2 years agonvme-tcp: fix an endianess miss-annotation
Christoph Hellwig [Fri, 15 Mar 2019 07:41:04 +0000 (08:41 +0100)]
nvme-tcp: fix an endianess miss-annotation

nvme_tcp_end_request just takes the status value and the converts
it to little endian as well as shifting for the phase bit.

Fixes: 43ce38a6d823 ("nvme-tcp: support C2HData with SUCCESS flag")
Signed-off-by: Christoph Hellwig <>
Reviewed-by: Sagi Grimberg <>
2 years agolibata: fix using DMA buffers on stack
raymond pang [Thu, 28 Mar 2019 12:19:25 +0000 (12:19 +0000)]
libata: fix using DMA buffers on stack

When CONFIG_VMAP_STACK=y, __pa() returns incorrect physical address for
a stack virtual address. Stack DMA buffers must be avoided.

Signed-off-by: raymond pang <>
Signed-off-by: Jens Axboe <>
2 years agoio_uring: offload write to async worker in case of -EAGAIN
Roman Penyaev [Mon, 25 Mar 2019 19:09:24 +0000 (20:09 +0100)]
io_uring: offload write to async worker in case of -EAGAIN

In case of direct write -EAGAIN will be returned if page cache was
previously populated.  To avoid immediate completion of a request
with -EAGAIN error write has to be offloaded to the async worker,
like io_read() does.

Signed-off-by: Roman Penyaev <>
Cc: Jens Axboe <>
Signed-off-by: Jens Axboe <>
2 years agosbitmap: order READ/WRITE freed instance and setting clear bit
Ming Lei [Fri, 22 Mar 2019 01:13:51 +0000 (09:13 +0800)]
sbitmap: order READ/WRITE freed instance and setting clear bit

Inside sbitmap_queue_clear(), once the clear bit is set, it will be
visiable to allocation path immediately. Meantime READ/WRITE on old
associated instance(such as request in case of blk-mq) may be
out-of-order with the setting clear bit, so race with re-allocation
may be triggered.

Adds one memory barrier for ordering READ/WRITE of the freed associated
instance with setting clear bit for avoiding race with re-allocation.

The following kernel oops triggerd by block/006 on aarch64 may be fixed:

[  142.330954] Unable to handle kernel NULL pointer dereference at virtual address 0000000000000330
[  142.338794] Mem abort info:
[  142.341554]   ESR = 0x96000005
[  142.344632]   Exception class = DABT (current EL), IL = 32 bits
[  142.350500]   SET = 0, FnV = 0
[  142.353544]   EA = 0, S1PTW = 0
[  142.356678] Data abort info:
[  142.359528]   ISV = 0, ISS = 0x00000005
[  142.363343]   CM = 0, WnR = 0
[  142.366305] user pgtable: 64k pages, 48-bit VAs, pgdp = 000000002a3c51c0
[  142.372983] [0000000000000330] pgd=0000000000000000, pud=0000000000000000
[  142.379777] Internal error: Oops: 96000005 [#1] SMP
[  142.384613] Modules linked in: null_blk ib_isert iscsi_target_mod ib_srpt target_core_mod ib_srp scsi_transport_srp vfat fat rpcrdma sunrpc rdma_ucm ib_iser rdma_cm iw_cm libiscsi ib_umad scsi_transport_iscsi ib_ipoib ib_cm mlx5_ib ib_uverbs ib_core sbsa_gwdt crct10dif_ce ghash_ce ipmi_ssif sha2_ce ipmi_devintf sha256_arm64 sg sha1_ce ipmi_msghandler ip_tables xfs libcrc32c mlx5_core sdhci_acpi mlxfw ahci_platform at803x sdhci libahci_platform qcom_emac mmc_core hdma hdma_mgmt i2c_dev [last unloaded: null_blk]
[  142.429753] CPU: 7 PID: 1983 Comm: fio Not tainted 5.0.0.cki #2
[  142.449458] pstate: 00400005 (nzcv daif +PAN -UAO)
[  142.454239] pc : __blk_mq_free_request+0x4c/0xa8
[  142.458830] lr : blk_mq_free_request+0xec/0x118
[  142.463344] sp : ffff00003360f6a0
[  142.466646] x29: ffff00003360f6a0 x28: ffff000010e70000
[  142.471941] x27: ffff801729a50048 x26: 0000000000010000
[  142.477232] x25: ffff00003360f954 x24: ffff7bdfff021440
[  142.482529] x23: 0000000000000000 x22: 00000000ffffffff
[  142.487830] x21: ffff801729810000 x20: 0000000000000000
[  142.493123] x19: ffff801729a50000 x18: 0000000000000000
[  142.498413] x17: 0000000000000000 x16: 0000000000000001
[  142.503709] x15: 00000000000000ff x14: ffff7fe000000000
[  142.509003] x13: ffff8017dcde09a0 x12: 0000000000000000
[  142.514308] x11: 0000000000000001 x10: 0000000000000008
[  142.519597] x9 : ffff8017dcde09a0 x8 : 0000000000002000
[  142.524889] x7 : ffff8017dcde0a00 x6 : 000000015388f9be
[  142.530187] x5 : 0000000000000001 x4 : 0000000000000000
[  142.535478] x3 : 0000000000000000 x2 : 0000000000000000
[  142.540777] x1 : 0000000000000001 x0 : ffff00001041b194
[  142.546071] Process fio (pid: 1983, stack limit = 0x000000006460a0ea)
[  142.552500] Call trace:
[  142.554926]  __blk_mq_free_request+0x4c/0xa8
[  142.559181]  blk_mq_free_request+0xec/0x118
[  142.563352]  blk_mq_end_request+0xfc/0x120
[  142.567444]  end_cmd+0x3c/0xa8 [null_blk]
[  142.571434]  null_complete_rq+0x20/0x30 [null_blk]
[  142.576194]  blk_mq_complete_request+0x108/0x148
[  142.580797]  null_handle_cmd+0x1d4/0x718 [null_blk]
[  142.585662]  null_queue_rq+0x60/0xa8 [null_blk]
[  142.590171]  blk_mq_try_issue_directly+0x148/0x280
[  142.594949]  blk_mq_try_issue_list_directly+0x9c/0x108
[  142.600064]  blk_mq_sched_insert_requests+0xb0/0xd0
[  142.604926]  blk_mq_flush_plug_list+0x16c/0x2a0
[  142.609441]  blk_flush_plug_list+0xec/0x118
[  142.613608]  blk_finish_plug+0x3c/0x4c
[  142.617348]  blkdev_direct_IO+0x3b4/0x428
[  142.621336]  generic_file_read_iter+0x84/0x180
[  142.625761]  blkdev_read_iter+0x50/0x78
[  142.629579]  aio_read.isra.6+0xf8/0x190
[  142.633409]  __io_submit_one.isra.8+0x148/0x738
[  142.637912]  io_submit_one.isra.9+0x88/0xb8
[  142.642078]  __arm64_sys_io_submit+0xe0/0x238
[  142.646428]  el0_svc_handler+0xa0/0x128
[  142.650238]  el0_svc+0x8/0xc
[  142.653104] Code: b9402a63 f9000a7f 3100047f 540000a0 (f9419a81)
[  142.659202] ---[ end trace 467586bc175eb09d ]---

Fixes: ea86ea2cdced20057da ("sbitmap: ammortize cost of clearing bits")
Reported-and-bisected_and_tested-by: Yi Zhang <>
Cc: Yi Zhang <>
Cc: "" <>
Reviewed-by: Omar Sandoval <>
Signed-off-by: Ming Lei <>
Signed-off-by: Jens Axboe <>
2 years agoblk-mq: fix sbitmap ws_active for shared tags
Jens Axboe [Mon, 25 Mar 2019 18:34:10 +0000 (12:34 -0600)]
blk-mq: fix sbitmap ws_active for shared tags

We now wrap sbitmap waitqueues in an active counter, so we can avoid
iterating wakeups unless we have waiters there. This works as long as
everyone that's manipulating the waitqueues use the proper helpers. For
the tag wait case for shared tags, however, we add ourselves to the
waitqueue without incrementing/decrementing the ->ws_active count. This
means that wakeups can take a long time to happen.

Fix this by manually doing the inc/dec as needed for the wait queue

Reported-by: Michael Leun <>
Tested-by: Michael Leun <>
Reviewed-by: Omar Sandoval <>
Fixes: 5d2ee7122c73 ("sbitmap: optimize wakeup check")
Signed-off-by: Jens Axboe <>
2 years agoio_uring: fix big-endian compat signal mask handling
Arnd Bergmann [Mon, 25 Mar 2019 14:34:53 +0000 (15:34 +0100)]
io_uring: fix big-endian compat signal mask handling

On big-endian architectures, the signal masks are differnet
between 32-bit and 64-bit tasks, so we have to use a different
function for reading them from user space.

io_cqring_wait() initially got this wrong, and always interprets
this as a native structure. This is ok on x86 and most arm64,
but not on s390, ppc64be, mips64be, sparc64 and parisc.

Signed-off-by: Arnd Bergmann <>
Signed-off-by: Jens Axboe <>
2 years agoblk-mq: update comment for blk_mq_hctx_has_pending()
Yufen Yu [Sun, 24 Mar 2019 09:57:08 +0000 (17:57 +0800)]
blk-mq: update comment for blk_mq_hctx_has_pending()

For now, blk_mq_hctx_has_pending() checks any of ctx, hctx->dispatch
or io scheduler have pending work. So, update the comment accordingly.

Signed-off-by: Yufen Yu <>
Signed-off-by: Jens Axboe <>
2 years agoblk-mq: use blk_mq_put_driver_tag() to put tag
Yufen Yu [Sun, 24 Mar 2019 09:57:07 +0000 (17:57 +0800)]
blk-mq: use blk_mq_put_driver_tag() to put tag

Expect arguments, blk_mq_put_driver_tag_hctx() and blk_mq_put_driver_tag()
is same. We can just use argument 'request' to put tag by blk_mq_put_driver_tag().
Then we can remove the unused blk_mq_put_driver_tag_hctx().

Signed-off-by: Yufen Yu <>
Signed-off-by: Jens Axboe <>
2 years agoMerge tag 'io_uring-20190323' of git://
Linus Torvalds [Sat, 23 Mar 2019 17:25:12 +0000 (10:25 -0700)]
Merge tag 'io_uring-20190323' of git://

Pull io_uring fixes and improvements from Jens Axboe:
 "The first five in this series are heavily inspired by the work Al did
  on the aio side to fix the races there.

  The last two re-introduce a feature that was in io_uring before it got
  merged, but which I pulled since we didn't have a good way to have
  BVEC iters that already have a stable reference. These aren't
  necessarily related to block, it's just how io_uring pins fixed

* tag 'io_uring-20190323' of git://
  block: add BIO_NO_PAGE_REF flag
  iov_iter: add ITER_BVEC_FLAG_NO_REF flag
  io_uring: mark me as the maintainer
  io_uring: retry bulk slab allocs as single allocs
  io_uring: fix poll races
  io_uring: fix fget/fput handling
  io_uring: add prepped flag
  io_uring: make io_read/write return an integer
  io_uring: use regular request ref counts

2 years agoMerge tag 'for-linus-20190323' of git://
Linus Torvalds [Sat, 23 Mar 2019 17:14:42 +0000 (10:14 -0700)]
Merge tag 'for-linus-20190323' of git://

Pull block fixes from Jens Axboe:
 "A set of fixes/changes that should go into this series. This contains:

   - Kernel doc / comment updates (Bart, Shenghui)

   - Un-export of core-only used function (Bart)

   - Fix race on loop file access (Dongli)

   - pf/pcd queue cleanup fixes (me)

   - Use appropriate helper for RESTART bit set (Yufen)

   - Use named identifier for classic poll (Yufen)"

* tag 'for-linus-20190323' of git://
  sbitmap: trivial - update comment for sbitmap_deferred_clear_bit
  blkcg: Fix kernel-doc warnings
  blk-iolatency: #include "blk.h"
  block: Unexport blk_mq_add_to_requeue_list()
  block: add BLK_MQ_POLL_CLASSIC for hybrid poll and return EINVAL for unexpected value
  blk-mq: remove unused 'nr_expired' from blk_mq_hw_ctx
  loop: access lo_backing_file only when the loop device is Lo_bound
  blk-mq: use blk_mq_sched_mark_restart_hctx to set RESTART
  paride/pcd: cleanup queues when detection fails
  paride/pf: cleanup queues when detection fails

2 years agoMerge tag 'ceph-for-5.1-rc2' of git://
Linus Torvalds [Sat, 23 Mar 2019 17:04:47 +0000 (10:04 -0700)]
Merge tag 'ceph-for-5.1-rc2' of git://

Pull ceph fixes from Ilya Dryomov:
 "A follow up for the new alloc_size logic and a blacklisting fix,
  marked for stable"

* tag 'ceph-for-5.1-rc2' of git://
  rbd: drop wait_for_latest_osdmap()
  libceph: wait for latest osdmap in ceph_monc_blacklist_add()
  rbd: set io_min, io_opt and discard_granularity to alloc_size

2 years agoMerge tag 'powerpc-5.1-3' of git://
Linus Torvalds [Fri, 22 Mar 2019 21:15:11 +0000 (14:15 -0700)]
Merge tag 'powerpc-5.1-3' of git://git./linux/kernel/git/powerpc/linux

Pull powerpc fixes from Michael Ellerman:
 "One fix for a boot failure on 32-bit, introduced during the merge

  A fix for our handling of CLOCK_MONOTONIC in the 64-bit VDSO. Changing
  the wall clock across the Y2038 boundary could cause CLOCK_MONOTONIC
  to jump forward and backward.

  Our spectre_v2 reporting was a bit confusing due to a bug I
  introduced. On some systems it was reporting that the count cache was
  disabled and also that we were flushing the count cache on context
  switch. Only the former is true, and given that the count cache is
  disabled it doesn't make any sense to flush it. No one reported it, so
  presumably the presence of any mitigation is all people check for.

  Finally a small build fix for zsmalloc on 32-bit.

  Thanks to: Ben Hutchings, Christophe Leroy, Diana Craciun, Guenter
  Roeck, Michael Neuling"

* tag 'powerpc-5.1-3' of git://
  powerpc/security: Fix spectre_v2 reporting
  powerpc/mm: Only define MAX_PHYSMEM_BITS in SPARSEMEM configurations
  powerpc/6xx: fix setup and use of SPRN_SPRG_PGDIR for hash32
  powerpc/vdso64: Fix CLOCK_MONOTONIC inconsistencies across Y2038

2 years agoMerge tag 'iommu-fixes-v5.1-rc2' of git://
Linus Torvalds [Fri, 22 Mar 2019 21:10:27 +0000 (14:10 -0700)]
Merge tag 'iommu-fixes-v5.1-rc2' of git://git./linux/kernel/git/joro/iommu

Pull iommu fixes from Joerg Roedel:

 - AMD IOMMU fix for sg-mapping with sg->offset > PAGE_SIZE

 - Fix for IOVA code to trigger the slow-path less often

 - Two fixes for Intel VT-d to avoid writing to read-only registers and
   to flush the right domain id for the default domains in scalable mode

* tag 'iommu-fixes-v5.1-rc2' of git://
  iommu/vt-d: Save the right domain ID used by hardware
  iommu/vt-d: Check capability before disabling protected memory
  iommu/iova: Fix tracking of recently failed iova address
  iommu/amd: fix sg->dma_address for sg->offset bigger than PAGE_SIZE

2 years agoMerge tag 'sound-5.1-rc2' of git://
Linus Torvalds [Fri, 22 Mar 2019 21:04:38 +0000 (14:04 -0700)]
Merge tag 'sound-5.1-rc2' of git://git./linux/kernel/git/tiwai/sound

Pull sound fixes from Takashi Iwai:
 "The only significant change is the regression fixes for the jack
  detection at resume on HD-audio, while others are all small or trivial
  fixes like the coverage of missing error code or usual HD-audio quirk"

* tag 'sound-5.1-rc2' of git://
  ALSA: hda/realtek: Enable headset MIC of Acer AIO with ALC286
  ALSA: hda - Enforces runtime_resume after S3 and S4 for each codec
  ALSA: hda - Don't trigger jackpoll_work in azx_resume
  ALSA: opl3: fix mismatch between snd_opl3_drum_switch definition and declaration
  ALSA: hda - add Lenovo IdeaCentre B550 to the power_save_blacklist
  ALSA: firewire-motu: use 'version' field of unit directory to identify model
  ALSA: sb8: add a check for request_region
  ALSA: echoaudio: add a check for ioremap_nocache

2 years agoMerge tag 'devprop-5.1-rc2' of git://
Linus Torvalds [Fri, 22 Mar 2019 19:08:52 +0000 (12:08 -0700)]
Merge tag 'devprop-5.1-rc2' of git://git./linux/kernel/git/rafael/linux-pm

Pull device properties framework fix from Rafael Wysocki:
 "Add missing 'static' in two places (YueHaibing)"

* tag 'devprop-5.1-rc2' of git://
  drivers: base: swnode: Make two functions static

2 years agoMerge tag 'acpi-5.1-rc2' of git://
Linus Torvalds [Fri, 22 Mar 2019 19:07:06 +0000 (12:07 -0700)]
Merge tag 'acpi-5.1-rc2' of git://git./linux/kernel/git/rafael/linux-pm

Pull ACPI fix from Rafael Wysocki:
 "Prevent device references acquired by bus_find_device() in
  acpi_dev_present() from being leaked (Andy Shevchenko)"

* tag 'acpi-5.1-rc2' of git://
  ACPI / utils: Drop reference in test for device presence

2 years agoMerge tag 'pm-5.1-rc2' of git://
Linus Torvalds [Fri, 22 Mar 2019 19:03:19 +0000 (12:03 -0700)]
Merge tag 'pm-5.1-rc2' of git://git./linux/kernel/git/rafael/linux-pm

Pull power management fixes from Rafael Wysocki:
 "These rearrange some code in the generic power domains (genpd)
  framework to avoid a potential deadlock and make the turbostat utility
  behave more as expected.


   - Rearrange the generic power domains (genpd) code to avoid a
     potential deadlock possible due to its interactions with the clock
     framework (Jiada Wang)

   - Make turbostat return the exit status of the command run under it
     if that command fails (David Arcari)"

* tag 'pm-5.1-rc2' of git://
  PM / Domains: Avoid a potential deadlock
  tools/power turbostat: return the exit status of a command

2 years agoMerge tag 'mmc-v5.1-rc1' of git://
Linus Torvalds [Fri, 22 Mar 2019 18:26:57 +0000 (11:26 -0700)]
Merge tag 'mmc-v5.1-rc1' of git://git./linux/kernel/git/ulfh/mmc

Pull MMC fixes from Ulf Hansson:
 "A couple of MMC host fixes intended for v5.1:

   - alcor: Fix DMA reads

   - renesas_sdhi: Limit block count to 16-bit for old revisions

   - sdhci-omap: Fixup support for read-only pins

   - mxcmmc: Revert support for highmem pages

   - davinci/pxamci: Fix clang build warnings"

* tag 'mmc-v5.1-rc1' of git://
  mmc: renesas_sdhi: limit block count to 16 bit for old revisions
  mmc: alcor: fix DMA reads
  mmc: sdhci-omap: Set caps2 to indicate no physical write protect pin
  mmc: mxcmmc: "Revert mmc: mxcmmc: handle highmem pages"
  mmc: davinci: remove extraneous __init annotation
  mmc: pxamci: fix enum type confusion

2 years agosbitmap: trivial - update comment for sbitmap_deferred_clear_bit
Shenghui Wang [Sat, 16 Mar 2019 08:24:37 +0000 (16:24 +0800)]
sbitmap: trivial - update comment for sbitmap_deferred_clear_bit

"sbitmap_batch_clear" should be "sbitmap_deferred_clear"

Acked-by: Omar Sandoval <>
Signed-off-by: Shenghui Wang <>
Signed-off-by: Jens Axboe <>
2 years agoiommu/vt-d: Save the right domain ID used by hardware
Lu Baolu [Wed, 20 Mar 2019 01:58:34 +0000 (09:58 +0800)]
iommu/vt-d: Save the right domain ID used by hardware

The driver sets a default domain id (FLPT_DEFAULT_DID) in the
first level only pasid entry, but saves a different domain id
in @sdev->did. The value saved in @sdev->did will be used to
invalidate the translation caches. Hence, the driver might
result in invalidating the caches with a wrong domain id.

Cc: Ashok Raj <>
Cc: Jacob Pan <>
Fixes: 1c4f88b7f1f92 ("iommu/vt-d: Shared virtual address in scalable mode")
Signed-off-by: Liu Yi L <>
Signed-off-by: Lu Baolu <>
Signed-off-by: Joerg Roedel <>
2 years agoiommu/vt-d: Check capability before disabling protected memory
Lu Baolu [Wed, 20 Mar 2019 01:58:33 +0000 (09:58 +0800)]
iommu/vt-d: Check capability before disabling protected memory

The spec states in 10.4.16 that the Protected Memory Enable
Register should be treated as read-only for implementations
not supporting protected memory regions (PLMR and PHMR fields
reported as Clear in the Capability register).

Cc: Jacob Pan <>
Cc: mark gross <>
Suggested-by: Ashok Raj <>
Fixes: f8bab73515ca5 ("intel-iommu: PMEN support")
Signed-off-by: Lu Baolu <>
Signed-off-by: Joerg Roedel <>
2 years agoiommu/iova: Fix tracking of recently failed iova address
Robert Richter [Wed, 20 Mar 2019 18:57:23 +0000 (18:57 +0000)]
iommu/iova: Fix tracking of recently failed iova address

If a 32 bit allocation request is too big to possibly succeed, it
early exits with a failure and then should never update max32_alloc_
size. This patch fixes current code, now the size is only updated if
the slow path failed while walking the tree. Without the fix the
allocation may enter the slow path again even if there was a failure
before of a request with the same or a smaller size.

Cc: <> # 4.20+
Fixes: bee60e94a1e2 ("iommu/iova: Optimise attempts to allocate iova from 32bit address range")
Reviewed-by: Robin Murphy <>
Signed-off-by: Robert Richter <>
Signed-off-by: Joerg Roedel <>
2 years agoMerge branch 'pm-domains' into pm
Rafael J. Wysocki [Fri, 22 Mar 2019 09:12:13 +0000 (10:12 +0100)]
Merge branch 'pm-domains' into pm

* pm-domains:
  PM / Domains: Avoid a potential deadlock

2 years agoMerge tag 'drm-fixes-2019-03-22' of git://
Linus Torvalds [Fri, 22 Mar 2019 03:40:05 +0000 (20:40 -0700)]
Merge tag 'drm-fixes-2019-03-22' of git://

Pull drm fixes from Dave Airlie:
 "i915, amdgpu, vmwgfx, exynos, nouveau and udl fixes.

  Seems to be lots of little minor ones for regressions in rc1, and some
  cleanups. The exynos one is the largest one, and is for a hw
  difference between exynos versions"

* tag 'drm-fixes-2019-03-22' of git://
  drm/nouveau/dmem: empty chunk do not have a buffer object associated with them.
  drm/nouveau/debugfs: Fix check of pm_runtime_get_sync failure
  drm/nouveau/dmem: Fix a NULL vs IS_ERR() check
  drm/nouveau/dmem: remove set but not used variable 'drm'
  drm/exynos/mixer: fix MIXER shadow registry synchronisation code
  drm/vmwgfx: Don't double-free the mode stored in par->set_mode
  drm/vmwgfx: Return 0 when gmrid::get_node runs out of ID's
  drm/amdgpu: fix invalid use of change_bit
  drm/amdgpu: revert "cleanup setting bulk_movable"
  drm/i915: Sanity check mmap length against object size
  drm/i915: Fix off-by-one in reporting hanging process
  drm/i915/bios: assume eDP is present on port A when there is no VBT
  drm/udl: use drm_gem_object_put_unlocked.

2 years agoMerge branch 'drm-fixes-5.1' of git:// into drm...
Dave Airlie [Fri, 22 Mar 2019 02:07:16 +0000 (12:07 +1000)]
Merge branch 'drm-fixes-5.1' of git:// into drm-fixes

- Parially revert a bulk move clean up change to fix a ref count bug
- Fix invalid use of change_bit that caused a crash on PPC64 and ARM64

Signed-off-by: Dave Airlie <>
From: Alex Deucher <>
2 years agoMerge branch 'vmwgfx-fixes-5.1' of git:// into...
Dave Airlie [Fri, 22 Mar 2019 01:53:28 +0000 (11:53 +1000)]
Merge branch 'vmwgfx-fixes-5.1' of git:// into drm-fixes

Two fixes CC'd stable. One fix for a long-standing a bit hard-to-trigger fbdev
modesetting bug and one out-of-bo-id fix.

Signed-off-by: Dave Airlie <>
From: Thomas Hellstrom <>
2 years agoMerge tag 'exynos-drm-fixes-for-5.1-rc2' of git://
Dave Airlie [Fri, 22 Mar 2019 01:52:40 +0000 (11:52 +1000)]
Merge tag 'exynos-drm-fixes-for-5.1-rc2' of git://git./linux/kernel/git/daeinki/drm-exynos into drm-fixes

- Fix page fault issue at Mixer device
  . This patch fixes the page fault issue by correcting sychronization
    method for updating shadow registers for Mixer device.

Signed-off-by: Dave Airlie <>
From: Inki Dae <>
2 years agoMerge tag 'drm-intel-fixes-2019-03-20' of git://
Dave Airlie [Fri, 22 Mar 2019 00:41:51 +0000 (10:41 +1000)]
Merge tag 'drm-intel-fixes-2019-03-20' of git:// into drm-fixes

A protection on our mmap against attempts to map past the end of the object;
plus a fix off-by-one in our hang report and a protection;
and a fix for eDP panels on Gen9 platforms on VBT absence.

Signed-off-by: Dave Airlie <>
From: Rodrigo Vivi <>
2 years agoMerge branch 'linux-5.1' of git:// into drm-fixes
Dave Airlie [Fri, 22 Mar 2019 00:39:27 +0000 (10:39 +1000)]
Merge branch 'linux-5.1' of git:// into drm-fixes

Some minor nouveau dmem and other fixes.

Signed-off-by: Dave Airlie <>
From: Ben Skeggs <>
2 years agodrm/nouveau/dmem: empty chunk do not have a buffer object associated with them.
Jérôme Glisse [Thu, 21 Mar 2019 19:42:18 +0000 (15:42 -0400)]
drm/nouveau/dmem: empty chunk do not have a buffer object associated with them.

Empty chunk do not have a bo associated with them so no need to pin/unpin
on suspend/resume.

This fix suspend/resume on 5.1rc1 when NOUVEAU_SVM is enabled.

Signed-off-by: Jérôme Glisse <>
Reviewed-by: Tobias Klausmann <>
Tested-by: Tobias Klausmann <>
Cc: Ben Skeggs <>
Cc: David Airlie <>
Cc: Daniel Vetter <>
Signed-off-by: Ben Skeggs <>
2 years agodrm/nouveau/debugfs: Fix check of pm_runtime_get_sync failure
YueHaibing [Thu, 28 Feb 2019 12:24:59 +0000 (20:24 +0800)]
drm/nouveau/debugfs: Fix check of pm_runtime_get_sync failure

pm_runtime_get_sync returns negative on failure.

Fixes: eaeb9010bb4b ("drm/nouveau/debugfs: Wake up GPU before doing any reclocking")
Signed-off-by: YueHaibing <>
Signed-off-by: Ben Skeggs <>
2 years agodrm/nouveau/dmem: Fix a NULL vs IS_ERR() check
Dan Carpenter [Fri, 22 Feb 2019 06:34:04 +0000 (09:34 +0300)]
drm/nouveau/dmem: Fix a NULL vs IS_ERR() check

The hmm_devmem_add() function doesn't return NULL, it returns error

Fixes: 5be73b690875 ("drm/nouveau/dmem: device memory helpers for SVM")
Signed-off-by: Dan Carpenter <>
Signed-off-by: Ben Skeggs <>
2 years agodrm/nouveau/dmem: remove set but not used variable 'drm'
YueHaibing [Thu, 21 Feb 2019 03:38:51 +0000 (03:38 +0000)]
drm/nouveau/dmem: remove set but not used variable 'drm'

Fixes gcc '-Wunused-but-set-variable' warning:

drivers/gpu/drm/nouveau/nouveau_dmem.c: In function 'nouveau_dmem_free':
drivers/gpu/drm/nouveau/nouveau_dmem.c:103:22: warning:
 variable 'drm' set but not used [-Wunused-but-set-variable]
  struct nouveau_drm *drm;

Signed-off-by: YueHaibing <>
Signed-off-by: Ben Skeggs <>
2 years agoMerge tag 'arm64-fixes' of git://
Linus Torvalds [Thu, 21 Mar 2019 18:48:06 +0000 (11:48 -0700)]
Merge tag 'arm64-fixes' of git://git./linux/kernel/git/arm64/linux

Pull arm64 fixes from Catalin Marinas:
 "Mostly fixes apart from the kprobe blacklist checking which was
  deferred because of conflicting with a fix merged after I pinned the
  arm64 for-next/core branch (f2b3d8566d81 "arm64: kprobe: Always
  blacklist the KVM world-switch code").


   - Update the kprobe blacklist checking for arm64. This was supposed
     to be queued during the merging window but, due to conflicts, it
     was deferred post -rc1

   - Extend the Fujitsu erratum 010001 workaround to A64FX v1r0

   - Whitelist HiSilicon Taishan v110 CPUs as not susceptible to

   - Export save_stack_trace_regs()

   - Remove obsolete selection of MULTI_IRQ_HANDLER"

* tag 'arm64-fixes' of git://
  arm64: remove obsolete selection of MULTI_IRQ_HANDLER
  arm64: kpti: Whitelist HiSilicon Taishan v110 CPUs
  arm64: Add MIDR encoding for HiSilicon Taishan CPUs
  arm64/stacktrace: Export save_stack_trace_regs()
  arm64: apply workaround on A64FX v1r0
  arm64: kprobes: Use arch_populate_kprobe_blacklist()
  arm64: kprobes: Move exception_text check in blacklist
  arm64: kprobes: Remove unneeded RODATA check
  arm64: kprobes: Move extable address check into arch_prepare_kprobe()

2 years agoMerge tag 'selinux-pr-20190321' of git://
Linus Torvalds [Thu, 21 Mar 2019 18:12:56 +0000 (11:12 -0700)]
Merge tag 'selinux-pr-20190321' of git://git./linux/kernel/git/pcmoore/selinux

Pull selinux fix from Paul Moore:
 "Another small SELinux fix for v5.1"

* tag 'selinux-pr-20190321' of git://
  selinux: fix NULL dereference in policydb_destroy()

2 years agoMerge tag 'fixes_for_v5.1-rc2' of git://
Linus Torvalds [Thu, 21 Mar 2019 17:31:55 +0000 (10:31 -0700)]
Merge tag 'fixes_for_v5.1-rc2' of git://git./linux/kernel/git/jack/linux-fs

Pull udf fixes from Jan Kara:
 "Two udf error handling fixes"

* tag 'fixes_for_v5.1-rc2' of git://
  udf: Propagate errors from udf_truncate_extents()
  udf: Fix crash on IO error during truncate

2 years agoMerge tag 'fsnotify_for_v5.1-rc2' of git://
Linus Torvalds [Thu, 21 Mar 2019 17:24:00 +0000 (10:24 -0700)]
Merge tag 'fsnotify_for_v5.1-rc2' of git://git./linux/kernel/git/jack/linux-fs

Pull fsnotify fixes from Jan Kara:
 "One inotify and one fanotify fix"

* tag 'fsnotify_for_v5.1-rc2' of git://
  fanotify: Allow copying of file handle to userspace
  inotify: Fix fsnotify_mark refcount leak in inotify_update_existing_watch()

2 years agommc: renesas_sdhi: limit block count to 16 bit for old revisions
Wolfram Sang [Tue, 19 Mar 2019 10:12:59 +0000 (11:12 +0100)]
mmc: renesas_sdhi: limit block count to 16 bit for old revisions

R-Car Gen2 has two different SDHI incarnations in the same chip. The
older one does not support the recently introduced 32 bit register
access to the block count register. Make sure we use this feature only
after the first known version.

Thanks to the Renesas Testing team for this bug report!

Fixes: 5603731a15ef ("mmc: tmio: fix access width of Block Count Register")
Reported-by: Yoshihiro Shimoda <>
Signed-off-by: Wolfram Sang <>
Reviewed-by: Simon Horman <>
Tested-by: Phong Hoang <>
Signed-off-by: Ulf Hansson <>
2 years agommc: alcor: fix DMA reads
Daniel Drake [Wed, 20 Mar 2019 06:36:53 +0000 (14:36 +0800)]
mmc: alcor: fix DMA reads

Setting max_blk_count to 1 here was causing the mmc block layer
to always use the MMC_READ_SINGLE_BLOCK command here, which the
driver does not DMA-accelerate.

Drop the max_blk_ settings here. The mmc host defaults suffice,
along with the max_segs and max_seg_size settings, which I have
now documented in more detail.

Now each MMC command reads 4 512-byte blocks, using DMA instead of
PIO. On my SD card, this increases read performance (measured with dd)
from 167kb/sec to 4.6mb/sec.

Signed-off-by: Daniel Drake <>
Reviewed-by: Oleksij Rempel <>
Fixes: c5413ad815a6 ("mmc: add new Alcor Micro Cardreader SD/MMC driver")
Signed-off-by: Ulf Hansson <>
2 years agommc: sdhci-omap: Set caps2 to indicate no physical write protect pin
Kishon Vijay Abraham I [Thu, 21 Mar 2019 06:15:44 +0000 (11:45 +0530)]
mmc: sdhci-omap: Set caps2 to indicate no physical write protect pin

After commit 6d5cd068ee59fba ("mmc: sdhci: use WP GPIO in
sdhci_check_ro()") and commit 39ee32ce486756f ("mmc: sdhci-omap: drop
->get_ro() implementation"), sdhci-omap relied on SDHCI_PRESENT_STATE
to check if the card is read-only, if wp-gpios is not populated
in device tree. However SDHCI_PRESENT_STATE in sdhci-omap does not have
correct read-only state.

sdhci-omap can be used by platforms with both micro SD slot and standard
SD slot with physical write protect pin (using GPIO). Set caps2 to
MMC_CAP2_NO_WRITE_PROTECT based on if wp-gpios property is populated or

This fix is required since existing device-tree node doesn't have
"disable-wp" property and to preserve old-dt compatibility.

Fixes: 6d5cd068ee59fba ("mmc: sdhci: use WP GPIO in sdhci_check_ro()")
Fixes: 39ee32ce486756f ("mmc: sdhci-omap: drop ->get_ro() implementation")
Signed-off-by: Kishon Vijay Abraham I <>
Signed-off-by: Ulf Hansson <>
2 years agopowerpc/security: Fix spectre_v2 reporting
Michael Ellerman [Thu, 21 Mar 2019 04:24:33 +0000 (15:24 +1100)]
powerpc/security: Fix spectre_v2 reporting

When I updated the spectre_v2 reporting to handle software count cache
flush I got the logic wrong when there's no software count cache
enabled at all.

The result is that on systems with the software count cache flush
disabled we print:

  Mitigation: Indirect branch cache disabled, Software count cache flush

Which correctly indicates that the count cache is disabled, but
incorrectly says the software count cache flush is enabled.

The root of the problem is that we are trying to handle all
combinations of options. But we know now that we only expect to see
the software count cache flush enabled if the other options are false.

So split the two cases, which simplifies the logic and fixes the bug.
We were also missing a space before "(hardware accelerated)".

The result is we see one of:

  Mitigation: Indirect branch serialisation (kernel only)
  Mitigation: Indirect branch cache disabled
  Mitigation: Software count cache flush
  Mitigation: Software count cache flush (hardware accelerated)

Fixes: ee13cb249fab ("powerpc/64s: Add support for software count cache flush")
Cc: # v4.19+
Signed-off-by: Michael Ellerman <>
Reviewed-by: Michael Neuling <>
Reviewed-by: Diana Craciun <>
Signed-off-by: Michael Ellerman <>
2 years agommc: mxcmmc: "Revert mmc: mxcmmc: handle highmem pages"
Alexander Shiyan [Sun, 17 Mar 2019 09:58:25 +0000 (12:58 +0300)]
mmc: mxcmmc: "Revert mmc: mxcmmc: handle highmem pages"

This reverts commit b189e7589f6d3411e85c6b7ae6eef158f08f388f.

Unable to handle kernel paging request at virtual address c8358000
pgd = efa405c3
[c8358000] *pgd=00000000
Internal error: Oops: 805 [#1] PREEMPT ARM
CPU: 0 PID: 711 Comm: kworker/0:2 Not tainted 4.20.0+ #30
Hardware name: Freescale i.MX27 (Device Tree Support)
Workqueue: events mxcmci_datawork
PC is at mxcmci_datawork+0xbc/0x2ac
LR is at mxcmci_datawork+0xac/0x2ac
pc : [<c04e33c8>]    lr : [<c04e33b8>]    psr: 60000013
sp : c6c93f08  ip : 24004180  fp : 00000008
r10: c8358000  r9 : c78b3e24  r8 : c6c92000
r7 : 00000000  r6 : c7bb8680  r5 : c7bb86d4  r4 : c78b3de0
r3 : 00002502  r2 : c090b2e0  r1 : 00000880  r0 : 00000000
Flags: nZCv  IRQs on  FIQs on  Mode SVC_32  ISA ARM  Segment user
Control: 0005317f  Table: a68a8000  DAC: 00000055
Process kworker/0:2 (pid: 711, stack limit = 0x389543bc)
Stack: (0xc6c93f08 to 0xc6c94000)
3f00:                   c7bb86d4 00000000 00000000 c6cbfde0 c7bb86d4 c7ee4200
3f20: 00000000 c0907ea8 00000000 c7bb86d8 c0907ea8 c012077c c6cbfde0 c7bb86d4
3f40: c6cbfde0 c6c92000 c6cbfdf4 c09280ba c0907ea8 c090b2e0 c0907ebc c0120c18
3f60: c6cbfde0 00000000 00000000 c6cbb580 c7ba7c40 c7837edc c6cbb598 00000000
3f80: c6cbfde0 c01208f8 00000000 c01254fc c7ba7c40 c0125400 00000000 00000000
3fa0: 00000000 00000000 00000000 c01010d0 00000000 00000000 00000000 00000000
3fc0: 00000000 00000000 00000000 00000000 00000000 00000000 00000000 00000000
3fe0: 00000000 00000000 00000000 00000000 00000013 00000000 00000000 00000000
[<c04e33c8>] (mxcmci_datawork) from [<c012077c>] (process_one_work+0x1f0/0x338)
[<c012077c>] (process_one_work) from [<c0120c18>] (worker_thread+0x320/0x474)
[<c0120c18>] (worker_thread) from [<c01254fc>] (kthread+0xfc/0x118)
[<c01254fc>] (kthread) from [<c01010d0>] (ret_from_fork+0x14/0x24)
Exception stack(0xc6c93fb0 to 0xc6c93ff8)
3fa0:                                     00000000 00000000 00000000 00000000
3fc0: 00000000 00000000 00000000 00000000 00000000 00000000 00000000 00000000
3fe0: 00000000 00000000 00000000 00000000 00000013 00000000
Code: e3500000 1a000059 e5153050 e5933038 (e48a3004)
---[ end trace 54ca629b75f0e737 ]---
note: kworker/0:2[711] exited with preempt_count 1

Signed-off-by: Alexander Shiyan <>
Fixes: b189e7589f6d ("mmc: mxcmmc: handle highmem pages")
Signed-off-by: Ulf Hansson <>
2 years agoALSA: hda/realtek: Enable headset MIC of Acer AIO with ALC286
Jian-Hong Pan [Fri, 15 Mar 2019 09:51:09 +0000 (17:51 +0800)]
ALSA: hda/realtek: Enable headset MIC of Acer AIO with ALC286

Some Acer AIO desktops like Veriton Z6860G, Z4860G and Z4660G cannot
record sound from headset MIC.  This patch adds the
ALC286_FIXUP_ACER_AIO_HEADSET_MIC quirk to fix this issue.

Fixes: 9f8aefed9623 ("ALSA: hda/realtek: Fix mic issue on Acer AIO Veriton Z4660G")
Fixes: b72f936f6b32 ("ALSA: hda/realtek: Fix mic issue on Acer AIO Veriton Z4860G/Z6860G")
Signed-off-by: Jian-Hong Pan <>
Reviewed-by: Kailang Yang <>
Cc: <>
Signed-off-by: Takashi Iwai <>
2 years agodrm/exynos/mixer: fix MIXER shadow registry synchronisation code
Andrzej Hajda [Tue, 19 Mar 2019 13:05:11 +0000 (14:05 +0100)]
drm/exynos/mixer: fix MIXER shadow registry synchronisation code

MIXER on Exynos5 SoCs uses different synchronisation method than Exynos4
to update internal state (shadow registers).
Apparently the driver implements it incorrectly. The rule should be
as follows:
- do not request updating registers until previous request was finished,
- before setting registers synchronisation on VSYNC should be turned off,
  ie. MXR_STATUS_SYNC_ENABLE should be reset,
- after finishing MXR_STATUS_SYNC_ENABLE should be set again.
The patch hopefully implements it correctly.
Below sample kernel log from page fault caused by the bug:

[   25.670038] exynos-sysmmu 14650000.sysmmu: 14450000.mixer: PAGE FAULT occurred at 0x2247b800
[   25.677888] ------------[ cut here ]------------
[   25.682164] kernel BUG at ../drivers/iommu/exynos-iommu.c:450!
[   25.687971] Internal error: Oops - BUG: 0 [#1] PREEMPT SMP ARM
[   25.693778] Modules linked in:
[   25.696816] CPU: 5 PID: 1553 Comm: fb-release_test Not tainted 5.0.0-rc7-01157-g5f86b1566bdd #136
[   25.705646] Hardware name: SAMSUNG EXYNOS (Flattened Device Tree)
[   25.711710] PC is at exynos_sysmmu_irq+0x1c0/0x264
[   25.716470] LR is at lock_is_held_type+0x44/0x64

v2: added missing MXR_CFG_LAYER_UPDATE bit setting in mixer_enable_sync

Reported-by: Marian Mihailescu <>
Signed-off-by: Andrzej Hajda <>
Signed-off-by: Inki Dae <>
2 years agoblkcg: Fix kernel-doc warnings
Bart Van Assche [Wed, 20 Mar 2019 20:18:45 +0000 (13:18 -0700)]
blkcg: Fix kernel-doc warnings

Avoid that the following warnings are reported when building with W=1:

block/blk-cgroup.c:1755: warning: Function parameter or member 'q' not described in 'blkcg_schedule_throttle'
block/blk-cgroup.c:1755: warning: Function parameter or member 'use_memdelay' not described in 'blkcg_schedule_throttle'
block/blk-cgroup.c:1779: warning: Function parameter or member 'blkg' not described in 'blkcg_add_delay'
block/blk-cgroup.c:1779: warning: Function parameter or member 'now' not described in 'blkcg_add_delay'
block/blk-cgroup.c:1779: warning: Function parameter or member 'delta' not described in 'blkcg_add_delay'

Signed-off-by: Bart Van Assche <>
Signed-off-by: Jens Axboe <>
2 years agoblk-iolatency: #include "blk.h"
Bart Van Assche [Wed, 20 Mar 2019 20:15:01 +0000 (13:15 -0700)]
blk-iolatency: #include "blk.h"

This patch avoids that the following warning is reported when building
with W=1:

block/blk-iolatency.c:734:5: warning: no previous prototype for 'blk_iolatency_init' [-Wmissing-prototypes]

Cc: Josef Bacik <>
Fixes: d70675121546 ("block: introduce blk-iolatency io controller") # v4.19
Signed-off-by: Bart Van Assche <>
Signed-off-by: Jens Axboe <>
2 years agoblock: Unexport blk_mq_add_to_requeue_list()
Bart Van Assche [Wed, 20 Mar 2019 20:14:37 +0000 (13:14 -0700)]
block: Unexport blk_mq_add_to_requeue_list()

This function is not used outside the block layer core. Hence unexport it.

Cc: Christoph Hellwig <>
Cc: Ming Lei <>
Signed-off-by: Bart Van Assche <>
Signed-off-by: Jens Axboe <>
2 years agoblock: add BLK_MQ_POLL_CLASSIC for hybrid poll and return EINVAL for unexpected value
Yufen Yu [Mon, 18 Mar 2019 14:44:41 +0000 (22:44 +0800)]
block: add BLK_MQ_POLL_CLASSIC for hybrid poll and return EINVAL for unexpected value

For q->poll_nsec == -1, means doing classic poll, not hybrid poll.
We introduce a new flag BLK_MQ_POLL_CLASSIC to replace -1, which
may make code much easier to read.

Additionally, since val is an int obtained with kstrtoint(), val can be
a negative value other than -1, so return -EINVAL for that case.

Thanks to Damien Le Moal for some good suggestion.

Reviewed-by: Damien Le Moal <>
Signed-off-by: Yufen Yu <>
Signed-off-by: Jens Axboe <>
2 years agoMerge tag 'arc-5.1-rc2' of git://
Linus Torvalds [Wed, 20 Mar 2019 18:01:52 +0000 (11:01 -0700)]
Merge tag 'arc-5.1-rc2' of git://git./linux/kernel/git/vgupta/arc

Pull ARC updates from Vineet Gupta:

 - unaligned access support for HS cores

 - Removed extra memory barrier around spinlock code

 - HSDK platform updates: enable dmac, reset

 - some more boot logging updates

 - misc minor fixes

* tag 'arc-5.1-rc2' of git://
  arch: arc: Kconfig: pedantic formatting
  ARCv2: spinlock: remove the extra smp_mb before lock, after unlock
  ARC: unaligned: relax the check for gcc supporting -mno-unaligned-access
  ARC: boot log: cut down on verbosity
  ARCv2: boot log: refurbish HS core/release identification
  arc: hsdk_defconfig: Enable CONFIG_BLK_DEV_RAM
  ARC: u-boot args: check that magic number is correct
  ARC: perf: bpok condition only exists for ARCompact
  ARCv2: Add explcit unaligned access support (and ability to disable too)
  ARCv2: lib: introduce memcpy optimized for unaligned access
  ARC: [plat-hsdk]: Enable AXI DW DMAC support
  ARC: [plat-hsdk]: Add reset controller handle to manage USB reset
  ARC: DTB: [scripted] fix node name and address spelling

2 years agoarm64: remove obsolete selection of MULTI_IRQ_HANDLER
Matthias Kaehlcke [Wed, 20 Mar 2019 17:20:56 +0000 (10:20 -0700)]
arm64: remove obsolete selection of MULTI_IRQ_HANDLER

The arm64 config selects MULTI_IRQ_HANDLER, which was renamed to
GENERIC_IRQ_MULTI_HANDLER by commit 4c301f9b6a94 ("ARM: Convert
to GENERIC_IRQ_MULTI_HANDLER"). The 'new' option is already
selected, so just remove the obsolete entry.

Signed-off-by: Matthias Kaehlcke <>
Signed-off-by: Catalin Marinas <>
2 years agorbd: drop wait_for_latest_osdmap()
Ilya Dryomov [Wed, 20 Mar 2019 09:58:05 +0000 (10:58 +0100)]
rbd: drop wait_for_latest_osdmap()

Signed-off-by: Ilya Dryomov <>
Reviewed-by: Jason Dillaman <>
2 years agolibceph: wait for latest osdmap in ceph_monc_blacklist_add()
Ilya Dryomov [Wed, 20 Mar 2019 08:46:58 +0000 (09:46 +0100)]
libceph: wait for latest osdmap in ceph_monc_blacklist_add()

Because map updates are distributed lazily, an OSD may not know about
the new blacklist for quite some time after "osd blacklist add" command
is completed.  This makes it possible for a blacklisted but still alive
client to overwrite a post-blacklist update, resulting in data

Waiting for latest osdmap in ceph_monc_blacklist_add() and thus using
the post-blacklist epoch for all post-blacklist requests ensures that
all such requests "wait" for the blacklist to come into force on their
respective OSDs.

Fixes: 6305a3b41515 ("libceph: support for blacklisting clients")
Signed-off-by: Ilya Dryomov <>
Reviewed-by: Jason Dillaman <>
2 years agopowerpc/mm: Only define MAX_PHYSMEM_BITS in SPARSEMEM configurations
Ben Hutchings [Sun, 17 Mar 2019 01:17:56 +0000 (01:17 +0000)]
powerpc/mm: Only define MAX_PHYSMEM_BITS in SPARSEMEM configurations

MAX_PHYSMEM_BITS only needs to be defined if CONFIG_SPARSEMEM is
enabled, and that was the case before commit 4ffe713b7587
("powerpc/mm: Increase the max addressable memory to 2PB").

On 32-bit systems, where CONFIG_SPARSEMEM is not enabled, we now
define it as 46.  That is larger than the real number of physical
address bits, and breaks calculations in zsmalloc:

  mm/zsmalloc.c:130:49: warning: right shift count is negative
  mm/zsmalloc.c:253:21: error: variably modified 'size_class' at file scope
    struct size_class *size_class[ZS_SIZE_CLASSES];

Fixes: 4ffe713b7587 ("powerpc/mm: Increase the max addressable memory to 2PB")
Cc: # v4.20+
Signed-off-by: Ben Hutchings <>
Signed-off-by: Michael Ellerman <>
2 years agodrm/vmwgfx: Don't double-free the mode stored in par->set_mode
Thomas Zimmermann [Mon, 18 Mar 2019 14:47:58 +0000 (15:47 +0100)]
drm/vmwgfx: Don't double-free the mode stored in par->set_mode

When calling vmw_fb_set_par(), the mode stored in par->set_mode gets free'd
twice. The first free is in vmw_fb_kms_detach(), the second is near the
end of vmw_fb_set_par() under the name of 'old_mode'. The mode-setting code
only works correctly if the mode doesn't actually change. Removing
'old_mode' in favor of using par->set_mode directly fixes the problem.

Cc: <>
Fixes: a278724aa23c ("drm/vmwgfx: Implement fbdev on kms v2")
Signed-off-by: Thomas Zimmermann <>
Reviewed-by: Deepak Rawat <>
Signed-off-by: Thomas Hellstrom <>
2 years agodrm/vmwgfx: Return 0 when gmrid::get_node runs out of ID's
Deepak Rawat [Thu, 28 Feb 2019 18:29:54 +0000 (10:29 -0800)]
drm/vmwgfx: Return 0 when gmrid::get_node runs out of ID's

If it's not a system error and get_node implementation accommodate the
buffer object then it should return 0 with memm::mm_node set to NULL.

v2: Test for id != -ENOMEM instead of id == -ENOSPC.

Cc: <>
Fixes: 4eb085e42fde ("drm/vmwgfx: Convert to new IDA API")
Signed-off-by: Deepak Rawat <>
Reviewed-by: Thomas Hellstrom <>
Signed-off-by: Thomas Hellstrom <>
2 years agoPM / Domains: Avoid a potential deadlock
Jiada Wang [Tue, 12 Mar 2019 06:51:28 +0000 (15:51 +0900)]
PM / Domains: Avoid a potential deadlock

Lockdep warns that prepare_lock and genpd->mlock can cause a deadlock
the deadlock scenario is like following:
First thread is probing cs2000
      clk_prepare_lock()                            ----> acquires prepare_lock
                            genpd_runtime_resume()   ----> acquires genpd->mlock

Second thread is attaching any device to the same PM domain
  genpd_lock()                                       ----> acquires genpd->mlock
            clk_prepare_lock()                       ----> acquires prepare_lock

Since currently no PM provider access genpd's critical section
in .attach_dev, and .detach_dev callbacks, so there is no need to protect
these two callbacks with genpd->mlock.
This patch avoids a potential deadlock by moving out .attach_dev and .detach_dev
from genpd->mlock, so that genpd->mlock won't be held when prepare_lock is acquired
in .attach_dev and .detach_dev

Signed-off-by: Jiada Wang <>
Reviewed-by: Ulf Hansson <>
Tested-by: Geert Uytterhoeven <>
Reviewed-by: Geert Uytterhoeven <>
Signed-off-by: Rafael J. Wysocki <>
2 years agoACPI / utils: Drop reference in test for device presence
Andy Shevchenko [Mon, 18 Mar 2019 18:47:09 +0000 (21:47 +0300)]
ACPI / utils: Drop reference in test for device presence

When commit 8661423eea1a ("ACPI / utils: Add new acpi_dev_present
helper") introduced acpi_dev_present(), it missed the fact that
bus_find_device() took a reference on the device found by it and
the callers of acpi_dev_present() don't drop that reference.

Drop the reference on the device in acpi_dev_present().

Fixes: 8661423eea1a ("ACPI / utils: Add new acpi_dev_present helper")
Signed-off-by: Andy Shevchenko <>
Reviewed-by: Hans de Goede <>
Signed-off-by: Rafael J. Wysocki <>
2 years agodrm/amdgpu: fix invalid use of change_bit
Christian König [Mon, 18 Mar 2019 10:09:54 +0000 (11:09 +0100)]
drm/amdgpu: fix invalid use of change_bit

We only need to clear the bit in a 32bit integer.

This fixes a crah on ARM64 and PPC64LE caused by
"drm/amdgpu: update the vm invalidation engine layout V2"

Signed-off-by: Christian König <>
Acked-by: Alex Deucher <>
Signed-off-by: Alex Deucher <>
2 years agodrm/amdgpu: revert "cleanup setting bulk_movable"
Christian König [Wed, 13 Mar 2019 08:36:52 +0000 (09:36 +0100)]
drm/amdgpu: revert "cleanup setting bulk_movable"

This reverts commit 8466cc61da89d33441e0d7a98de1ba98697cd465.

It can trigger a reference counter bug in TTM. Need to investigate further, but
for now revert the offending change.

Signed-off-by: Christian König <>
Reviewed-by: Michel Dänzer <>
Tested-by: Michel Dänzer <>
Signed-off-by: Alex Deucher <>
2 years agoMerge tag 'tag-chrome-platform-fixes-for-v5.1-rc2' of git://
Linus Torvalds [Tue, 19 Mar 2019 18:28:15 +0000 (11:28 -0700)]
Merge tag 'tag-chrome-platform-fixes-for-v5.1-rc2' of git://git./linux/kernel/git/chrome-platform/linux

Pull chrome platform fixes from Benson Leung:
 "Two fixes:

   - Fix locking and close a potential race condition in the new
     wilco_ec driver.

   - Fix a warning in cros_ec_debugfs on systems that do not support
     console logging, such as the Asus C201"

* tag 'tag-chrome-platform-fixes-for-v5.1-rc2' of git://
  platform/chrome: cros_ec_debugfs: cancel/schedule logging work only if supported
  platform/chrome: Fix locking pattern in wilco_ec_mailbox()

2 years agoplatform/chrome: cros_ec_debugfs: cancel/schedule logging work only if supported
Guenter Roeck [Tue, 19 Mar 2019 16:32:36 +0000 (09:32 -0700)]
platform/chrome: cros_ec_debugfs: cancel/schedule logging work only if supported

The following traceback was reported on ASUS C201, which does not support
console logging.

------------[ cut here ]------------
WARNING: CPU: 2 PID: 361 at kernel/workqueue.c:3030 __flush_work+0x38/0x154
Modules linked in: snd_soc_hdmi_codec cros_ec_debugfs cros_ec_sysfs uvcvideo dw_hdmi_cec dw_hdmi_i2s_audio videobuf2_vmalloc cfg80211 gpio_charger rk_crypto rfkill videobuf2_memops videobuf2_v4l2 des_generic videobuf2_common ofpart m25p80 spi_nor tpm_i2c_infineon sbs_battery mtd tpm joydev cros_ec_dev coreboot_table evdev mousedev ip_tables x_tables [last unloaded: brcmutil]
CPU: 2 PID: 361 Comm: systemd-sleep Not tainted 5.1.0-rc1-1-ARCH+ #1
Hardware name: Rockchip (Device Tree)
[<c020e4b0>] (unwind_backtrace) from [<c020ac18>] (show_stack+0x10/0x14)
[<c020ac18>] (show_stack) from [<c07a3e04>] (dump_stack+0x7c/0x9c)
[<c07a3e04>] (dump_stack) from [<c0222748>] (__warn+0xd0/0xec)
[<c0222748>] (__warn) from [<c022279c>] (warn_slowpath_null+0x38/0x44)
[<c022279c>] (warn_slowpath_null) from [<c02365d0>] (__flush_work+0x38/0x154)
[<c02365d0>] (__flush_work) from [<c023786c>] (__cancel_work_timer+0x114/0x1a4)
[<c023786c>] (__cancel_work_timer) from [<bf33233c>] (cros_ec_debugfs_suspend+0x14/0x1c [cros_ec_debugfs])
[<bf33233c>] (cros_ec_debugfs_suspend [cros_ec_debugfs]) from [<c056a888>] (dpm_run_callback+0x64/0xcc)
[<c056a888>] (dpm_run_callback) from [<c056ad2c>] (__device_suspend+0x174/0x3a8)
[<c056ad2c>] (__device_suspend) from [<c056b9e0>] (dpm_suspend+0x174/0x1e0)
[<c056b9e0>] (dpm_suspend) from [<c026b3e0>] (suspend_devices_and_enter+0x6c/0x50c)
[<c026b3e0>] (suspend_devices_and_enter) from [<c026ba8c>] (pm_suspend+0x20c/0x274)
[<c026ba8c>] (pm_suspend) from [<c026a628>] (state_store+0x54/0x88)
[<c026a628>] (state_store) from [<c03cd2d0>] (kernfs_fop_write+0x114/0x180)
[<c03cd2d0>] (kernfs_fop_write) from [<c035d48c>] (__vfs_write+0x1c/0x154)
[<c035d48c>] (__vfs_write) from [<c035f9e8>] (vfs_write+0xb8/0x198)
[<c035f9e8>] (vfs_write) from [<c035fbc0>] (ksys_write+0x3c/0x74)
[<c035fbc0>] (ksys_write) from [<c0201000>] (ret_fast_syscall+0x0/0x4c)
Exception stack(0xe9365fa8 to 0xe9365ff0)
5fa0: 00000004 beef8b28 00000004 beef8b28 00000004 00000000
5fc0: 00000004 beef8b28 02319170 00000004 beef8b28 00000004 b6f3d900 beef8b74
5fe0: 0000006c beef8a98 b6c0adac b6c66534
---[ end trace f4ee5df14e8ea0ec ]---

If console logging is not supported, the work structure is never
initialized, resulting in the traceback. Calling cancel/schedule functions
conditionally fixes the problem.

While at it, also fix error handling in the probe function.

Reported-by: Urja Rannikko <>
Cc: Urja Rannikko <>
Fixes: 6fce0a2cf5a05 ("mfd / platform: cros_ec: Move debugfs attributes to its own driver")
Signed-off-by: Guenter Roeck <>
Signed-off-by: Benson Leung <>
2 years agoMerge tag 'mips_fixes_5.1_1' of git://
Linus Torvalds [Tue, 19 Mar 2019 17:50:15 +0000 (10:50 -0700)]
Merge tag 'mips_fixes_5.1_1' of git://git./linux/kernel/git/mips/linux

Pull MIPS fixes from Paul Burton:
 "A small batch of MIPS fixes for 5.1:

   - An interrupt masking fix for Loongson-based Lemote 2F systems
     (fixing a regression from v3.19)

   - A relocation fix for configurations in which the devicetree is
     stored in an ELF section (fixing a regression from v4.7)

   - Fix jump labels for MIPSr6 kernels where they previously could
     inadvertently place a control transfer instruction in a forbidden
     slot & take unexpected exceptions (fixing MIPSr6 support added in

   - Extend an existing USB power workaround for the Netgear WNDR3400 to
     v2 boards in addition to the v3 ones that already used it

   - Remove the custom MIPS32 definition of __kernel_fsid_t to make it
     consistent with MIPS64 & every other architecture, in particular
     resolving issues for code which tries to print the val field whose
     type previously differed (though had identical memory layout)"

* tag 'mips_fixes_5.1_1' of git://
  MIPS: Remove custom MIPS32 __kernel_fsid_t type
  mips: bcm47xx: Enable USB power on Netgear WNDR3400v2
  MIPS: Fix kernel crash for R6 in jump label branch function
  MIPS: Ensure ELF appended dtb is relocated
  mips: loongson64: lemote-2f: Add IRQF_NO_SUSPEND to "cascade" irqaction.

2 years agotools/power turbostat: return the exit status of a command
David Arcari [Tue, 12 Feb 2019 14:34:39 +0000 (09:34 -0500)]
tools/power turbostat: return the exit status of a command

turbostat failed to return a non-zero exit status even though the
supplied command (turbostat <command>) failed.  Currently when turbostat
forks a command it returns zero instead of the actual exit status of the
command.  Modify the code to return the exit status.

Signed-off-by: David Arcari <>
Acked-by: Len Brown <>
Signed-off-by: Rafael J. Wysocki <>
2 years agodrivers: base: swnode: Make two functions static
YueHaibing [Tue, 19 Mar 2019 15:20:42 +0000 (23:20 +0800)]
drivers: base: swnode: Make two functions static

Fix sparse warning:

drivers/base/swnode.c:475:22: warning: symbol 'software_node_get_parent' was not declared. Should it be static?
drivers/base/swnode.c:484:22: warning: symbol 'software_node_get_next_child' was not declared. Should it be static?

Signed-off-by: YueHaibing <>
Signed-off-by: Rafael J. Wysocki <>
2 years agoblk-mq: remove unused 'nr_expired' from blk_mq_hw_ctx
Dongli Zhang [Tue, 19 Mar 2019 15:05:18 +0000 (23:05 +0800)]
blk-mq: remove unused 'nr_expired' from blk_mq_hw_ctx

There is no usage of 'nr_expired'.

The 'nr_expired' was introduced by commit 1d9bd5161ba3 ("blk-mq: replace
timeout synchronization with a RCU and generation based scheme"). Its usage
was removed since commit 12f5b9314545 ("blk-mq: Remove generation

Signed-off-by: Dongli Zhang <>
Signed-off-by: Jens Axboe <>
2 years agoarm64: kpti: Whitelist HiSilicon Taishan v110 CPUs
Hanjun Guo [Tue, 5 Mar 2019 13:40:58 +0000 (21:40 +0800)]
arm64: kpti: Whitelist HiSilicon Taishan v110 CPUs

HiSilicon Taishan v110 CPUs didn't implement CSV3 field of the
ID_AA64PFR0_EL1 and are not susceptible to Meltdown, so whitelist
the MIDR in kpti_safe_list[] table.

Signed-off-by: Hanjun Guo <>
Reviewed-by: John Garry <>
Reviewed-by: Zhangshaokun <>
Signed-off-by: Catalin Marinas <>
2 years agoarm64: Add MIDR encoding for HiSilicon Taishan CPUs
Hanjun Guo [Tue, 5 Mar 2019 13:40:57 +0000 (21:40 +0800)]
arm64: Add MIDR encoding for HiSilicon Taishan CPUs

Adding the MIDR encodings for HiSilicon Taishan v110 CPUs,
which is used in Kunpeng ARM64 server SoCs. TSV110 is the
abbreviation of Taishan v110.

Signed-off-by: Hanjun Guo <>
Reviewed-by: John Garry <>
Reviewed-by: Zhangshaokun <>
Signed-off-by: Catalin Marinas <>
2 years agoarm64/stacktrace: Export save_stack_trace_regs()
William Cohen [Fri, 1 Mar 2019 20:00:41 +0000 (15:00 -0500)]
arm64/stacktrace: Export save_stack_trace_regs()

The ARM64 implements the save_stack_trace_regs function, but it is
unusable for any diagnostic tooling compiled as a kernel module due
the missing EXPORT_SYMBOL_GPL for the function.  Export
save_stack_trace_regs() to align with other architectures such as
s390, openrisc, and powerpc. This is similar to the ARM64 export of
save_stack_trace_tsk() added in git commit e27c7fa015d6.

Signed-off-by: William Cohen <>
Signed-off-by: Catalin Marinas <>
2 years agoarm64: apply workaround on A64FX v1r0
Mark Rutland [Fri, 15 Mar 2019 12:22:36 +0000 (12:22 +0000)]
arm64: apply workaround on A64FX v1r0

Fujitsu erratum 010001 applies to A64FX v0r0 and v1r0, and we try to
handle either by masking MIDR with MIDR_FUJITSU_ERRATUM_010001_MASK
before comparing it to MIDR_FUJITSU_ERRATUM_010001.

Unfortunately, MIDR_FUJITSU_ERRATUM_010001 is constructed incorrectly
using MIDR_VARIANT(), which is intended to extract the variant field
from MIDR_EL1, rather than generate the field in-place. This results in
MIDR_FUJITSU_ERRATUM_010001 being all-ones, and we only match A64FX

This patch uses MIDR_CPU_VAR_REV() to generate an in-place mask for the
variant field, ensuring the we match both v0r0 and v1r0.

Fixes: 3e32131abc311a5c ("arm64: Add workaround for Fujitsu A64FX erratum 010001")
Reported-by: "Okamoto, Takayuki" <>
Signed-off-by: Mark Rutland <>
[ fixed the patch author]
Signed-off-by: Catalin Marinas <>
2 years agoarm64: kprobes: Use arch_populate_kprobe_blacklist()
Masami Hiramatsu [Tue, 12 Feb 2019 15:44:48 +0000 (00:44 +0900)]
arm64: kprobes: Use arch_populate_kprobe_blacklist()

Use arch_populate_kprobe_blacklist() instead of
arch_within_kprobe_blacklist() so that we can see the full
blacklisted symbols under the debugfs.

Acked-by: Will Deacon <>
Signed-off-by: Masami Hiramatsu <>
[ Add arch_populate_kprobe_blacklist() comment]
Signed-off-by: Catalin Marinas <>
2 years agoarm64: kprobes: Move exception_text check in blacklist
Masami Hiramatsu [Tue, 12 Feb 2019 15:44:19 +0000 (00:44 +0900)]
arm64: kprobes: Move exception_text check in blacklist

Move exception/irqentry text address check in blacklist,
since those are symbol based rejection.

If we prohibit probing on the symbols in exception_text,
those should be blacklisted.

Acked-by: Will Deacon <>
Signed-off-by: Masami Hiramatsu <>
Signed-off-by: Catalin Marinas <>
2 years agoarm64: kprobes: Remove unneeded RODATA check
Masami Hiramatsu [Tue, 12 Feb 2019 15:43:51 +0000 (00:43 +0900)]
arm64: kprobes: Remove unneeded RODATA check

Remove unneeded RODATA check from arch_prepare_kprobe().

Since check_kprobe_address_safe() already ensured that
the probe address is in kernel text, we don't need to
check whether the address in RODATA or not. That must
be always false.

Acked-by: Will Deacon <>
Signed-off-by: Masami Hiramatsu <>
Signed-off-by: Catalin Marinas <>
2 years agoarm64: kprobes: Move extable address check into arch_prepare_kprobe()
Masami Hiramatsu [Tue, 12 Feb 2019 15:43:22 +0000 (00:43 +0900)]
arm64: kprobes: Move extable address check into arch_prepare_kprobe()

Move extable address check into arch_prepare_kprobe() from
The blacklist is exposed via debugfs as a list of symbols.
The extable entries are smaller, so must be filtered out
by arch_prepare_kprobe().

Acked-by: Will Deacon <>
Reviewed-by: James Morse <>
Signed-off-by: Masami Hiramatsu <>
Signed-off-by: Catalin Marinas <>
2 years agofanotify: Allow copying of file handle to userspace
Jan Kara [Tue, 12 Mar 2019 11:42:37 +0000 (12:42 +0100)]
fanotify: Allow copying of file handle to userspace

When file handle is embedded inside fanotify_event and usercopy checks
are enabled, we get a warning like:

Bad or missing usercopy whitelist? Kernel memory exposure attempt detected
from SLAB object 'fanotify_event' (offset 40, size 8)!
WARNING: CPU: 1 PID: 7649 at mm/usercopy.c:78 usercopy_warn+0xeb/0x110

Annotate handling in fanotify_event properly to mark copying it to
userspace is fine.

Fixes: a8b13aa20afb ("fanotify: enable FAN_REPORT_FID init flag")
Signed-off-by: Kees Cook <>
Reviewed-by: Amir Goldstein <>
Signed-off-by: Jan Kara <>
2 years agoALSA: hda - Enforces runtime_resume after S3 and S4 for each codec
Hui Wang [Tue, 19 Mar 2019 01:28:44 +0000 (09:28 +0800)]
ALSA: hda - Enforces runtime_resume after S3 and S4 for each codec

Recently we found the audio jack detection stop working after suspend
on many machines with Realtek codec. Sometimes the audio selection
dialogue didn't show up after users plugged headhphone/headset into
the headset jack, sometimes after uses plugged headphone/headset, then
click the sound icon on the upper-right corner of gnome-desktop, it
also showed the speaker rather than the headphone.

The root cause is that before suspend, the codec already call the
runtime_suspend since this codec is not used by any apps, then in
resume, it will not call runtime_resume for this codec. But for some
realtek codec (so far, alc236, alc255 and alc891) with the specific
BIOS, if it doesn't run runtime_resume after suspend, all codec
functions including jack detection stop working anymore.

This problem existed for a long time, but it was not exposed, that is
because when problem happens, if users play sound or open
sound-setting to check audio device, this will trigger calling to
runtime_resume (via snd_hda_power_up), then the codec starts working
again before users notice this problem.

Since we don't know how many codec and BIOS combinations have this
problem, to fix it, let the driver call runtime_resume for all codecs
in pm_resume, maybe for some codecs, this is not needed, but it is
harmless. After a codec is runtime resumed, if it is not used by any
apps, it will be runtime suspended soon and furthermore we don't run
suspend frequently, this change will not add much power consumption.

Fixes: cc72da7d4d06 ("ALSA: hda - Use standard runtime PM for codec power-save control")
Signed-off-by: Hui Wang <>
Signed-off-by: Takashi Iwai <>
2 years agoALSA: hda - Don't trigger jackpoll_work in azx_resume
Hui Wang [Tue, 19 Mar 2019 01:28:43 +0000 (09:28 +0800)]
ALSA: hda - Don't trigger jackpoll_work in azx_resume

The commit 3baffc4a84d7 (ALSA: hda/intel: Refactoring PM code) changed
the behaviour of azx_resume(), it triggers the jackpoll_work after
applying this commit.

This change introduced a new issue, all codecs are runtime active
after S3, and will not call runtime_suspend() automatically.

The root cause is the jackpoll_work calls snd_hda_power_up/down_pm,
and it calls up_pm before snd_hdac_enter_pm is called, while calls
the down_pm in the middle of enter_pm and leave_pm is called. This
makes the dev->power.usage_count unbalanced after S3.

To fix it, let azx_resume() don't trigger jackpoll_work as before
it did.

Fixes: 3baffc4a84d7 ("ALSA: hda/intel: Refactoring PM code")
Signed-off-by: Hui Wang <>
Signed-off-by: Takashi Iwai <>
2 years agodrm/i915: Sanity check mmap length against object size
Chris Wilson [Thu, 14 Mar 2019 07:58:29 +0000 (07:58 +0000)]
drm/i915: Sanity check mmap length against object size

We assumed that vm_mmap() would reject an attempt to mmap past the end of
the filp (our object), but we were wrong.

Applications that tried to use the mmap beyond the end of the object
would be greeted by a SIGBUS. After this patch, those applications will
be told about the error on creating the mmap, rather than at a random
moment on later access.

Reported-by: Antonio Argenziano <>
Testcase: igt/gem_mmap/bad-size
Signed-off-by: Chris Wilson <>
Cc: Antonio Argenziano <>
Cc: Joonas Lahtinen <>
Cc: Tvrtko Ursulin <>
Reviewed-by: Tvrtko Ursulin <>
Reviewed-by: Joonas Lahtinen <>
(cherry picked from commit 794a11cb67201ad1bb61af510bb8460280feb3f3)
Signed-off-by: Rodrigo Vivi <>
2 years agodrm/i915: Fix off-by-one in reporting hanging process
Chris Wilson [Fri, 15 Mar 2019 16:39:33 +0000 (16:39 +0000)]
drm/i915: Fix off-by-one in reporting hanging process

ffs() is 1-indexed, but we want to use it as an index into an array, so
use __ffs() instead.

Fixes: eb8d0f5af4ec ("drm/i915: Remove GPU reset dependence on struct_mutex")
Signed-off-by: Chris Wilson <>
Cc: Mika Kuoppala <>
Reviewed-by: Rodrigo Vivi <>
(cherry picked from commit 9073e5b26743b8b675cc44a9c0c8f8c3d584e1c0)
Signed-off-by: Rodrigo Vivi <>
2 years agodrm/i915/bios: assume eDP is present on port A when there is no VBT
Thomas Preston [Wed, 6 Mar 2019 20:06:18 +0000 (20:06 +0000)]
drm/i915/bios: assume eDP is present on port A when there is no VBT

We rely on VBT DDI port info for eDP detection on GEN9 platforms and
above. This breaks GEN9 platforms which don't have VBT because port A
eDP now defaults to false. Fix this by defaulting to true when VBT is

Fixes: a98d9c1d7e9b ("drm/i915/ddi: Rely on VBT DDI port info for eDP detection")
Signed-off-by: Thomas Preston <>
Signed-off-by: Jani Nikula <>
(cherry picked from commit 2131bc0ced6088648e47f126566c3da58b07e4ef)
Signed-off-by: Rodrigo Vivi <>
2 years agorbd: set io_min, io_opt and discard_granularity to alloc_size
Ilya Dryomov [Fri, 15 Mar 2019 13:50:04 +0000 (14:50 +0100)]
rbd: set io_min, io_opt and discard_granularity to alloc_size

Now that we have alloc_size that controls our discard behavior, it
doesn't make sense to have these set to object (set) size.  alloc_size
defaults to 64k, but because discard_granularity is likely 4M, only
ranges that are equal to or bigger than 4M can be considered during
fstrim.  A smaller io_min is also more likely to be met, resulting in
fewer deferred writes on bluestore OSDs.

Signed-off-by: Ilya Dryomov <>
Reviewed-by: Jason Dillaman <>
2 years agoplatform/chrome: Fix locking pattern in wilco_ec_mailbox()
Nick Crews [Mon, 11 Mar 2019 15:58:38 +0000 (09:58 -0600)]
platform/chrome: Fix locking pattern in wilco_ec_mailbox()

Before, ec->data_buffer could be written to from multiple
contexts at the same time. Since the ec is shared data,
it needs to be inside the mutex as well.

Fixes: 7b3d4f44abf0 ("platform/chrome: Add new driver for Wilco EC")
Signed-off-by: Nick Crews <>
Signed-off-by: Enric Balletbo i Serra <>
Signed-off-by: Benson Leung <>
2 years agoMerge tag 'for-linus' of git://
Linus Torvalds [Mon, 18 Mar 2019 17:19:12 +0000 (10:19 -0700)]
Merge tag 'for-linus' of git://git./linux/kernel/git/rdma/rdma

Pull rdma fixes from Jason Gunthorpe:
 "Several driver bug fixes post in the last three weeks

   - first part of a race condition fix in mlx4 with CATAS errors

   - bad interaction with FW causing resource leaks in the mlx5 DCT flow

   - bad reporting of link speed/width in new mlx5 devices

   - user triggable OOPS in i40iw"

* tag 'for-linus' of git://
  i40iw: Avoid panic when handling the inetdev event
  IB/mlx5: Fix mapping of link-mode to IB width and speed
  IB/mlx5: Use mlx5 core to create/destroy a DEVX DCT
  net/mlx5: Fix DCT creation bad flow
  IB/mlx4: Fix race condition between catas error reset and aliasguid flows

2 years agoblock: add BIO_NO_PAGE_REF flag
Jens Axboe [Wed, 27 Feb 2019 20:13:23 +0000 (13:13 -0700)]
block: add BIO_NO_PAGE_REF flag

If bio_iov_iter_get_pages() is called on an iov_iter that is flagged
with NO_REF, then we don't need to add a page reference for the pages
that we add.

Add BIO_NO_PAGE_REF to track this in the bio, so IO completion knows
not to drop a reference to these pages.

Signed-off-by: Jens Axboe <>
2 years agoiov_iter: add ITER_BVEC_FLAG_NO_REF flag
Jens Axboe [Wed, 27 Feb 2019 20:05:25 +0000 (13:05 -0700)]
iov_iter: add ITER_BVEC_FLAG_NO_REF flag

For ITER_BVEC, if we're holding on to kernel pages, the caller
doesn't need to grab a reference to the bvec pages, and drop that
same reference on IO completion. This is essentially safe for any
ITER_BVEC, but some use cases end up reusing pages and uncondtionally
dropping a page reference on completion. And example of that is
sendfile(2), that ends up being a splice_in + splice_out on the
pipe pages.

Add a flag that tells us it's fine to not grab a page reference
to the bvec pages, since that caller knows not to drop a reference
when it's done with the pages.

Signed-off-by: Jens Axboe <>
2 years agoio_uring: mark me as the maintainer
Jens Axboe [Thu, 14 Mar 2019 21:22:18 +0000 (15:22 -0600)]
io_uring: mark me as the maintainer

And io_uring as maintained in general.

Signed-off-by: Jens Axboe <>
2 years agoio_uring: retry bulk slab allocs as single allocs
Jens Axboe [Thu, 14 Mar 2019 22:30:06 +0000 (16:30 -0600)]
io_uring: retry bulk slab allocs as single allocs

I've seen cases where bulk alloc fails, since the bulk alloc API
is all-or-nothing - either we get the number we ask for, or it
returns 0 as number of entries.

If we fail a batch bulk alloc, retry a "normal" kmem_cache_alloc()
and just use that instead of failing with -EAGAIN.

While in there, ensure we use GFP_KERNEL. That was an oversight in
the original code, when we switched away from GFP_ATOMIC.

Signed-off-by: Jens Axboe <>
2 years agoselinux: fix NULL dereference in policydb_destroy()
Ondrej Mosnacek [Sun, 17 Mar 2019 13:46:53 +0000 (14:46 +0100)]
selinux: fix NULL dereference in policydb_destroy()

The conversion to kvmalloc() forgot to account for the possibility that
p->type_attr_map_array might be null in policydb_destroy().

Fix this by destroying its contents only if it is not NULL.

Also make sure ebitmap_init() is called on all entries before
policydb_destroy() can be called. Right now this is a no-op, because
both kvcalloc() and ebitmap_init() just zero out the whole struct, but
let's rather not rely on a specific implementation.

Fixes: acdf52d97f82 ("selinux: convert to kvmalloc")
Signed-off-by: Ondrej Mosnacek <>
Acked-by: Stephen Smalley <>
Signed-off-by: Paul Moore <>
2 years agoudf: Propagate errors from udf_truncate_extents()
Jan Kara [Mon, 11 Mar 2019 14:27:02 +0000 (15:27 +0100)]
udf: Propagate errors from udf_truncate_extents()

Make udf_truncate_extents() properly propagate errors to its callers and
let udf_setsize() handle the error properly as well. This lets userspace
know in case there's some error when truncating blocks.

Signed-off-by: Jan Kara <>
2 years agoudf: Fix crash on IO error during truncate
Jan Kara [Mon, 11 Mar 2019 14:04:18 +0000 (15:04 +0100)]
udf: Fix crash on IO error during truncate

When truncate(2) hits IO error when reading indirect extent block the
code just bugs with:

kernel BUG at linux-4.15.0/fs/udf/truncate.c:249!

Fix the problem by bailing out cleanly in case of IO error.

Reported-by: jean-luc malet <>
Signed-off-by: Jan Kara <>
2 years agoloop: access lo_backing_file only when the loop device is Lo_bound
Dongli Zhang [Mon, 18 Mar 2019 12:23:17 +0000 (20:23 +0800)]
loop: access lo_backing_file only when the loop device is Lo_bound

Commit 758a58d0bc67 ("loop: set GENHD_FL_NO_PART_SCAN after
blkdev_reread_part()") separates "lo->lo_backing_file = NULL" and
"lo->lo_state = Lo_unbound" into different critical regions protected by

However, there is below race that the NULL lo->lo_backing_file would be
accessed when the backend of a loop is another loop device, e.g., loop0's
backend is a file, while loop1's backend is loop0.

loop0's backend is file            loop1's backend is loop0

  lo->lo_backing_file = NULL; --> set to NULL
                                       f = l->lo_backing_file; --> NULL
                                         access if loop0 is not Lo_unbound
  lo->lo_state = Lo_unbound;

lo->lo_backing_file should be accessed only when the loop device is

In fact, the problem has been introduced already in commit 7ccd0791d985
("loop: Push loop_ctl_mutex down into loop_clr_fd()") after which
loop_validate_file() could see devices in Lo_rundown state with which it
did not count. It was harmless at that point but still.

Fixes: 7ccd0791d985 ("loop: Push loop_ctl_mutex down into loop_clr_fd()")
Signed-off-by: Dongli Zhang <>
Reviewed-by: Jan Kara <>
Signed-off-by: Jens Axboe <>
2 years agoblk-mq: use blk_mq_sched_mark_restart_hctx to set RESTART
Yufen Yu [Fri, 15 Mar 2019 03:05:10 +0000 (11:05 +0800)]
blk-mq: use blk_mq_sched_mark_restart_hctx to set RESTART

Let blk_mq_mark_tag_wait() use the blk_mq_sched_mark_restart_hctx()

Signed-off-by: Yufen Yu <>
Signed-off-by: Jens Axboe <>
2 years agoparide/pcd: cleanup queues when detection fails
Jens Axboe [Mon, 18 Mar 2019 14:10:32 +0000 (08:10 -0600)]
paride/pcd: cleanup queues when detection fails

The driver allocates queues for all the units it potentially
supports. But if we fail to detect any drives, then we fail
loading the module without cleaning up those queues. This is
now evident with the switch to blk-mq, though the bug has
been there forever as far as I can tell.

Also fix cleanup through regular module exit.

Reported-by: Randy Dunlap <>
Tested-by: Randy Dunlap <>
Signed-off-by: Jens Axboe <>
2 years agoparide/pf: cleanup queues when detection fails
Jens Axboe [Mon, 18 Mar 2019 14:08:43 +0000 (08:08 -0600)]
paride/pf: cleanup queues when detection fails

The driver allocates queues for all the units it potentially
supports. But if we fail to detect any drives, then we fail
loading the module without cleaning up those queues. This is
now evident with the switch to blk-mq, though the bug has
been there forever as far as I can tell.

Also fix cleanup through regular module exit.

Reported-by: Randy Dunlap <>
Tested-by: Randy Dunlap <>
Signed-off-by: Jens Axboe <>
2 years agopowerpc/6xx: fix setup and use of SPRN_SPRG_PGDIR for hash32
Christophe Leroy [Mon, 11 Mar 2019 08:30:27 +0000 (08:30 +0000)]
powerpc/6xx: fix setup and use of SPRN_SPRG_PGDIR for hash32

Not only the 603 but all 6xx need SPRN_SPRG_PGDIR to be initialised at
startup. This patch move it from __setup_cpu_603() to start_here()
and __secondary_start(), close to the initialisation of SPRN_THREAD.

Previously, virt addr of PGDIR was retrieved from thread struct.
Now that it is the phys addr which is stored in SPRN_SPRG_PGDIR,
hash_page() shall not convert it to phys anymore.
This patch removes the conversion.

Fixes: 93c4a162b014 ("powerpc/6xx: Store PGDIR physical address in a SPRG")
Reported-by: Guenter Roeck <>
Tested-by: Guenter Roeck <>
Signed-off-by: Christophe Leroy <>
Signed-off-by: Michael Ellerman <>
2 years agoALSA: opl3: fix mismatch between snd_opl3_drum_switch definition and declaration
Colin Ian King [Sun, 17 Mar 2019 23:21:24 +0000 (23:21 +0000)]
ALSA: opl3: fix mismatch between snd_opl3_drum_switch definition and declaration

The function snd_opl3_drum_switch declaration in the header file
has the order of the two arguments on_off and vel swapped when
compared to the definition arguments of vel and on_off.  Fix this
by swapping them around to match the definition.

This error predates the git history, so no idea when this error
was introduced.

Signed-off-by: Colin Ian King <>
Signed-off-by: Takashi Iwai <>
2 years agoALSA: hda - add Lenovo IdeaCentre B550 to the power_save_blacklist
Jaroslav Kysela [Mon, 18 Mar 2019 12:45:43 +0000 (13:45 +0100)]
ALSA: hda - add Lenovo IdeaCentre B550 to the power_save_blacklist

Another machine which does not like the power saving (noise):

Also, reorder the Lenovo C50 entry to keep the table sorted.

Signed-off-by: Jaroslav Kysela <>
Cc: <>
Signed-off-by: Takashi Iwai <>
2 years agoiommu/amd: fix sg->dma_address for sg->offset bigger than PAGE_SIZE
Stanislaw Gruszka [Wed, 13 Mar 2019 09:03:17 +0000 (10:03 +0100)]
iommu/amd: fix sg->dma_address for sg->offset bigger than PAGE_SIZE

Take into account that sg->offset can be bigger than PAGE_SIZE when
setting segment sg->dma_address. Otherwise sg->dma_address will point
at diffrent page, what makes DMA not possible with erros like this:

xhci_hcd 0000:38:00.3: AMD-Vi: Event logged [IO_PAGE_FAULT domain=0x0000 address=0x00000000fdaa70c0 flags=0x0020]
xhci_hcd 0000:38:00.3: AMD-Vi: Event logged [IO_PAGE_FAULT domain=0x0000 address=0x00000000fdaa7040 flags=0x0020]
xhci_hcd 0000:38:00.3: AMD-Vi: Event logged [IO_PAGE_FAULT domain=0x0000 address=0x00000000fdaa7080 flags=0x0020]
xhci_hcd 0000:38:00.3: AMD-Vi: Event logged [IO_PAGE_FAULT domain=0x0000 address=0x00000000fdaa7100 flags=0x0020]
xhci_hcd 0000:38:00.3: AMD-Vi: Event logged [IO_PAGE_FAULT domain=0x0000 address=0x00000000fdaa7000 flags=0x0020]

Additinally with wrong sg->dma_address unmap_sg will free wrong pages,
what what can cause crashes like this:

Feb 28 19:27:45 kernel: BUG: Bad page state in process cinnamon  pfn:39e8b1
Feb 28 19:27:45 kernel: Disabling lock debugging due to kernel taint
Feb 28 19:27:45 kernel: flags: 0x2ffff0000000000()
Feb 28 19:27:45 kernel: raw: 02ffff0000000000 0000000000000000 ffffffff00000301 0000000000000000
Feb 28 19:27:45 kernel: raw: 0000000000000000 0000000000000000 00000001ffffffff 0000000000000000
Feb 28 19:27:45 kernel: page dumped because: nonzero _refcount
Feb 28 19:27:45 kernel: Modules linked in: ccm fuse arc4 nct6775 hwmon_vid amdgpu nls_iso8859_1 nls_cp437 edac_mce_amd vfat fat kvm_amd ccp rng_core kvm mt76x0u mt76x0_common mt76x02_usb irqbypass mt76_usb mt76x02_lib mt76 crct10dif_pclmul crc32_pclmul chash mac80211 amd_iommu_v2 ghash_clmulni_intel gpu_sched i2c_algo_bit ttm wmi_bmof snd_hda_codec_realtek snd_hda_codec_generic drm_kms_helper snd_hda_codec_hdmi snd_hda_intel drm snd_hda_codec aesni_intel snd_hda_core snd_hwdep aes_x86_64 crypto_simd snd_pcm cfg80211 cryptd mousedev snd_timer glue_helper pcspkr r8169 input_leds realtek agpgart libphy rfkill snd syscopyarea sysfillrect sysimgblt fb_sys_fops soundcore sp5100_tco k10temp i2c_piix4 wmi evdev gpio_amdpt pinctrl_amd mac_hid pcc_cpufreq acpi_cpufreq sg ip_tables x_tables ext4(E) crc32c_generic(E) crc16(E) mbcache(E) jbd2(E) fscrypto(E) sd_mod(E) hid_generic(E) usbhid(E) hid(E) dm_mod(E) serio_raw(E) atkbd(E) libps2(E) crc32c_intel(E) ahci(E) libahci(E) libata(E) xhci_pci(E) xhci_hcd(E)
Feb 28 19:27:45 kernel:  scsi_mod(E) i8042(E) serio(E) bcache(E) crc64(E)
Feb 28 19:27:45 kernel: CPU: 2 PID: 896 Comm: cinnamon Tainted: G    B   W   E     4.20.12-arch1-1-custom #1
Feb 28 19:27:45 kernel: Hardware name: To Be Filled By O.E.M. To Be Filled By O.E.M./B450M Pro4, BIOS P1.20 06/26/2018
Feb 28 19:27:45 kernel: Call Trace:
Feb 28 19:27:45 kernel:  dump_stack+0x5c/0x80
Feb 28 19:27:45 kernel:  bad_page.cold.29+0x7f/0xb2
Feb 28 19:27:45 kernel:  __free_pages_ok+0x2c0/0x2d0
Feb 28 19:27:45 kernel:  skb_release_data+0x96/0x180
Feb 28 19:27:45 kernel:  __kfree_skb+0xe/0x20
Feb 28 19:27:45 kernel:  tcp_recvmsg+0x894/0xc60
Feb 28 19:27:45 kernel:  ? reuse_swap_page+0x120/0x340
Feb 28 19:27:45 kernel:  ? ptep_set_access_flags+0x23/0x30
Feb 28 19:27:45 kernel:  inet_recvmsg+0x5b/0x100
Feb 28 19:27:45 kernel:  __sys_recvfrom+0xc3/0x180
Feb 28 19:27:45 kernel:  ? handle_mm_fault+0x10a/0x250
Feb 28 19:27:45 kernel:  ? syscall_trace_enter+0x1d3/0x2d0
Feb 28 19:27:45 kernel:  ? __audit_syscall_exit+0x22a/0x290
Feb 28 19:27:45 kernel:  __x64_sys_recvfrom+0x24/0x30
Feb 28 19:27:45 kernel:  do_syscall_64+0x5b/0x170
Feb 28 19:27:45 kernel:  entry_SYSCALL_64_after_hwframe+0x44/0xa9

Reported-and-tested-by: Jan Viktorin <>
Reviewed-by: Alexander Duyck <>
Signed-off-by: Stanislaw Gruszka <>
Fixes: 80187fd39dcb ('iommu/amd: Optimize map_sg and unmap_sg')
Signed-off-by: Joerg Roedel <>
2 years agommc: davinci: remove extraneous __init annotation
Arnd Bergmann [Thu, 7 Mar 2019 10:10:11 +0000 (11:10 +0100)]
mmc: davinci: remove extraneous __init annotation

Building with clang finds a mistaken __init tag:

WARNING: vmlinux.o(.text+0x5e4250): Section mismatch in reference from the function davinci_mmcsd_probe() to the function .init.text:init_mmcsd_host()
The function davinci_mmcsd_probe() references
the function __init init_mmcsd_host().
This is often because davinci_mmcsd_probe lacks a __init
annotation or the annotation of init_mmcsd_host is wrong.

Signed-off-by: Arnd Bergmann <>
Acked-by: Wolfram Sang <>
Reviewed-by: Nathan Chancellor <>
Signed-off-by: Ulf Hansson <>
2 years agommc: pxamci: fix enum type confusion
Arnd Bergmann [Thu, 7 Mar 2019 10:09:19 +0000 (11:09 +0100)]
mmc: pxamci: fix enum type confusion

clang points out several instances of mismatched types in this drivers,
all coming from a single declaration:

drivers/mmc/host/pxamci.c:193:15: error: implicit conversion from enumeration type 'enum dma_transfer_direction' to
      different enumeration type 'enum dma_data_direction' [-Werror,-Wenum-conversion]
                direction = DMA_DEV_TO_MEM;
                          ~ ^~~~~~~~~~~~~~
drivers/mmc/host/pxamci.c:212:62: error: implicit conversion from enumeration type 'enum dma_data_direction' to
      different enumeration type 'enum dma_transfer_direction' [-Werror,-Wenum-conversion]
        tx = dmaengine_prep_slave_sg(chan, data->sg, host->dma_len, direction,

The behavior is correct, so this must be a simply typo from
dma_data_direction and dma_transfer_direction being similarly named
types with a similar purpose.

Fixes: 6464b7140951 ("mmc: pxamci: switch over to dmaengine use")
Signed-off-by: Arnd Bergmann <>
Reviewed-by: Nathan Chancellor <>
Acked-by: Robert Jarzmik <>
Signed-off-by: Ulf Hansson <>
2 years agopowerpc/vdso64: Fix CLOCK_MONOTONIC inconsistencies across Y2038
Michael Ellerman [Wed, 13 Mar 2019 13:14:38 +0000 (00:14 +1100)]
powerpc/vdso64: Fix CLOCK_MONOTONIC inconsistencies across Y2038

Jakub Drnec reported:
  Setting the realtime clock can sometimes make the monotonic clock go
  back by over a hundred years. Decreasing the realtime clock across
  the y2k38 threshold is one reliable way to reproduce. Allegedly this
  can also happen just by running ntpd, I have not managed to
  reproduce that other than booting with rtc at >2038 and then running
  ntp. When this happens, anything with timers (e.g. openjdk) breaks
  rather badly.

And included a test case (slightly edited for brevity):
  #define _POSIX_C_SOURCE 199309L
  #include <stdio.h>
  #include <time.h>
  #include <stdlib.h>
  #include <unistd.h>

  long get_time(void) {
    struct timespec tp;
    clock_gettime(CLOCK_MONOTONIC, &tp);
    return tp.tv_sec + tp.tv_nsec / 1000000000;

  int main(void) {
    long last = get_time();
    while(1) {
      long now = get_time();
      if (now < last) {
        printf("clock went backwards by %ld seconds!\n", last - now);
      last = now;
    return 0;

Which when run concurrently with:
 # date -s 2040-1-1
 # date -s 2037-1-1

Will detect the clock going backward.

The root cause is that wtom_clock_sec in struct vdso_data is only a
32-bit signed value, even though we set its value to be equal to
tk->wall_to_monotonic.tv_sec which is 64-bits.

Because the monotonic clock starts at zero when the system boots the
wall_to_montonic.tv_sec offset is negative for current and future
dates. Currently on a freshly booted system the offset will be in the
vicinity of negative 1.5 billion seconds.

However if the wall clock is set past the Y2038 boundary, the offset
from wall to monotonic becomes less than negative 2^31, and no longer
fits in 32-bits. When that value is assigned to wtom_clock_sec it is
truncated and becomes positive, causing the VDSO assembly code to
calculate CLOCK_MONOTONIC incorrectly.

That causes CLOCK_MONOTONIC to jump ahead by ~4 billion seconds which
it is not meant to do. Worse, if the time is then set back before the
Y2038 boundary CLOCK_MONOTONIC will jump backward.

We can fix it simply by storing the full 64-bit offset in the
vdso_data, and using that in the VDSO assembly code. We also shuffle
some of the fields in vdso_data to avoid creating a hole.

The original commit that added the CLOCK_MONOTONIC support to the VDSO
did actually use a 64-bit value for wtom_clock_sec, see commit
a7f290dad32e ("[PATCH] powerpc: Merge vdso's and add vdso support to
32 bits kernel") (Nov 2005). However just 3 days later it was
converted to 32-bits in commit 0c37ec2aa88b ("[PATCH] powerpc: vdso
fixes (take #2)"), and the bug has existed since then AFAICS.

Fixes: 0c37ec2aa88b ("[PATCH] powerpc: vdso fixes (take #2)")
Cc: # v2.6.15+
Reported-by: Jakub Drnec <>
Signed-off-by: Michael Ellerman <>