for lethal signals first, we need to check for
them after we've checked the actual bit that we were waiting for. Even
if that might then delay the killing of the process.
This matches the order of the old "wait_on_bit_lock()" infrastructure
that the page locking used to use (and is still used in a few other
areas).
Note that if we still return an error after having unsuccessfully tried
to acquire the page lock, that is ok: that means that some other thread
was able to get ahead of us and lock the page, and when that other
thread then unlocks the page, the wakeup event will be repeated. So any
other pending waiters will now get properly woken up.
Fixes: 6290602709 ("mm: add PageWaiters indicating tasks are waiting for a page bit")
Cc: Nick Piggin <npiggin@gmail.com>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: Jan Kara <jack@suse.cz>
Cc: Davidlohr Bueso <dave@stgolabs.net>
Cc: Andi Kleen <ak@linux.intel.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
Tim Chen and Kan Liang have been battling a customer load that shows
extremely long page wakeup lists. The cause seems to be constant NUMA
migration of a hot page that is shared across a lot of threads, but the
actual root cause for the exact behavior has not been found.
Tim has a patch that batches the wait list traversal at wakeup time, so
that we at least don't get long uninterruptible cases where we traverse
and wake up thousands of processes and get nasty latency spikes. That
is likely 4.14 material, but we're still discussing the page waitqueue
specific parts of it.
In the meantime, I've tried to look at making the page wait queues less
expensive, and failing miserably. If you have thousands of threads
waiting for the same page, it will be painful. We'll need to try to
figure out the NUMA balancing issue some day, in addition to avoiding
the excessive spinlock hold times.
That said, having tried to rewrite the page wait queues, I can at least
fix up some of the braindamage in the current situation. In particular:
(a) we don't want to continue walking the page wait list if the bit
we're waiting for already got set again (which seems to be one of
the patterns of the bad load). That makes no progress and just
causes pointless cache pollution chasing the pointers.
(b) we don't want to put the non-locking waiters always on the front of
the queue, and the locking waiters always on the back. Not only is
that unfair, it means that we wake up thousands of reading threads
that will just end up being blocked by the writer later anyway.
Also add a comment about the layout of 'struct wait_page_key' - there is
an external user of it in the cachefiles code that means that it has to
match the layout of 'struct wait_bit_key' in the two first members. It
so happens to match, because 'struct page *' and 'unsigned long *' end
up having the same values simply because the page flags are the first
member in struct page.
Cc: Tim Chen <tim.c.chen@linux.intel.com>
Cc: Kan Liang <kan.liang@intel.com>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: Christopher Lameter <cl@linux.com>
Cc: Andi Kleen <ak@linux.intel.com>
Cc: Davidlohr Bueso <dave@stgolabs.net>
Cc: Peter Zijlstra <peterz@infradead.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
