Pull parisc updates from Helge Deller:

 - add Kernel address space layout randomization support

 - re-enable interrupts earlier now that we have a working IRQ stack

 - optimize the timer interrupt function to better cope with missed
   timer irqs

 - fix error return code in parisc perf code (by Dan Carpenter)

 - fix PAT debug code

* 'parisc-4.10-1' of git://git.kernel.org/pub/scm/linux/kernel/git/deller/parisc-linux:
  parisc: Optimize timer interrupt function
  parisc: perf: return -EFAULT on error
  parisc: Enhance CPU detection code on PAT machines
  parisc: Re-enable interrupts early
  parisc: Enable KASLR

Pull more NFS client updates from Trond Myklebust:
 "Highlights include:

   - further attribute cache improvements to make revalidation more fine
     grained

   - NFSv4 locking improvements

  Bugfixes:

   - nfs4_fl_prepare_ds must be careful about reporting success in files
     layout

   - pNFS/flexfiles: Instead of marking a device inactive, remove it
     from the cache"

* tag 'nfs-for-4.10-2' of git://git.linux-nfs.org/projects/trondmy/linux-nfs:
  NFSv4: Retry the DELEGRETURN if the embedded GETATTR is rejected with EACCES
  NFS: Retry the CLOSE if the embedded GETATTR is rejected with EACCES
  NFSv4: Place the GETATTR operation before the CLOSE
  NFSv4: Also ask for attributes when downgrading to a READ-only state
  NFS: Don't abuse NFS_INO_REVAL_FORCED in nfs_post_op_update_inode_locked()
  pNFS: Return RW layouts on OPEN_DOWNGRADE
  NFSv4: Add encode/decode of the layoutreturn op in OPEN_DOWNGRADE
  NFS: Don't disconnect open-owner on NFS4ERR_BAD_SEQID
  NFSv4: ensure __nfs4_find_lock_state returns consistent result.
  NFSv4.1: nfs4_fl_prepare_ds must be careful about reporting success.
  pNFS/flexfiles: delete deviceid, don't mark inactive
  NFS: Clean up nfs_attribute_timeout()
  NFS: Remove unused function nfs_revalidate_inode_rcu()
  NFS: Fix and clean up the access cache validity checking
  NFS: Only look at the change attribute cache state in nfs_weak_revalidate()
  NFS: Clean up cache validity checking
  NFS: Don't revalidate the file on close if we hold a delegation
  NFSv4: Don't discard the attributes returned by asynchronous DELEGRETURN
  NFSv4: Update the attribute cache info in update_changeattr

Merge branch 'scsi-target-for-v4.10' of git://git.kernel.org/pub/scm/linux/kernel/git/bvanassche/linux

Pull scsi target cleanups from Bart Van Assche:
 "The changes here are:

   - a few small bug fixes for the iSCSI and user space target drivers.

   - minimize the target build time by about 30% by rearranging #include
     directives

   - fix the second argument passed to percpu_ida_alloc()

   - reduce the number of false positive warnings reported by sparse

  These patches pass Wu Fengguang's build bot tests and also the
  linux-next tests"

* 'scsi-target-for-v4.10' of git://git.kernel.org/pub/scm/linux/kernel/git/bvanassche/linux:
  iscsi-target: Return error if unable to add network portal
  target: Fix spelling mistake and unwrap multi-line text
  target/iscsi: Fix double free in lio_target_tiqn_addtpg()
  target/user: Fix use-after-free of tcmu_cmds if they are expired
  target: Minimize #include directives
  target/user: Add an #include directive
  cxgbit: Add an #include directive
  ibmvscsi_tgt: Add two #include directives
  sbp-target: Add an #include directive
  qla2xxx: Add an #include directive
  configfs: Minimize #include directives
  usb: gadget: Fix second argument of percpu_ida_alloc()
  sbp-target: Fix second argument of percpu_ida_alloc()
  target/user: Fix a data type in tcmu_queue_cmd()
  target: Use NULL instead of 0 to represent a pointer

This reverts commit 16200948.

The commit was intended to cover the race condition, but it introduced
yet another regression for devices with the implicit feedback, leading
to a kernel panic due to NULL-dereference in an irq context.

As the race condition that was addressed by the commit is very rare
and the regression is much worse, let's revert the commit for rc1, and
fix the issue properly in a later patch.

Fixes: 16200948 ("ALSA: usb-audio: Fix race at stopping the stream")
Reported-by: Ioan-Adrian Ratiu <adi@adirat.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

Pull networking fixes and cleanups from David Miller:

 1) Use rb_entry() instead of hardcoded container_of(), from Geliang
    Tang.

 2) Use correct memory barriers in stammac driver, from Pavel Machek.

 3) Fix assoc bind address handling in SCTP, from Xin Long.

 4) Make the length check for UFO handling consistent between
    __ip_append_data() and ip_finish_output(), from Zheng Li.

 5) HSI driver compatible strings were busted fro hix5hd2, from Dongpo
    Li.

 6) Handle devm_ioremap() errors properly in cavium driver, from Arvind
    Yadav.

* git://git.kernel.org/pub/scm/linux/kernel/git/davem/net: (22 commits)
  RDS: use rb_entry()
  net_sched: sch_netem: use rb_entry()
  net_sched: sch_fq: use rb_entry()
  net/mlx5: use rb_entry()
  ethernet: sfc: Add Kconfig entry for vendor Solarflare
  sctp: not copying duplicate addrs to the assoc's bind address list
  sctp: reduce indent level in sctp_copy_local_addr_list
  ARM: dts: hix5hd2: don't change the existing compatible string
  net: hix5hd2_gmac: fix compatible strings name
  openvswitch: Add a missing break statement.
  net: netcp: ethss: fix 10gbe host port tx pri map configuration
  net: netcp: ethss: fix errors in ethtool ops
  fsl/fman: enable compilation on ARM64
  fsl/fman: A007273 only applies to PPC SoCs
  powerpc: fsl/fman: remove fsl,fman from of_device_ids[]
  fsl/fman: fix 1G support for QSGMII interfaces
  dt: bindings: net: use boolean dt properties for eee broken modes
  net: phy: use boolean dt properties for eee broken modes
  net: phy: fix sign type error in genphy_config_eee_advert
  ipv4: Should use consistent conditional judgement for ip fragment in __ip_append_data and ip_finish_output
  ...

Merge final set of updates from Andrew Morton:

 - a series to make IMA play better across kexec

 - a handful of random fixes

* emailed patches from Andrew Morton <akpm@linux-foundation.org>:
  printk: fix typo in CONSOLE_LOGLEVEL_DEFAULT help text
  ratelimit: fix WARN_ON_RATELIMIT return value
  kcov: make kcov work properly with KASLR enabled
  arm64: setup: introduce kaslr_offset()
  mm: fadvise: avoid expensive remote LRU cache draining after FADV_DONTNEED
  ima: platform-independent hash value
  ima: define a canonical binary_runtime_measurements list format
  ima: support restoring multiple template formats
  ima: store the builtin/custom template definitions in a list
  ima: on soft reboot, save the measurement list
  powerpc: ima: send the kexec buffer to the next kernel
  ima: maintain memory size needed for serializing the measurement list
  ima: permit duplicate measurement list entries
  ima: on soft reboot, restore the measurement list
  powerpc: ima: get the kexec buffer passed by the previous kernel

Pull mailbox updates from Jassi Brar:

 - new features (poll and SRAM usage) added to the mailbox-test driver

 - major update of Broadcom's PDC controller driver

 - minor fix for auto-loading test and STI driver modules

* 'mailbox-for-next' of git://git.linaro.org/landing-teams/working/fujitsu/integration:
  mailbox: mailbox-test: allow reserved areas in SRAM
  mailbox: mailbox-test: add support for fasync/poll
  mailbox: bcm-pdc: Remove unnecessary void* casts
  mailbox: bcm-pdc: Simplify interrupt handler logic
  mailbox: bcm-pdc: Performance improvements
  mailbox: bcm-pdc: Don't use iowrite32 to write DMA descriptors
  mailbox: bcm-pdc: Convert from threaded IRQ to tasklet
  mailbox: bcm-pdc: Try to improve branch prediction
  mailbox: bcm-pdc: streamline rx code
  mailbox: bcm-pdc: Convert from interrupts to poll for tx done
  mailbox: bcm-pdc: PDC driver leaves debugfs files after removal
  mailbox: bcm-pdc: Changes so mbox client can be removed / re-inserted
  mailbox: bcm-pdc: Use octal permissions rather than symbolic
  mailbox: sti: Fix module autoload for OF registration
  mailbox: mailbox-test: Fix module autoload

Pull i2c fixes from Wolfram Sang.

* 'i2c/for-current' of git://git.kernel.org/pub/scm/linux/kernel/git/wsa/linux:
  i2c: mux: mlxcpld: fix i2c mux selection caching
  i2c: designware: fix wrong Tx/Rx FIFO for ACPI
  i2c: xgene: Fix missing code of DTB support
  i2c: mux: pca954x: fix i2c mux selection caching
  i2c: octeon: thunderx: Limit register access retries

Pull documentation fix from Jonathan Corbet:
 "A single fix for the build system.

  It would appear that the docutils developers, in their wisdom, broke
  the API in the 0.13 release. This fix detects the breakage and allows
  the docs to be built with both the old and new versions"

* tag 'doc-4.10-3' of git://git.lwn.net/linux:
  docs: sphinx-extensions: make rstFlatTable work with docutils 0.13

Pull arch/microblaze updates from Michal Simek:

 - wire-up new syscalls

 - add new codes and fpga families

 - fix a return value

* tag 'microblaze-4.10-rc1' of git://git.monstr.eu/linux-2.6-microblaze:
  microblaze: Add new fpga families
  microblaze: Add missing release version code v9.6 and v10
  microblaze: Add missing syscalls
  microblaze: Fix return value from xilinx_timer_init

Pull Xtensa updates from Max Filippov:

 - enable HAVE_DMA_CONTIGUOUS, configure shared DMA pool reservation in
   kc705 DTS

 - update xtensa DMA-related Documentation/features entries

 - clean up arch/xtensa/kernel/setup.c: move S32C1I self-test out of it,
   remove unused declarations, fix screen_info definition

* tag 'xtensa-20161219' of git://github.com/jcmvbkbc/linux-xtensa:
  xtensa: update DMA-related Documentation/features entries
  xtensa: configure shared DMA pool reservation in kc705 DTS
  xtensa: enable HAVE_DMA_CONTIGUOUS
  xtensa: move S32C1I self-test to a separate file
  xtensa: fix screen_info, clean up unused declarations in setup.c

Restructure the timer interrupt function to better cope with missed timer irqs.
Optimize the calculation when the next interrupt should happen and skip irqs if
they would happen too shortly after exit of the irq function.

The update_process_times() call is done anyway at every timer irq, so we can
safely drop the prof_counter and prof_multiplier variables from the per_cpu
structure.
Signed-off-by: Helge Deller <deller@gmx.de>

To make the code clearer, use rb_entry() instead of container_of() to
deal with rbtree.
Signed-off-by: Geliang Tang <geliangtang@gmail.com>
Reviewed-by: Leon Romanovsky <leonro@mellanox.com>
Acked-by: Santosh Shilimkar <santosh.shilimkar@oracle.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

To make the code clearer, use rb_entry() instead of container_of() to
deal with rbtree.
Signed-off-by: Geliang Tang <geliangtang@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

To make the code clearer, use rb_entry() instead of container_of() to
deal with rbtree.
Signed-off-by: Geliang Tang <geliangtang@gmail.com>
Acked-by: Eric Dumazet <edumazet@google.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

To make the code clearer, use rb_entry() instead of container_of() to
deal with rbtree.
Signed-off-by: Geliang Tang <geliangtang@gmail.com>
Acked-by: Leon Romanovsky <leonro@mellanox.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Since commit

  5a6681e2 ("sfc: separate out SFC4000 ("Falcon") support into new sfc-falcon driver")

there are two drivers for Solarflare devices, but both still show up
directly beneath "Ethernet driver support" in the Kconfig. Follow the
pattern of other vendors and group them beneath an own vendor Kconfig
entry for Solarflare.

Cc: Edward Cree <ecree@solarflare.com>
Signed-off-by: Tobias Klauser <tklauser@distanz.ch>
Acked-by: Edward Cree <ecree@solarflare.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Xin Long says:

====================
sctp: fix the issue that may copy duplicate addrs into assoc's bind address list

Patch 1/2 is to fix some indent level.

Given that we have kernels out there with this issue, patch 2/2 also
fix sctp_raw_to_bind_addrs.

v1 -> v2:
  Explain why we didn't filter the duplicate addresses when global
  address list gets updated in patch 2/2 changelog.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

sctp.local_addr_list is a global address list that is supposed to include
all the local addresses. sctp updates this list according to NETDEV_UP/
NETDEV_DOWN notifications.

However, if multiple NICs have the same address, the global list would
have duplicate addresses. Even if for one NIC, promote secondaries in
__inet_del_ifa can also lead to accumulating duplicate addresses.

When sctp binds address 'ANY' and creates a connection, it copies all
the addresses from global list into asoc's bind addr list, which makes
sctp pack the duplicate addresses into INIT/INIT_ACK packets.

This patch is to filter the duplicate addresses when copying the addrs
from global list in sctp_copy_local_addr_list and unpacking addr_param
from cookie in sctp_raw_to_bind_addrs to asoc's bind addr list.

Note that we can't filter the duplicate addrs when global address list
gets updated, As NETDEV_DOWN event may remove an addr that still exists
in another NIC.
Signed-off-by: Xin Long <lucien.xin@gmail.com>
Acked-by: Marcelo Ricardo Leitner <marcelo.leitner@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

This patch is to reduce indent level by using continue when the addr
is not allowed, and also drop end_copy by using break.
Signed-off-by: Xin Long <lucien.xin@gmail.com>
Acked-by: Marcelo Ricardo Leitner <marcelo.leitner@gmail.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Dongpo Li says:

====================
net: hix5hd2_gmac: keep the compatible string not changed

This patch series fix the patch:
d0fb6ba7 ("net: hix5hd2_gmac: add generic compatible string")

The SoC hix5hd2 compatible string has the suffix "-gmac" and
we should not change its compatible string.
So we should name all the compatible string with the suffix "-gmac".
Creating a new name suffix "-gemac" is unnecessary.
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

The SoC hix5hd2 compatible string has the suffix "-gmac" and
we should not change it.
We should only add the generic compatible string "hisi-gmac-v1".

Fixes: 0855950b ("ARM: dts: hix5hd2: add gmac generic compatible and clock names")
Signed-off-by: Dongpo Li <lidongpo@hisilicon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

The SoC hix5hd2 compatible string has the suffix "-gmac" and
we should not change its compatible string.
So we should name all the compatible string with the suffix "-gmac".
Creating a new name suffix "-gemac" is unnecessary.

We also add another SoC compatible string in dt binding documentation
and describe which generic version the SoC belongs to.

Fixes: d0fb6ba7 ("net: hix5hd2_gmac: add generic compatible string")
Signed-off-by: Dongpo Li <lidongpo@hisilicon.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Add a break statement to prevent fall-through from
OVS_KEY_ATTR_ETHERNET to OVS_KEY_ATTR_TUNNEL.  Without the break
actions setting ethernet addresses fail to validate with log messages
complaining about invalid tunnel attributes.

Fixes: 0a6410fb ("openvswitch: netlink: support L3 packets")
Signed-off-by: Jarno Rajahalme <jarno@ovn.org>
Acked-by: Pravin B Shelar <pshelar@ovn.org>
Acked-by: Jiri Benc <jbenc@redhat.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

This patch adds the missing 10gbe host port tx priority map
configurations.
Signed-off-by: WingMan Kwok <w-kwok2@ti.com>
Signed-off-by: Murali Karicheri <m-karicheri2@ti.com>
Signed-off-by: Sekhar Nori <nsekhar@ti.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

In ethtool ops, it needs to retrieve the corresponding
ethss module (gbe or xgbe) from the net_device structure.
Prior to this patch, the retrieving procedure only
checks for the gbe module.  This patch fixes the issue
by checking the xgbe module if the net_device structure
does not correspond to the gbe module.
Signed-off-by: WingMan Kwok <w-kwok2@ti.com>
Signed-off-by: Murali Karicheri <m-karicheri2@ti.com>
Signed-off-by: Sekhar Nori <nsekhar@ti.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Madalin Bucur says:

====================
fsl/fman: fixes for ARM

The patch set fixes advertised speeds for QSGMII interfaces, disables
A007273 erratum workaround on non-PowerPC platforms where it does not
apply, enables compilation on ARM64 and addresses a probing issue on
non PPC platforms.

Changes from v3: removed redundant comment, added ack by Scott
Changes from v2: merged fsl/fman changes to avoid a point of failure
Changes from v1: unifying probing on all supported platforms
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

Signed-off-by: Madalin Bucur <madalin.bucur@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Signed-off-by: Madalin Bucur <madalin.bucur@nxp.com>
Reviewed-by: Camelia Groza <camelia.groza@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

The fsl/fman drivers will use of_platform_populate() on all
supported platforms. Call of_platform_populate() to probe the
FMan sub-nodes.
Signed-off-by: Igal Liberman <igal.liberman@freescale.com>
Signed-off-by: Madalin Bucur <madalin.bucur@nxp.com>
Acked-by: Scott Wood <oss@buserror.net>
Signed-off-by: David S. Miller <davem@davemloft.net>

QSGMII ports were not advertising 1G speed.
Signed-off-by: Madalin Bucur <madalin.bucur@nxp.com>
Reviewed-by: Camelia Groza <camelia.groza@nxp.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

Jerome Brunet says:

====================
phy: Fix integration of eee-broken-modes

The purpose of this series is to fix the integration of the ethernet phy
property "eee-broken-modes" [0]

The v3 of this series has been merged, missing a fix (error reported by
kbuild robot) available in the v4 [1]

More importantly, Florian opposed adding a DT property mapping a device
register this directly [2]. The concern was that the property could be
abused to implement platform configuration policy. After discussing it,
I think we agreed that such information about the HW (defect) should appear
in the platform DT. However, the preferred way is to add a boolean property
for each EEE broken mode.

[0]: http://lkml.kernel.org/r/1480326409-25419-1-git-send-email-jbrunet@baylibre.com
[1]: http://lkml.kernel.org/r/1480348229-25672-1-git-send-email-jbrunet@baylibre.com
[2]: http://lkml.kernel.org/r/e14a3b0c-dc34-be14-48b3-518a0ad0c080@gmail.com
====================
Signed-off-by: David S. Miller <davem@davemloft.net>

The patches regarding eee-broken-modes was merged before all people
involved could find an agreement on the best way to move forward.

While we agreed on having a DT property to mark particular modes as broken,
the value used for eee-broken-modes mapped the phy register in very direct
way. Because of this, the concern is that it could be used to implement
configuration policies instead of describing a broken HW.

In the end, having a boolean property for each mode seems to be preferred
over one bit field value mapping the register (too) directly.

Cc: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: Jerome Brunet <jbrunet@baylibre.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

The patches regarding eee-broken-modes was merged before all people
involved could find an agreement on the best way to move forward.

While we agreed on having a DT property to mark particular modes as broken,
the value used for eee-broken-modes mapped the phy register in very direct
way. Because of this, the concern is that it could be used to implement
configuration policies instead of describing a broken HW.

In the end, having a boolean property for each mode seems to be preferred
over one bit field value mapping the register (too) directly.

Cc: Florian Fainelli <f.fainelli@gmail.com>
Signed-off-by: Jerome Brunet <jbrunet@baylibre.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

In genphy_config_eee_advert, the return value of phy_read_mmd_indirect is
checked to know if the register could be accessed but the result is
assigned to a 'u32'.
Changing to 'int' to correctly get errors from phy_read_mmd_indirect.

Fixes: d853d145 ("net: phy: add an option to disable EEE advertisement")
Reported-by: Julia Lawall <julia.lawall@lip6.fr>
Signed-off-by: Jerome Brunet <jbrunet@baylibre.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

s/prink/printk/

Link: http://lkml.kernel.org/r/20161215170111.19075-1-bp@alien8.deSigned-off-by: Borislav Petkov <bp@suse.de>
Cc: Olof Johansson <olof@lixom.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

The macro is to be used similarly as WARN_ON as:

  if (WARN_ON_RATELIMIT(condition, state))
	do_something();

One would expect only 'condition' to affect the 'if', but
WARN_ON_RATELIMIT does internally only:

  WARN_ON((condition) && __ratelimit(state))

So the 'if' is affected by the ratelimiting state too.  Fix this by
returning 'condition' in any case.

Note that nobody uses WARN_ON_RATELIMIT yet, so there is nothing to
worry about.  But I was about to use it and was a bit surprised.

Link: http://lkml.kernel.org/r/20161215093224.23126-1-jslaby@suse.czSigned-off-by: Jiri Slaby <jslaby@suse.cz>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

Subtract KASLR offset from the kernel addresses reported by kcov.
Tested on x86_64 and AArch64 (Hikey LeMaker).

Link: http://lkml.kernel.org/r/1481417456-28826-3-git-send-email-alex.popov@linux.comSigned-off-by: Alexander Popov <alex.popov@linux.com>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Will Deacon <will.deacon@arm.com>
Cc: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Rob Herring <robh@kernel.org>
Cc: Kefeng Wang <wangkefeng.wang@huawei.com>
Cc: AKASHI Takahiro <takahiro.akashi@linaro.org>
Cc: Jon Masters <jcm@redhat.com>
Cc: David Daney <david.daney@cavium.com>
Cc: Ganapatrao Kulkarni <gkulkarni@caviumnetworks.com>
Cc: Dmitry Vyukov <dvyukov@google.com>
Cc: Nicolai Stange <nicstange@gmail.com>
Cc: James Morse <james.morse@arm.com>
Cc: Andrey Ryabinin <aryabinin@virtuozzo.com>
Cc: Andrey Konovalov <andreyknvl@google.com>
Cc: Alexander Popov <alex.popov@linux.com>
Cc: syzkaller <syzkaller@googlegroups.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

Introduce kaslr_offset() similar to x86_64 to fix kcov.

[ Updated by Will Deacon ]

Link: http://lkml.kernel.org/r/1481417456-28826-2-git-send-email-alex.popov@linux.comSigned-off-by: Alexander Popov <alex.popov@linux.com>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Ard Biesheuvel <ard.biesheuvel@linaro.org>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Rob Herring <robh@kernel.org>
Cc: Kefeng Wang <wangkefeng.wang@huawei.com>
Cc: AKASHI Takahiro <takahiro.akashi@linaro.org>
Cc: Jon Masters <jcm@redhat.com>
Cc: David Daney <david.daney@cavium.com>
Cc: Ganapatrao Kulkarni <gkulkarni@caviumnetworks.com>
Cc: Dmitry Vyukov <dvyukov@google.com>
Cc: Nicolai Stange <nicstange@gmail.com>
Cc: James Morse <james.morse@arm.com>
Cc: Andrey Ryabinin <aryabinin@virtuozzo.com>
Cc: Andrey Konovalov <andreyknvl@google.com>
Cc: Alexander Popov <alex.popov@linux.com>
Cc: syzkaller <syzkaller@googlegroups.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Will Deacon <will.deacon@arm.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

When FADV_DONTNEED cannot drop all pages in the range, it observes that
some pages might still be on per-cpu LRU caches after recent
instantiation and so initiates remote calls to all CPUs to flush their
local caches.  However, in most cases, the fadvise happens from the same
context that instantiated the pages, and any pre-LRU pages in the
specified range are most likely sitting on the local CPU's LRU cache,
and so in many cases this results in unnecessary remote calls, which, in
a loaded system, can hold up the fadvise() call significantly.

[ I didn't record it in the extreme case we observed at Facebook,
  unfortunately. We had a slow-to-respond system and noticed it
  lru_add_drain_all() leading the profile during fadvise calls. This
  patch came out of thinking about the code and how we commonly call
  FADV_DONTNEED.

  FWIW, I wrote a silly directory tree walker/searcher that recurses
  through /usr to read and FADV_DONTNEED each file it finds. On a 2
  socket 40 ht machine, over 1% is spent in lru_add_drain_all(). With
  the patch, that cost is gone; the local drain cost shows at 0.09%. ]

Try to avoid the remote call by flushing the local LRU cache before even
attempting to invalidate anything.  It's a cheap operation, and the
local LRU cache is the most likely to hold any pre-LRU pages in the
specified fadvise range.

Link: http://lkml.kernel.org/r/20161214210017.GA1465@cmpxchg.orgSigned-off-by: Johannes Weiner <hannes@cmpxchg.org>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Acked-by: Mel Gorman <mgorman@suse.de>
Acked-by: Hillf Danton <hillf.zj@alibaba-inc.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>