5 years agocpufreq: intel_pstate: constify attribute_group structures
Arvind Yadav [Mon, 3 Jul 2017 08:10:33 +0000 (13:40 +0530)]
cpufreq: intel_pstate: constify attribute_group structures

attribute_groups are not supposed to change at runtime. All functions
working with attribute_groups provided by <linux/sysfs.h> work with const
attribute_group. So mark the non-const structs as const.

File size before:
   text    data     bss     dec     hex filename
  15197    2552      40   17789    457d drivers/cpufreq/intel_pstate.o

File size After adding 'const':
   text    data     bss     dec     hex filename
  15261    2488      40   17789    457d drivers/cpufreq/intel_pstate.o

Signed-off-by: Arvind Yadav <>
Acked-by: Viresh Kumar <>
Signed-off-by: Rafael J. Wysocki <>
5 years agocpufreq: cpufreq_stats: constify attribute_group structures
Arvind Yadav [Mon, 3 Jul 2017 07:59:04 +0000 (13:29 +0530)]
cpufreq: cpufreq_stats: constify attribute_group structures

attribute_groups are not supposed to change at runtime. All functions
working with attribute_groups provided by <linux/sysfs.h> work with const
attribute_group. So mark the non-const structs as const.

File size before:
   text    data     bss     dec     hex filename
   1655     256       4    1915     77b drivers/cpufreq/cpufreq_stats.o

File size After adding 'const':
   text    data     bss     dec     hex filename
   1695     192       4    1891     763 drivers/cpufreq/cpufreq_stats.o

Signed-off-by: Arvind Yadav <>
Acked-by: Viresh Kumar <>
Signed-off-by: Rafael J. Wysocki <>
5 years agocpufreq / CPPC: Initialize policy->min to lowest nonlinear performance
Prakash, Prashanth [Thu, 11 May 2017 22:39:44 +0000 (16:39 -0600)]
cpufreq / CPPC: Initialize policy->min to lowest nonlinear performance

Description of Lowest Perfomance in ACPI 6.1 specification states:
"Lowest Performance is the absolute lowest performance level of
the platform. Selecting a performance level lower than the lowest
nonlinear performance level may actually cause an efficiency penalty,
but should reduce the instantaneous power consumption of the processor.
In traditional terms, this represents the T-state range of performance

Set the default value of policy->min to Lowest Nonlinear Performance
to avoid any potential efficiency penalty.

Signed-off-by: Prashanth Prakash <>
Acked-by: Viresh Kumar <>
Acked-by: Alexey Klimov <>
Signed-off-by: Rafael J. Wysocki <>
5 years agocpufreq: sfi: make freq_table static
Colin Ian King [Thu, 15 Jun 2017 09:55:59 +0000 (10:55 +0100)]
cpufreq: sfi: make freq_table static

pointer freq_table can be made static as it does not need to be in
global scope.

Cleans up sparse warning:
"symbol 'freq_table' was not declared. Should it be static?"

Signed-off-by: Colin Ian King <>
Acked-by: Viresh Kumar <>
Signed-off-by: Rafael J. Wysocki <>
5 years agocpufreq: exynos5440: Fix inconsistent indenting
Krzysztof Kozlowski [Wed, 7 Jun 2017 18:13:24 +0000 (20:13 +0200)]
cpufreq: exynos5440: Fix inconsistent indenting

Fix inconsistent indenting and unneeded white space in assignment.

Signed-off-by: Krzysztof Kozlowski <>
Acked-by: Viresh Kumar <>
Signed-off-by: Rafael J. Wysocki <>
5 years agocpufreq: imx6q: imx6ull should use the same flow as imx6ul
Octavian Purdila [Tue, 30 May 2017 15:57:18 +0000 (18:57 +0300)]
cpufreq: imx6q: imx6ull should use the same flow as imx6ul

This fixes an issue with imx6ull where setting the frequency to 528Mhz
would actually set the ARM clock to 324Mhz.

Signed-off-by: Octavian Purdila <>
Signed-off-by: Leonard Crestez <>
Acked-by: Viresh Kumar <>
Reviewed-by: Fabio Estevam <>
Signed-off-by: Rafael J. Wysocki <>
5 years agocpufreq: dt: Add support for hi3660
Tao Wang [Tue, 23 May 2017 08:13:18 +0000 (16:13 +0800)]
cpufreq: dt: Add support for hi3660

Add the compatible string for supporting the generic device tree cpufreq-dt
driver on Hisilicon's 3660 SoC.

Signed-off-by: Tao Wang <>
Acked-by: Viresh Kumar <>
Signed-off-by: Rafael J. Wysocki <>
5 years agoLinux 4.12-rc6 v4.12-rc6
Linus Torvalds [Mon, 19 Jun 2017 14:19:37 +0000 (22:19 +0800)]
Linux 4.12-rc6

5 years agomm: larger stack guard gap, between vmas
Hugh Dickins [Mon, 19 Jun 2017 11:03:24 +0000 (04:03 -0700)]
mm: larger stack guard gap, between vmas

Stack guard page is a useful feature to reduce a risk of stack smashing
into a different mapping. We have been using a single page gap which
is sufficient to prevent having stack adjacent to a different mapping.
But this seems to be insufficient in the light of the stack usage in
userspace. E.g. glibc uses as large as 64kB alloca() in many commonly
used functions. Others use constructs liks gid_t buffer[NGROUPS_MAX]
which is 256kB or stack strings with MAX_ARG_STRLEN.

This will become especially dangerous for suid binaries and the default
no limit for the stack size limit because those applications can be
tricked to consume a large portion of the stack and a single glibc call
could jump over the guard page. These attacks are not theoretical,

Make those attacks less probable by increasing the stack guard gap
to 1MB (on systems with 4k pages; but make it depend on the page size
because systems with larger base pages might cap stack allocations in
the PAGE_SIZE units) which should cover larger alloca() and VLA stack
allocations. It is obviously not a full fix because the problem is
somehow inherent, but it should reduce attack space a lot.

One could argue that the gap size should be configurable from userspace,
but that can be done later when somebody finds that the new 1MB is wrong
for some special case applications.  For now, add a kernel command line
option (stack_guard_gap) to specify the stack gap size (in page units).

Implementation wise, first delete all the old code for stack guard page:
because although we could get away with accounting one extra page in a
stack vma, accounting a larger gap can break userspace - case in point,
a program run with "ulimit -S -v 20000" failed when the 1MB gap was
counted for RLIMIT_AS; similar problems could come with RLIMIT_MLOCK
and strict non-overcommit mode.

Instead of keeping gap inside the stack vma, maintain the stack guard
gap as a gap between vmas: using vm_start_gap() in place of vm_start
(or vm_end_gap() in place of vm_end if VM_GROWSUP) in just those few
places which need to respect the gap - mainly arch_get_unmapped_area(),
and and the vma tree's subtree_gap support for that.

Original-patch-by: Oleg Nesterov <>
Original-patch-by: Michal Hocko <>
Signed-off-by: Hugh Dickins <>
Acked-by: Michal Hocko <>
Tested-by: Helge Deller <> # parisc
Signed-off-by: Linus Torvalds <>
5 years agoMerge tag 'armsoc-fixes' of git://
Linus Torvalds [Mon, 19 Jun 2017 08:50:09 +0000 (16:50 +0800)]
Merge tag 'armsoc-fixes' of git://git./linux/kernel/git/arm/arm-soc

Pull ARM SoC fixes from Olof Johansson:
 "Stream of fixes has slowed down, only a few this week:

   - Some DT fixes for Allwinner platforms, and addition of a clock to
     the R_CCU clock controller that had been missed.

   - A couple of small DT fixes for am335x-sl50"

* tag 'armsoc-fixes' of git://
  arm64: allwinner: a64: Add PLL_PERIPH0 clock to the R_CCU
  ARM: sunxi: h3-h5: Add PLL_PERIPH0 clock to the R_CCU
  ARM: dts: am335x-sl50: Fix cannot claim requested pins for spi0
  ARM: dts: am335x-sl50: Fix card detect pin for mmc1
  arm64: allwinner: h5: Remove syslink to shared DTSI
  ARM: sunxi: h3/h5: fix the compatible of R_CCU

5 years agoMerge tag 'sunxi-fixes-for-4.12' of
Olof Johansson [Mon, 19 Jun 2017 03:42:21 +0000 (20:42 -0700)]
Merge tag 'sunxi-fixes-for-4.12' of https://git./linux/kernel/git/sunxi/linux into fixes

Allwinner fixes for 4.12

A few fixes around the PRCM support that got in 4.12 with a wrong
compatible, and a missing clock in the binding.

* tag 'sunxi-fixes-for-4.12' of
  arm64: allwinner: a64: Add PLL_PERIPH0 clock to the R_CCU
  ARM: sunxi: h3-h5: Add PLL_PERIPH0 clock to the R_CCU
  arm64: allwinner: h5: Remove syslink to shared DTSI
  ARM: sunxi: h3/h5: fix the compatible of R_CCU

Signed-off-by: Olof Johansson <>
5 years agoMerge tag 'omap-for-v4.12/fixes-sl50' of git://
Olof Johansson [Mon, 19 Jun 2017 01:55:12 +0000 (18:55 -0700)]
Merge tag 'omap-for-v4.12/fixes-sl50' of git://git./linux/kernel/git/tmlind/linux-omap into fixes

Two fixes for am335x-sl50 to fix a boot time error
for claiming SPI pins, and to fix a SDIO card detect
pin for production version of the device.

* tag 'omap-for-v4.12/fixes-sl50' of git://
  ARM: dts: am335x-sl50: Fix cannot claim requested pins for spi0
  ARM: dts: am335x-sl50: Fix card detect pin for mmc1

Signed-off-by: Olof Johansson <>
5 years agoMerge tag 'for_linus' of git://
Linus Torvalds [Mon, 19 Jun 2017 00:25:05 +0000 (09:25 +0900)]
Merge tag 'for_linus' of git://git./linux/kernel/git/mst/vhost

Pull virtio bugfix from Michael Tsirkin:
 "It turns out balloon does not handle IOMMUs correctly. We should fix
  that at some point, for now let's just disable this configuration"

* tag 'for_linus' of git://
  virtio_balloon: disable VIOMMU support

5 years agoMerge branch 'i2c/for-current-fixed' of git://
Linus Torvalds [Mon, 19 Jun 2017 00:20:25 +0000 (09:20 +0900)]
Merge branch 'i2c/for-current-fixed' of git://git./linux/kernel/git/wsa/linux

Pull i2c fixes from Wolfram Sang:
 "Two driver bugfixes"

* 'i2c/for-current-fixed' of git://
  i2c: ismt: fix wrong device address when unmap the data buffer
  i2c: rcar: use correct length when unmapping DMA

5 years agoMerge branch 'upstream' of git://
Linus Torvalds [Mon, 19 Jun 2017 00:01:01 +0000 (09:01 +0900)]
Merge branch 'upstream' of git://

Pull MIPS fixes from Ralf Baechle:

 - Three highmem fixes:
    + Fixed mapping initialization
    + Adjust the pkmap location
    + Ensure we use at most one page for PTEs

 - Fix makefile dependencies for .its targets to depend on vmlinux

 - Fix reversed condition in BNEZC and JIALC software branch emulation

 - Only flush initialized flush_insn_slot to avoid NULL pointer

 - perf: Remove incorrect odd/even counter handling for I6400

 - ftrace: Fix init functions tracing

* 'upstream' of git://
  MIPS: .its targets depend on vmlinux
  MIPS: Fix bnezc/jialc return address calculation
  MIPS: kprobes: flush_insn_slot should flush only if probe initialised
  MIPS: ftrace: fix init functions tracing
  MIPS: mm: adjust PKMAP location
  MIPS: highmem: ensure that we don't use more than one page for PTEs
  MIPS: mm: fixed mappings: correct initialisation
  MIPS: perf: Remove incorrect odd/even counter handling for I6400

5 years agovirtio_balloon: disable VIOMMU support
Michael S. Tsirkin [Tue, 13 Jun 2017 17:56:44 +0000 (20:56 +0300)]
virtio_balloon: disable VIOMMU support

virtio balloon bypasses the DMA API entirely so does not support the
VIOMMU right now.  It's not clear we need that support, for now let's
just make sure we don't pretend to support it.

Cc: Wei Wang <>
Fixes: 1a937693993f ("virtio: new feature to detect IOMMU device quirk")
Signed-off-by: Michael S. Tsirkin <>
Acked-by: Jason Wang <>
5 years agoMerge branch 'x86-urgent-for-linus' of git://
Linus Torvalds [Sun, 18 Jun 2017 09:49:12 +0000 (18:49 +0900)]
Merge branch 'x86-urgent-for-linus' of git://git./linux/kernel/git/tip/tip

Pull x86 fixes from Thomas Gleixner:
 "Two fixlets for x86:

   - Handle WARN_ONs proper with the new UD based WARN implementation

   - Disable 1G mappings when 2M mappings are disabled by kmemleak or
     debug_pagealloc. Otherwise 1G mappings might still be used,
     confusing the debug mechanisms"

* 'x86-urgent-for-linus' of git://
  x86/mm: Disable 1GB direct mappings when disabling 2MB mappings
  x86/debug: Handle early WARN_ONs proper

5 years agoMerge branch 'timers-urgent-for-linus' of git://
Linus Torvalds [Sun, 18 Jun 2017 09:46:51 +0000 (18:46 +0900)]
Merge branch 'timers-urgent-for-linus' of git://git./linux/kernel/git/tip/tip

Pull timer fixes from Thomas Gleixner:
 "Three fixlets for timers:

   - Two hot-fixes for the alarmtimer based posix timers, which prevent
     a nasty DOS by self rescheduling timers. The proper cleanup of that
     mess is queued for 4.13

   - Make a function static"

* 'timers-urgent-for-linus' of git://
  tick/broadcast: Make tick_broadcast_setup_oneshot() static
  alarmtimer: Rate limit periodic intervals
  alarmtimer: Prevent overflow of relative timers

5 years agoMerge branch 'sched-urgent-for-linus' of git://
Linus Torvalds [Sun, 18 Jun 2017 09:45:17 +0000 (18:45 +0900)]
Merge branch 'sched-urgent-for-linus' of git://git./linux/kernel/git/tip/tip

Pull scheduler fixes from Thomas Gleixner:
 "Two small fixes for the schedulre core:

   - Use the proper switch_mm() variant in idle_task_exit() because that
     code is not called with interrupts disabled.

   - Fix a confusing typo in a printk"

* 'sched-urgent-for-linus' of git://
  sched/core: Idle_task_exit() shouldn't use switch_mm_irqs_off()
  sched/fair: Fix typo in printk message

5 years agoMerge branch 'perf-urgent-for-linus' of git://
Linus Torvalds [Sun, 18 Jun 2017 09:42:31 +0000 (18:42 +0900)]
Merge branch 'perf-urgent-for-linus' of git://git./linux/kernel/git/tip/tip

Pull perf fixes from Thomas Gleixner:
 "Three fixes for the perf user space side:

   - Fix the probing of precise_ip level, which got broken recently for

   - Unbreak the ARCH=x86_64 build

   - Report module before trying to unwind into the module code, which
     avoids broken stack frames displayed"

* 'perf-urgent-for-linus' of git://
  perf unwind: Report module before querying isactivation in dwfl unwind
  perf tools: Fix build with ARCH=x86_64
  perf evsel: Fix probing of precise_ip level for default cycles event

5 years agoMerge branch 'irq-urgent-for-linus' of git://
Linus Torvalds [Sun, 18 Jun 2017 09:40:41 +0000 (18:40 +0900)]
Merge branch 'irq-urgent-for-linus' of git://git./linux/kernel/git/tip/tip

Pull irq fix from Thomas Gleixner:
 "Add a missing resource release to an error path"

* 'irq-urgent-for-linus' of git://
  genirq: Release resources in __setup_irq() error path

5 years agoMerge branch 'core-urgent-for-linus' of git://
Linus Torvalds [Sun, 18 Jun 2017 09:38:42 +0000 (18:38 +0900)]
Merge branch 'core-urgent-for-linus' of git://git./linux/kernel/git/tip/tip

Pull objtool fix from Thomas Gleixner:
 "A single fix which adds fortify_panic to the list of no return

* 'core-urgent-for-linus' of git://
  objtool: Add fortify_panic as __noreturn function

5 years agoMerge tag 'led_fixes_for_4.12-rc6' of git://
Linus Torvalds [Sat, 17 Jun 2017 23:51:35 +0000 (08:51 +0900)]
Merge tag 'led_fixes_for_4.12-rc6' of git://git./linux/kernel/git/j.anaszewski/linux-leds

Pull LED fixes from Jacek Anaszewski:
 "Two LED fixes:

   - fix signal source assignment for leds-bcm6328

   - revert patch that intended to fix LED behavior on suspend but it
     had a side effect preventing suspend at all due to uevent being
     sent on trigger removal"

* tag 'led_fixes_for_4.12-rc6' of git://
  Revert "leds: handle suspend/resume in heartbeat trigger"
  leds: bcm6328: fix signal source assignment for leds 4 to 7

5 years agoMerge tag 'usb-4.12-rc6' of git://
Linus Torvalds [Sat, 17 Jun 2017 23:39:54 +0000 (08:39 +0900)]
Merge tag 'usb-4.12-rc6' of git://git./linux/kernel/git/gregkh/usb

Pull USB fixes from Greg KH:
 "Here are some small gadget and xhci USB fixes for 4.12-rc6.

  Nothing major, but one of the gadget patches does fix a reported oops,
  and the xhci ones resolve reported problems. All have been in
  linux-next with no reported issues"

* tag 'usb-4.12-rc6' of git://
  USB: gadgetfs, dummy-hcd, net2280: fix locking for callbacks
  usb: xhci: ASMedia ASM1042A chipset need shorts TX quirk
  usb: xhci: Fix USB 3.1 supported protocol parsing
  USB: gadget: fix GPF in gadgetfs
  usb: gadget: composite: make sure to reactivate function on unbind

5 years agoMerge tag 'staging-4.12-rc6' of git://
Linus Torvalds [Sat, 17 Jun 2017 23:36:30 +0000 (08:36 +0900)]
Merge tag 'staging-4.12-rc6' of git://git./linux/kernel/git/gregkh/staging

Pull staging and IIO fixes from Greg KH:
 "Here are some small staging and IIO driver fixes for 4.12-rc6.

  Nothing huge, just a few small driver fixes for reported issues. All
  have been in linux-next with no reported issues"

* tag 'staging-4.12-rc6' of git://
  Staging: rtl8723bs: fix an error code in isFileReadable()
  iio: buffer-dmaengine: Add missing header buffer_impl.h
  iio: buffer-dma: Add missing header buffer_impl.h
  iio: adc: meson-saradc: fix potential crash in meson_sar_adc_clear_fifo
  iio: adc: mxs-lradc: Fix return value check in mxs_lradc_adc_probe()
  iio: imu: inv_mpu6050: add accel lpf setting for chip >= MPU6500
  staging: iio: ad7152: Fix deadlock in ad7152_write_raw_samp_freq()

5 years agoMerge tag 'ceph-for-4.12-rc6' of git://
Linus Torvalds [Sat, 17 Jun 2017 23:23:02 +0000 (08:23 +0900)]
Merge tag 'ceph-for-4.12-rc6' of git://

Pull ceph fixes from Ilya Dryomov:
 "A fix for an old ceph ->fh_to_* bug from Luis and two timestamp fixups
  from Zheng, prompted by the ongoing y2038 work"

* tag 'ceph-for-4.12-rc6' of git://
  ceph: unify inode i_ctime update
  ceph: use current_kernel_time() to get request time stamp
  ceph: check i_nlink while converting a file handle to dentry

5 years agoMerge tag 'xfs-4.12-fixes-4' of git://
Linus Torvalds [Sat, 17 Jun 2017 08:34:41 +0000 (17:34 +0900)]
Merge tag 'xfs-4.12-fixes-4' of git://git./fs/xfs/xfs-linux

Pull xfs fix from Darrick Wong:
 "One more bugfix for you for 4.12-rc6 to fix something that came up in
  an earlier rc:

   - Fix some bogus ASSERT failures on CONFIG_SMP=n and CONFIG_XFS_DEBUG=y"

* tag 'xfs-4.12-fixes-4' of git://
  xfs: fix spurious spin_is_locked() assert failures on non-smp kernels

5 years agoMerge branch 'ufs-fixes' of git://
Linus Torvalds [Sat, 17 Jun 2017 08:30:07 +0000 (17:30 +0900)]
Merge branch 'ufs-fixes' of git://git./linux/kernel/git/viro/vfs

Pull ufs fixes from Al Viro:
 "Fix assorted ufs bugs: a couple of deadlocks, fs corruption in
  truncate(), oopsen on tail unpacking and truncate when racing with
  vmscan, mild fs corruption (free blocks stats summary buggered, *BSD
  fsck would complain and fix), several instances of broken logics
  around reserved blocks (starting with "check almost never triggers
  when it should" and then there are issues with sufficiently large

[ Note: ufs hasn't gotten any loving in a long time, because nobody
  really seems to use it. These ufs fixes are triggered by people
  actually caring now, not some sudden influx of new bugs.  - Linus ]

* 'ufs-fixes' of git://
  ufs_truncate_blocks(): fix the case when size is in the last direct block
  ufs: more deadlock prevention on tail unpacking
  ufs: avoid grabbing ->truncate_mutex if possible
  ufs_get_locked_page(): make sure we have buffer_heads
  ufs: fix s_size/s_dsize users
  ufs: fix reserved blocks check
  ufs: make ufs_freespace() return signed
  ufs: fix logics in "ufs: make fsck -f happy"

5 years agoMerge branch 'for-linus' of git://
Linus Torvalds [Sat, 17 Jun 2017 08:26:53 +0000 (17:26 +0900)]
Merge branch 'for-linus' of git://git./linux/kernel/git/viro/vfs

Pull vfs fixes from Al Viro:
 "A couple of fixes; a leak in mntns_install() caught by Andrei (this
  cycle regression) + d_invalidate() softlockup fix - that had been
  reported by a bunch of people lately, but the problem is pretty old"

* 'for-linus' of git://
  fs: don't forget to put old mntns in mntns_install
  Hang/soft lockup in d_invalidate with simultaneous calls

5 years agoMerge tag 'pci-v4.12-fixes-2' of git://
Linus Torvalds [Fri, 16 Jun 2017 21:53:20 +0000 (06:53 +0900)]
Merge tag 'pci-v4.12-fixes-2' of git://git./linux/kernel/git/helgaas/pci

Pull PCI fixes from Bjorn Helgaas:

 - fix another PCI_ENDPOINT build error (merged for v4.12)

 - fix error codes added to config accessors for v4.12

* tag 'pci-v4.12-fixes-2' of git://
  PCI: endpoint: Select CRC32 to fix test build error
  PCI: Make error code types consistent in pci_{read,write}_config_*

5 years agoMerge tag 'fbdev-v4.12-rc6' of git://
Linus Torvalds [Fri, 16 Jun 2017 21:51:25 +0000 (06:51 +0900)]
Merge tag 'fbdev-v4.12-rc6' of git://

Pull fbdev fixes from Bartlomiej Zolnierkiewicz:

 - fix udlfb driver to stop spamming logs (Mike Gerow)

 - add missing endianness conversions in smscufx & udlfb drivers (Johan

 - fix few gcc warnings/errors (Arnd Bergmann)

* tag 'fbdev-v4.12-rc6' of git://
  video: fbdev: udlfb: drop log level for blanking
  video: fbdev: via: remove possibly unused variables
  video: fbdev: add missing USB-descriptor endianness conversions
  video: fbdev: avoid int-in-bool-context warning

5 years agoMerge branch 'akpm' (patches from Andrew)
Linus Torvalds [Fri, 16 Jun 2017 21:49:34 +0000 (06:49 +0900)]
Merge branch 'akpm' (patches from Andrew)

Merge misc fixes from Andrew Morton:
 "5 fixes"

* emailed patches from Andrew Morton <>:
  mm: correct the comment when reclaimed pages exceed the scanned pages
  userfaultfd: shmem: handle coredumping in handle_userfault()
  mm: numa: avoid waiting on freed migrated pages
  swap: cond_resched in swap_cgroup_prepare()
  mm/memory-failure.c: use compound_head() flags for huge pages

5 years agomm: correct the comment when reclaimed pages exceed the scanned pages
zhongjiang [Fri, 16 Jun 2017 21:02:40 +0000 (14:02 -0700)]
mm: correct the comment when reclaimed pages exceed the scanned pages

Commit e1587a494540 ("mm: vmpressure: fix sending wrong events on
underflow") declared that reclaimed pages exceed the scanned pages due
to the thp reclaim.

That is incorrect because THP will be spilt to normal page and loop
again, which will result in the scanned pages increment.

[ tweak comment text]
Signed-off-by: zhongjiang <>
Acked-by: Minchan Kim <>
Signed-off-by: Andrew Morton <>
Signed-off-by: Linus Torvalds <>
5 years agouserfaultfd: shmem: handle coredumping in handle_userfault()
Andrea Arcangeli [Fri, 16 Jun 2017 21:02:37 +0000 (14:02 -0700)]
userfaultfd: shmem: handle coredumping in handle_userfault()

Anon and hugetlbfs handle FOLL_DUMP set by get_dump_page() internally to

shmem as opposed has no special FOLL_DUMP handling there so
handle_mm_fault() is invoked without mmap_sem and ends up calling
handle_userfault() that isn't expecting to be invoked without mmap_sem

This makes handle_userfault() fail immediately if invoked through
shmem_vm_ops->fault during coredumping and solves the problem.

The side effect is a BUG_ON with no lock held triggered by the
coredumping process which exits.  Only 4.11 is affected, pre-4.11 anon
memory holes are skipped in __get_user_pages by checking FOLL_DUMP
explicitly against empty pagetables (mm/gup.c:no_page_table()).

It's zero cost as we already had a check for current->flags to prevent
futex to trigger userfaults during exit (PF_EXITING).

Signed-off-by: Andrea Arcangeli <>
Reported-by: "Dr. David Alan Gilbert" <>
Cc: <> [4.11+]
Signed-off-by: Andrew Morton <>
Signed-off-by: Linus Torvalds <>
5 years agomm: numa: avoid waiting on freed migrated pages
Mark Rutland [Fri, 16 Jun 2017 21:02:34 +0000 (14:02 -0700)]
mm: numa: avoid waiting on freed migrated pages

In do_huge_pmd_numa_page(), we attempt to handle a migrating thp pmd by
waiting until the pmd is unlocked before we return and retry.  However,
we can race with migrate_misplaced_transhuge_page():

    // do_huge_pmd_numa_page                // migrate_misplaced_transhuge_page()
    // Holds 0 refs on page                 // Holds 2 refs on page

    vmf->ptl = pmd_lock(vma->vm_mm, vmf->pmd);
    /* ... */
    if (pmd_trans_migrating(*vmf->pmd)) {
            page = pmd_page(*vmf->pmd);
                                            ptl = pmd_lock(mm, pmd);
                                            if (page_count(page) != 2)) {
                                                    /* roll back */
                                            /* ... */
                                            mlock_migrate_page(new_page, page);
                                            /* ... */
                                            put_page(page); // page freed here
            goto out;

This can result in the freed page having its waiters flag set
unexpectedly, which trips the PAGE_FLAGS_CHECK_AT_PREP checks in the
page alloc/free functions.  This has been observed on arm64 KVM guests.

We can avoid this by having do_huge_pmd_numa_page() take a reference on
the page before dropping the pmd lock, mirroring what we do in

When we hit the race, migrate_misplaced_transhuge_page() will see the
reference and abort the migration, as it may do today in other cases.

Fixes: b8916634b77bffb2 ("mm: Prevent parallel splits during THP migration")
Signed-off-by: Mark Rutland <>
Signed-off-by: Will Deacon <>
Acked-by: Steve Capper <>
Acked-by: Kirill A. Shutemov <>
Acked-by: Vlastimil Babka <>
Cc: Mel Gorman <>
Cc: <>
Signed-off-by: Andrew Morton <>
Signed-off-by: Linus Torvalds <>
5 years agoswap: cond_resched in swap_cgroup_prepare()
Yu Zhao [Fri, 16 Jun 2017 21:02:31 +0000 (14:02 -0700)]
swap: cond_resched in swap_cgroup_prepare()

I saw need_resched() warnings when swapping on large swapfile (TBs)
because continuously allocating many pages in swap_cgroup_prepare() took
too long.

We already cond_resched when freeing page in swap_cgroup_swapoff().  Do
the same for the page allocation.

Signed-off-by: Yu Zhao <>
Acked-by: Michal Hocko <>
Acked-by: Vladimir Davydov <>
Cc: <>
Signed-off-by: Andrew Morton <>
Signed-off-by: Linus Torvalds <>
5 years agomm/memory-failure.c: use compound_head() flags for huge pages
James Morse [Fri, 16 Jun 2017 21:02:29 +0000 (14:02 -0700)]
mm/memory-failure.c: use compound_head() flags for huge pages

memory_failure() chooses a recovery action function based on the page
flags.  For huge pages it uses the tail page flags which don't have
anything interesting set, resulting in:

> Memory failure: 0x9be3b4: Unknown page state
> Memory failure: 0x9be3b4: recovery action for unknown page: Failed

Instead, save a copy of the head page's flags if this is a huge page,
this means if there are no relevant flags for this tail page, we use the
head pages flags instead.  This results in the me_huge_page() recovery
action being called:

> Memory failure: 0x9b7969: recovery action for huge page: Delayed

For hugepages that have not yet been allocated, this allows the hugepage
to be dequeued.

Fixes: 524fca1e7356 ("HWPOISON: fix misjudgement of page_action() for errors on mlocked pages")
Signed-off-by: James Morse <>
Tested-by: Punit Agrawal <>
Acked-by: Punit Agrawal <>
Acked-by: Naoya Horiguchi <>
Cc: <>
Signed-off-by: Andrew Morton <>
Signed-off-by: Linus Torvalds <>
5 years agoMerge tag 'powerpc-4.12-6' of git://
Linus Torvalds [Fri, 16 Jun 2017 20:57:54 +0000 (05:57 +0900)]
Merge tag 'powerpc-4.12-6' of git://git./linux/kernel/git/powerpc/linux

Pull powerpc fixes from Michael Ellerman:
 "Three small fixes for recently merged code:

   - remove a spurious WARN_ON when a PCI device has no of_node, it's
     allowed in some circumstances for there to be no of_node.

   - fix the offset for store EOI MMIOs in the XIVE interrupt

   - fix non-const WARN_ONs which were becoming BUGs due to them losing
     BUGFLAG_WARNING in a recent cleanup patch.

  Thanks to: Alexey Kardashevskiy, Alistair Popple, Benjamin

* tag 'powerpc-4.12-6' of git://
  powerpc/debug: Add missing warn flag to WARN_ON's non-builtin path
  powerpc/xive: Fix offset for store EOI MMIOs
  powerpc/npu-dma: Remove spurious WARN_ON when a PCI device has no of_node

5 years agoMerge tag 'perf-urgent-for-mingo-4.12-20170616' of git://
Ingo Molnar [Fri, 16 Jun 2017 19:33:48 +0000 (21:33 +0200)]
Merge tag 'perf-urgent-for-mingo-4.12-20170616' of git://git./linux/kernel/git/acme/linux into perf/urgent

Pull perf/urgent fixes from Arnaldo Carvalho de Melo:

- Fix probing of precise_ip level for default cycles event, that
  got broken recently on x86_64 when its arch code started
  considering invalid requesting precise samples when not sampling
  (i.e. when attr.sample_period == 0).

  This also fixes another problem in s/390 where the precision
  probing with sample_period == 0 returned precise_ip > 0, that
  then, when setting up the real cycles event (not probing) would
  return EOPNOTSUPP for precise_ip > 0 (as determined previously
  by probing) and sample_period > 0.

  These problems resulted in attr_precise not being set to the
  highest precision available on x86.64 when no event was specified,
  i.e. the canonical:

perf record ./workload

  would end up using attr.precise_ip = 0. As a workaround this would
  need to be done:

perf record -e cycles:P ./workload

  And on s/390 it would plain not work, requiring using:

        perf record -e cycles ./workload

  as a workaround.  (Arnaldo Carvalho de Melo)

- Fix perf build with ARCH=x86_64, when ARCH should be transformed
  into ARCH=x86, just like with the main kernel Makefile and
  tools/objtool's, i.e. use SRCARCH. (Jiada Wang)

- Avoid accessing uninitialized data structures when unwinding with
  elfutils's libdw, making it more closely mimic libunwind's unwinder.
  (Milian Wolff)

Signed-off-by: Arnaldo Carvalho de Melo <>
Signed-off-by: Ingo Molnar <>
5 years agoperf unwind: Report module before querying isactivation in dwfl unwind
Milian Wolff [Fri, 2 Jun 2017 14:37:53 +0000 (16:37 +0200)]
perf unwind: Report module before querying isactivation in dwfl unwind

The PC returned by dwfl_frame_pc() may map into a not-yet-reported
module. We have to report it before we continue unwinding. But when we
query for the isactivation flag in dwfl_frame_pc, libdw will actually do
one more unwinding step internally which can then break and lead to
missed frames or broken stacks.

With libunwind we get e.g.:

  heaptrack_gui  2228 135073.400474:     613969 cycles:
          108c8e [unknown] (/usr/lib/
          1093bc [unknown] (/usr/lib/
          109e7b QLocale::QLocale (/usr/lib/
          1470ff [unknown] (/usr/lib/
          147f67 QSystemLocale::query (/usr/lib/
          109fbf QLocalePrivate::updateSystemPrivate (/usr/lib/
          10aa27 QLocale::QLocale (/usr/lib/
          1e02c3 [unknown] (/usr/lib/
          2113bb [unknown] (/usr/lib/
          211505 [unknown] (/usr/lib/
          1b5df0 QFileInfo::exists (/usr/lib/
           92eb2 [unknown] (/usr/lib/
           93423 [unknown] (/usr/lib/
           93d2a QLibraryInfo::location (/usr/lib/
          2170af [unknown] (/usr/lib/
          297c53 QCoreApplicationPrivate::init (/usr/lib/
           f7cde QGuiApplicationPrivate::init (/usr/lib/
          1589e8 QApplicationPrivate::init (/usr/lib/
           78622 main (/home/milian/projects/compiled/other/bin/heaptrack_gui)
           20439 __libc_start_main (/usr/lib/
           78299 _start (/home/milian/projects/compiled/other/bin/heaptrack_gui)

  heaptrack_gui  2228 135073.401156:     569521 cycles:
          131633 QString::endsWith (/usr/lib/
          1a0701 QDir::cleanPath (/usr/lib/
          21b82d [unknown] (/usr/lib/
          1b3727 QFileInfo::canonicalFilePath (/usr/lib/
          2780c7 QFactoryLoader::update (/usr/lib/
          279525 QFactoryLoader::QFactoryLoader (/usr/lib/
           e5bd0 QPlatformIntegrationFactory::create (/usr/lib/
           f5a1c QGuiApplicationPrivate::createPlatformIntegration (/usr/lib/
           f650c QGuiApplicationPrivate::createEventDispatcher (/usr/lib/
          298524 QCoreApplicationPrivate::init (/usr/lib/
           f7cde QGuiApplicationPrivate::init (/usr/lib/
          1589e8 QApplicationPrivate::init (/usr/lib/
           78622 main (/home/milian/projects/compiled/other/bin/heaptrack_gui)
           20439 __libc_start_main (/usr/lib/
           78299 _start (/home/milian/projects/compiled/other/bin/heaptrack_gui)

Note the two frames 1589e8 and 78622 in the first sample. These are
missing when unwinding with libdw. The second sample's breakage is
more obvious:

  heaptrack_gui  2228 135073.400474:     613969 cycles:
          108c8e [unknown] (/usr/lib/
          1093bc [unknown] (/usr/lib/
          109e7b QLocale::QLocale (/usr/lib/
          1470ff [unknown] (/usr/lib/
          147f67 QSystemLocale::query (/usr/lib/
          109fbf QLocalePrivate::updateSystemPrivate (/usr/lib/
          10aa27 QLocale::QLocale (/usr/lib/
          1e02c3 [unknown] (/usr/lib/
          2113bb [unknown] (/usr/lib/
          211505 [unknown] (/usr/lib/
          1b5df0 QFileInfo::exists (/usr/lib/
           92eb2 [unknown] (/usr/lib/
           93423 [unknown] (/usr/lib/
           93d2a QLibraryInfo::location (/usr/lib/
          2170af [unknown] (/usr/lib/
          297c53 QCoreApplicationPrivate::init (/usr/lib/
           f7cde QGuiApplicationPrivate::init (/usr/lib/
           20439 __libc_start_main (/usr/lib/
           78299 _start (/home/milian/projects/compiled/other/bin/heaptrack_gui)

heaptrack_gui  2228 135073.401156:     569521 cycles:
          131633 QString::endsWith (/usr/lib/
          1a0701 QDir::cleanPath (/usr/lib/
          21b82d [unknown] (/usr/lib/
          1b3727 QFileInfo::canonicalFilePath (/usr/lib/
          2780c7 QFactoryLoader::update (/usr/lib/
          279525 QFactoryLoader::QFactoryLoader (/usr/lib/
           e5bd0 QPlatformIntegrationFactory::create (/usr/lib/
          723dbf [unknown] ([unknown])

This patch fixes this issue and the libdw unwinder mimicks the libunwind
behavior more closely.

Signed-off-by: Milian Wolff <>
Acked-by: Jan Kratochvil <>
Cc: Jiri Olsa <>
Cc: Namhyung Kim <>
Signed-off-by: Arnaldo Carvalho de Melo <>
5 years agoMerge tag 'configfs-for-4.12' of git://
Linus Torvalds [Fri, 16 Jun 2017 09:45:47 +0000 (18:45 +0900)]
Merge tag 'configfs-for-4.12' of git://

Pull configfs updates from Christoph Hellwig:
 "A fix from Nic for a race seen in production (including a stable tag).

  And while I'm sending you this I'm also sneaking in a trivial new
  helper from Bart so that we don't need inter-tree dependencies for the
  next merge window"

* tag 'configfs-for-4.12' of git://
  configfs: Introduce config_item_get_unless_zero()
  configfs: Fix race between create_link and configfs_rmdir

5 years agofs: pass on flags in compat_writev
Christoph Hellwig [Fri, 16 Jun 2017 09:08:24 +0000 (11:08 +0200)]
fs: pass on flags in compat_writev

Fixes: 793b80ef14af ("vfs: pass a flags argument to vfs_readv/vfs_writev")
Signed-off-by: Christoph Hellwig <>
Signed-off-by: Linus Torvalds <>
5 years agoobjtool: Add fortify_panic as __noreturn function
Kees Cook [Thu, 15 Jun 2017 13:20:35 +0000 (08:20 -0500)]
objtool: Add fortify_panic as __noreturn function

CONFIG_FORTIFY_SOURCE=y implements fortify_panic() as a __noreturn function,
so objtool needs to know about it too.

Suggested-by: Daniel Micay <>
Tested-by: Stephen Rothwell <>
Signed-off-by: Kees Cook <>
Signed-off-by: Josh Poimboeuf <>
Cc: Linus Torvalds <>
Cc: Peter Zijlstra <>
Cc: Thomas Gleixner <>
Signed-off-by: Ingo Molnar <>
5 years agoMerge tag 'mmc-v4.12-rc5' of git://
Linus Torvalds [Fri, 16 Jun 2017 08:49:52 +0000 (17:49 +0900)]
Merge tag 'mmc-v4.12-rc5' of git://git./linux/kernel/git/ulfh/mmc

Pull MMC fix from Ulf Hansson:
 "MMC meson-gx host: work around broken SDIO with certain WiFi chips"

* tag 'mmc-v4.12-rc5' of git://
  mmc: meson-gx: work around broken SDIO with certain WiFi chips

5 years agoMerge tag 'drm-fixes-for-v4.12-rc6' of git://
Linus Torvalds [Fri, 16 Jun 2017 08:46:47 +0000 (17:46 +0900)]
Merge tag 'drm-fixes-for-v4.12-rc6' of git://

Pull drm fixes from Dave Airlie:
 "This is the main fixes pull for 4.12-rc6, all pretty normal for this
  stage, nothing really stands out. The mxsfb one is probably the
  largest and it's for a black screen boot problem.

  AMD, i915, mgag200, msxfb, tegra fixes"

* tag 'drm-fixes-for-v4.12-rc6' of git://
  drm: mxsfb_crtc: Reset the eLCDIF controller
  drm/mgag200: Fix to always set HiPri for G200e4 V2
  drm/tegra: Correct idr_alloc() minimum id
  drm/tegra: Fix lockup on a use of staging API
  gpu: host1x: Fix error handling
  drm/radeon: Fix overflow of watermark calcs at > 4k resolutions.
  drm/amdgpu: Fix overflow of watermark calcs at > 4k resolutions.
  drm/radeon: fix "force the UVD DPB into VRAM as well"
  drm/i915: Fix GVT-g PVINFO version compatibility check
  drm/i915: Fix SKL+ watermarks for 90/270 rotation
  drm/i915: Fix scaling check for 90/270 degree plane rotation
  drm: dw-hdmi: Fix compilation breakage by selecting REGMAP_MMIO

5 years agoMerge tag 'for-linus' of git://
Linus Torvalds [Fri, 16 Jun 2017 08:38:23 +0000 (17:38 +0900)]
Merge tag 'for-linus' of git://git./linux/kernel/git/dledford/rdma

Pull rdma fixes from Doug Ledford:

 "I had thought at the time of the last pull request that there wouldn't
  be much more to go, but several things just kept trickling in over the
  last week.

  Instead of just the six patches to bnxt_re that I had anticipated,
  there are another five IPoIB patches, two qedr patches, and a few
  other miscellaneous patches.

  The bnxt_re patches are more lines of diff than I like to submit this
  late in the game. That's mostly because of the first two patches in
  the series of six. I almost dropped them just because of the lines of
  churn, but on a close review, a lot of the churn came from removing
  duplicated code sections and consolidating them into callable
  routines. I felt like this made the number of lines of change more
  acceptable, and they address problems, so I left them. The remainder
  of the patches are all small, well contained, and well understood.

  These have passed 0day testing, but have not been submitted to
  linux-next (but a local merge test with your current master was
  without any conflicts).


   - A fix for fix eea40b8f624 ("infiniband: call ipv6 route lookup via
     the stub interface")

   - Six patches against bnxt_re...the first two are considerably larger
     than I would like, but as they address real issues I went ahead and
     submitted them (it also helped that a good deal of the churn was
     removing code repeated in multiple places and consolidating it to
     one common function)

   - Two fixes against qedr that just came in

   - One fix against rxe that took a few revisions to get right plus
     time to get the proper reviews

   - Five late breaking IPoIB fixes

   - One late cxgb4 fix"

* tag 'for-linus' of git://
  rdma/cxgb4: Fix memory leaks during module exit
  IB/ipoib: Fix memory leak in create child syscall
  IB/ipoib: Fix access to un-initialized napi struct
  IB/ipoib: Delete napi in device uninit default
  IB/ipoib: Limit call to free rdma_netdev for capable devices
  IB/ipoib: Fix memory leaks for child interfaces priv
  rxe: Fix a sleep-in-atomic bug in post_one_send
  RDMA/qedr: Add 64KB PAGE_SIZE support to user-space queues
  RDMA/qedr: Initialize byte_len in WC of READ and SEND commands
  RDMA/bnxt_re: Remove FMR support
  RDMA/bnxt_re: Fix RQE posting logic
  RDMA/bnxt_re: Add HW workaround for avoiding stall for UD QPs
  RDMA/bnxt_re: Dereg MR in FW before freeing the fast_reg_page_list
  RDMA/bnxt_re: HW workarounds for handling specific conditions
  RDMA/bnxt_re: Fixing the Control path command and response handling
  IB/addr: Fix setting source address in addr6_resolve()

5 years agoMerge tag 'platform-drivers-x86-v4.12-2' of git://
Linus Torvalds [Fri, 16 Jun 2017 08:30:44 +0000 (17:30 +0900)]
Merge tag 'platform-drivers-x86-v4.12-2' of git://

Pull x86 platform driver fix from Darren Hart:
 "Just a single patch to fix an oops in the intel_telemetry_debugfs
  module load/unload"

* tag 'platform-drivers-x86-v4.12-2' of git://
  platform/x86: intel_telemetry_debugfs: fix oops when load/unload module

5 years agoMerge branch 'for-linus' of git://
Linus Torvalds [Fri, 16 Jun 2017 08:26:10 +0000 (17:26 +0900)]
Merge branch 'for-linus' of git://

Pull block layer fix from Jens Axboe:
 "Just a single fix this week, fixing a regression introduced in this

  When we put the final reference to the queue, we may need to block.
  Ensure that we can safely do so. From Bart"

* 'for-linus' of git://
  block: Fix a blk_exit_rl() regression

5 years agoMerge branch 'dmi-for-linus' of git://
Linus Torvalds [Fri, 16 Jun 2017 08:13:06 +0000 (17:13 +0900)]
Merge branch 'dmi-for-linus' of git://git./linux/kernel/git/jdelvare/staging

Pull dmi fixes from Jean Delvare.

* 'dmi-for-linus' of git://
  firmware: dmi_scan: Check DMI structure length
  firmware: dmi: Fix permissions of product_family
  firmware: dmi_scan: Make dmi_walk and dmi_walk_early return real error codes
  firmware: dmi_scan: Look for SMBIOS 3 entry point first

5 years agoMerge branch 'for-linus' of git://
Linus Torvalds [Fri, 16 Jun 2017 08:00:29 +0000 (17:00 +0900)]
Merge branch 'for-linus' of git://git./linux/kernel/git/jmorris/linux-security

Pull selinux fix from James Morris:
 "Fix for a double free bug in SELinux"

* 'for-linus' of git://
  selinux: fix double free in selinux_parse_opts_str()

5 years agopowerpc/debug: Add missing warn flag to WARN_ON's non-builtin path
Alexey Kardashevskiy [Wed, 14 Jun 2017 03:01:25 +0000 (13:01 +1000)]
powerpc/debug: Add missing warn flag to WARN_ON's non-builtin path

When trapped on WARN_ON(), report_bug() is expected to return
BUG_TRAP_TYPE_WARN so the caller will increment NIP by 4 and continue.
The __builtin_constant_p() path of the PPC's WARN_ON()
calls (indirectly) __WARN_FLAGS() which has BUGFLAG_WARNING set,
however the other branch does not which makes report_bug() report a
bug rather than a warning.

Fixes: f26dee15103f ("debug: Avoid setting BUGFLAG_WARNING twice")
Signed-off-by: Alexey Kardashevskiy <>
Signed-off-by: Michael Ellerman <>
5 years agoMerge tag 'drm-intel-fixes-2017-06-15' of git://
Dave Airlie [Fri, 16 Jun 2017 00:01:52 +0000 (10:01 +1000)]
Merge tag 'drm-intel-fixes-2017-06-15' of git:// into drm-fixes

drm/i915 fixes for v4.12-rc6

* tag 'drm-intel-fixes-2017-06-15' of git://
  drm/i915: Fix GVT-g PVINFO version compatibility check
  drm/i915: Fix SKL+ watermarks for 90/270 rotation
  drm/i915: Fix scaling check for 90/270 degree plane rotation

5 years agoMerge tag 'drm-misc-fixes-2017-06-15' of git://
Dave Airlie [Fri, 16 Jun 2017 00:01:04 +0000 (10:01 +1000)]
Merge tag 'drm-misc-fixes-2017-06-15' of git:// into drm-fixes

Driver Changes:
- dw-hdmi: Fix compilation error if REGMAP_MMIO not selected (Laurent)
- host1x: Fix incorrect return value (Christophe)
- tegra: Shore up idr API usage in tegra staging code (Dmitry)
- mgag200: Always use HiPri mode for G200e4v2 and limit max bandwidth (Mathieu)
- mxsfb: Ensure display can be lit up without bootloader initialization (Fabio)

Cc: Laurent Pinchart <>
Cc: Christophe JAILLET <>
Cc: Dmitry Osipenko <>
Cc: Mathieu Larouche <>
Cc: Fabio Estevam <>
* tag 'drm-misc-fixes-2017-06-15' of git://
  drm: mxsfb_crtc: Reset the eLCDIF controller
  drm/mgag200: Fix to always set HiPri for G200e4 V2
  drm/tegra: Correct idr_alloc() minimum id
  drm/tegra: Fix lockup on a use of staging API
  gpu: host1x: Fix error handling
  drm: dw-hdmi: Fix compilation breakage by selecting REGMAP_MMIO

5 years agoMerge branch 'drm-fixes-4.12' of git:// into drm...
Dave Airlie [Fri, 16 Jun 2017 00:00:11 +0000 (10:00 +1000)]
Merge branch 'drm-fixes-4.12' of git:// into drm-fixes

A few fixes for 4.12:
- fix a UVD regression on SI
- fix overflow in watermark calcs on large modes

* 'drm-fixes-4.12' of git://
  drm/radeon: Fix overflow of watermark calcs at > 4k resolutions.
  drm/amdgpu: Fix overflow of watermark calcs at > 4k resolutions.
  drm/radeon: fix "force the UVD DPB into VRAM as well"

5 years agoUSB: gadgetfs, dummy-hcd, net2280: fix locking for callbacks
Alan Stern [Tue, 13 Jun 2017 19:23:42 +0000 (15:23 -0400)]
USB: gadgetfs, dummy-hcd, net2280: fix locking for callbacks

Using the syzkaller kernel fuzzer, Andrey Konovalov generated the
following error in gadgetfs:

> BUG: KASAN: use-after-free in __lock_acquire+0x3069/0x3690
> kernel/locking/lockdep.c:3246
> Read of size 8 at addr ffff88003a2bdaf8 by task kworker/3:1/903
> CPU: 3 PID: 903 Comm: kworker/3:1 Not tainted 4.12.0-rc4+ #35
> Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS Bochs 01/01/2011
> Workqueue: usb_hub_wq hub_event
> Call Trace:
>  __dump_stack lib/dump_stack.c:16 [inline]
>  dump_stack+0x292/0x395 lib/dump_stack.c:52
>  print_address_description+0x78/0x280 mm/kasan/report.c:252
>  kasan_report_error mm/kasan/report.c:351 [inline]
>  kasan_report+0x230/0x340 mm/kasan/report.c:408
>  __asan_report_load8_noabort+0x19/0x20 mm/kasan/report.c:429
>  __lock_acquire+0x3069/0x3690 kernel/locking/lockdep.c:3246
>  lock_acquire+0x22d/0x560 kernel/locking/lockdep.c:3855
>  __raw_spin_lock include/linux/spinlock_api_smp.h:142 [inline]
>  _raw_spin_lock+0x2f/0x40 kernel/locking/spinlock.c:151
>  spin_lock include/linux/spinlock.h:299 [inline]
>  gadgetfs_suspend+0x89/0x130 drivers/usb/gadget/legacy/inode.c:1682
>  set_link_state+0x88e/0xae0 drivers/usb/gadget/udc/dummy_hcd.c:455
>  dummy_hub_control+0xd7e/0x1fb0 drivers/usb/gadget/udc/dummy_hcd.c:2074
>  rh_call_control drivers/usb/core/hcd.c:689 [inline]
>  rh_urb_enqueue drivers/usb/core/hcd.c:846 [inline]
>  usb_hcd_submit_urb+0x92f/0x20b0 drivers/usb/core/hcd.c:1650
>  usb_submit_urb+0x8b2/0x12c0 drivers/usb/core/urb.c:542
>  usb_start_wait_urb+0x148/0x5b0 drivers/usb/core/message.c:56
>  usb_internal_control_msg drivers/usb/core/message.c:100 [inline]
>  usb_control_msg+0x341/0x4d0 drivers/usb/core/message.c:151
>  usb_clear_port_feature+0x74/0xa0 drivers/usb/core/hub.c:412
>  hub_port_disable+0x123/0x510 drivers/usb/core/hub.c:4177
>  hub_port_init+0x1ed/0x2940 drivers/usb/core/hub.c:4648
>  hub_port_connect drivers/usb/core/hub.c:4826 [inline]
>  hub_port_connect_change drivers/usb/core/hub.c:4999 [inline]
>  port_event drivers/usb/core/hub.c:5105 [inline]
>  hub_event+0x1ae1/0x3d40 drivers/usb/core/hub.c:5185
>  process_one_work+0xc08/0x1bd0 kernel/workqueue.c:2097
>  process_scheduled_works kernel/workqueue.c:2157 [inline]
>  worker_thread+0xb2b/0x1860 kernel/workqueue.c:2233
>  kthread+0x363/0x440 kernel/kthread.c:231
>  ret_from_fork+0x2a/0x40 arch/x86/entry/entry_64.S:424
> Allocated by task 9958:
>  save_stack_trace+0x1b/0x20 arch/x86/kernel/stacktrace.c:59
>  save_stack+0x43/0xd0 mm/kasan/kasan.c:513
>  set_track mm/kasan/kasan.c:525 [inline]
>  kasan_kmalloc+0xad/0xe0 mm/kasan/kasan.c:617
>  kmem_cache_alloc_trace+0x87/0x280 mm/slub.c:2745
>  kmalloc include/linux/slab.h:492 [inline]
>  kzalloc include/linux/slab.h:665 [inline]
>  dev_new drivers/usb/gadget/legacy/inode.c:170 [inline]
>  gadgetfs_fill_super+0x24f/0x540 drivers/usb/gadget/legacy/inode.c:1993
>  mount_single+0xf6/0x160 fs/super.c:1192
>  gadgetfs_mount+0x31/0x40 drivers/usb/gadget/legacy/inode.c:2019
>  mount_fs+0x9c/0x2d0 fs/super.c:1223
>  vfs_kern_mount.part.25+0xcb/0x490 fs/namespace.c:976
>  vfs_kern_mount fs/namespace.c:2509 [inline]
>  do_new_mount fs/namespace.c:2512 [inline]
>  do_mount+0x41b/0x2d90 fs/namespace.c:2834
>  SYSC_mount fs/namespace.c:3050 [inline]
>  SyS_mount+0xb0/0x120 fs/namespace.c:3027
>  entry_SYSCALL_64_fastpath+0x1f/0xbe
> Freed by task 9960:
>  save_stack_trace+0x1b/0x20 arch/x86/kernel/stacktrace.c:59
>  save_stack+0x43/0xd0 mm/kasan/kasan.c:513
>  set_track mm/kasan/kasan.c:525 [inline]
>  kasan_slab_free+0x72/0xc0 mm/kasan/kasan.c:590
>  slab_free_hook mm/slub.c:1357 [inline]
>  slab_free_freelist_hook mm/slub.c:1379 [inline]
>  slab_free mm/slub.c:2961 [inline]
>  kfree+0xed/0x2b0 mm/slub.c:3882
>  put_dev+0x124/0x160 drivers/usb/gadget/legacy/inode.c:163
>  gadgetfs_kill_sb+0x33/0x60 drivers/usb/gadget/legacy/inode.c:2027
>  deactivate_locked_super+0x8d/0xd0 fs/super.c:309
>  deactivate_super+0x21e/0x310 fs/super.c:340
>  cleanup_mnt+0xb7/0x150 fs/namespace.c:1112
>  __cleanup_mnt+0x1b/0x20 fs/namespace.c:1119
>  task_work_run+0x1a0/0x280 kernel/task_work.c:116
>  exit_task_work include/linux/task_work.h:21 [inline]
>  do_exit+0x18a8/0x2820 kernel/exit.c:878
>  do_group_exit+0x14e/0x420 kernel/exit.c:982
>  get_signal+0x784/0x1780 kernel/signal.c:2318
>  do_signal+0xd7/0x2130 arch/x86/kernel/signal.c:808
>  exit_to_usermode_loop+0x1ac/0x240 arch/x86/entry/common.c:157
>  prepare_exit_to_usermode arch/x86/entry/common.c:194 [inline]
>  syscall_return_slowpath+0x3ba/0x410 arch/x86/entry/common.c:263
>  entry_SYSCALL_64_fastpath+0xbc/0xbe
> The buggy address belongs to the object at ffff88003a2bdae0
>  which belongs to the cache kmalloc-1024 of size 1024
> The buggy address is located 24 bytes inside of
>  1024-byte region [ffff88003a2bdae0ffff88003a2bdee0)
> The buggy address belongs to the page:
> page:ffffea0000e8ae00 count:1 mapcount:0 mapping:          (null)
> index:0x0 compound_mapcount: 0
> flags: 0x100000000008100(slab|head)
> raw: 0100000000008100 0000000000000000 0000000000000000 0000000100170017
> raw: ffffea0000ed3020 ffffea0000f5f820 ffff88003e80efc0 0000000000000000
> page dumped because: kasan: bad access detected
> Memory state around the buggy address:
>  ffff88003a2bd980: fb fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc
>  ffff88003a2bda00: fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc fc
> >ffff88003a2bda80: fc fc fc fc fc fc fc fc fc fc fc fc fb fb fb fb
>                                                                 ^
>  ffff88003a2bdb00: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb
>  ffff88003a2bdb80: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb
> ==================================================================

What this means is that the gadgetfs_suspend() routine was trying to
access dev->lock after it had been deallocated.  The root cause is a
race in the dummy_hcd driver; the dummy_udc_stop() routine can race
with the rest of the driver because it contains no locking.  And even
when proper locking is added, it can still race with the
set_link_state() function because that function incorrectly drops the
private spinlock before invoking any gadget driver callbacks.

The result of this race, as seen above, is that set_link_state() can
invoke a callback in gadgetfs even after gadgetfs has been unbound
from dummy_hcd's UDC and its private data structures have been

include/linux/usb/gadget.h documents that the ->reset, ->disconnect,
->suspend, and ->resume callbacks may be invoked in interrupt context.
In general this is necessary, to prevent races with gadget driver
removal.  This patch fixes dummy_hcd to retain the spinlock across
these calls, and it adds a spinlock acquisition to dummy_udc_stop() to
prevent the race.

The net2280 driver makes the same mistake of dropping the private
spinlock for its ->disconnect and ->reset callback invocations.  The
patch fixes it too.

Lastly, since gadgetfs_suspend() may be invoked in interrupt context,
it cannot assume that interrupts are enabled when it runs.  It must
use spin_lock_irqsave() instead of spin_lock_irq().  The patch fixes
that bug as well.

Signed-off-by: Alan Stern <>
Reported-and-tested-by: Andrey Konovalov <>
CC: <>
Acked-by: Felipe Balbi <>
Signed-off-by: Greg Kroah-Hartman <>
5 years agodrm: mxsfb_crtc: Reset the eLCDIF controller
Fabio Estevam [Fri, 5 May 2017 18:01:41 +0000 (15:01 -0300)]
drm: mxsfb_crtc: Reset the eLCDIF controller

According to the eLCDIF initialization steps listed in the MX6SX
Reference Manual the eLCDIF block reset is mandatory.

Without performing the eLCDIF reset the display shows garbage content
when the kernel boots.

In earlier tests this issue has not been observed because the bootloader
was previously showing a splash screen and the bootloader display driver
does properly implement the eLCDIF reset.

Add the eLCDIF reset to the driver, so that it can operate correctly
independently of the bootloader.

Tested on a imx6sx-sdb board.

Cc: <>
Signed-off-by: Fabio Estevam <>
Signed-off-by: Sean Paul <>
5 years agodrm/mgag200: Fix to always set HiPri for G200e4 V2
Mathieu Larouche [Wed, 14 Jun 2017 14:39:42 +0000 (10:39 -0400)]
drm/mgag200: Fix to always set HiPri for G200e4 V2

  - Changed the HiPri value for G200e4 to always be 0.
  - Added Bandwith limitation to block resolution above 1920x1200x60Hz

Signed-off-by: Mathieu Larouche <>
Acked-by: Dave Airlie <>
[seanpaul removed some trailing whitespace from the patch]
Signed-off-by: Sean Paul <>
5 years agoi2c: ismt: fix wrong device address when unmap the data buffer
Liwei Song [Tue, 13 Jun 2017 04:59:53 +0000 (00:59 -0400)]
i2c: ismt: fix wrong device address when unmap the data buffer

Fix the following kernel bug:

kernel BUG at drivers/iommu/intel-iommu.c:3260!
invalid opcode: 0000 [#5] PREEMPT SMP
Hardware name: Intel Corp. Harcuvar/Server, BIOS HAVLCRB0.X64.0013.D39.1608311820 08/31/2016
task: ffff880175389950 ti: ffff880176bec000 task.ti: ffff880176bec000
RIP: 0010:[<ffffffff8150a83b>]  [<ffffffff8150a83b>] intel_unmap+0x25b/0x260
RSP: 0018:ffff880176bef5e8  EFLAGS: 00010296
RAX: 0000000000000024 RBX: ffff8800773c7c88 RCX: 000000000000ce04
RDX: 0000000080000000 RSI: 0000000000000000 RDI: 0000000000000009
RBP: ffff880176bef638 R08: 0000000000000010 R09: 0000000000000004
R10: ffff880175389c78 R11: 0000000000000a4f R12: ffff8800773c7868
R13: 00000000ffffac88 R14: ffff8800773c7818 R15: 0000000000000001
FS:  00007fef21258700(0000) GS:ffff88017b5c0000(0000) knlGS:0000000000000000
CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 000000000066d6d8 CR3: 000000007118c000 CR4: 00000000003406e0
 00000000ffffac88 ffffffff8199867f ffff880176bef5f8 ffff880100000030
 ffff880176bef668 ffff8800773c7c88 ffff880178288098 ffff8800772c0010
 ffff8800773c7818 0000000000000001 ffff880176bef648 ffffffff8150a86e
Call Trace:
 [<ffffffff8199867f>] ? printk+0x46/0x48
 [<ffffffff8150a86e>] intel_unmap_page+0xe/0x10
 [<ffffffffa039d99b>] ismt_access+0x27b/0x8fa [i2c_ismt]
 [<ffffffff81554420>] ? __pm_runtime_suspend+0xa0/0xa0
 [<ffffffff815544a0>] ? pm_suspend_timer_fn+0x80/0x80
 [<ffffffff81554420>] ? __pm_runtime_suspend+0xa0/0xa0
 [<ffffffff815544a0>] ? pm_suspend_timer_fn+0x80/0x80
 [<ffffffff8143dfd0>] ? pci_bus_read_dev_vendor_id+0xf0/0xf0
 [<ffffffff8172b36c>] i2c_smbus_xfer+0xec/0x4b0
 [<ffffffff810aa4d5>] ? vprintk_emit+0x345/0x530
 [<ffffffffa038936b>] i2cdev_ioctl_smbus+0x12b/0x240 [i2c_dev]
 [<ffffffff810aa829>] ? vprintk_default+0x29/0x40
 [<ffffffffa0389b33>] i2cdev_ioctl+0x63/0x1ec [i2c_dev]
 [<ffffffff811b04c8>] do_vfs_ioctl+0x328/0x5d0
 [<ffffffff8119d8ec>] ? vfs_write+0x11c/0x190
 [<ffffffff8109d449>] ? rt_up_read+0x19/0x20
 [<ffffffff811b07f1>] SyS_ioctl+0x81/0xa0
 [<ffffffff819a351b>] system_call_fastpath+0x16/0x6e

This happen When run "i2cdetect -y 0" detect SMBus iSMT adapter.

After finished I2C block read/write, when unmap the data buffer,
a wrong device address was pass to dma_unmap_single().

To fix this, give dma_unmap_single() the "dev" parameter, just like
what dma_map_single() does, then unmap can find the right devices.

Fixes: 13f35ac14cd0 ("i2c: Adding support for Intel iSMT SMBus 2.0 host controller")
Signed-off-by: Liwei Song <>
Reviewed-by: Andy Shevchenko <>
Signed-off-by: Wolfram Sang <>
5 years agoi2c: rcar: use correct length when unmapping DMA
Wolfram Sang [Sun, 28 May 2017 07:52:17 +0000 (09:52 +0200)]
i2c: rcar: use correct length when unmapping DMA

Because we need to transfer some bytes with PIO, the msg length is not
the length of the DMA buffer. Use the correct value which we used when
doing the mapping.

Fixes: 73e8b0528346e8 ("i2c: rcar: add DMA support")
Signed-off-by: Wolfram Sang <>
Reviewed-by: Geert Uytterhoeven <>
Reviewed-by: Niklas Söderlund <>
Signed-off-by: Wolfram Sang <>
5 years agopowerpc/xive: Fix offset for store EOI MMIOs
Benjamin Herrenschmidt [Wed, 14 Jun 2017 00:19:25 +0000 (10:19 +1000)]
powerpc/xive: Fix offset for store EOI MMIOs

Architecturally we should apply a 0x400 offset for these. Not doing
it will break future HW implementations.

The offset of 0 is supposed to remain for "triggers" though not all
sources support both trigger and store EOI, and in P9 specifically,
some sources will treat 0 as a store EOI. But future chips will not.
So this makes us use the properly architected offset which should work

Fixes: 243e25112d06 ("powerpc/xive: Native exploitation of the XIVE interrupt controller")
Signed-off-by: Benjamin Herrenschmidt <>
Signed-off-by: Michael Ellerman <>
5 years agodrm/tegra: Correct idr_alloc() minimum id
Dmitry Osipenko [Wed, 14 Jun 2017 23:18:25 +0000 (02:18 +0300)]
drm/tegra: Correct idr_alloc() minimum id

The client ID 0 is reserved by the host1x/cdma to mark the timeout timer
work as already been scheduled and context ID is used as the clients one.
This fixes spurious CDMA timeouts.

Fixes: bdd2f9cd10eb ("drm/tegra: Don't leak kernel pointer to userspace")
Signed-off-by: Dmitry Osipenko <>
Reviewed-by: Mikko Perttunen <>
Signed-off-by: Thierry Reding <>
5 years agodrm/tegra: Fix lockup on a use of staging API
Dmitry Osipenko [Wed, 14 Jun 2017 23:18:24 +0000 (02:18 +0300)]
drm/tegra: Fix lockup on a use of staging API

Commit bdd2f9cd10eb ("Don't leak kernel pointer to userspace") added a
mutex around staging IOCTL's, some of those mutexes are taken twice.

Fixes: bdd2f9cd10eb ("drm/tegra: Don't leak kernel pointer to userspace")
Signed-off-by: Dmitry Osipenko <>
Reviewed-by: Mikko Perttunen <>
Reviewed-by: Erik Faye-Lund <>
Signed-off-by: Thierry Reding <>
5 years agogpu: host1x: Fix error handling
Christophe JAILLET [Mon, 10 Apr 2017 20:29:22 +0000 (22:29 +0200)]
gpu: host1x: Fix error handling

If 'devm_reset_control_get' returns an error, then we erroneously return
success because error code is taken from 'host->clk' instead of

Fixes: b386c6b73ac6 ("gpu: host1x: Support module reset")
Signed-off-by: Christophe JAILLET <>
Reviewed-by: Mikko Perttunen <>
Signed-off-by: Thierry Reding <>
5 years agofirmware: dmi_scan: Check DMI structure length
Jean Delvare [Thu, 15 Jun 2017 11:46:01 +0000 (13:46 +0200)]
firmware: dmi_scan: Check DMI structure length

Before accessing DMI data to record it for later, we should ensure
that the DMI structures are large enough to contain the data in

Signed-off-by: Jean Delvare <>
Reviewed-by: Mika Westerberg <>
Cc: Dmitry Torokhov <>
Cc: Andy Shevchenko <>
Cc: Linus Walleij <>
5 years agofirmware: dmi: Fix permissions of product_family
Jean Delvare [Thu, 15 Jun 2017 11:46:01 +0000 (13:46 +0200)]
firmware: dmi: Fix permissions of product_family

This is not sensitive information like serial numbers, we can allow
all users to read it.

Fix odd alignment while we're here.

Signed-off-by: Jean Delvare <>
Fixes: c61872c9833d ("firmware: dmi: Add DMI_PRODUCT_FAMILY identification string")
Reviewed-by: Andy Shevchenko <>
Reviewed-by: Mika Westerberg <>
Cc: Dmitry Torokhov <>
Cc: Linus Walleij <>
5 years agofirmware: dmi_scan: Make dmi_walk and dmi_walk_early return real error codes
Andy Lutomirski [Thu, 15 Jun 2017 11:46:00 +0000 (13:46 +0200)]
firmware: dmi_scan: Make dmi_walk and dmi_walk_early return real error codes

Currently they return -1 on error, which will confuse callers if
they try to interpret it as a normal negative error code.

Signed-off-by: Andy Lutomirski <>
Signed-off-by: Darren Hart (VMware) <>
Signed-off-by: Jean Delvare <>
5 years agofirmware: dmi_scan: Look for SMBIOS 3 entry point first
Jean Delvare [Thu, 15 Jun 2017 11:46:00 +0000 (13:46 +0200)]
firmware: dmi_scan: Look for SMBIOS 3 entry point first

Since version 3.0.0 of the SMBIOS specification, there can be
multiple entry points in memory, pointing to one or two DMI tables.
If both a 32-bit ("_SM_") entry point and a 64-bit ("_SM3_") entry
point are present, the specification requires that the latter points
to a table which is a super-set of the table pointed to by the
former. Therefore we should give preference to the 64-bit ("_SM3_")
entry point.

However, currently the code is picking the first valid entry point
it finds. Per specification, we should look for a 64-bit ("_SM3_")
entry point first, and if we can't find any, look for a 32-bit
("_SM_" or "_DMI_") entry point. Modify the code to do that.

Signed-off-by: Jean Delvare <>
5 years agofs: don't forget to put old mntns in mntns_install
Andrei Vagin [Fri, 9 Jun 2017 00:32:29 +0000 (17:32 -0700)]
fs: don't forget to put old mntns in mntns_install

Fixes: 4f757f3cbf54 ("make sure that mntns_install() doesn't end up with referral for root")
Cc: Al Viro <>
Signed-off-by: Andrei Vagin <>
Signed-off-by: Al Viro <>
5 years agoHang/soft lockup in d_invalidate with simultaneous calls
Al Viro [Sat, 3 Jun 2017 06:20:09 +0000 (07:20 +0100)]
Hang/soft lockup in d_invalidate with simultaneous calls

It's not hard to trigger a bunch of d_invalidate() on the same
dentry in parallel.  They end up fighting each other - any
dentry picked for removal by one will be skipped by the rest
and we'll go for the next iteration through the entire
subtree, even if everything is being skipped.  Morevoer, we
immediately go back to scanning the subtree.  The only thing
we really need is to dissolve all mounts in the subtree and
as soon as we've nothing left to do, we can just unhash the
dentry and bugger off.

Signed-off-by: Al Viro <>
5 years agoMIPS: .its targets depend on vmlinux
Paul Burton [Fri, 2 Jun 2017 19:02:08 +0000 (12:02 -0700)]
MIPS: .its targets depend on vmlinux

The .its targets require information about the kernel binary, such as
its entry point, which is extracted from the vmlinux ELF. We therefore
require that the ELF is built before the .its files are generated.
Declare this requirement in the Makefile such that make will ensure this
is always the case, otherwise in corner cases we can hit issues as the
.its is generated with an incorrect (either invalid or stale) entry

Signed-off-by: Paul Burton <>
Fixes: cf2a5e0bb4c6 ("MIPS: Support generating Flattened Image Trees (.itb)")
Cc: stable <> # v4.9+
Signed-off-by: Ralf Baechle <>
5 years agoMIPS: Fix bnezc/jialc return address calculation
Paul Burton [Fri, 2 Jun 2017 18:35:01 +0000 (11:35 -0700)]
MIPS: Fix bnezc/jialc return address calculation

The code handling the pop76 opcode (ie. bnezc & jialc instructions) in
__compute_return_epc_for_insn() needs to set the value of $31 in the
jialc case, which is encoded with rs = 0. However its check to
differentiate bnezc (rs != 0) from jialc (rs = 0) was unfortunately
backwards, meaning that if we emulate a bnezc instruction we clobber $31
& if we emulate a jialc instruction it actually behaves like a jic

Fix this by inverting the check of rs to match the way the instructions
are actually encoded.

Signed-off-by: Paul Burton <>
Fixes: 28d6f93d201d ("MIPS: Emulate the new MIPS R6 BNEZC and JIALC instructions")
Cc: stable <> # v4.0+
Signed-off-by: Ralf Baechle <>
5 years agoMerge git://
Linus Torvalds [Thu, 15 Jun 2017 09:09:47 +0000 (18:09 +0900)]
Merge git://git./linux/kernel/git/davem/net

Pull networking fixes from David Miller:

 1) The netlink attribute passed in to dev_set_alias() is not
    necessarily NULL terminated, don't use strlcpy() on it. From
    Alexander Potapenko.

 2) Fix implementation of atomics in arm64 bpf JIT, from Daniel

 3) Correct the release of netdevs and driver private data in certain

 4) Sanitize netlink message length properly in decnet, from Mateusz

 5) Don't leak kernel data in rtnl_fill_vfinfo() netlink blobs. From
    Yuval Mintz.

 6) Hash secret is never initialized in ipv6 ILA translation code, from
    Arnd Bergmann. I guess those clang warnings about unused inline
    functions are useful for something!

 7) Fix endian selection in bpf_endian.h, from Daniel Borkmann.

 8) Sanitize sockaddr length before dereferncing any fields in AF_UNIX
    and CAIF. From Mateusz Jurczyk.

 9) Fix timestamping for GMAC3 chips in stmmac driver, from Mario

10) Do not leak netdev on dev_alloc_name() errors in mac80211, from
    Johannes Berg.

11) Fix locking in sctp_for_each_endpoint(), from Xin Long.

12) Fix wrong memset size on 32-bit in snmp6, from Christian Perle.

13) Fix use after free in ip_mc_clear_src(), from WANG Cong.

14) Fix regressions caused by ICMP rate limiting changes in 4.11, from
    Jesper Dangaard Brouer.

* git:// (91 commits)
  i40e: Fix a sleep-in-atomic bug
  net: don't global ICMP rate limit packets originating from loopback
  net/act_pedit: fix an error code
  net: update undefined ->ndo_change_mtu() comment
  net_sched: move tcf_lock down after gen_replace_estimator()
  caif: Add sockaddr length check before accessing sa_family in connect handler
  qed: fix dump of context data
  qmi_wwan: new Telewell and Sierra device IDs
  net: phy: Fix MDIO_THUNDER dependencies
  netconsole: Remove duplicate "netconsole: " logging prefix
  igmp: acquire pmc lock for ip_mc_clear_src()
  r8152: give the device version
  net: rps: fix uninitialized symbol warning
  mac80211: don't send SMPS action frame in AP mode when not needed
  mac80211/wpa: use constant time memory comparison for MACs
  mac80211: set bss_info data before configuring the channel
  mac80211: remove 5/10 MHz rate code from station MLME
  mac80211: Fix incorrect condition when checking rx timestamp
  mac80211: don't look at the PM bit of BAR frames
  i40e: fix handling of HW ATR eviction

5 years agoMerge branch 'linus' of git://
Linus Torvalds [Thu, 15 Jun 2017 08:54:51 +0000 (17:54 +0900)]
Merge branch 'linus' of git://git./linux/kernel/git/herbert/crypto-2.6

Pull crypto fix from Herbert Xu:
 "This fixes a bug on sparc where we may dereference freed stack memory"

* 'linus' of git://
  crypto: Work around deallocated stack frame reference gcc bug on sparc.

5 years agoMerge tag 'acpi-4.12-rc6' of git://
Linus Torvalds [Thu, 15 Jun 2017 08:51:19 +0000 (17:51 +0900)]
Merge tag 'acpi-4.12-rc6' of git://git./linux/kernel/git/rafael/linux-pm

Pull ACPI fixes from Rafael Wysocki:
 "These revert an ACPICA commit from the 4.11 cycle that causes problems
  to happen on some systems and add a protection against possible kernel
  crashes due to table reference counter imbalance.


   - Revert a 4.11 ACPICA change that made assumptions which are not
     satisfied on some systems and caused the enumeration of resources
     to fail on them (Rafael Wysocki).

   - Add a mechanism to prevent tables from being unmapped prematurely
     due to reference counter overflows (Lv Zheng)"

* tag 'acpi-4.12-rc6' of git://
  ACPICA: Tables: Mechanism to handle late stage acpi_get_table() imbalance
  Revert "ACPICA: Disassembler: Enhance resource descriptor detection"

5 years agoMerge tag 'pm-4.12-rc6' of git://
Linus Torvalds [Thu, 15 Jun 2017 08:47:46 +0000 (17:47 +0900)]
Merge tag 'pm-4.12-rc6' of git://git./linux/kernel/git/rafael/linux-pm

Pull power management fixes from Rafael Wysocki:
 "These revert a recent cpufreq schedutil governor change that turned
  out to be problematic and fix a few minor issues in cpufreq, cpuidle
  and the Exynos devfreq drivers.


   - Revert a recent cpufreq schedutil governor change that caused some
     systems to behave undesirably (Rafael Wysocki).

   - Fix a cpufreq conservative governor issue introduced during the
     3.10 cycle that prevents it from working as expected in some
     situations (Tomasz Wilczyński).

   - Fix an error code path in the generic cpuidle driver for DT-based
     systems (Christophe Jaillet).

   - Fix three minor issues in devfreq drivers for Exynos (Arvind Yadav,
     Krzysztof Kozlowski)"

* tag 'pm-4.12-rc6' of git://
  cpuidle: dt: Add missing 'of_node_put()'
  cpufreq: conservative: Allow down_threshold to take values from 1 to 10
  Revert "cpufreq: schedutil: Reduce frequencies slower"
  PM / devfreq: exynos-ppmu: Staticize event list
  PM / devfreq: exynos-ppmu: Handle return value of clk_prepare_enable
  PM / devfreq: exynos-nocp: Handle return value of clk_prepare_enable

5 years agoMerge branch 'for-4.12/driver-matching-fix' of git://
Linus Torvalds [Thu, 15 Jun 2017 08:44:41 +0000 (17:44 +0900)]
Merge branch 'for-4.12/driver-matching-fix' of git://git./linux/kernel/git/jikos/hid

Pull HID fix from Jiri Kosina:

 - ifdef-based bandaid for a long-standing issue with HID driver
   matching, avoiding regressions in cases where specific driver is not
   enabled in kernel .config, from Jiri Kosina

* 'for-4.12/driver-matching-fix' of git://
  HID: let generic driver yield control iff specific driver has been enabled

5 years agoMerge tag 'media/v4.12-3' of git://
Linus Torvalds [Thu, 15 Jun 2017 08:37:40 +0000 (17:37 +0900)]
Merge tag 'media/v4.12-3' of git://git./linux/kernel/git/mchehab/linux-media

Pull media fixes from Mauro Carvalho Chehab:

 - some build dependency issues at CEC core with randconfigs

 - fix an off by one error at vb2

 - a race fix at cec core

 - driver fixes at tc358743, sir_ir and rainshadow-cec

* tag 'media/v4.12-3' of git://
  [media] media/cec.h: use IS_REACHABLE instead of IS_ENABLED
  [media] cec: race fix: don't return -ENONET in cec_receive()
  [media] sir_ir: infinite loop in interrupt handler
  [media] cec-notifier.h: handle unreachable CONFIG_CEC_CORE
  [media] cec: improve MEDIA_CEC_RC dependencies
  [media] vb2: Fix an off by one error in 'vb2_plane_vaddr'
  [media] rainshadow-cec: Fix missing spin_lock_init()
  [media] tc358743: fix register i2c_rd/wr function fix

5 years agoufs_truncate_blocks(): fix the case when size is in the last direct block
Al Viro [Thu, 15 Jun 2017 07:57:46 +0000 (03:57 -0400)]
ufs_truncate_blocks(): fix the case when size is in the last direct block

The logics when deciding whether we need to do anything with direct blocks
is broken when new size is within the last direct block.  It's better to
find the path to the last byte _not_ to be removed and use that instead
of the path to the beginning of the first block to be freed...

Signed-off-by: Al Viro <>
5 years agoufs: more deadlock prevention on tail unpacking
Al Viro [Thu, 15 Jun 2017 04:42:56 +0000 (00:42 -0400)]
ufs: more deadlock prevention on tail unpacking

->s_lock is not needed for ufs_change_blocknr()

Signed-off-by: Al Viro <>
5 years agoufs: avoid grabbing ->truncate_mutex if possible
Al Viro [Thu, 15 Jun 2017 04:17:30 +0000 (00:17 -0400)]
ufs: avoid grabbing ->truncate_mutex if possible

tail unpacking is done in a wrong place; the deadlocks galore
is best dealt with by doing that in ->write_iter() (and switching
to iomap, while we are at it), but that's rather painful to
backport.  The trouble comes from grabbing pages that cover
the beginning of tail from inside of ufs_new_fragments(); ongoing
pageout of any of those is going to deadlock on ->truncate_mutex
with process that got around to extending the tail holding that
and waiting for page to get unlocked, while ->writepage() on
that page is waiting on ->truncate_mutex.

The thing is, we don't need ->truncate_mutex when the fragment
we are trying to map is within the tail - the damn thing is
allocated (tail can't contain holes).

Let's do a plain lookup and if the fragment is present, we can
just pretend that we'd won the race in almost all cases.  The
only exception is a fragment between the end of tail and the
end of block containing tail.

Protect ->i_lastfrag with ->meta_lock - read_seqlock_excl() is

Signed-off-by: Al Viro <>
5 years agoi40e: Fix a sleep-in-atomic bug
Jia-Ju Bai [Wed, 14 Jun 2017 23:35:31 +0000 (16:35 -0700)]
i40e: Fix a sleep-in-atomic bug

The driver may sleep under a spin lock, and the function call path is:
i40e_ndo_set_vf_port_vlan (acquire the lock by spin_lock_bh)
          mutex_lock --> may sleep

To fixed it, the spin lock is released before "i40e_vsi_remove_pvid", and
the lock is acquired again after this function.

Signed-off-by: Jia-Ju Bai <>
Tested-by: Andrew Bowers <>
Signed-off-by: Jeff Kirsher <>
Signed-off-by: David S. Miller <>
5 years agoufs_get_locked_page(): make sure we have buffer_heads
Al Viro [Thu, 15 Jun 2017 03:32:19 +0000 (23:32 -0400)]
ufs_get_locked_page(): make sure we have buffer_heads

callers rely upon that, but find_lock_page() racing with attempt of
page eviction by memory pressure might have left us with
* try_to_free_buffers() successfully done
* __remove_mapping() failed, leaving the page in our mapping
* find_lock_page() returning an uptodate page with no
buffer_heads attached.

Signed-off-by: Al Viro <>
5 years agoMerge branch 'acpica-fixes'
Rafael J. Wysocki [Wed, 14 Jun 2017 23:52:32 +0000 (01:52 +0200)]
Merge branch 'acpica-fixes'

* acpica-fixes:
  ACPICA: Tables: Mechanism to handle late stage acpi_get_table() imbalance
  Revert "ACPICA: Disassembler: Enhance resource descriptor detection"

5 years agoMerge branches 'pm-cpufreq', 'pm-cpuidle' and 'pm-devfreq'
Rafael J. Wysocki [Wed, 14 Jun 2017 23:51:33 +0000 (01:51 +0200)]
Merge branches 'pm-cpufreq', 'pm-cpuidle' and 'pm-devfreq'

* pm-cpufreq:
  cpufreq: conservative: Allow down_threshold to take values from 1 to 10
  Revert "cpufreq: schedutil: Reduce frequencies slower"

* pm-cpuidle:
  cpuidle: dt: Add missing 'of_node_put()'

* pm-devfreq:
  PM / devfreq: exynos-ppmu: Staticize event list
  PM / devfreq: exynos-ppmu: Handle return value of clk_prepare_enable
  PM / devfreq: exynos-nocp: Handle return value of clk_prepare_enable

5 years agoufs: fix s_size/s_dsize users
Al Viro [Wed, 14 Jun 2017 20:36:29 +0000 (16:36 -0400)]
ufs: fix s_size/s_dsize users

For UFS2 we need 64bit variants; we even store them in uspi, but
use 32bit ones instead.  One wrinkle is in handling of reserved
space - recalculating it every time had been stupid all along, but
now it would become really ugly.  Just calculate it once...

Signed-off-by: Al Viro <>
5 years agoufs: fix reserved blocks check
Al Viro [Wed, 14 Jun 2017 19:41:17 +0000 (15:41 -0400)]
ufs: fix reserved blocks check

a) honour ->s_minfree; don't just go with default (5)
b) don't bother with capability checks until we know we'll need them

Signed-off-by: Al Viro <>
5 years agoufs: make ufs_freespace() return signed
Al Viro [Wed, 14 Jun 2017 19:36:31 +0000 (15:36 -0400)]
ufs: make ufs_freespace() return signed

as it is, checking that its return value is <= 0 is useless and
that's how it's being used.

Signed-off-by: Al Viro <>
5 years agonet: don't global ICMP rate limit packets originating from loopback
Jesper Dangaard Brouer [Wed, 14 Jun 2017 11:27:37 +0000 (13:27 +0200)]
net: don't global ICMP rate limit packets originating from loopback

Florian Weimer seems to have a glibc test-case which requires that
loopback interfaces does not get ICMP ratelimited.  This was broken by
commit c0303efeab73 ("net: reduce cycles spend on ICMP replies that
gets rate limited").

An ICMP response will usually be routed back-out the same incoming
interface.  Thus, take advantage of this and skip global ICMP
ratelimit when the incoming device is loopback.  In the unlikely event
that the outgoing it not loopback, due to strange routing policy
rules, ICMP rate limiting still works via peer ratelimiting via
icmpv4_xrlim_allow().  Thus, we should still comply with RFC1812
(section "Rate Limiting").

This seems to fix the reproducer given by Florian.  While still
avoiding to perform expensive and unneeded outgoing route lookup for
rate limited packets (in the non-loopback case).

Fixes: c0303efeab73 ("net: reduce cycles spend on ICMP replies that gets rate limited")
Reported-by: Florian Weimer <>
Reported-by: "H.J. Lu" <>
Signed-off-by: Jesper Dangaard Brouer <>
Signed-off-by: David S. Miller <>
5 years agoblock: Fix a blk_exit_rl() regression
Bart Van Assche [Wed, 14 Jun 2017 19:27:50 +0000 (13:27 -0600)]
block: Fix a blk_exit_rl() regression

Avoid that the following complaint is reported:

 BUG: sleeping function called from invalid context at kernel/workqueue.c:2790
 in_atomic(): 1, irqs_disabled(): 0, pid: 41, name: rcuop/3
 1 lock held by rcuop/3/41:
  #0:  (rcu_callback){......}, at: [<ffffffff8111f9a2>] rcu_nocb_kthread+0x282/0x500
 Call Trace:

This happens since we invoke callbacks that need to block from the
queue release handler. Fix this by pushing the final release to
a workqueue.

Reported-by: Ross Zwisler <>
Fixes: commit b425e5049258 ("block: Avoid that blk_exit_rl() triggers a use-after-free")
Signed-off-by: Bart Van Assche <>
Tested-by: Ross Zwisler <>
Updated changelog
Signed-off-by: Jens Axboe <>
5 years agordma/cxgb4: Fix memory leaks during module exit
Raju Rangoju [Fri, 9 Jun 2017 16:47:49 +0000 (22:17 +0530)]
rdma/cxgb4: Fix memory leaks during module exit

Fix memory leaks of iw_cxgb4 module in the exit path

Signed-off-by: Raju Rangoju <>
Reviewed-by: Steve Wise <>
Signed-off-by: Doug Ledford <>
5 years agonet/act_pedit: fix an error code
Dan Carpenter [Wed, 14 Jun 2017 10:29:31 +0000 (13:29 +0300)]
net/act_pedit: fix an error code

I'm reviewing static checker warnings where we do ERR_PTR(0), which is
the same as NULL.  I'm pretty sure we intended to return ERR_PTR(-EINVAL)
here.  Sometimes these bugs lead to a NULL dereference but I don't
immediately see that problem here.

Fixes: 71d0ed7079df ("net/act_pedit: Support using offset relative to the conventional network headers")
Signed-off-by: Dan Carpenter <>
Acked-by: Amir Vadai <>
Signed-off-by: David S. Miller <>
5 years agoufs: fix logics in "ufs: make fsck -f happy"
Al Viro [Wed, 14 Jun 2017 19:17:32 +0000 (15:17 -0400)]
ufs: fix logics in "ufs: make fsck -f happy"

Storing stats _only_ at new locations is wrong for UFS1; old
locations should always be kept updated.  The check for "has
been converted to use of new locations" is also wrong - it
should be "->fs_maxbsize is equal to ->fs_bsize".

Signed-off-by: Al Viro <>
5 years agoIB/ipoib: Fix memory leak in create child syscall
Feras Daoud [Wed, 14 Jun 2017 06:59:09 +0000 (09:59 +0300)]
IB/ipoib: Fix memory leak in create child syscall

The flow of creating a new child goes through ipoib_vlan_add
which allocates a new interface and checks the rtnl_lock.

If the lock is taken, restart_syscall will be called to restart
the system call again. In this case we are not releasing the
already allocated interface, causing a leak.

Fixes: 9baa0b036410 ("IB/ipoib: Add rtnl_link_ops support")
Signed-off-by: Feras Daoud <>
Signed-off-by: Leon Romanovsky <>
Signed-off-by: Doug Ledford <>
5 years agoIB/ipoib: Fix access to un-initialized napi struct
Alex Vesker [Wed, 14 Jun 2017 06:59:08 +0000 (09:59 +0300)]
IB/ipoib: Fix access to un-initialized napi struct

There is no need to re-enable napi since we set the initialized
flag before calling ipoib_ib_dev_stop which will disable napi,
disabling napi twice is harmless in case it was already disabled.

One more reason for this fix is that when using IPoIB new device
driver napi is not added to priv, this can lead to kernel panic
when rn_ops ndo_open fails.

[ 289.755840] invalid opcode: 0000 [#1] SMP
[ 289.757111] task: ffff880036964440 ti: ffff880178ee8000 task.ti: ffff880178ee8000
[ 289.757111] RIP: 0010:[<ffffffffa05368d6>] [<ffffffffa05368d6>] napi_enable.part.24+0x4/0x6 [ib_ipoib]
[ 289.757111] RSP: 0018:ffff880178eeb6d8 EFLAGS: 00010246
[ 289.757111] RAX: 0000000000000000 RBX: ffff880177a80010 RCX: 000000007fffffff
[ 289.757111] RDX: ffffffff81d5f118 RSI: 0000000000000000 RDI: ffff880177a80010
[ 289.757111] RBP: ffff880178eeb6d8 R08: 0000000000000082 R09: 0000000000000283
[ 289.757111] R10: 0000000000000000 R11: 0000000000000000 R12: ffff880175a00000
[ 289.757111] R13: ffff880177a80080 R14: 0000000000000000 R15: 0000000000000001
[ 289.757111] FS: 00007fe2ee346880(0000) GS:ffff88017fc00000(0000) knlGS:0000000000000000
[ 289.757111] CS: 0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[ 289.757111] CR2: 00007fffca979020 CR3: 00000001792e4000 CR4: 00000000000006f0
[ 289.757111] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
[ 289.757111] DR3: 0000000000000000 DR6: 00000000ffff0ff0 DR7: 0000000000000400
[ 289.757111] Stack:
[ 289.796027] ffff880178eeb6f0 ffffffffa05251f5 ffff880177a80000 ffff880178eeb718
[ 289.796027] ffffffffa0528505 ffff880175a00000 ffff880177a80000 0000000000000000
[ 289.796027] ffff880178eeb748 ffffffffa051f0ab ffff880175a00000 ffffffffa0537d60
[ 289.796027] Call Trace:
[ 289.796027] [<ffffffffa05251f5>] napi_enable+0x25/0x30 [ib_ipoib]
[ 289.796027] [<ffffffffa0528505>] ipoib_ib_dev_open+0x175/0x190 [ib_ipoib]
[ 289.796027] [<ffffffffa051f0ab>] ipoib_open+0x4b/0x160 [ib_ipoib]
[ 289.796027] [<ffffffff814fe33f>] _dev_open+0xbf/0x130
[ 289.796027] [<ffffffff814fe62d>] __dev_change_flags+0x9d/0x170
[ 289.796027] [<ffffffff814fe729>] dev_change_flags+0x29/0x60
[ 289.796027] [<ffffffff8150caf7>] do_setlink+0x397/0xa40

Fixes: cd565b4b51e5 ('IB/IPoIB: Support acceleration options callbacks')
Signed-off-by: Alex Vesker <>
Signed-off-by: Leon Romanovsky <>
Signed-off-by: Doug Ledford <>
5 years agoIB/ipoib: Delete napi in device uninit default
Alex Vesker [Wed, 14 Jun 2017 06:59:07 +0000 (09:59 +0300)]
IB/ipoib: Delete napi in device uninit default

This patch mekas init_default and uninit_default symmetric
with a call to delete napi. Additionally, the uninit_default
gained delete napi call in case of init_default fails.

Fixes: 515ed4f3aab4 ('IB/IPoIB: Separate control and data related initializations')
Signed-off-by: Alex Vesker <>
Signed-off-by: Leon Romanovsky <>
Signed-off-by: Doug Ledford <>
5 years agoIB/ipoib: Limit call to free rdma_netdev for capable devices
Alex Vesker [Wed, 14 Jun 2017 06:59:06 +0000 (09:59 +0300)]
IB/ipoib: Limit call to free rdma_netdev for capable devices

Limit calls to free_rdma_netdev() for capable devices only.

Fixes: cd565b4b51e5 ('IB/IPoIB: Support acceleration options callbacks')
Signed-off-by: Alex Vesker <>
Signed-off-by: Leon Romanovsky <>
Signed-off-by: Doug Ledford <>
5 years agoIB/ipoib: Fix memory leaks for child interfaces priv
Alex Vesker [Wed, 14 Jun 2017 06:59:05 +0000 (09:59 +0300)]
IB/ipoib: Fix memory leaks for child interfaces priv

There is a need to free priv explicitly and not just to release
the device, child priv is freed explicitly on remove flow and this
patch also includes priv free on error flow in P_key creation
and also in add_port.

Fixes: cd565b4b51e5 ('IB/IPoIB: Support acceleration options callbacks')
Signed-off-by: Alex Vesker <>
Signed-off-by: Leon Romanovsky <>
Signed-off-by: Doug Ledford <>
5 years agonet: update undefined ->ndo_change_mtu() comment
Magnus Damm [Wed, 14 Jun 2017 07:15:24 +0000 (16:15 +0900)]
net: update undefined ->ndo_change_mtu() comment

Update ->ndo_change_mtu() callback comment to remove text
about returning error in case of undefined callback. This
change makes the comment match the existing code behavior.

Signed-off-by: Magnus Damm <>
Signed-off-by: David S. Miller <>
5 years agoperf tools: Fix build with ARCH=x86_64
Jiada Wang [Mon, 10 Apr 2017 03:02:37 +0000 (20:02 -0700)]
perf tools: Fix build with ARCH=x86_64

With commit: 0a943cb10ce78 (tools build: Add HOSTARCH Makefile variable)
when building for ARCH=x86_64, ARCH=x86_64 is passed to perf instead of
ARCH=x86, so the perf build process searchs header files from
tools/arch/x86_64/include, which doesn't exist.

The following build failure is seen:

  In file included from util/event.c:2:0:
    tools/include/uapi/linux/mman.h:4:27: fatal error: uapi/asm/mman.h: No such file or directory
    compilation terminated.

Fix this issue by using SRCARCH instead of ARCH in perf, just like the
main kernel Makefile and tools/objtool's.

Signed-off-by: Jiada Wang <>
Tested-by: Arnaldo Carvalho de Melo <>
Acked-by: Jiri Olsa <>
Cc: Alexander Shishkin <>
Cc: Andi Kleen <>
Cc: Eugeniu Rosca <>
Cc: Jan Stancek <>
Cc: Masami Hiramatsu <>
Cc: Peter Zijlstra <>
Cc: Ravi Bangoria <>
Cc: Rui Teng <>
Cc: Sukadev Bhattiprolu <>
Cc: Wang Nan <>
Fixes: 0a943cb10ce7 ("tools build: Add HOSTARCH Makefile variable")
Signed-off-by: Arnaldo Carvalho de Melo <>
5 years agoperf evsel: Fix probing of precise_ip level for default cycles event
Arnaldo Carvalho de Melo [Fri, 9 Jun 2017 19:54:28 +0000 (16:54 -0300)]
perf evsel: Fix probing of precise_ip level for default cycles event

Since commit 18e7a45af91a ("perf/x86: Reject non sampling events with
precise_ip") returns -EINVAL for sys_perf_event_open() with an attribute
with (attr.precise_ip > 0 && attr.sample_period == 0), just like is done
in the routine used to probe the max precise level when no events were
passed to 'perf record' or 'perf top', i.e.:


The x86 code, in x86_pmu_hw_config(), which is called all the way from
sys_perf_event_open() did, starting with the aforementioned commit:

                /* There's no sense in having PEBS for non sampling events: */
                if (!is_sampling_event(event))
                        return -EINVAL;

Which makes it fail for cycles:ppp, cycles:pp and cycles:p, always using
just the non precise cycles variant.

To make sure that this is the case, I tested it, before this patch,

  # perf probe -L x86_pmu_hw_config
        0  int x86_pmu_hw_config(struct perf_event *event)
        1  {
        2         if (event->attr.precise_ip) {
       17                 if (event->attr.precise_ip > precise)
       18                         return -EOPNOTSUPP;

                          /* There's no sense in having PEBS for non sampling events: */
       21                 if (!is_sampling_event(event))
       22                         return -EINVAL;
  # perf probe x86_pmu_hw_config:22
  Added new events:
    probe:x86_pmu_hw_config (on x86_pmu_hw_config:22)
    probe:x86_pmu_hw_config_1 (on x86_pmu_hw_config:22)

  You can now use it in all perf tools, such as:

        perf record -e probe:x86_pmu_hw_config_1 -aR sleep 1

  # perf trace -e perf_event_open,probe:x86_pmu_hwconfig*/max-stack=16/ perf record usleep 1
     0.000 ( 0.015 ms): perf/4150 perf_event_open(attr_uptr: 0x7ffebc8ba110, cpu: -1, group_fd: -1      ) ...
     0.015 (         ): probe:x86_pmu_hw_config:(ffffffff9c0065e1))
                                       x86_pmu_hw_config ([kernel.kallsyms])
                                       hsw_hw_config ([kernel.kallsyms])
                                       x86_pmu_event_init ([kernel.kallsyms])
                                       perf_try_init_event ([kernel.kallsyms])
                                       perf_event_alloc ([kernel.kallsyms])
                                       SYSC_perf_event_open ([kernel.kallsyms])
                                       sys_perf_event_open ([kernel.kallsyms])
                                       do_syscall_64 ([kernel.kallsyms])
                                       return_from_SYSCALL_64 ([kernel.kallsyms])
                                       syscall (/usr/lib64/
                                       perf_event_attr__set_max_precise_ip (/home/acme/bin/perf)
                                       perf_evsel__new_cycles (/home/acme/bin/perf)
                                       perf_evlist__add_default (/home/acme/bin/perf)
                                       cmd_record (/home/acme/bin/perf)
                                       run_builtin (/home/acme/bin/perf)
                                       handle_internal_command (/home/acme/bin/perf)
     0.000 ( 0.021 ms): perf/4150  ... [continued]: perf_event_open()) = -1 EINVAL Invalid argument
     0.023 ( 0.002 ms): perf/4150 perf_event_open(attr_uptr: 0x7ffebc8ba110, cpu: -1, group_fd: -1      ) ...
     0.025 (         ): probe:x86_pmu_hw_config:(ffffffff9c0065e1))
                                       x86_pmu_hw_config ([kernel.kallsyms])
                                       hsw_hw_config ([kernel.kallsyms])
                                       x86_pmu_event_init ([kernel.kallsyms])
                                       perf_try_init_event ([kernel.kallsyms])
                                       perf_event_alloc ([kernel.kallsyms])
                                       SYSC_perf_event_open ([kernel.kallsyms])
                                       sys_perf_event_open ([kernel.kallsyms])
                                       do_syscall_64 ([kernel.kallsyms])
                                       return_from_SYSCALL_64 ([kernel.kallsyms])
                                       syscall (/usr/lib64/
                                       perf_event_attr__set_max_precise_ip (/home/acme/bin/perf)
                                       perf_evsel__new_cycles (/home/acme/bin/perf)
                                       perf_evlist__add_default (/home/acme/bin/perf)
                                       cmd_record (/home/acme/bin/perf)
                                       run_builtin (/home/acme/bin/perf)
                                       handle_internal_command (/home/acme/bin/perf)
     0.023 ( 0.004 ms): perf/4150  ... [continued]: perf_event_open()) = -1 EINVAL Invalid argument
     0.028 ( 0.002 ms): perf/4150 perf_event_open(attr_uptr: 0x7ffebc8ba110, cpu: -1, group_fd: -1      ) ...
     0.030 (         ): probe:x86_pmu_hw_config:(ffffffff9c0065e1))
                                       x86_pmu_hw_config ([kernel.kallsyms])
                                       hsw_hw_config ([kernel.kallsyms])
                                       x86_pmu_event_init ([kernel.kallsyms])
                                       perf_try_init_event ([kernel.kallsyms])
                                       perf_event_alloc ([kernel.kallsyms])
                                       SYSC_perf_event_open ([kernel.kallsyms])
                                       sys_perf_event_open ([kernel.kallsyms])
                                       do_syscall_64 ([kernel.kallsyms])
                                       return_from_SYSCALL_64 ([kernel.kallsyms])
                                       syscall (/usr/lib64/
                                       perf_event_attr__set_max_precise_ip (/home/acme/bin/perf)
                                       perf_evsel__new_cycles (/home/acme/bin/perf)
                                       perf_evlist__add_default (/home/acme/bin/perf)
                                       cmd_record (/home/acme/bin/perf)
                                       run_builtin (/home/acme/bin/perf)
                                       handle_internal_command (/home/acme/bin/perf)
     0.028 ( 0.004 ms): perf/4150  ... [continued]: perf_event_open()) = -1 EINVAL Invalid argument
    41.018 ( 0.012 ms): perf/4150 perf_event_open(attr_uptr: 0x7ffebc8b5dd0, pid: -1, group_fd: -1, flags: FD_CLOEXEC) = 4
    41.065 ( 0.011 ms): perf/4150 perf_event_open(attr_uptr: 0x3c7db78, pid: -1, group_fd: -1, flags: FD_CLOEXEC) = 4
    41.080 ( 0.006 ms): perf/4150 perf_event_open(attr_uptr: 0x3c7db78, pid: -1, group_fd: -1, flags: FD_CLOEXEC) = 4
    41.103 ( 0.010 ms): perf/4150 perf_event_open(attr_uptr: 0x3c4e748, pid: 4151 (perf), group_fd: -1, flags: FD_CLOEXEC) = 4
    41.115 ( 0.006 ms): perf/4150 perf_event_open(attr_uptr: 0x3c4e748, pid: 4151 (perf), cpu: 1, group_fd: -1, flags: FD_CLOEXEC) = 5
    41.122 ( 0.004 ms): perf/4150 perf_event_open(attr_uptr: 0x3c4e748, pid: 4151 (perf), cpu: 2, group_fd: -1, flags: FD_CLOEXEC) = 6
    41.128 ( 0.008 ms): perf/4150 perf_event_open(attr_uptr: 0x3c4e748, pid: 4151 (perf), cpu: 3, group_fd: -1, flags: FD_CLOEXEC) = 8
  [ perf record: Woken up 1 times to write data ]
  [ perf record: Captured and wrote 0.017 MB (2 samples) ]

I.e. that return -EINVAL in x86_pmu_hw_config() is hit three times.

So fix it by just setting attr.sample_period

Now, after this patch:

  # perf trace --max-stack=2 -e perf_event_open,probe:x86_pmu_hw_config* perf record usleep 1
  [ perf record: Woken up 1 times to write data ]
     0.000 ( 0.017 ms): perf/8469 perf_event_open(attr_uptr: 0x7ffe36c27d10, pid: -1, cpu: 3, group_fd: -1, flags: FD_CLOEXEC) = 4
                                       syscall (/usr/lib64/
                                       perf_event_open_cloexec_flag (/home/acme/bin/perf)
     0.050 ( 0.031 ms): perf/8469 perf_event_open(attr_uptr: 0x24ebb78, pid: -1, group_fd: -1, flags: FD_CLOEXEC) = 4
                                       syscall (/usr/lib64/
                                       perf_evlist__config (/home/acme/bin/perf)
     0.092 ( 0.040 ms): perf/8469 perf_event_open(attr_uptr: 0x24ebb78, pid: -1, group_fd: -1, flags: FD_CLOEXEC) = 4
                                       syscall (/usr/lib64/
                                       perf_evlist__config (/home/acme/bin/perf)
     0.143 ( 0.007 ms): perf/8469 perf_event_open(attr_uptr: 0x24bc748, cpu: -1, group_fd: -1           ) = 4
                                       syscall (/usr/lib64/
                                       perf_event_attr__set_max_precise_ip (/home/acme/bin/perf)
     0.161 ( 0.007 ms): perf/8469 perf_event_open(attr_uptr: 0x24bc748, pid: 8470 (perf), group_fd: -1, flags: FD_CLOEXEC) = 4
                                       syscall (/usr/lib64/
                                       perf_evsel__open (/home/acme/bin/perf)
     0.171 ( 0.005 ms): perf/8469 perf_event_open(attr_uptr: 0x24bc748, pid: 8470 (perf), cpu: 1, group_fd: -1, flags: FD_CLOEXEC) = 5
                                       syscall (/usr/lib64/
                                       perf_evsel__open (/home/acme/bin/perf)
     0.180 ( 0.007 ms): perf/8469 perf_event_open(attr_uptr: 0x24bc748, pid: 8470 (perf), cpu: 2, group_fd: -1, flags: FD_CLOEXEC) = 6
                                       syscall (/usr/lib64/
                                       perf_evsel__open (/home/acme/bin/perf)
     0.190 ( 0.005 ms): perf/8469 perf_event_open(attr_uptr: 0x24bc748, pid: 8470 (perf), cpu: 3, group_fd: -1, flags: FD_CLOEXEC) = 8
                                       syscall (/usr/lib64/
                                       perf_evsel__open (/home/acme/bin/perf)
  [ perf record: Captured and wrote 0.017 MB (7 samples) ]

The probe one called from perf_event_attr__set_max_precise_ip() works
the first time, with attr.precise_ip = 3, wit hthe next ones being the
per cpu ones for the cycles:ppp event.

And here is the text from a report and alternative proposed patch by
Thomas-Mich Richter:


On s390 the counter and sampling facility do not support a precise IP
skid level and sometimes returns EOPNOTSUPP when structure member
precise_ip in struct perf_event_attr is not set to zero.

On s390 commnd 'perf record -- true' fails with error EOPNOTSUPP.  This
happens only when no events are specified on command line.

The functions called are
  --> perf_evlist__add_default
      --> perf_evsel__new_cycles
          --> perf_event_attr__set_max_precise_ip

The last function determines the value of structure member precise_ip by
invoking the perf_event_open() system call and checking the return code.
The first successful open is the value for precise_ip.

However the value is determined without setting member sample_period and
indicates no sampling.

On s390 the counter facility and sampling facility are different.  The
above procedure determines a precise_ip value of 3 using the counter
facility. Later it uses the sampling facility with a value of 3 and
fails with EOPNOTSUPP.


v2: Older compilers (e.g. gcc 4.4.7) don't support referencing members
    of unnamed union members in the container struct initialization, so
    move from:

struct perf_event_attr attr = {
.sample_period = 1,

to right after it as:

struct perf_event_attr attr = {

attr.sample_period = 1;

v3: We need to reset .sample_period to 0 to let the users of
perf_evsel__new_cycles() to properly setup attr.sample_period or
attr.sample_freq. Reported by Ingo Molnar.

Reported-and-Acked-by: Thomas-Mich Richter <>
Acked-by: Hendrik Brueckner <>
Acked-by: Jiri Olsa <>
Cc: Adrian Hunter <>
Cc: Alexander Shishkin <>
Cc: David Ahern <>
Cc: Namhyung Kim <>
Cc: Peter Zijlstra <>
Cc: Wang Nan <>
Fixes: 18e7a45af91a ("perf/x86: Reject non sampling events with precise_ip")
Signed-off-by: Arnaldo Carvalho de Melo <>