linux-2.6-microblaze.git
2 years agoubd: remove dead code in ubd_setup_common
Christoph Hellwig [Mon, 28 Jun 2021 09:39:37 +0000 (11:39 +0200)]
ubd: remove dead code in ubd_setup_common

Remove some leftovers of the fake major number parsing that cause
complains from some compilers.

Fixes: 2933a1b2c6f3 ("ubd: remove the code to register as the legacy IDE driver")
Signed-off-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20210628093937.1325608-1-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 years agonvme: use return value from blk_execute_rq()
Keith Busch [Thu, 10 Jun 2021 21:44:37 +0000 (14:44 -0700)]
nvme: use return value from blk_execute_rq()

We don't have an nvme status to report if the driver's .queue_rq()
returns an error without dispatching the requested nvme command. Check
the return value from blk_execute_rq() for all passthrough commands so
the caller may know their command was not successful.

If the command is from the target passthrough interface and fails to
dispatch, synthesize the response back to the host as a internal target
error.

Signed-off-by: Keith Busch <kbusch@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Link: https://lore.kernel.org/r/20210610214437.641245-5-kbusch@kernel.org
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 years agoblock: return errors from blk_execute_rq()
Keith Busch [Thu, 10 Jun 2021 21:44:36 +0000 (14:44 -0700)]
block: return errors from blk_execute_rq()

The synchronous blk_execute_rq() had not provided a way for its callers
to know if its request was successful or not. Return the blk_status_t
result of the request.

Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Keith Busch <kbusch@kernel.org>
Reviewed-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Link: https://lore.kernel.org/r/20210610214437.641245-4-kbusch@kernel.org
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 years agonvme: use blk_execute_rq() for passthrough commands
Keith Busch [Thu, 10 Jun 2021 21:44:35 +0000 (14:44 -0700)]
nvme: use blk_execute_rq() for passthrough commands

The generic blk_execute_rq() knows how to handle polled completions. Use
that instead of implementing an nvme specific handler.

Signed-off-by: Keith Busch <kbusch@kernel.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Link: https://lore.kernel.org/r/20210610214437.641245-3-kbusch@kernel.org
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 years agoblock: support polling through blk_execute_rq
Keith Busch [Thu, 10 Jun 2021 21:44:34 +0000 (14:44 -0700)]
block: support polling through blk_execute_rq

Poll for completions if the request's hctx is a polling type.

Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Signed-off-by: Keith Busch <kbusch@kernel.org>
Reviewed-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Link: https://lore.kernel.org/r/20210610214437.641245-2-kbusch@kernel.org
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 years agoblock: remove REQ_OP_SCSI_{IN,OUT}
Christoph Hellwig [Thu, 24 Jun 2021 12:39:34 +0000 (14:39 +0200)]
block: remove REQ_OP_SCSI_{IN,OUT}

With the legacy IDE driver gone drivers now use either REQ_OP_DRV_*
or REQ_OP_SCSI_*, so unify the two concepts of passthrough requests
into a single one.

Reviewed-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 years agoblock: mark blk_mq_init_queue_data static
Christoph Hellwig [Thu, 24 Jun 2021 08:10:12 +0000 (10:10 +0200)]
block: mark blk_mq_init_queue_data static

All driver uses are gone now.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Bart Van Assche <bvanassche@acm.org>
Link: https://lore.kernel.org/r/20210624081012.256464-1-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 years agoloop: rewrite loop_exit using idr_for_each_entry
Christoph Hellwig [Wed, 23 Jun 2021 14:59:08 +0000 (16:59 +0200)]
loop: rewrite loop_exit using idr_for_each_entry

Use idr_for_each_entry to simplify removing all devices.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Link: https://lore.kernel.org/r/20210623145908.92973-10-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 years agoloop: split loop_lookup
Christoph Hellwig [Wed, 23 Jun 2021 14:59:07 +0000 (16:59 +0200)]
loop: split loop_lookup

loop_lookup has two callers - one wants to do the a find by index and the
other wants any unbound loop device.  Open code the respective
functionality in each caller.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20210623145908.92973-9-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 years agoloop: don't allow deleting an unspecified loop device
Christoph Hellwig [Wed, 23 Jun 2021 14:59:06 +0000 (16:59 +0200)]
loop: don't allow deleting an unspecified loop device

Passing a negative index to loop_lookup while return any unbound device.
Doing that for a delete does not make much sense, so add check to
explicitly reject that case.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20210623145908.92973-8-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 years agoloop: move loop_ctl_mutex locking into loop_add
Christoph Hellwig [Wed, 23 Jun 2021 14:59:05 +0000 (16:59 +0200)]
loop: move loop_ctl_mutex locking into loop_add

Move acquiring and releasing loop_ctl_mutex from the callers into
loop_add.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Link: https://lore.kernel.org/r/20210623145908.92973-7-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 years agoloop: split loop_control_ioctl
Christoph Hellwig [Wed, 23 Jun 2021 14:59:04 +0000 (16:59 +0200)]
loop: split loop_control_ioctl

Split loop_control_ioctl into a helper for each command.  This keeps the
code nicely separated for the upcoming locking changes.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Link: https://lore.kernel.org/r/20210623145908.92973-6-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 years agoloop: don't call loop_lookup before adding a loop device
Christoph Hellwig [Wed, 23 Jun 2021 14:59:03 +0000 (16:59 +0200)]
loop: don't call loop_lookup before adding a loop device

loop_add returns the right error if the slot wasn't available.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20210623145908.92973-5-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 years agoloop: remove the l argument to loop_add
Christoph Hellwig [Wed, 23 Jun 2021 14:59:02 +0000 (16:59 +0200)]
loop: remove the l argument to loop_add

None of the callers cares about the allocated struct loop_device.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Link: https://lore.kernel.org/r/20210623145908.92973-4-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 years agoloop: reduce loop_ctl_mutex coverage in loop_exit
Christoph Hellwig [Wed, 23 Jun 2021 14:59:01 +0000 (16:59 +0200)]
loop: reduce loop_ctl_mutex coverage in loop_exit

loop_ctl_mutex is only needed to iterate the IDR for removing the loop
devices, so reduce the coverage.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Link: https://lore.kernel.org/r/20210623145908.92973-3-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 years agoloop: reorder loop_exit
Christoph Hellwig [Wed, 23 Jun 2021 14:59:00 +0000 (16:59 +0200)]
loop: reorder loop_exit

Unregister the misc and blockdevice first to prevent further access,
and only then iterate to remove the devices.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>
Link: https://lore.kernel.org/r/20210623145908.92973-2-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 years agommc: initialized disk->minors
Christoph Hellwig [Mon, 21 Jun 2021 08:01:44 +0000 (10:01 +0200)]
mmc: initialized disk->minors

Fix a let hunk from the blk_mq_alloc_disk conversion.

Fixes: 281ea6a5bfdc ("mmc: switch to blk_mq_alloc_disk")
Reported-by: Marek Szyprowski <m.szyprowski@samsung.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Tested-by: Marek Szyprowski <m.szyprowski@samsung.com>
Acked-by: Ulf Hansson <ulf.hansson@linaro.org>
Link: https://lore.kernel.org/r/20210621080144.3655131-1-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 years agommc: switch to blk_mq_alloc_disk
Christoph Hellwig [Wed, 16 Jun 2021 05:39:34 +0000 (07:39 +0200)]
mmc: switch to blk_mq_alloc_disk

Use the blk_mq_alloc_disk to allocate the request_queue and gendisk
together.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Ulf Hansson <ulf.hansson@linaro.org>
Link: https://lore.kernel.org/r/20210616053934.880951-3-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 years agommc: remove an extra blk_{get,put}_queue pair
Christoph Hellwig [Wed, 16 Jun 2021 05:39:33 +0000 (07:39 +0200)]
mmc: remove an extra blk_{get,put}_queue pair

The gendisk already acquires a reference to the queue when add_disk
is called, which dropped on put_disk.  So remove the superflous
extra refcounting.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Ulf Hansson <ulf.hansson@linaro.org>
Link: https://lore.kernel.org/r/20210616053934.880951-2-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 years agonbd: provide a way for userspace processes to identify device backends
Prasanna Kumar Kalever [Thu, 29 Apr 2021 10:28:28 +0000 (15:58 +0530)]
nbd: provide a way for userspace processes to identify device backends

Problem:
On reconfigure of device, there is no way to defend if the backend
storage is matching with the initial backend storage.

Say, if an initial connect request for backend "pool1/image1" got
mapped to /dev/nbd0 and the userspace process is terminated. A next
reconfigure request within NBD_ATTR_DEAD_CONN_TIMEOUT is allowed to
use /dev/nbd0 for a different backend "pool1/image2"

For example, an operation like below could be dangerous:

$ sudo rbd-nbd map --try-netlink rbd-pool/ext4-image
/dev/nbd0
$ sudo blkid /dev/nbd0
/dev/nbd0: UUID="bfc444b4-64b1-418f-8b36-6e0d170cfc04" TYPE="ext4"
$ sudo pkill -9 rbd-nbd
$ sudo rbd-nbd attach --try-netlink --device /dev/nbd0 rbd-pool/xfs-image
/dev/nbd0
$ sudo blkid /dev/nbd0
/dev/nbd0: UUID="d29bf343-6570-4069-a9ea-2fa156ced908" TYPE="xfs"

Solution:
Provide a way for userspace processes to keep some metadata to identify
between the device and the backend, so that when a reconfigure request is
made, we can compare and avoid such dangerous operations.

With this solution, as part of the initial connect request, backend
path can be stored in the sysfs per device config, so that on a reconfigure
request it's easy to check if the backend path matches with the initial
connect backend path.

Please note, ioctl interface to nbd will not have these changes, as there
won't be any reconfigure.

Signed-off-by: Prasanna Kumar Kalever <prasanna.kalever@redhat.com>
Reviewed-by: Xiubo Li <xiubli@redhat.com>
Reviewed-by: Ming Lei <ming.lei@redhat.com>
Link: https://lore.kernel.org/r/20210429102828.31248-1-prasanna.kalever@redhat.com
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 years agoubd: use blk_mq_alloc_disk and blk_cleanup_disk
Christoph Hellwig [Mon, 14 Jun 2021 06:07:59 +0000 (08:07 +0200)]
ubd: use blk_mq_alloc_disk and blk_cleanup_disk

Use blk_mq_alloc_disk and blk_cleanup_disk to simplify the gendisk and
request_queue allocation.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20210614060759.3965724-3-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 years agoubd: remove the code to register as the legacy IDE driver
Christoph Hellwig [Mon, 14 Jun 2021 06:07:58 +0000 (08:07 +0200)]
ubd: remove the code to register as the legacy IDE driver

With the legacy IDE driver long deprecated, and modern userspace being
much more flexible about dev_t assignments there is no reason to fake
a registration as the legacy IDE driver in ubd.  This registeration
is a little problematic as it registers the same request_queue for
multiple gendisks, so just remove it.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Acked-By: Anton Ivanov <anton.ivanov@cambridgegreys.com>
Link: https://lore.kernel.org/r/20210614060759.3965724-2-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 years agonull_blk: remove an unused variable assignment in null_add_dev
Christoph Hellwig [Mon, 14 Jun 2021 06:02:31 +0000 (08:02 +0200)]
null_blk: remove an unused variable assignment in null_add_dev

Fix up the recent blk_alloc_disk conversion.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20210614060231.3965278-1-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 years agomtip32xx: use blk_mq_alloc_disk and blk_cleanup_disk
Christoph Hellwig [Mon, 14 Jun 2021 06:03:43 +0000 (08:03 +0200)]
mtip32xx: use blk_mq_alloc_disk and blk_cleanup_disk

Use blk_mq_alloc_disk and blk_cleanup_disk to simplify the gendisk and
request_queue allocation.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20210614060343.3965416-3-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 years agomtip32xx: simplify sysfs setup
Christoph Hellwig [Mon, 14 Jun 2021 06:03:42 +0000 (08:03 +0200)]
mtip32xx: simplify sysfs setup

Pass the driver specific attributes directly to device_add_disk instead
of manually creating them after the disk registration.

Signed-off-by: Christoph Hellwig <hch@lst.de>
Link: https://lore.kernel.org/r/20210614060343.3965416-2-hch@lst.de
Signed-off-by: Jens Axboe <axboe@kernel.dk>
2 years agoMerge tag 'for-5.14/drivers-2021-06-29' of git://git.kernel.dk/linux-block
Linus Torvalds [Wed, 30 Jun 2021 19:21:16 +0000 (12:21 -0700)]
Merge tag 'for-5.14/drivers-2021-06-29' of git://git.kernel.dk/linux-block

Pull block driver updates from Jens Axboe:
 "Pretty calm round, mostly just NVMe and a bit of MD:

   - NVMe updates (via Christoph)
        - improve the APST configuration algorithm (Alexey Bogoslavsky)
        - look for StorageD3Enable on companion ACPI device
          (Mario Limonciello)
        - allow selecting the network interface for TCP connections
          (Martin Belanger)
        - misc cleanups (Amit Engel, Chaitanya Kulkarni, Colin Ian King,
          Christoph)
        - move the ACPI StorageD3 code to drivers/acpi/ and add quirks
          for certain AMD CPUs (Mario Limonciello)
        - zoned device support for nvmet (Chaitanya Kulkarni)
        - fix the rules for changing the serial number in nvmet
          (Noam Gottlieb)
        - various small fixes and cleanups (Dan Carpenter, JK Kim,
          Chaitanya Kulkarni, Hannes Reinecke, Wesley Sheng, Geert
          Uytterhoeven, Daniel Wagner)

   - MD updates (Via Song)
        - iostats rewrite (Guoqing Jiang)
        - raid5 lock contention optimization (Gal Ofri)

   - Fall through warning fix (Gustavo)

   - Misc fixes (Gustavo, Jiapeng)"

* tag 'for-5.14/drivers-2021-06-29' of git://git.kernel.dk/linux-block: (78 commits)
  nvmet: use NVMET_MAX_NAMESPACES to set nn value
  loop: Fix missing discard support when using LOOP_CONFIGURE
  nvme.h: add missing nvme_lba_range_type endianness annotations
  nvme: remove zeroout memset call for struct
  nvme-pci: remove zeroout memset call for struct
  nvmet: remove zeroout memset call for struct
  nvmet: add ZBD over ZNS backend support
  nvmet: add Command Set Identifier support
  nvmet: add nvmet_req_bio put helper for backends
  nvmet: add req cns error complete helper
  block: export blk_next_bio()
  nvmet: remove local variable
  nvmet: use nvme status value directly
  nvmet: use u32 type for the local variable nsid
  nvmet: use u32 for nvmet_subsys max_nsid
  nvmet: use req->cmd directly in file-ns fast path
  nvmet: use req->cmd directly in bdev-ns fast path
  nvmet: make ver stable once connection established
  nvmet: allow mn change if subsys not discovered
  nvmet: make sn stable once connection was established
  ...

2 years agoMerge tag 'for-5.14/block-2021-06-29' of git://git.kernel.dk/linux-block
Linus Torvalds [Wed, 30 Jun 2021 19:12:56 +0000 (12:12 -0700)]
Merge tag 'for-5.14/block-2021-06-29' of git://git.kernel.dk/linux-block

Pull core block updates from Jens Axboe:

 - disk events cleanup (Christoph)

 - gendisk and request queue allocation simplifications (Christoph)

 - bdev_disk_changed cleanups (Christoph)

 - IO priority improvements (Bart)

 - Chained bio completion trace fix (Edward)

 - blk-wbt fixes (Jan)

 - blk-wbt enable/disable fix (Zhang)

 - Scheduler dispatch improvements (Jan, Ming)

 - Shared tagset scheduler improvements (John)

 - BFQ updates (Paolo, Luca, Pietro)

 - BFQ lock inversion fix (Jan)

 - Documentation improvements (Kir)

 - CLONE_IO block cgroup fix (Tejun)

 - Remove of ancient and deprecated block dump feature (zhangyi)

 - Discard merge fix (Ming)

 - Misc fixes or followup fixes (Colin, Damien, Dan, Long, Max, Thomas,
   Yang)

* tag 'for-5.14/block-2021-06-29' of git://git.kernel.dk/linux-block: (129 commits)
  block: fix discard request merge
  block/mq-deadline: Remove a WARN_ON_ONCE() call
  blk-mq: update hctx->dispatch_busy in case of real scheduler
  blk: Fix lock inversion between ioc lock and bfqd lock
  bfq: Remove merged request already in bfq_requests_merged()
  block: pass a gendisk to bdev_disk_changed
  block: move bdev_disk_changed
  block: add the events* attributes to disk_attrs
  block: move the disk events code to a separate file
  block: fix trace completion for chained bio
  block/partitions/msdos: Fix typo inidicator -> indicator
  block, bfq: reset waker pointer with shared queues
  block, bfq: check waker only for queues with no in-flight I/O
  block, bfq: avoid delayed merge of async queues
  block, bfq: boost throughput by extending queue-merging times
  block, bfq: consider also creation time in delayed stable merge
  block, bfq: fix delayed stable merge check
  block, bfq: let also stably merged queues enjoy weight raising
  blk-wbt: make sure throttle is enabled properly
  blk-wbt: introduce a new disable state to prevent false positive by rwb_enabled()
  ...

2 years agoMerge branch 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/hid/hid
Linus Torvalds [Wed, 30 Jun 2021 18:31:32 +0000 (11:31 -0700)]
Merge branch 'for-linus' of git://git./linux/kernel/git/hid/hid

Pull HID updates from Jiri Kosina:

 - patch series that ensures that hid-multitouch driver disables touch
   and button-press reporting on hid-mt devices during suspend when the
   device is not configured as a wakeup-source, from Hans de Goede

 - support for ISH DMA on Intel EHL platform, from Even Xu

 - support for Renoir and Cezanne SoCs, Ambient Light Sensor and Human
   Presence Detection sensor for amd-sfh driver, from Basavaraj Natikar

 - other assorted code cleanups and device-specific fixes/quirks

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/hid/hid: (45 commits)
  HID: thrustmaster: Switch to kmemdup() when allocate change_request
  HID: multitouch: Disable event reporting on suspend when the device is not a wakeup-source
  HID: logitech-dj: Implement may_wakeup ll-driver callback
  HID: usbhid: Implement may_wakeup ll-driver callback
  HID: core: Add hid_hw_may_wakeup() function
  HID: input: Add support for Programmable Buttons
  HID: wacom: Correct base usage for capacitive ExpressKey status bits
  HID: amd_sfh: Add initial support for HPD sensor
  HID: amd_sfh: Extend ALS support for newer AMD platform
  HID: amd_sfh: Extend driver capabilities for multi-generation support
  HID: surface-hid: Fix get-report request
  HID: sony: fix freeze when inserting ghlive ps3/wii dongles
  HID: usbkbd: Avoid GFP_ATOMIC when GFP_KERNEL is possible
  HID: amd_sfh: change in maintainer
  HID: intel-ish-hid: ipc: Specify that EHL no cache snooping
  HID: intel-ish-hid: ishtp: Add dma_no_cache_snooping() callback
  HID: intel-ish-hid: Set ISH driver depends on x86
  HID: hid-input: add Surface Go battery quirk
  HID: intel-ish-hid: Fix minor typos in comments
  HID: usbmouse: Avoid GFP_ATOMIC when GFP_KERNEL is possible
  ...

2 years agoMerge tag 'edac_updates_for_v5.14' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Wed, 30 Jun 2021 18:27:49 +0000 (11:27 -0700)]
Merge tag 'edac_updates_for_v5.14' of git://git./linux/kernel/git/ras/ras

Pull EDAC updates from Tony Luck:
 "Various fixes and support for new CPUs:

   - Clean up error messages from thunderx_edac

   - Add MODULE_DEVICE_TABLE to ti_edac so it will autoload

   - Use %pR to print resources in aspeed_edac

   - Add Yazen Ghannam as MAINTAINER for AMD edac drivers

   - Fix Ice Lake and Sapphire Rapids drivers to report correct "near"
     or "far" device for errors in 2LM configurations

   - Add support of on package high bandwidth memory in Sapphire Rapids

   - New CPU support for three CPUs supporting in-band ECC (IOT SKUs for
     ICL-NNPI, Tiger Lake and Alder Lake)

   - Don't even try to load Intel EDAC drivers when running as a guest

   - Fix Kconfig dependency on X86_MCE_INTEL for EDAC_IGEN6"

* tag 'edac_updates_for_v5.14' of git://git.kernel.org/pub/scm/linux/kernel/git/ras/ras:
  EDAC/igen6: fix core dependency
  EDAC/Intel: Do not load EDAC driver when running as a guest
  EDAC/igen6: Add Intel Alder Lake SoC support
  EDAC/igen6: Add Intel Tiger Lake SoC support
  EDAC/igen6: Add Intel ICL-NNPI SoC support
  EDAC/i10nm: Add support for high bandwidth memory
  EDAC/i10nm: Add detection of memory levels for ICX/SPR servers
  EDAC/skx_common: Add new ADXL components for 2-level memory
  MAINTAINERS: Make Yazen Ghannam maintainer for EDAC-AMD64
  EDAC/aspeed: Use proper format string for printing resource
  EDAC/ti: Add missing MODULE_DEVICE_TABLE
  EDAC/thunderx: Remove irrelevant variable from error messages

2 years agoMerge tag 'tpmdd-next-v5.14-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Wed, 30 Jun 2021 18:23:33 +0000 (11:23 -0700)]
Merge tag 'tpmdd-next-v5.14-rc1' of git://git./linux/kernel/git/jarkko/linux-tpmdd

Pull tpm driver updates from Jarkko Sakkinen:
 "Bug fixes for TPM"

[ This isn't actually the whole contents of the tag and thus doesn't
  contain Jarkko's signature - I dropped the two top commits that added
  support for signing modules using elliptic curve keys because there's
  a new series for that that fixes a few confising things   - Linus ]

* tag 'tpmdd-next-v5.14-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/jarkko/linux-tpmdd:
  tpm: Replace WARN_ONCE() with dev_err_once() in tpm_tis_status()
  tpm_tis: Use DEFINE_RES_MEM() to simplify code
  tpm: fix some doc warnings in tpm1-cmd.c
  tpm_tis_spi: add missing SPI device ID entries
  tpm: add longer timeout for TPM2_CC_VERIFY_SIGNATURE
  char: tpm: move to use request_irq by IRQF_NO_AUTOEN flag
  tpm_tis_spi: set default probe function if device id not match
  tpm_crb: Use IOMEM_ERR_PTR when function returns iomem

2 years agoMerge tag 'platform-drivers-x86-v5.14-1' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Wed, 30 Jun 2021 18:15:39 +0000 (11:15 -0700)]
Merge tag 'platform-drivers-x86-v5.14-1' of git://git./linux/kernel/git/pdx86/platform-drivers-x86

Pull x86 platform driver updates from Hans de Goede:
 "Highlights:

   - New think-lmi driver adding support for changing Lenovo Thinkpad
     BIOS settings from within Linux using the standard firmware-
     attributes class sysfs API

   - MS Surface aggregator-cdev now also supports forwarding events to
     user-space (for debugging / new driver development purposes only)

   - New intel_skl_int3472 driver this provides the necessary glue to
     translate ACPI table information to GPIOs, regulators, etc. for
     camera sensors on Intel devices with IPU3 attached MIPI cameras

   - A whole bunch of other fixes + device-specific quirk additions

   - New devm_work_autocancel() devm-helpers.h function"

* tag 'platform-drivers-x86-v5.14-1' of git://git.kernel.org/pub/scm/linux/kernel/git/pdx86/platform-drivers-x86: (83 commits)
  platform/x86: dell-wmi-sysman: Change user experience when Admin/System Password is modified
  platform/x86: intel_skl_int3472: Uninitialized variable in skl_int3472_handle_gpio_resources()
  platform/x86: think-lmi: Move kfree(setting->possible_values) to tlmi_attr_setting_release()
  platform/x86: think-lmi: Split current_value to reflect only the value
  platform/x86: think-lmi: Fix issues with duplicate attributes
  platform/x86: think-lmi: Return EINVAL when kbdlang gets set to a 0 length string
  platform/x86: intel_cht_int33fe: Move to its own subfolder
  platform/x86: intel_skl_int3472: Move to intel/ subfolder
  platform/x86: intel_skl_int3472: Provide skl_int3472_unregister_clock()
  platform/x86: intel_skl_int3472: Provide skl_int3472_unregister_regulator()
  platform/x86: intel_skl_int3472: Use ACPI GPIO resource directly
  platform/x86: intel_skl_int3472: Fix dependencies (drop CLKDEV_LOOKUP)
  platform/x86: intel_skl_int3472: Free ACPI device resources after use
  platform/x86: Remove "default n" entries
  platform/x86: ISST: Use numa node id for cpu pci dev mapping
  platform/x86: ISST: Optimize CPU to PCI device mapping
  tools/power/x86/intel-speed-select: v1.10 release
  tools/power/x86/intel-speed-select: Fix uncore memory frequency display
  extcon: extcon-max8997: Simplify driver using devm
  extcon: extcon-max8997: Fix IRQ freeing at error path
  ...

2 years agoMerge tag 'mailbox-v5.14' of git://git.linaro.org/landing-teams/working/fujitsu/integ...
Linus Torvalds [Wed, 30 Jun 2021 18:11:47 +0000 (11:11 -0700)]
Merge tag 'mailbox-v5.14' of git://git.linaro.org/landing-teams/working/fujitsu/integration

Pull mailbox updates from Jassi Brar:

 - imx: add support for i.MX8ULP

 - mtk: code change around callback struct

 - qcom: add sm6125, MSM8939 fix for channel exhaustion

 - microchip: add support for polarfire controller

 - misc: cosmetic changes to bcm-2835,flexrm,pdc, arm-mhu and hisilicon

* tag 'mailbox-v5.14' of git://git.linaro.org/landing-teams/working/fujitsu/integration: (26 commits)
  MAINTAINERS: add entry for polarfire soc mailbox
  dt-bindings: add bindings for polarfire soc system controller
  mbox: add polarfire soc system controller mailbox
  dt-bindings: add bindings for polarfire soc mailbox
  mailbox: imx: Avoid using val uninitialized in imx_mu_isr()
  mailbox: qcom: Add MSM8939 APCS support
  mailbox: qcom: Use PLATFORM_DEVID_AUTO to register platform device
  dt-bindings: mailbox: qcom: Add MSM8939 APCS compatible
  mailbox: qcom-apcs: Add SM6125 compatible
  dt-bindings: mailbox: Add binding for sm6125
  mailbox: mtk-cmdq: Fix uninitialized variable in cmdq_mbox_flush()
  mailbox: bcm-flexrm-mailbox: Remove redundant dev_err call in flexrm_mbox_probe()
  mailbox: bcm2835: Remove redundant dev_err call in bcm2835_mbox_probe()
  mailbox: qcom-ipcc: Fix IPCC mbox channel exhaustion
  mailbox: mtk-cmdq: Add struct cmdq_pkt in struct cmdq_cb_data
  mailbox: mtk-cmdq: Use mailbox rx_callback
  mailbox: mtk-cmdq: Remove cmdq_cb_status
  mailbox: imx-mailbox: support i.MX8ULP MU
  mailbox: imx: add xSR/xCR register array
  mailbox: imx: replace the xTR/xRR array with single register
  ...

2 years agoMerge tag 'for-linus-5.14-1' of git://github.com/cminyard/linux-ipmi
Linus Torvalds [Wed, 30 Jun 2021 18:09:37 +0000 (11:09 -0700)]
Merge tag 'for-linus-5.14-1' of git://github.com/cminyard/linux-ipmi

Pull IPMI driver updates from Corey Minyard:
 "Mostly a restructure of the kcs_bmc driver to make it easier to use
  with different types of devices, and just to clean things up and
  improve things.

  Also some bug fixes for the kcs_bmc driver.

  One fix to the IPMI watchdog to stop the timer when the action is
  none. Not a big deal, but it's the right thing to do"

* tag 'for-linus-5.14-1' of git://github.com/cminyard/linux-ipmi:
  ipmi: kcs_bmc_aspeed: Fix less than zero comparison of a unsigned int
  ipmi: kcs_bmc_aspeed: Optionally apply status address
  ipmi: kcs_bmc_aspeed: Fix IBFIE typo from datasheet
  ipmi: kcs_bmc_aspeed: Implement KCS SerIRQ configuration
  dt-bindings: ipmi: Add optional SerIRQ property to ASPEED KCS devices
  dt-bindings: ipmi: Convert ASPEED KCS binding to schema
  ipmi: kcs_bmc: Add serio adaptor
  ipmi: kcs_bmc: Enable IBF on open
  ipmi: kcs_bmc: Allow clients to control KCS IRQ state
  ipmi: kcs_bmc: Decouple the IPMI chardev from the core
  ipmi: kcs_bmc: Strip private client data from struct kcs_bmc
  ipmi: kcs_bmc: Split headers into device and client
  ipmi: kcs_bmc: Turn the driver data-structures inside-out
  ipmi: kcs_bmc: Split out kcs_bmc_cdev_ipmi
  ipmi: kcs_bmc: Rename {read,write}_{status,data}() functions
  ipmi: kcs_bmc: Make status update atomic
  ipmi: kcs_bmc_aspeed: Use of match data to extract KCS properties
  ipmi/watchdog: Stop watchdog timer when the current action is 'none'

2 years agoMerge branch 'for-5.14/multitouch' into for-linus
Jiri Kosina [Wed, 30 Jun 2021 07:15:15 +0000 (09:15 +0200)]
Merge branch 'for-5.14/multitouch' into for-linus

- patch series that ensures that hid-multitouch driver disables touch and
  button-press reporting on hid-mt devices during suspend when the device is
  not configured as a wakeup-source, from Hans de Goede

2 years agoMerge branch 'for-5.14/logitech' into for-linus
Jiri Kosina [Wed, 30 Jun 2021 07:12:44 +0000 (09:12 +0200)]
Merge branch 'for-5.14/logitech' into for-linus

- support for LCD menu keys + LCD brightness control on the Logitech Z-10
  speakers (with LCD) which use the same protocol as the G15 keyboards
  from Hans de Goede

2 years agoMerge branch 'for-5.14/intel-ish' into for-linus
Jiri Kosina [Wed, 30 Jun 2021 07:06:53 +0000 (09:06 +0200)]
Merge branch 'for-5.14/intel-ish' into for-linus

- support for ISH DMA on EHL platform from Even Xu
- various code style fixes and cleanups from Lee Jones and Uwe Kleine-König

2 years agoMerge branch 'for-5.14/google' into for-linus
Jiri Kosina [Wed, 30 Jun 2021 07:05:31 +0000 (09:05 +0200)]
Merge branch 'for-5.14/google' into for-linus

- device tree match for Google Whiskers device from Ikjoon Jang

2 years agoMerge branch 'for-5.14/core' into for-linus
Jiri Kosina [Wed, 30 Jun 2021 07:03:51 +0000 (09:03 +0200)]
Merge branch 'for-5.14/core' into for-linus

- device unbinding locking fix from Dmitry Torokhov
- support for programmable buttons (mapping to KEY_MACRO# event codes)
  from Thomas Weißschuh
- various other small fixes and code style improvements

2 years agoMerge branch 'for-5.14/amd-sfh' into for-linus
Jiri Kosina [Wed, 30 Jun 2021 07:02:28 +0000 (09:02 +0200)]
Merge branch 'for-5.14/amd-sfh' into for-linus

- support for Renoir and Cezanne SoCs
- support for Ambient Light Sensor
- support for Human Presence Detection sensor

all from Basavaraj Natikar

2 years agoMerge tag 'dlm-5.14' of git://git.kernel.org/pub/scm/linux/kernel/git/teigland/linux-dlm
Linus Torvalds [Wed, 30 Jun 2021 03:30:00 +0000 (20:30 -0700)]
Merge tag 'dlm-5.14' of git://git./linux/kernel/git/teigland/linux-dlm

Pull dlm updates from David Teigland:
 "This is a major dlm networking enhancement that adds message
  retransmission so that the dlm can reliably continue operating when
  network connections fail and nodes reconnect.

  Previously, this would result in lost messages which could only be
  handled as a node failure"

* tag 'dlm-5.14' of git://git.kernel.org/pub/scm/linux/kernel/git/teigland/linux-dlm: (26 commits)
  fs: dlm: invalid buffer access in lookup error
  fs: dlm: fix race in mhandle deletion
  fs: dlm: rename socket and app buffer defines
  fs: dlm: introduce proto values
  fs: dlm: move dlm allow conn
  fs: dlm: use alloc_ordered_workqueue
  fs: dlm: fix memory leak when fenced
  fs: dlm: fix lowcomms_start error case
  fs: dlm: Fix spelling mistake "stucked" -> "stuck"
  fs: dlm: Fix memory leak of object mh
  fs: dlm: don't allow half transmitted messages
  fs: dlm: add midcomms debugfs functionality
  fs: dlm: add reliable connection if reconnect
  fs: dlm: add union in dlm header for lockspace id
  fs: dlm: move out some hash functionality
  fs: dlm: add functionality to re-transmit a message
  fs: dlm: make buffer handling per msg
  fs: dlm: add more midcomms hooks
  fs: dlm: public header in out utility
  fs: dlm: fix connection tcp EOF handling
  ...

2 years agoMerge tag 'gfs2-v5.13-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/gfs2...
Linus Torvalds [Wed, 30 Jun 2021 03:23:08 +0000 (20:23 -0700)]
Merge tag 'gfs2-v5.13-fixes' of git://git./linux/kernel/git/gfs2/linux-gfs2

Pull gfs2 updates from Andreas Gruenbacher:
 "Various minor gfs2 cleanups and fixes"

* tag 'gfs2-v5.13-fixes' of git://git.kernel.org/pub/scm/linux/kernel/git/gfs2/linux-gfs2:
  gfs2: Clean up gfs2_unstuff_dinode
  gfs2: Unstuff before locking page in gfs2_page_mkwrite
  gfs2: Clean up the error handling in gfs2_page_mkwrite
  gfs2: Fix error handling in init_statfs
  gfs2: Fix underflow in gfs2_page_mkwrite
  gfs2: Use list_move_tail instead of list_del/list_add_tail
  gfs2: Fix do_gfs2_set_flags description

2 years agoMerge tag '5.14-rc-smb3-fixes-part1' of git://git.samba.org/sfrench/cifs-2.6
Linus Torvalds [Wed, 30 Jun 2021 03:18:16 +0000 (20:18 -0700)]
Merge tag '5.14-rc-smb3-fixes-part1' of git://git.samba.org/sfrench/cifs-2.6

Pull cifs updates from Steve French:

 - improve fallocate emulation

 - DFS fixes

 - minor multichannel fixes

 - various cleanup patches, many to address Coverity warnings

* tag '5.14-rc-smb3-fixes-part1' of git://git.samba.org/sfrench/cifs-2.6: (38 commits)
  smb3: prevent races updating CurrentMid
  cifs: fix missing spinlock around update to ses->status
  cifs: missing null pointer check in cifs_mount
  smb3: fix possible access to uninitialized pointer to DACL
  cifs: missing null check for newinode pointer
  cifs: remove two cases where rc is set unnecessarily in sid_to_id
  SMB3: Add new info level for query directory
  cifs: fix NULL dereference in smb2_check_message()
  smbdirect: missing rc checks while waiting for rdma events
  cifs: Avoid field over-reading memcpy()
  smb311: remove dead code for non compounded posix query info
  cifs: fix SMB1 error path in cifs_get_file_info_unix
  smb3: fix uninitialized value for port in witness protocol move
  cifs: fix unneeded null check
  cifs: use SPDX-Licence-Identifier
  cifs: convert list_for_each to entry variant in cifs_debug.c
  cifs: convert list_for_each to entry variant in smb2misc.c
  cifs: avoid extra calls in posix_info_parse
  cifs: retry lookup and readdir when EAGAIN is returned.
  cifs: fix check of dfs interlinks
  ...

2 years agoMerge tag 'fs.openat2.unknown_flags.v5.14' of git://git.kernel.org/pub/scm/linux...
Linus Torvalds [Wed, 30 Jun 2021 03:10:16 +0000 (20:10 -0700)]
Merge tag 'fs.openat2.unknown_flags.v5.14' of git://git./linux/kernel/git/brauner/linux

Pull openat2 fixes from Christian Brauner:

 - Remove the unused VALID_UPGRADE_FLAGS define we carried from an
   extension to openat2() that we haven't merged. Aleksa might be
   getting back to it at some point but just not right now.

 - openat2() used to accidently ignore unknown flag values in the upper
   32 bits.

   The new openat2() syscall verifies that no unknown O-flag values are
   set and returns an error to userspace if they are while the older
   open syscalls like open() and openat() simply ignore unknown flag
   values:

      #define O_FLAG_CURRENTLY_INVALID (1 << 31)
      struct open_how how = {
            .flags = O_RDONLY | O_FLAG_CURRENTLY_INVALID,
            .resolve = 0,
      };

      /* fails */
      fd = openat2(-EBADF, "/dev/null", &how, sizeof(how));

      /* succeeds */
      fd = openat(-EBADF, "/dev/null", O_RDONLY | O_FLAG_CURRENTLY_INVALID);

   However, openat2() silently truncates the upper 32 bits meaning:

      #define O_FLAG_CURRENTLY_INVALID_LOWER32 (1 << 31)
      #define O_FLAG_CURRENTLY_INVALID_UPPER32 (1 << 40)

      struct open_how how_lowe32 = {
            .flags = O_RDONLY | O_FLAG_CURRENTLY_INVALID_LOWER32,
      };

      struct open_how how_upper32 = {
            .flags = O_RDONLY | O_FLAG_CURRENTLY_INVALID_UPPER32,
      };

      /* fails */
      fd = openat2(-EBADF, "/dev/null", &how_lower32, sizeof(how_lower32));

      /* succeeds */
      fd = openat2(-EBADF, "/dev/null", &how_upper32, sizeof(how_upper32));

   Fix this by preventing the immediate truncation in build_open_flags()
   and add a compile-time check to catch when we add flags in the upper
   32 bit range.

* tag 'fs.openat2.unknown_flags.v5.14' of git://git.kernel.org/pub/scm/linux/kernel/git/brauner/linux:
  test: add openat2() test for invalid upper 32 bit flag value
  open: don't silently ignore unknown O-flags in openat2()
  fcntl: remove unused VALID_UPGRADE_FLAGS

2 years agoMerge tag 'fs.mount_setattr.nosymfollow.v5.14' of git://git.kernel.org/pub/scm/linux...
Linus Torvalds [Wed, 30 Jun 2021 03:07:06 +0000 (20:07 -0700)]
Merge tag 'fs.mount_setattr.nosymfollow.v5.14' of git://git./linux/kernel/git/brauner/linux

Pull mount_setattr updates from Christian Brauner:
 "A few releases ago the old mount API gained support for a mount
  options which prevents following symlinks on a given mount. This adds
  support for it in the new mount api through the MOUNT_ATTR_NOSYMFOLLOW
  flag via mount_setattr() and fsmount(). With mount_setattr() that flag
  can even be applied recursively.

  There's an additional ack from Ross Zwisler who originally authored
  the nosymfollow patch. As I've already had the patches in my for-next
  I didn't add his ack explicitly"

* tag 'fs.mount_setattr.nosymfollow.v5.14' of git://git.kernel.org/pub/scm/linux/kernel/git/brauner/linux:
  tests: test MOUNT_ATTR_NOSYMFOLLOW with mount_setattr()
  mount: Support "nosymfollow" in new mount api

2 years agoMerge branch 'akpm' (patches from Andrew)
Linus Torvalds [Wed, 30 Jun 2021 00:29:11 +0000 (17:29 -0700)]
Merge branch 'akpm' (patches from Andrew)

Merge misc updates from Andrew Morton:
 "191 patches.

  Subsystems affected by this patch series: kthread, ia64, scripts,
  ntfs, squashfs, ocfs2, kernel/watchdog, and mm (gup, pagealloc, slab,
  slub, kmemleak, dax, debug, pagecache, gup, swap, memcg, pagemap,
  mprotect, bootmem, dma, tracing, vmalloc, kasan, initialization,
  pagealloc, and memory-failure)"

* emailed patches from Andrew Morton <akpm@linux-foundation.org>: (191 commits)
  mm,hwpoison: make get_hwpoison_page() call get_any_page()
  mm,hwpoison: send SIGBUS with error virutal address
  mm/page_alloc: split pcp->high across all online CPUs for cpuless nodes
  mm/page_alloc: allow high-order pages to be stored on the per-cpu lists
  mm: replace CONFIG_FLAT_NODE_MEM_MAP with CONFIG_FLATMEM
  mm: replace CONFIG_NEED_MULTIPLE_NODES with CONFIG_NUMA
  docs: remove description of DISCONTIGMEM
  arch, mm: remove stale mentions of DISCONIGMEM
  mm: remove CONFIG_DISCONTIGMEM
  m68k: remove support for DISCONTIGMEM
  arc: remove support for DISCONTIGMEM
  arc: update comment about HIGHMEM implementation
  alpha: remove DISCONTIGMEM and NUMA
  mm/page_alloc: move free_the_page
  mm/page_alloc: fix counting of managed_pages
  mm/page_alloc: improve memmap_pages dbg msg
  mm: drop SECTION_SHIFT in code comments
  mm/page_alloc: introduce vm.percpu_pagelist_high_fraction
  mm/page_alloc: limit the number of pages on PCP lists when reclaim is active
  mm/page_alloc: scale the number of pages that are batch freed
  ...

2 years agoMerge tag 'devprop-5.14-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael...
Linus Torvalds [Tue, 29 Jun 2021 21:04:37 +0000 (14:04 -0700)]
Merge tag 'devprop-5.14-rc1' of git://git./linux/kernel/git/rafael/linux-pm

Pull device properties framework updates from Rafael Wysocki:
 "These unify device properties access in some pieces of code and make
  related changes.

  Specifics:

   - Handle device properties with software node API in the ACPI IORT
     table parsing code (Heikki Krogerus).

   - Unify of_node access in the common device properties code, constify
     the acpi_dma_supported() argument pointer and fix up CONFIG_ACPI=n
     stubs of some functions related to device properties (Andy
     Shevchenko)"

* tag 'devprop-5.14-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael/linux-pm:
  device property: Unify access to of_node
  ACPI: scan: Constify acpi_dma_supported() helper function
  ACPI: property: Constify stubs for CONFIG_ACPI=n case
  ACPI: IORT: Handle device properties with software node API
  device property: Retrieve fwnode from of_node via accessor

2 years agoMerge tag 'pnp-5.14-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael...
Linus Torvalds [Tue, 29 Jun 2021 20:50:21 +0000 (13:50 -0700)]
Merge tag 'pnp-5.14-rc1' of git://git./linux/kernel/git/rafael/linux-pm

Pull PNP updates from Rafael Wysocki:
 "These get rid of unnecessary local variables and function, reduce code
  duplication and clean up message printing.

  Specifics:

   - Remove unnecessary local variables from isapnp_proc_attach_device()
     (Anupama K Patil).

   - Make the callers of pnp_alloc() use kzalloc() directly and drop the
     former (Heiner Kallweit).

   - Make two pieces of code use dev_dbg() instead of dev_printk() with
     the KERN_DEBUG message level (Heiner Kallweit).

   - Use DEVICE_ATTR_RO() instead of full DEVICE_ATTR() in some places
     in card.c (Zhen Lei).

   - Use list_for_each_entry() instead of list_for_each() in
     insert_device() (Zou Wei)"

* tag 'pnp-5.14-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael/linux-pm:
  PNP: pnpbios: Use list_for_each_entry() instead of list_for_each()
  PNP: use DEVICE_ATTR_RO macro
  PNP: Switch over to dev_dbg()
  PNP: Remove pnp_alloc()
  drivers: pnp: isapnp: proc.c: Remove unnecessary local variables

2 years agoMerge tag 'acpi-5.14-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael...
Linus Torvalds [Tue, 29 Jun 2021 20:39:41 +0000 (13:39 -0700)]
Merge tag 'acpi-5.14-rc1' of git://git./linux/kernel/git/rafael/linux-pm

Pull ACPI updates from Rafael Wysocki:
 "These update the ACPICA code in the kernel to the 20210604 upstream
  revision, add preliminary support for the Platform Runtime Mechanism
  (PRM), address issues related to the handling of device dependencies
  in the ACPI device eunmeration code, improve the tracking of ACPI
  power resource states, improve the ACPI support for suspend-to-idle on
  AMD systems, continue the unification of message printing in the ACPI
  code, address assorted issues and clean up the code in a number of
  places.

  Specifics:

   - Update ACPICA code in the kernel to upstrea revision 20210604
     including the following changes:

      - Add defines for the CXL Host Bridge Structureand and add the
        CFMWS structure definition to CEDT (Alison Schofield).
      - iASL: Finish support for the IVRS ACPI table (Bob Moore).
      - iASL: Add support for the SVKL table (Bob Moore).
      - iASL: Add full support for RGRT ACPI table (Bob Moore).
      - iASL: Add support for the BDAT ACPI table (Bob Moore).
      - iASL: add disassembler support for PRMT (Erik Kaneda).
      - Fix memory leak caused by _CID repair function (Erik Kaneda).
      - Add support for PlatformRtMechanism OpRegion (Erik Kaneda).
      - Add PRMT module header to facilitate parsing (Erik Kaneda).
      - Add _PLD panel positions (Fabian Wüthrich).
      - MADT: add Multiprocessor Wakeup Mailbox Structure and the SVKL
        table headers (Kuppuswamy Sathyanarayanan).
      - Use ACPI_FALLTHROUGH (Wei Ming Chen).

   - Add preliminary support for the Platform Runtime Mechanism (PRM) to
     allow the AML interpreter to call PRM functions (Erik Kaneda).

   - Address some issues related to the handling of device dependencies
     reported by _DEP in the ACPI device enumeration code and clean up
     some related pieces of it (Rafael Wysocki).

   - Improve the tracking of states of ACPI power resources (Rafael
     Wysocki).

   - Improve ACPI support for suspend-to-idle on AMD systems (Alex
     Deucher, Mario Limonciello, Pratik Vishwakarma).

   - Continue the unification and cleanup of message printing in the
     ACPI code (Hanjun Guo, Heiner Kallweit).

   - Fix possible buffer overrun issue with the description_show() sysfs
     attribute method (Krzysztof WilczyÅ„ski).

   - Improve the acpi_mask_gpe kernel command line parameter handling
     and clean up the core ACPI code related to sysfs (Andy Shevchenko,
     Baokun Li, Clayton Casciato).

   - Postpone bringing devices in the general ACPI PM domain to D0
     during resume from system-wide suspend until they are really needed
     (Dmitry Torokhov).

   - Make the ACPI processor driver fix up C-state latency if not
     ordered (Mario Limonciello).

   - Add support for identifying devices depening on the given one that
     are not its direct descendants with the help of _DEP (Daniel
     Scally).

   - Extend the checks related to ACPI IRQ overrides on x86 in order to
     avoid false-positives (Hui Wang).

   - Add battery DPTF participant for Intel SoCs (Sumeet Pawnikar).

   - Rearrange the ACPI fan driver and device power management code to
     use a common list of device IDs (Rafael Wysocki).

   - Fix clang CFI violation in the ACPI BGRT table parsing code and
     clean it up (Nathan Chancellor).

   - Add GPE-related quirks for some laptops to the EC driver (Chris
     Chiu, Zhang Rui).

   - Make the ACPI PPTT table parsing code populate the cache-id value
     if present in the firmware (James Morse).

   - Remove redundant clearing of context->ret.pointer from
     acpi_run_osc() (Hans de Goede).

   - Add missing acpi_put_table() in acpi_init_fpdt() (Jing Xiangfeng).

   - Make ACPI APEI handle ARM Processor Error CPER records like Memory
     Error ones to avoid user space task lockups (Xiaofei Tan).

   - Stop warning about disabled ACPI in APEI (Jon Hunter).

   - Fix fall-through warning for Clang in the SBSHC driver (Gustavo A.
     R. Silva).

   - Add custom DSDT file as Makefile prerequisite (Richard Fitzgerald).

   - Initialize local variable to avoid garbage being returned (Colin
     Ian King).

   - Simplify assorted pieces of code, address assorted coding style and
     documentation issues and comment typos (Baokun Li, Christophe
     JAILLET, Clayton Casciato, Liu Shixin, Shaokun Zhang, Wei Yongjun,
     Yang Li, Zhen Lei)"

* tag 'acpi-5.14-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael/linux-pm: (97 commits)
  ACPI: PM: postpone bringing devices to D0 unless we need them
  ACPI: tables: Add custom DSDT file as makefile prerequisite
  ACPI: bgrt: Use sysfs_emit
  ACPI: bgrt: Fix CFI violation
  ACPI: EC: trust DSDT GPE for certain HP laptop
  ACPI: scan: Simplify acpi_table_events_fn()
  ACPI: PM: Adjust behavior for field problems on AMD systems
  ACPI: PM: s2idle: Add support for new Microsoft UUID
  ACPI: PM: s2idle: Add support for multiple func mask
  ACPI: PM: s2idle: Refactor common code
  ACPI: PM: s2idle: Use correct revision id
  ACPI: sysfs: Remove tailing return statement in void function
  ACPI: sysfs: Use __ATTR_RO() and __ATTR_RW() macros
  ACPI: sysfs: Sort headers alphabetically
  ACPI: sysfs: Refactor param_get_trace_state() to drop dead code
  ACPI: sysfs: Unify pattern of memory allocations
  ACPI: sysfs: Allow bitmap list to be supplied to acpi_mask_gpe
  ACPI: sysfs: Make sparse happy about address space in use
  ACPI: scan: Fix race related to dropping dependencies
  ACPI: scan: Reorganize acpi_device_add()
  ...

2 years agoMerge tag 'pm-5.14-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael/linux-pm
Linus Torvalds [Tue, 29 Jun 2021 20:36:06 +0000 (13:36 -0700)]
Merge tag 'pm-5.14-rc1' of git://git./linux/kernel/git/rafael/linux-pm

Pull power management updates from Rafael Wysocki:
 "These add hybrid processors support to the intel_pstate driver and
  make it work with more processor models when HWP is disabled, make the
  intel_idle driver use special C6 idle state paremeters when package
  C-states are disabled, add cooling support to the tegra30 devfreq
  driver, rework the TEO (timer events oriented) cpuidle governor,
  extend the OPP (operating performance points) framework to use the
  required-opps DT property in more cases, fix some issues and clean up
  a number of assorted pieces of code.

  Specifics:

   - Make intel_pstate support hybrid processors using abstract
     performance units in the HWP interface (Rafael Wysocki).

   - Add Icelake servers and Cometlake support in no-HWP mode to
     intel_pstate (Giovanni Gherdovich).

   - Make cpufreq_online() error path be consistent with the CPU device
     removal path in cpufreq (Rafael Wysocki).

   - Clean up 3 cpufreq drivers and the statistics code (Hailong Liu,
     Randy Dunlap, Shaokun Zhang).

   - Make intel_idle use special idle state parameters for C6 when
     package C-states are disabled (Chen Yu).

   - Rework the TEO (timer events oriented) cpuidle governor to address
     some theoretical shortcomings in it (Rafael Wysocki).

   - Drop unneeded semicolon from the TEO governor (Wan Jiabing).

   - Modify the runtime PM framework to accept unassigned suspend and
     resume callback pointers (Ulf Hansson).

   - Improve pm_runtime_get_sync() documentation (Krzysztof Kozlowski).

   - Improve device performance states support in the generic power
     domains (genpd) framework (Ulf Hansson).

   - Fix some documentation issues in genpd (Yang Yingliang).

   - Make the operating performance points (OPP) framework use the
     required-opps DT property in use cases that are not related to
     genpd (Hsin-Yi Wang).

   - Make lazy_link_required_opp_table() use list_del_init instead of
     list_del/INIT_LIST_HEAD (Yang Yingliang).

   - Simplify wake IRQs handling in the core system-wide sleep support
     code and clean up some coding style inconsistencies in it (Tian
     Tao, Zhen Lei).

   - Add cooling support to the tegra30 devfreq driver and improve its
     DT bindings (Dmitry Osipenko).

   - Fix some assorted issues in the devfreq core and drivers (Chanwoo
     Choi, Dong Aisheng, YueHaibing)"

* tag 'pm-5.14-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael/linux-pm: (39 commits)
  PM / devfreq: passive: Fix get_target_freq when not using required-opp
  cpufreq: Make cpufreq_online() call driver->offline() on errors
  opp: Allow required-opps to be used for non genpd use cases
  cpuidle: teo: remove unneeded semicolon in teo_select()
  dt-bindings: devfreq: tegra30-actmon: Add cooling-cells
  dt-bindings: devfreq: tegra30-actmon: Convert to schema
  PM / devfreq: userspace: Use DEVICE_ATTR_RW macro
  PM: runtime: Clarify documentation when callbacks are unassigned
  PM: runtime: Allow unassigned ->runtime_suspend|resume callbacks
  PM: runtime: Improve path in rpm_idle() when no callback
  PM: hibernate: remove leading spaces before tabs
  PM: sleep: remove trailing spaces and tabs
  PM: domains: Drop/restore performance state votes for devices at runtime PM
  PM: domains: Return early if perf state is already set for the device
  PM: domains: Split code in dev_pm_genpd_set_performance_state()
  cpuidle: teo: Use kerneldoc documentation in admin-guide
  cpuidle: teo: Rework most recent idle duration values treatment
  cpuidle: teo: Change the main idle state selection logic
  cpuidle: teo: Cosmetic modification of teo_select()
  cpuidle: teo: Cosmetic modifications of teo_update()
  ...

2 years agoMerge tag 'x86-entry-2021-06-29' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Tue, 29 Jun 2021 19:44:51 +0000 (12:44 -0700)]
Merge tag 'x86-entry-2021-06-29' of git://git./linux/kernel/git/tip/tip

Pull x86 entry code related updates from Thomas Gleixner:

 - Consolidate the macros for .byte ... opcode sequences

 - Deduplicate register offset defines in include files

 - Simplify the ia32,x32 compat handling of the related syscall tables
   to get rid of #ifdeffery.

 - Clear all EFLAGS which are not required for syscall handling

 - Consolidate the syscall tables and switch the generation over to the
   generic shell script and remove the CFLAGS tweaks which are not
   longer required.

 - Use 'int' type for system call numbers to match the generic code.

 - Add more selftests for syscalls

* tag 'x86-entry-2021-06-29' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  x86/syscalls: Don't adjust CFLAGS for syscall tables
  x86/syscalls: Remove -Wno-override-init for syscall tables
  x86/uml/syscalls: Remove array index from syscall initializers
  x86/syscalls: Clear 'offset' and 'prefix' in case they are set in env
  x86/entry: Use int everywhere for system call numbers
  x86/entry: Treat out of range and gap system calls the same
  x86/entry/64: Sign-extend system calls on entry to int
  selftests/x86/syscall: Add tests under ptrace to syscall_numbering_64
  selftests/x86/syscall: Simplify message reporting in syscall_numbering
  selftests/x86/syscall: Update and extend syscall_numbering_64
  x86/syscalls: Switch to generic syscallhdr.sh
  x86/syscalls: Use __NR_syscalls instead of __NR_syscall_max
  x86/unistd: Define X32_NR_syscalls only for 64-bit kernel
  x86/syscalls: Stop filling syscall arrays with *_sys_ni_syscall
  x86/syscalls: Switch to generic syscalltbl.sh
  x86/entry/x32: Rename __x32_compat_sys_* to __x64_compat_sys_*

2 years agoMerge tag 'x86-irq-2021-06-29' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip
Linus Torvalds [Tue, 29 Jun 2021 19:36:59 +0000 (12:36 -0700)]
Merge tag 'x86-irq-2021-06-29' of git://git./linux/kernel/git/tip/tip

Pull x86 interrupt related updates from Thomas Gleixner:

 - Consolidate the VECTOR defines and the usage sites.

 - Cleanup GDT/IDT related code and replace open coded ASM with proper
   native helper functions.

* tag 'x86-irq-2021-06-29' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  x86/kexec: Set_[gi]dt() -> native_[gi]dt_invalidate() in machine_kexec_*.c
  x86: Add native_[ig]dt_invalidate()
  x86/idt: Remove address argument from idt_invalidate()
  x86/irq: Add and use NR_EXTERNAL_VECTORS and NR_SYSTEM_VECTORS
  x86/irq: Remove unused vectors defines

2 years agoMerge tag 'timers-core-2021-06-29' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Tue, 29 Jun 2021 19:31:16 +0000 (12:31 -0700)]
Merge tag 'timers-core-2021-06-29' of git://git./linux/kernel/git/tip/tip

Pull timer updates from Thomas Gleixner:
 "Time and clocksource/clockevent related updates:

  Core changes:

   - Infrastructure to support per CPU "broadcast" devices for per CPU
     clockevent devices which stop in deep idle states. This allows us
     to utilize the more efficient architected timer on certain ARM SoCs
     for normal operation instead of permanentely using the slow to
     access SoC specific clockevent device.

   - Print the name of the broadcast/wakeup device in /proc/timer_list

   - Make the clocksource watchdog more robust against delays between
     reading the current active clocksource and the watchdog
     clocksource. Such delays can be caused by NMIs, SMIs and vCPU
     preemption.

     Handle this by reading the watchdog clocksource twice, i.e. before
     and after reading the current active clocksource. In case that the
     two watchdog reads shows an excessive time delta, the read sequence
     is repeated up to 3 times.

   - Improve the debug output and add a test module for the watchdog
     mechanism.

   - Reimplementation of the venerable time64_to_tm() function with a
     faster and significantly smaller version. Straight from the source,
     i.e. the author of the related research paper contributed this!

  Driver changes:

   - No new drivers, not even new device tree bindings!

   - Fixes, improvements and cleanups and all over the place"

* tag 'timers-core-2021-06-29' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: (30 commits)
  time/kunit: Add missing MODULE_LICENSE()
  time: Improve performance of time64_to_tm()
  clockevents: Use list_move() instead of list_del()/list_add()
  clocksource: Print deviation in nanoseconds when a clocksource becomes unstable
  clocksource: Provide kernel module to test clocksource watchdog
  clocksource: Reduce clocksource-skew threshold
  clocksource: Limit number of CPUs checked for clock synchronization
  clocksource: Check per-CPU clock synchronization when marked unstable
  clocksource: Retry clock read if long delays detected
  clockevents: Add missing parameter documentation
  clocksource/drivers/timer-ti-dm: Drop unnecessary restore
  clocksource/arm_arch_timer: Improve Allwinner A64 timer workaround
  clocksource/drivers/arm_global_timer: Remove duplicated argument in arm_global_timer
  clocksource/drivers/arm_global_timer: Make symbol 'gt_clk_rate_change_nb' static
  arm: zynq: don't disable CONFIG_ARM_GLOBAL_TIMER due to CONFIG_CPU_FREQ anymore
  clocksource/drivers/arm_global_timer: Implement rate compensation whenever source clock changes
  clocksource/drivers/ingenic: Rename unreasonable array names
  clocksource/drivers/timer-ti-dm: Save and restore timer TIOCP_CFG
  clocksource/drivers/mediatek: Ack and disable interrupts on suspend
  clocksource/drivers/samsung_pwm: Constify source IO memory
  ...

2 years agoMerge tag 'irq-core-2021-06-29' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Tue, 29 Jun 2021 19:25:04 +0000 (12:25 -0700)]
Merge tag 'irq-core-2021-06-29' of git://git./linux/kernel/git/tip/tip

Pull irq updates from Thomas Gleixner:
 "Updates for the interrupt subsystem:

  Core changes:

   - Cleanup and simplification of common code to invoke the low level
     interrupt flow handlers when this invocation requires irqdomain
     resolution. Add the necessary core infrastructure.

   - Provide a proper interface for modular PMU drivers to set the
     interrupt affinity.

   - Add a request flag which allows to exclude interrupts from spurious
     interrupt detection. Useful especially for IPI handlers which
     always return IRQ_HANDLED which turns the spurious interrupt
     detection into a pointless waste of CPU cycles.

  Driver changes:

   - Bulk convert interrupt chip drivers to the new irqdomain low level
     flow handler invocation mechanism.

   - Add device tree bindings for the Renesas R-Car M3-W+ SoC

   - Enable modular build of the Qualcomm PDC driver

   - The usual small fixes and improvements"

* tag 'irq-core-2021-06-29' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip: (38 commits)
  dt-bindings: interrupt-controller: arm,gic-v3: Describe GICv3 optional properties
  irqchip: gic-pm: Remove redundant error log of clock bulk
  irqchip/sun4i: Remove unnecessary oom message
  irqchip/irq-imx-gpcv2: Remove unnecessary oom message
  irqchip/imgpdc: Remove unnecessary oom message
  irqchip/gic-v3-its: Remove unnecessary oom message
  irqchip/gic-v2m: Remove unnecessary oom message
  irqchip/exynos-combiner: Remove unnecessary oom message
  irqchip: Bulk conversion to generic_handle_domain_irq()
  genirq: Move non-irqdomain handle_domain_irq() handling into ARM's handle_IRQ()
  genirq: Add generic_handle_domain_irq() helper
  irqchip/nvic: Convert from handle_IRQ() to handle_domain_irq()
  irqdesc: Fix __handle_domain_irq() comment
  genirq: Use irq_resolve_mapping() to implement __handle_domain_irq() and co
  irqdomain: Introduce irq_resolve_mapping()
  irqdomain: Protect the linear revmap with RCU
  irqdomain: Cache irq_data instead of a virq number in the revmap
  irqdomain: Use struct_size() helper when allocating irqdomain
  irqdomain: Make normal and nomap irqdomains exclusive
  powerpc: Move the use of irq_domain_add_nomap() behind a config option
  ...

2 years agoMerge tag 'smp-urgent-2021-06-29' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Tue, 29 Jun 2021 19:23:02 +0000 (12:23 -0700)]
Merge tag 'smp-urgent-2021-06-29' of git://git./linux/kernel/git/tip/tip

Pull CPU hotplug fix from Thomas Gleixner:
 "A fix for the CPU hotplug and cpusets interaction:

  cpusets delegate the hotplug work to a workqueue to prevent a lock
  order inversion vs. the CPU hotplug lock. The work is not flushed
  before the hotplug operation returns which creates user visible
  inconsistent state. Prevent this by flushing the work after dropping
  CPU hotplug lock and before releasing the outer mutex which serializes
  the CPU hotplug related sysfs interface operations"

* tag 'smp-urgent-2021-06-29' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  cpu/hotplug: Cure the cpusets trainwreck

2 years agoMerge tag 'smp-core-2021-06-29' of git://git.kernel.org/pub/scm/linux/kernel/git...
Linus Torvalds [Tue, 29 Jun 2021 19:21:21 +0000 (12:21 -0700)]
Merge tag 'smp-core-2021-06-29' of git://git./linux/kernel/git/tip/tip

Pull CPU hotplug cleanup from Thomas Gleixner:
 "A simple cleanup for the CPU hotplug code to avoid per_cpu_ptr()
  reevaluation"

* tag 'smp-core-2021-06-29' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  cpu/hotplug: Simplify access to percpu cpuhp_state

2 years agoMerge tag 'printk-for-5.14' of git://git.kernel.org/pub/scm/linux/kernel/git/printk...
Linus Torvalds [Tue, 29 Jun 2021 19:07:18 +0000 (12:07 -0700)]
Merge tag 'printk-for-5.14' of git://git./linux/kernel/git/printk/linux

Pull printk updates from Petr Mladek:

 - Add %pt[RT]s modifier to vsprintf(). It overrides ISO 8601 separator
   by using ' ' (space). It produces "YYYY-mm-dd HH:MM:SS" instead of
   "YYYY-mm-ddTHH:MM:SS".

 - Correctly parse long row of numbers by sscanf() when using the field
   width. Add extensive sscanf() selftest.

 - Generalize re-entrant CPU lock that has already been used to
   serialize dump_stack() output. It is part of the ongoing printk
   rework. It will allow to remove the obsoleted printk_safe buffers and
   introduce atomic consoles.

 - Some code clean up and sparse warning fixes.

* tag 'printk-for-5.14' of git://git.kernel.org/pub/scm/linux/kernel/git/printk/linux:
  printk: fix cpu lock ordering
  lib/dump_stack: move cpu lock to printk.c
  printk: Remove trailing semicolon in macros
  random32: Fix implicit truncation warning in prandom_seed_state()
  lib: test_scanf: Remove pointless use of type_min() with unsigned types
  selftests: lib: Add wrapper script for test_scanf
  lib: test_scanf: Add tests for sscanf number conversion
  lib: vsprintf: Fix handling of number field widths in vsscanf
  lib: vsprintf: scanf: Negative number must have field width > 1
  usb: host: xhci-tegra: Switch to use %ptTs
  nilfs2: Switch to use %ptTs
  kdb: Switch to use %ptTs
  lib/vsprintf: Allow to override ISO 8601 date and time separator

2 years agoMerge tag 'hyperv-next-signed-20210629' of git://git.kernel.org/pub/scm/linux/kernel...
Linus Torvalds [Tue, 29 Jun 2021 18:21:35 +0000 (11:21 -0700)]
Merge tag 'hyperv-next-signed-20210629' of git://git./linux/kernel/git/hyperv/linux

Pull hyperv updates from Wei Liu:
 "Just a few minor enhancement patches and bug fixes"

* tag 'hyperv-next-signed-20210629' of git://git.kernel.org/pub/scm/linux/kernel/git/hyperv/linux:
  PCI: hv: Add check for hyperv_initialized in init_hv_pci_drv()
  Drivers: hv: Move Hyper-V extended capability check to arch neutral code
  drivers: hv: Fix missing error code in vmbus_connect()
  x86/hyperv: fix logical processor creation
  hv_utils: Fix passing zero to 'PTR_ERR' warning
  scsi: storvsc: Use blk_mq_unique_tag() to generate requestIDs
  Drivers: hv: vmbus: Copy packets sent by Hyper-V out of the ring buffer
  hv_balloon: Remove redundant assignment to region_start

2 years agomm,hwpoison: make get_hwpoison_page() call get_any_page()
Naoya Horiguchi [Tue, 29 Jun 2021 02:43:17 +0000 (19:43 -0700)]
mm,hwpoison: make get_hwpoison_page() call get_any_page()

__get_hwpoison_page() could fail to grab refcount by some race condition,
so it's helpful if we can handle it by retrying.  We already have retry
logic, so make get_hwpoison_page() call get_any_page() when called from
memory_failure().

As a result, get_hwpoison_page() can return negative values (i.e.  error
code), so some callers are also changed to handle error cases.
soft_offline_page() does nothing for -EBUSY because that's enough and
users in userspace can easily handle it.  unpoison_memory() is also
unchanged because it's broken and need thorough fixes (will be done
later).

Link: https://lkml.kernel.org/r/20210603233632.2964832-3-nao.horiguchi@gmail.com
Signed-off-by: Naoya Horiguchi <naoya.horiguchi@nec.com>
Cc: Oscar Salvador <osalvador@suse.de>
Cc: Muchun Song <songmuchun@bytedance.com>
Cc: Mike Kravetz <mike.kravetz@oracle.com>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Tony Luck <tony.luck@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2 years agomm,hwpoison: send SIGBUS with error virutal address
Naoya Horiguchi [Tue, 29 Jun 2021 02:43:14 +0000 (19:43 -0700)]
mm,hwpoison: send SIGBUS with error virutal address

Now an action required MCE in already hwpoisoned address surely sends a
SIGBUS to current process, but the SIGBUS doesn't convey error virtual
address.  That's not optimal for hwpoison-aware applications.

To fix the issue, make memory_failure() call kill_accessing_process(),
that does pagetable walk to find the error virtual address.  It could find
multiple virtual addresses for the same error page, and it seems hard to
tell which virtual address is correct one.  But that's rare and sending
incorrect virtual address could be better than no address.  So let's
report the first found virtual address for now.

[naoya.horiguchi@nec.com: fix walk_page_range() return]
Link: https://lkml.kernel.org/r/20210603051055.GA244241@hori.linux.bs1.fc.nec.co.jp
Link: https://lkml.kernel.org/r/20210521030156.2612074-4-nao.horiguchi@gmail.com
Signed-off-by: Naoya Horiguchi <naoya.horiguchi@nec.com>
Cc: Tony Luck <tony.luck@intel.com>
Cc: Aili Yao <yaoaili@kingsoft.com>
Cc: Oscar Salvador <osalvador@suse.de>
Cc: David Hildenbrand <david@redhat.com>
Cc: Borislav Petkov <bp@alien8.de>
Cc: Andy Lutomirski <luto@kernel.org>
Cc: Jue Wang <juew@google.com>
Cc: Borislav Petkov <bp@suse.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2 years agomm/page_alloc: split pcp->high across all online CPUs for cpuless nodes
Mel Gorman [Tue, 29 Jun 2021 02:43:11 +0000 (19:43 -0700)]
mm/page_alloc: split pcp->high across all online CPUs for cpuless nodes

Dave Hansen reported the following about Feng Tang's tests on a machine
with persistent memory onlined as a DRAM-like device.

  Feng Tang tossed these on a "Cascade Lake" system with 96 threads and
  ~512G of persistent memory and 128G of DRAM.  The PMEM is in "volatile
  use" mode and being managed via the buddy just like the normal RAM.

  The PMEM zones are big ones:

        present  65011712 = 248 G
        high       134595 = 525 M

  The PMEM nodes, of course, don't have any CPUs in them.

  With your series, the pcp->high value per-cpu is 69584 pages or about
  270MB per CPU.  Scaled up by the 96 CPU threads, that's ~26GB of
  worst-case memory in the pcps per zone, or roughly 10% of the size of
  the zone.

This should not cause a problem as such although it could trigger reclaim
due to pages being stored on per-cpu lists for CPUs remote to a node.  It
is not possible to treat cpuless nodes exactly the same as normal nodes
but the worst-case scenario can be mitigated by splitting pcp->high across
all online CPUs for cpuless memory nodes.

Link: https://lkml.kernel.org/r/20210616110743.GK30378@techsingularity.net
Suggested-by: Dave Hansen <dave.hansen@intel.com>
Signed-off-by: Mel Gorman <mgorman@techsingularity.net>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Acked-by: Dave Hansen <dave.hansen@intel.com>
Cc: Hillf Danton <hdanton@sina.com>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: "Tang, Feng" <feng.tang@intel.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2 years agomm/page_alloc: allow high-order pages to be stored on the per-cpu lists
Mel Gorman [Tue, 29 Jun 2021 02:43:08 +0000 (19:43 -0700)]
mm/page_alloc: allow high-order pages to be stored on the per-cpu lists

The per-cpu page allocator (PCP) only stores order-0 pages.  This means
that all THP and "cheap" high-order allocations including SLUB contends on
the zone->lock.  This patch extends the PCP allocator to store THP and
"cheap" high-order pages.  Note that struct per_cpu_pages increases in
size to 256 bytes (4 cache lines) on x86-64.

Note that this is not necessarily a universal performance win because of
how it is implemented.  High-order pages can cause pcp->high to be
exceeded prematurely for lower-orders so for example, a large number of
THP pages being freed could release order-0 pages from the PCP lists.
Hence, much depends on the allocation/free pattern as observed by a single
CPU to determine if caching helps or hurts a particular workload.

That said, basic performance testing passed.  The following is a netperf
UDP_STREAM test which hits the relevant patches as some of the network
allocations are high-order.

netperf-udp
                                 5.13.0-rc2             5.13.0-rc2
                           mm-pcpburst-v3r4   mm-pcphighorder-v1r7
Hmean     send-64         261.46 (   0.00%)      266.30 *   1.85%*
Hmean     send-128        516.35 (   0.00%)      536.78 *   3.96%*
Hmean     send-256       1014.13 (   0.00%)     1034.63 *   2.02%*
Hmean     send-1024      3907.65 (   0.00%)     4046.11 *   3.54%*
Hmean     send-2048      7492.93 (   0.00%)     7754.85 *   3.50%*
Hmean     send-3312     11410.04 (   0.00%)    11772.32 *   3.18%*
Hmean     send-4096     13521.95 (   0.00%)    13912.34 *   2.89%*
Hmean     send-8192     21660.50 (   0.00%)    22730.72 *   4.94%*
Hmean     send-16384    31902.32 (   0.00%)    32637.50 *   2.30%*

Functionally, a patch like this is necessary to make bulk allocation of
high-order pages work with similar performance to order-0 bulk
allocations.  The bulk allocator is not updated in this series as it would
have to be determined by bulk allocation users how they want to track the
order of pages allocated with the bulk allocator.

Link: https://lkml.kernel.org/r/20210611135753.GC30378@techsingularity.net
Signed-off-by: Mel Gorman <mgorman@techsingularity.net>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Zi Yan <ziy@nvidia.com>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Jesper Dangaard Brouer <brouer@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2 years agomm: replace CONFIG_FLAT_NODE_MEM_MAP with CONFIG_FLATMEM
Mike Rapoport [Tue, 29 Jun 2021 02:43:05 +0000 (19:43 -0700)]
mm: replace CONFIG_FLAT_NODE_MEM_MAP with CONFIG_FLATMEM

After removal of the DISCONTIGMEM memory model the FLAT_NODE_MEM_MAP
configuration option is equivalent to FLATMEM.

Drop CONFIG_FLAT_NODE_MEM_MAP and use CONFIG_FLATMEM instead.

Link: https://lkml.kernel.org/r/20210608091316.3622-10-rppt@kernel.org
Signed-off-by: Mike Rapoport <rppt@linux.ibm.com>
Acked-by: Arnd Bergmann <arnd@arndb.de>
Acked-by: David Hildenbrand <david@redhat.com>
Cc: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: Ivan Kokshaysky <ink@jurassic.park.msu.ru>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Matt Turner <mattst88@gmail.com>
Cc: Richard Henderson <rth@twiddle.net>
Cc: Vineet Gupta <vgupta@synopsys.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2 years agomm: replace CONFIG_NEED_MULTIPLE_NODES with CONFIG_NUMA
Mike Rapoport [Tue, 29 Jun 2021 02:43:01 +0000 (19:43 -0700)]
mm: replace CONFIG_NEED_MULTIPLE_NODES with CONFIG_NUMA

After removal of DISCINTIGMEM the NEED_MULTIPLE_NODES and NUMA
configuration options are equivalent.

Drop CONFIG_NEED_MULTIPLE_NODES and use CONFIG_NUMA instead.

Done with

$ sed -i 's/CONFIG_NEED_MULTIPLE_NODES/CONFIG_NUMA/' \
$(git grep -wl CONFIG_NEED_MULTIPLE_NODES)
$ sed -i 's/NEED_MULTIPLE_NODES/NUMA/' \
$(git grep -wl NEED_MULTIPLE_NODES)

with manual tweaks afterwards.

[rppt@linux.ibm.com: fix arm boot crash]
Link: https://lkml.kernel.org/r/YMj9vHhHOiCVN4BF@linux.ibm.com
Link: https://lkml.kernel.org/r/20210608091316.3622-9-rppt@kernel.org
Signed-off-by: Mike Rapoport <rppt@linux.ibm.com>
Acked-by: Arnd Bergmann <arnd@arndb.de>
Acked-by: David Hildenbrand <david@redhat.com>
Cc: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: Ivan Kokshaysky <ink@jurassic.park.msu.ru>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Matt Turner <mattst88@gmail.com>
Cc: Richard Henderson <rth@twiddle.net>
Cc: Vineet Gupta <vgupta@synopsys.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2 years agodocs: remove description of DISCONTIGMEM
Mike Rapoport [Tue, 29 Jun 2021 02:42:58 +0000 (19:42 -0700)]
docs: remove description of DISCONTIGMEM

Remove description of DISCONTIGMEM from the "Memory Models" document and
update VM sysctl description so that it won't mention DISCONIGMEM.

Link: https://lkml.kernel.org/r/20210608091316.3622-8-rppt@kernel.org
Signed-off-by: Mike Rapoport <rppt@linux.ibm.com>
Acked-by: Arnd Bergmann <arnd@arndb.de>
Reviewed-by: David Hildenbrand <david@redhat.com>
Cc: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: Ivan Kokshaysky <ink@jurassic.park.msu.ru>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Matt Turner <mattst88@gmail.com>
Cc: Richard Henderson <rth@twiddle.net>
Cc: Vineet Gupta <vgupta@synopsys.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2 years agoarch, mm: remove stale mentions of DISCONIGMEM
Mike Rapoport [Tue, 29 Jun 2021 02:42:55 +0000 (19:42 -0700)]
arch, mm: remove stale mentions of DISCONIGMEM

There are several places that mention DISCONIGMEM in comments or have
stale code guarded by CONFIG_DISCONTIGMEM.

Remove the dead code and update the comments.

Link: https://lkml.kernel.org/r/20210608091316.3622-7-rppt@kernel.org
Signed-off-by: Mike Rapoport <rppt@linux.ibm.com>
Acked-by: Arnd Bergmann <arnd@arndb.de>
Reviewed-by: David Hildenbrand <david@redhat.com>
Cc: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: Ivan Kokshaysky <ink@jurassic.park.msu.ru>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Matt Turner <mattst88@gmail.com>
Cc: Richard Henderson <rth@twiddle.net>
Cc: Vineet Gupta <vgupta@synopsys.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2 years agomm: remove CONFIG_DISCONTIGMEM
Mike Rapoport [Tue, 29 Jun 2021 02:42:52 +0000 (19:42 -0700)]
mm: remove CONFIG_DISCONTIGMEM

There are no architectures that support DISCONTIGMEM left.

Remove the configuration option and the dead code it was guarding in the
generic memory management code.

Link: https://lkml.kernel.org/r/20210608091316.3622-6-rppt@kernel.org
Signed-off-by: Mike Rapoport <rppt@linux.ibm.com>
Acked-by: Arnd Bergmann <arnd@arndb.de>
Acked-by: David Hildenbrand <david@redhat.com>
Cc: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: Ivan Kokshaysky <ink@jurassic.park.msu.ru>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Matt Turner <mattst88@gmail.com>
Cc: Richard Henderson <rth@twiddle.net>
Cc: Vineet Gupta <vgupta@synopsys.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2 years agom68k: remove support for DISCONTIGMEM
Mike Rapoport [Tue, 29 Jun 2021 02:42:49 +0000 (19:42 -0700)]
m68k: remove support for DISCONTIGMEM

DISCONTIGMEM was replaced by FLATMEM with freeing of the unused memory map
in v5.11.

Remove the support for DISCONTIGMEM entirely.

Link: https://lkml.kernel.org/r/20210608091316.3622-5-rppt@kernel.org
Signed-off-by: Mike Rapoport <rppt@linux.ibm.com>
Reviewed-by: Geert Uytterhoeven <geert@linux-m68k.org>
Acked-by: Geert Uytterhoeven <geert@linux-m68k.org>
Acked-by: Arnd Bergmann <arnd@arndb.de>
Cc: Ivan Kokshaysky <ink@jurassic.park.msu.ru>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Matt Turner <mattst88@gmail.com>
Cc: Richard Henderson <rth@twiddle.net>
Cc: Vineet Gupta <vgupta@synopsys.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2 years agoarc: remove support for DISCONTIGMEM
Mike Rapoport [Tue, 29 Jun 2021 02:42:46 +0000 (19:42 -0700)]
arc: remove support for DISCONTIGMEM

DISCONTIGMEM was replaced by FLATMEM with freeing of the unused memory map
in v5.11.

Remove the support for DISCONTIGMEM entirely.

Link: https://lkml.kernel.org/r/20210608091316.3622-4-rppt@kernel.org
Signed-off-by: Mike Rapoport <rppt@linux.ibm.com>
Acked-by: Vineet Gupta <vgupta@synopsys.com>
Acked-by: Arnd Bergmann <arnd@arndb.de>
Acked-by: David Hildenbrand <david@redhat.com>
Cc: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: Ivan Kokshaysky <ink@jurassic.park.msu.ru>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Matt Turner <mattst88@gmail.com>
Cc: Richard Henderson <rth@twiddle.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2 years agoarc: update comment about HIGHMEM implementation
Mike Rapoport [Tue, 29 Jun 2021 02:42:43 +0000 (19:42 -0700)]
arc: update comment about HIGHMEM implementation

Arc does not use DISCONTIGMEM to implement high memory, update the comment
describing how high memory works to reflect this.

Link: https://lkml.kernel.org/r/20210608091316.3622-3-rppt@kernel.org
Signed-off-by: Mike Rapoport <rppt@linux.ibm.com>
Acked-by: Vineet Gupta <vgupta@synopsys.com>
Acked-by: Arnd Bergmann <arnd@arndb.de>
Reviewed-by: David Hildenbrand <david@redhat.com>
Cc: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: Ivan Kokshaysky <ink@jurassic.park.msu.ru>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Matt Turner <mattst88@gmail.com>
Cc: Richard Henderson <rth@twiddle.net>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2 years agoalpha: remove DISCONTIGMEM and NUMA
Mike Rapoport [Tue, 29 Jun 2021 02:42:39 +0000 (19:42 -0700)]
alpha: remove DISCONTIGMEM and NUMA

Patch series "Remove DISCONTIGMEM memory model", v3.

SPARSEMEM memory model was supposed to entirely replace DISCONTIGMEM a
(long) while ago.  The last architectures that used DISCONTIGMEM were
updated to use other memory models in v5.11 and it is about the time to
entirely remove DISCONTIGMEM from the kernel.

This set removes DISCONTIGMEM from alpha, arc and m68k, simplifies memory
model selection in mm/Kconfig and replaces usage of redundant
CONFIG_NEED_MULTIPLE_NODES and CONFIG_FLAT_NODE_MEM_MAP with CONFIG_NUMA
and CONFIG_FLATMEM respectively.

I've also removed NUMA support on alpha that was BROKEN for more than 15
years.

There were also minor updates all over arch/ to remove mentions of
DISCONTIGMEM in comments and #ifdefs.

This patch (of 9):

NUMA is marked broken on alpha for more than 15 years and DISCONTIGMEM was
replaced with SPARSEMEM in v5.11.

Remove both NUMA and DISCONTIGMEM support from alpha.

Link: https://lkml.kernel.org/r/20210608091316.3622-1-rppt@kernel.org
Link: https://lkml.kernel.org/r/20210608091316.3622-2-rppt@kernel.org
Signed-off-by: Mike Rapoport <rppt@linux.ibm.com>
Acked-by: Arnd Bergmann <arnd@arndb.de>
Acked-by: David Hildenbrand <david@redhat.com>
Cc: Geert Uytterhoeven <geert@linux-m68k.org>
Cc: Ivan Kokshaysky <ink@jurassic.park.msu.ru>
Cc: Jonathan Corbet <corbet@lwn.net>
Cc: Matt Turner <mattst88@gmail.com>
Cc: Richard Henderson <rth@twiddle.net>
Cc: Vineet Gupta <vgupta@synopsys.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2 years agomm/page_alloc: move free_the_page
Mel Gorman [Tue, 29 Jun 2021 02:42:36 +0000 (19:42 -0700)]
mm/page_alloc: move free_the_page

Patch series "Allow high order pages to be stored on PCP", v2.

The per-cpu page allocator (PCP) only handles order-0 pages.  With the
series "Use local_lock for pcp protection and reduce stat overhead" and
"Calculate pcp->high based on zone sizes and active CPUs", it's now
feasible to store high-order pages on PCP lists.

This small series allows PCP to store "cheap" orders where cheap is
determined by PAGE_ALLOC_COSTLY_ORDER and THP-sized allocations.

This patch (of 2):

In the next page, free_compount_page is going to use the common helper
free_the_page.  This patch moves the definition to ease review.  No
functional change.

Link: https://lkml.kernel.org/r/20210603142220.10851-1-mgorman@techsingularity.net
Link: https://lkml.kernel.org/r/20210603142220.10851-2-mgorman@techsingularity.net
Signed-off-by: Mel Gorman <mgorman@techsingularity.net>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Jesper Dangaard Brouer <brouer@redhat.com>
Cc: Michal Hocko <mhocko@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2 years agomm/page_alloc: fix counting of managed_pages
Liu Shixin [Tue, 29 Jun 2021 02:42:33 +0000 (19:42 -0700)]
mm/page_alloc: fix counting of managed_pages

commit f63661566fad ("mm/page_alloc.c: clear out zone->lowmem_reserve[] if
the zone is empty") clears out zone->lowmem_reserve[] if zone is empty.
But when zone is not empty and sysctl_lowmem_reserve_ratio[i] is set to
zero, zone_managed_pages(zone) is not counted in the managed_pages either.
This is inconsistent with the description of lowmem_reserve, so fix it.

Link: https://lkml.kernel.org/r/20210527125707.3760259-1-liushixin2@huawei.com
Fixes: f63661566fad ("mm/page_alloc.c: clear out zone->lowmem_reserve[] if the zone is empty")
Signed-off-by: Liu Shixin <liushixin2@huawei.com>
Reported-by: yangerkun <yangerkun@huawei.com>
Reviewed-by: Baoquan He <bhe@redhat.com>
Acked-by: David Hildenbrand <david@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2 years agomm/page_alloc: improve memmap_pages dbg msg
Dong Aisheng [Tue, 29 Jun 2021 02:42:30 +0000 (19:42 -0700)]
mm/page_alloc: improve memmap_pages dbg msg

Make debug message more accurate.

Link: https://lkml.kernel.org/r/20210531091908.1738465-6-aisheng.dong@nxp.com
Signed-off-by: Dong Aisheng <aisheng.dong@nxp.com>
Reviewed-by: David Hildenbrand <david@redhat.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2 years agomm: drop SECTION_SHIFT in code comments
Dong Aisheng [Tue, 29 Jun 2021 02:42:27 +0000 (19:42 -0700)]
mm: drop SECTION_SHIFT in code comments

Actually SECTIONS_SHIFT is used in the kernel code, so the code comments
is strictly incorrect.  And since commit bbeae5b05ef6 ("mm: move page
flags layout to separate header"), SECTIONS_SHIFT definition has been
moved to include/linux/page-flags-layout.h, since code itself looks quite
straighforward, instead of moving the code comment into the new place as
well, we just simply remove it.

This also fixed a checkpatch complain derived from the original code:
WARNING: please, no space before tabs
+ * SECTIONS_SHIFT    ^I^I#bits space required to store a section #$

Link: https://lkml.kernel.org/r/20210531091908.1738465-2-aisheng.dong@nxp.com
Signed-off-by: Dong Aisheng <aisheng.dong@nxp.com>
Suggested-by: Yu Zhao <yuzhao@google.com>
Reviewed-by: Yu Zhao <yuzhao@google.com>
Cc: Andrey Konovalov <andreyknvl@gmail.com>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Kees Cook <keescook@chromium.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2 years agomm/page_alloc: introduce vm.percpu_pagelist_high_fraction
Mel Gorman [Tue, 29 Jun 2021 02:42:24 +0000 (19:42 -0700)]
mm/page_alloc: introduce vm.percpu_pagelist_high_fraction

This introduces a new sysctl vm.percpu_pagelist_high_fraction.  It is
similar to the old vm.percpu_pagelist_fraction.  The old sysctl increased
both pcp->batch and pcp->high with the higher pcp->high potentially
reducing zone->lock contention.  However, the higher pcp->batch value also
potentially increased allocation latency while the PCP was refilled.  This
sysctl only adjusts pcp->high so that zone->lock contention is potentially
reduced but allocation latency during a PCP refill remains the same.

  # grep -E "high:|batch" /proc/zoneinfo | tail -2
              high:  649
              batch: 63

  # sysctl vm.percpu_pagelist_high_fraction=8
  # grep -E "high:|batch" /proc/zoneinfo | tail -2
              high:  35071
              batch: 63

  # sysctl vm.percpu_pagelist_high_fraction=64
              high:  4383
              batch: 63

  # sysctl vm.percpu_pagelist_high_fraction=0
              high:  649
              batch: 63

[mgorman@techsingularity.net: fix documentation]
Link: https://lkml.kernel.org/r/20210528151010.GQ30378@techsingularity.net
Link: https://lkml.kernel.org/r/20210525080119.5455-7-mgorman@techsingularity.net
Signed-off-by: Mel Gorman <mgorman@techsingularity.net>
Acked-by: Dave Hansen <dave.hansen@linux.intel.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Hillf Danton <hdanton@sina.com>
Cc: Michal Hocko <mhocko@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2 years agomm/page_alloc: limit the number of pages on PCP lists when reclaim is active
Mel Gorman [Tue, 29 Jun 2021 02:42:21 +0000 (19:42 -0700)]
mm/page_alloc: limit the number of pages on PCP lists when reclaim is active

When kswapd is active then direct reclaim is potentially active.  In
either case, it is possible that a zone would be balanced if pages were
not trapped on PCP lists.  Instead of draining remote pages, simply limit
the size of the PCP lists while kswapd is active.

Link: https://lkml.kernel.org/r/20210525080119.5455-6-mgorman@techsingularity.net
Signed-off-by: Mel Gorman <mgorman@techsingularity.net>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Hillf Danton <hdanton@sina.com>
Cc: Michal Hocko <mhocko@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2 years agomm/page_alloc: scale the number of pages that are batch freed
Mel Gorman [Tue, 29 Jun 2021 02:42:18 +0000 (19:42 -0700)]
mm/page_alloc: scale the number of pages that are batch freed

When a task is freeing a large number of order-0 pages, it may acquire the
zone->lock multiple times freeing pages in batches.  This may
unnecessarily contend on the zone lock when freeing very large number of
pages.  This patch adapts the size of the batch based on the recent
pattern to scale the batch size for subsequent frees.

As the machines I used were not large enough to test this are not large
enough to illustrate a problem, a debugging patch shows patterns like the
following (slightly editted for clarity)

Baseline vanilla kernel
  time-unmap-14426   [...] free_pcppages_bulk: free   63 count  378 high  378
  time-unmap-14426   [...] free_pcppages_bulk: free   63 count  378 high  378
  time-unmap-14426   [...] free_pcppages_bulk: free   63 count  378 high  378
  time-unmap-14426   [...] free_pcppages_bulk: free   63 count  378 high  378
  time-unmap-14426   [...] free_pcppages_bulk: free   63 count  378 high  378

With patches
  time-unmap-7724    [...] free_pcppages_bulk: free  126 count  814 high  814
  time-unmap-7724    [...] free_pcppages_bulk: free  252 count  814 high  814
  time-unmap-7724    [...] free_pcppages_bulk: free  504 count  814 high  814
  time-unmap-7724    [...] free_pcppages_bulk: free  751 count  814 high  814
  time-unmap-7724    [...] free_pcppages_bulk: free  751 count  814 high  814

Link: https://lkml.kernel.org/r/20210525080119.5455-5-mgorman@techsingularity.net
Signed-off-by: Mel Gorman <mgorman@techsingularity.net>
Acked-by: Dave Hansen <dave.hansen@linux.intel.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Hillf Danton <hdanton@sina.com>
Cc: Michal Hocko <mhocko@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2 years agomm/page_alloc: adjust pcp->high after CPU hotplug events
Mel Gorman [Tue, 29 Jun 2021 02:42:15 +0000 (19:42 -0700)]
mm/page_alloc: adjust pcp->high after CPU hotplug events

The PCP high watermark is based on the number of online CPUs so the
watermarks must be adjusted during CPU hotplug.  At the time of
hot-remove, the number of online CPUs is already adjusted but during
hot-add, a delta needs to be applied to update PCP to the correct value.
After this patch is applied, the high watermarks are adjusted correctly.

  # grep high: /proc/zoneinfo  | tail -1
              high:  649
  # echo 0 > /sys/devices/system/cpu/cpu4/online
  # grep high: /proc/zoneinfo  | tail -1
              high:  664
  # echo 1 > /sys/devices/system/cpu/cpu4/online
  # grep high: /proc/zoneinfo  | tail -1
              high:  649

Link: https://lkml.kernel.org/r/20210525080119.5455-4-mgorman@techsingularity.net
Signed-off-by: Mel Gorman <mgorman@techsingularity.net>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Hillf Danton <hdanton@sina.com>
Cc: Michal Hocko <mhocko@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2 years agomm/page_alloc: disassociate the pcp->high from pcp->batch
Mel Gorman [Tue, 29 Jun 2021 02:42:12 +0000 (19:42 -0700)]
mm/page_alloc: disassociate the pcp->high from pcp->batch

The pcp high watermark is based on the batch size but there is no
relationship between them other than it is convenient to use early in
boot.

This patch takes the first step and bases pcp->high on the zone low
watermark split across the number of CPUs local to a zone while the batch
size remains the same to avoid increasing allocation latencies.  The
intent behind the default pcp->high is "set the number of PCP pages such
that if they are all full that background reclaim is not started
prematurely".

Note that in this patch the pcp->high values are adjusted after memory
hotplug events, min_free_kbytes adjustments and watermark scale factor
adjustments but not CPU hotplug events which is handled later in the
series.

On a test KVM instance;

Before grep -E "high:|batch" /proc/zoneinfo | tail -2
              high:  378
              batch: 63

After grep -E "high:|batch" /proc/zoneinfo | tail -2
              high:  649
              batch: 63

[mgorman@techsingularity.net:  fix __setup_per_zone_wmarks for parallel memory
hotplug]
Link: https://lkml.kernel.org/r/20210528105925.GN30378@techsingularity.net
Link: https://lkml.kernel.org/r/20210525080119.5455-3-mgorman@techsingularity.net
Signed-off-by: Mel Gorman <mgorman@techsingularity.net>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Dave Hansen <dave.hansen@linux.intel.com>
Cc: Hillf Danton <hdanton@sina.com>
Cc: Michal Hocko <mhocko@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2 years agomm/page_alloc: delete vm.percpu_pagelist_fraction
Mel Gorman [Tue, 29 Jun 2021 02:42:09 +0000 (19:42 -0700)]
mm/page_alloc: delete vm.percpu_pagelist_fraction

Patch series "Calculate pcp->high based on zone sizes and active CPUs", v2.

The per-cpu page allocator (PCP) is meant to reduce contention on the zone
lock but the sizing of batch and high is archaic and neither takes the
zone size into account or the number of CPUs local to a zone.  With larger
zones and more CPUs per node, the contention is getting worse.
Furthermore, the fact that vm.percpu_pagelist_fraction adjusts both batch
and high values means that the sysctl can reduce zone lock contention but
also increase allocation latencies.

This series disassociates pcp->high from pcp->batch and then scales
pcp->high based on the size of the local zone with limited impact to
reclaim and accounting for active CPUs but leaves pcp->batch static.  It
also adapts the number of pages that can be on the pcp list based on
recent freeing patterns.

The motivation is partially to adjust to larger memory sizes but is also
driven by the fact that large batches of page freeing via release_pages()
often shows zone contention as a major part of the problem.  Another is a
bug report based on an older kernel where a multi-terabyte process can
takes several minutes to exit.  A workaround was to use
vm.percpu_pagelist_fraction to increase the pcp->high value but testing
indicated that a production workload could not use the same values because
of an increase in allocation latencies.  Unfortunately, I cannot reproduce
this test case myself as the multi-terabyte machines are in active use but
it should alleviate the problem.

The series aims to address both and partially acts as a pre-requisite.
pcp only works with order-0 which is useless for SLUB (when using high
orders) and THP (unconditionally).  To store high-order pages on PCP, the
pcp->high values need to be increased first.

This patch (of 6):

The vm.percpu_pagelist_fraction is used to increase the batch and high
limits for the per-cpu page allocator (PCP).  The intent behind the sysctl
is to reduce zone lock acquisition when allocating/freeing pages but it
has a problem.  While it can decrease contention, it can also increase
latency on the allocation side due to unreasonably large batch sizes.
This leads to games where an administrator adjusts
percpu_pagelist_fraction on the fly to work around contention and
allocation latency problems.

This series aims to alleviate the problems with zone lock contention while
avoiding the allocation-side latency problems.  For the purposes of
review, it's easier to remove this sysctl now and reintroduce a similar
sysctl later in the series that deals only with pcp->high.

Link: https://lkml.kernel.org/r/20210525080119.5455-1-mgorman@techsingularity.net
Link: https://lkml.kernel.org/r/20210525080119.5455-2-mgorman@techsingularity.net
Signed-off-by: Mel Gorman <mgorman@techsingularity.net>
Acked-by: Dave Hansen <dave.hansen@linux.intel.com>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Hillf Danton <hdanton@sina.com>
Cc: Michal Hocko <mhocko@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2 years agomm: page_alloc: dump migrate-failed pages only at -EBUSY
Minchan Kim [Tue, 29 Jun 2021 02:42:06 +0000 (19:42 -0700)]
mm: page_alloc: dump migrate-failed pages only at -EBUSY

alloc_contig_dump_pages() aims for helping debugging page migration
failure by elevated page refcount compared to expected_count.  (for the
detail, please look at migrate_page_move_mapping)

However, -ENOMEM is just the case that system is under memory pressure
state, not relevant with page refcount at all.  Thus, the dumping page
list is not helpful for the debugging point of view.

Link: https://lkml.kernel.org/r/YKa2Wyo9xqIErpfa@google.com
Signed-off-by: Minchan Kim <minchan@kernel.org>
Reviewed-by: David Hildenbrand <david@redhat.com>
Cc: Suren Baghdasaryan <surenb@google.com>
Cc: John Dias <joaodias@google.com>
Cc: Michal Hocko <mhocko@suse.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2 years agomm/page_alloc: update PGFREE outside the zone lock in __free_pages_ok
Mel Gorman [Tue, 29 Jun 2021 02:42:03 +0000 (19:42 -0700)]
mm/page_alloc: update PGFREE outside the zone lock in __free_pages_ok

VM events do not need explicit protection by disabling IRQs so update the
counter with IRQs enabled in __free_pages_ok.

Link: https://lkml.kernel.org/r/20210512095458.30632-10-mgorman@techsingularity.net
Signed-off-by: Mel Gorman <mgorman@techsingularity.net>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Chuck Lever <chuck.lever@oracle.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Jesper Dangaard Brouer <brouer@redhat.com>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2 years agomm/page_alloc: avoid conflating IRQs disabled with zone->lock
Mel Gorman [Tue, 29 Jun 2021 02:42:00 +0000 (19:42 -0700)]
mm/page_alloc: avoid conflating IRQs disabled with zone->lock

Historically when freeing pages, free_one_page() assumed that callers had
IRQs disabled and the zone->lock could be acquired with spin_lock().  This
confuses the scope of what local_lock_irq is protecting and what
zone->lock is protecting in free_unref_page_list in particular.

This patch uses spin_lock_irqsave() for the zone->lock in free_one_page()
instead of relying on callers to have disabled IRQs.
free_unref_page_commit() is changed to only deal with PCP pages protected
by the local lock.  free_unref_page_list() then first frees isolated pages
to the buddy lists with free_one_page() and frees the rest of the pages to
the PCP via free_unref_page_commit().  The end result is that
free_one_page() is no longer depending on side-effects of local_lock to be
correct.

Note that this may incur a performance penalty while memory hot-remove is
running but that is not a common operation.

[lkp@intel.com: Ensure CMA pages get addded to correct pcp list]

Link: https://lkml.kernel.org/r/20210512095458.30632-9-mgorman@techsingularity.net
Signed-off-by: Mel Gorman <mgorman@techsingularity.net>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Chuck Lever <chuck.lever@oracle.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Jesper Dangaard Brouer <brouer@redhat.com>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2 years agomm/page_alloc: explicitly acquire the zone lock in __free_pages_ok
Mel Gorman [Tue, 29 Jun 2021 02:41:57 +0000 (19:41 -0700)]
mm/page_alloc: explicitly acquire the zone lock in __free_pages_ok

__free_pages_ok() disables IRQs before calling a common helper
free_one_page() that acquires the zone lock.  This is not safe according
to Documentation/locking/locktypes.rst and in this context, IRQ disabling
is not protecting a per_cpu_pages structure either or a local_lock would
be used.

This patch explicitly acquires the lock with spin_lock_irqsave instead of
relying on a helper.  This removes the last instance of local_irq_save()
in page_alloc.c.

Link: https://lkml.kernel.org/r/20210512095458.30632-8-mgorman@techsingularity.net
Signed-off-by: Mel Gorman <mgorman@techsingularity.net>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Chuck Lever <chuck.lever@oracle.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Jesper Dangaard Brouer <brouer@redhat.com>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2 years agomm/page_alloc: reduce duration that IRQs are disabled for VM counters
Mel Gorman [Tue, 29 Jun 2021 02:41:54 +0000 (19:41 -0700)]
mm/page_alloc: reduce duration that IRQs are disabled for VM counters

IRQs are left disabled for the zone and node VM event counters.  This is
unnecessary as the affected counters are allowed to race for preemmption
and IRQs.

This patch reduces the scope of IRQs being disabled via
local_[lock|unlock]_irq on !PREEMPT_RT kernels.  One
__mod_zone_freepage_state is still called with IRQs disabled.  While this
could be moved out, it's not free on all architectures as some require
IRQs to be disabled for mod_zone_page_state on !PREEMPT_RT kernels.

Link: https://lkml.kernel.org/r/20210512095458.30632-7-mgorman@techsingularity.net
Signed-off-by: Mel Gorman <mgorman@techsingularity.net>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Chuck Lever <chuck.lever@oracle.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Jesper Dangaard Brouer <brouer@redhat.com>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2 years agomm/page_alloc: batch the accounting updates in the bulk allocator
Mel Gorman [Tue, 29 Jun 2021 02:41:50 +0000 (19:41 -0700)]
mm/page_alloc: batch the accounting updates in the bulk allocator

Now that the zone_statistics are simple counters that do not require
special protection, the bulk allocator accounting updates can be batch
updated without adding too much complexity with protected RMW updates or
using xchg.

Link: https://lkml.kernel.org/r/20210512095458.30632-6-mgorman@techsingularity.net
Signed-off-by: Mel Gorman <mgorman@techsingularity.net>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Chuck Lever <chuck.lever@oracle.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Jesper Dangaard Brouer <brouer@redhat.com>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2 years agomm/vmstat: inline NUMA event counter updates
Mel Gorman [Tue, 29 Jun 2021 02:41:47 +0000 (19:41 -0700)]
mm/vmstat: inline NUMA event counter updates

__count_numa_event is small enough to be treated similarly to
__count_vm_event so inline it.

Link: https://lkml.kernel.org/r/20210512095458.30632-5-mgorman@techsingularity.net
Signed-off-by: Mel Gorman <mgorman@techsingularity.net>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Chuck Lever <chuck.lever@oracle.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Jesper Dangaard Brouer <brouer@redhat.com>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2 years agomm/vmstat: convert NUMA statistics to basic NUMA counters
Mel Gorman [Tue, 29 Jun 2021 02:41:44 +0000 (19:41 -0700)]
mm/vmstat: convert NUMA statistics to basic NUMA counters

NUMA statistics are maintained on the zone level for hits, misses, foreign
etc but nothing relies on them being perfectly accurate for functional
correctness.  The counters are used by userspace to get a general overview
of a workloads NUMA behaviour but the page allocator incurs a high cost to
maintain perfect accuracy similar to what is required for a vmstat like
NR_FREE_PAGES.  There even is a sysctl vm.numa_stat to allow userspace to
turn off the collection of NUMA statistics like NUMA_HIT.

This patch converts NUMA_HIT and friends to be NUMA events with similar
accuracy to VM events.  There is a possibility that slight errors will be
introduced but the overall trend as seen by userspace will be similar.
The counters are no longer updated from vmstat_refresh context as it is
unnecessary overhead for counters that may never be read by userspace.
Note that counters could be maintained at the node level to save space but
it would have a user-visible impact due to /proc/zoneinfo.

[lkp@intel.com: Fix misplaced closing brace for !CONFIG_NUMA]

Link: https://lkml.kernel.org/r/20210512095458.30632-4-mgorman@techsingularity.net
Signed-off-by: Mel Gorman <mgorman@techsingularity.net>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Chuck Lever <chuck.lever@oracle.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Jesper Dangaard Brouer <brouer@redhat.com>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2 years agomm/page_alloc: convert per-cpu list protection to local_lock
Mel Gorman [Tue, 29 Jun 2021 02:41:41 +0000 (19:41 -0700)]
mm/page_alloc: convert per-cpu list protection to local_lock

There is a lack of clarity of what exactly
local_irq_save/local_irq_restore protects in page_alloc.c .  It conflates
the protection of per-cpu page allocation structures with per-cpu vmstat
deltas.

This patch protects the PCP structure using local_lock which for most
configurations is identical to IRQ enabling/disabling.  The scope of the
lock is still wider than it should be but this is decreased later.

It is possible for the local_lock to be embedded safely within struct
per_cpu_pages but it adds complexity to free_unref_page_list.

[akpm@linux-foundation.org: coding style fixes]
[mgorman@techsingularity.net: work around a pahole limitation with zero-sized struct pagesets]
Link: https://lkml.kernel.org/r/20210526080741.GW30378@techsingularity.net
[lkp@intel.com: Make pagesets static]

Link: https://lkml.kernel.org/r/20210512095458.30632-3-mgorman@techsingularity.net
Signed-off-by: Mel Gorman <mgorman@techsingularity.net>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Chuck Lever <chuck.lever@oracle.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Jesper Dangaard Brouer <brouer@redhat.com>
Cc: Michal Hocko <mhocko@kernel.org>
Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Cc: Thomas Gleixner <tglx@linutronix.de>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2 years agomm/page_alloc: split per cpu page lists and zone stats
Mel Gorman [Tue, 29 Jun 2021 02:41:38 +0000 (19:41 -0700)]
mm/page_alloc: split per cpu page lists and zone stats

The PCP (per-cpu page allocator in page_alloc.c) shares locking
requirements with vmstat and the zone lock which is inconvenient and
causes some issues.  For example, the PCP list and vmstat share the same
per-cpu space meaning that it's possible that vmstat updates dirty cache
lines holding per-cpu lists across CPUs unless padding is used.  Second,
PREEMPT_RT does not want to disable IRQs for too long in the page
allocator.

This series splits the locking requirements and uses locks types more
suitable for PREEMPT_RT, reduces the time when special locking is required
for stats and reduces the time when IRQs need to be disabled on
!PREEMPT_RT kernels.

Why local_lock?  PREEMPT_RT considers the following sequence to be unsafe
as documented in Documentation/locking/locktypes.rst

   local_irq_disable();
   spin_lock(&lock);

The pcp allocator has this sequence for rmqueue_pcplist (local_irq_save)
-> __rmqueue_pcplist -> rmqueue_bulk (spin_lock).  While it's possible to
separate this out, it generally means there are points where we enable
IRQs and reenable them again immediately.  To prevent a migration and the
per-cpu pointer going stale, migrate_disable is also needed.  That is a
custom lock that is similar, but worse, than local_lock.  Furthermore, on
PREEMPT_RT, it's undesirable to leave IRQs disabled for too long.  By
converting to local_lock which disables migration on PREEMPT_RT, the
locking requirements can be separated and start moving the protections for
PCP, stats and the zone lock to PREEMPT_RT-safe equivalent locking.  As a
bonus, local_lock also means that PROVE_LOCKING does something useful.

After that, it's obvious that zone_statistics incurs too much overhead and
leaves IRQs disabled for longer than necessary on !PREEMPT_RT kernels.
zone_statistics uses perfectly accurate counters requiring IRQs be
disabled for parallel RMW sequences when inaccurate ones like vm_events
would do.  The series makes the NUMA statistics (NUMA_HIT and friends)
inaccurate counters that then require no special protection on
!PREEMPT_RT.

The bulk page allocator can then do stat updates in bulk with IRQs enabled
which should improve the efficiency.  Technically, this could have been
done without the local_lock and vmstat conversion work and the order
simply reflects the timing of when different series were implemented.

Finally, there are places where we conflate IRQs being disabled for the
PCP with the IRQ-safe zone spinlock.  The remainder of the series reduces
the scope of what is protected by disabled IRQs on !PREEMPT_RT kernels.
By the end of the series, page_alloc.c does not call local_irq_save so the
locking scope is a bit clearer.  The one exception is that modifying
NR_FREE_PAGES still happens in places where it's known the IRQs are
disabled as it's harmless for PREEMPT_RT and would be expensive to split
the locking there.

No performance data is included because despite the overhead of the stats,
it's within the noise for most workloads on !PREEMPT_RT.  However, Jesper
Dangaard Brouer ran a page allocation microbenchmark on a E5-1650 v4 @
3.60GHz CPU on the first version of this series.  Focusing on the array
variant of the bulk page allocator reveals the following.

(CPU: Intel(R) Xeon(R) CPU E5-1650 v4 @ 3.60GHz)
ARRAY variant: time_bulk_page_alloc_free_array: step=bulk size

         Baseline        Patched
 1       56.383          54.225 (+3.83%)
 2       40.047          35.492 (+11.38%)
 3       37.339          32.643 (+12.58%)
 4       35.578          30.992 (+12.89%)
 8       33.592          29.606 (+11.87%)
 16      32.362          28.532 (+11.85%)
 32      31.476          27.728 (+11.91%)
 64      30.633          27.252 (+11.04%)
 128     30.596          27.090 (+11.46%)

While this is a positive outcome, the series is more likely to be
interesting to the RT people in terms of getting parts of the PREEMPT_RT
tree into mainline.

This patch (of 9):

The per-cpu page allocator lists and the per-cpu vmstat deltas are stored
in the same struct per_cpu_pages even though vmstats have no direct impact
on the per-cpu page lists.  This is inconsistent because the vmstats for a
node are stored on a dedicated structure.  The bigger issue is that the
per_cpu_pages structure is not cache-aligned and stat updates either cache
conflict with adjacent per-cpu lists incurring a runtime cost or padding
is required incurring a memory cost.

This patch splits the per-cpu pagelists and the vmstat deltas into
separate structures.  It's mostly a mechanical conversion but some
variable renaming is done to clearly distinguish the per-cpu pages
structure (pcp) from the vmstats (pzstats).

Superficially, this appears to increase the size of the per_cpu_pages
structure but the movement of expire fills a structure hole so there is no
impact overall.

[mgorman@techsingularity.net: make it W=1 cleaner]
Link: https://lkml.kernel.org/r/20210514144622.GA3735@techsingularity.net
[mgorman@techsingularity.net: make it W=1 even cleaner]
Link: https://lkml.kernel.org/r/20210516140705.GB3735@techsingularity.net
[lkp@intel.com: check struct per_cpu_zonestat has a non-zero size]
[vbabka@suse.cz: Init zone->per_cpu_zonestats properly]

Link: https://lkml.kernel.org/r/20210512095458.30632-1-mgorman@techsingularity.net
Link: https://lkml.kernel.org/r/20210512095458.30632-2-mgorman@techsingularity.net
Signed-off-by: Mel Gorman <mgorman@techsingularity.net>
Acked-by: Vlastimil Babka <vbabka@suse.cz>
Acked-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Chuck Lever <chuck.lever@oracle.com>
Cc: Jesper Dangaard Brouer <brouer@redhat.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Michal Hocko <mhocko@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2 years agokbuild: skip per-CPU BTF generation for pahole v1.18-v1.21
Andrii Nakryiko [Tue, 29 Jun 2021 02:41:34 +0000 (19:41 -0700)]
kbuild: skip per-CPU BTF generation for pahole v1.18-v1.21

Commit "mm/page_alloc: convert per-cpu list protection to local_lock" will
introduce a zero-sized per-CPU variable, which causes pahole to generate
invalid BTF.  Only pahole versions 1.18 through 1.21 are impacted, as
before 1.18 pahole doesn't know anything about per-CPU variables, and 1.22
contains the proper fix for the issue.

Luckily, pahole 1.18 got --skip_encoding_btf_vars option disabling BTF
generation for per-CPU variables in anticipation of some unanticipated
problems.  So use this escape hatch to disable per-CPU var BTF info on
those problematic pahole versions.  Users relying on availability of
per-CPU var BTFs would need to upgrade to pahole 1.22+, but everyone won't
notice any regressions.

Link: https://lkml.kernel.org/r/20210530002536.3193829-1-andrii@kernel.org
Signed-off-by: Andrii Nakryiko <andrii@kernel.org>
Acked-by: Mel Gorman <mgorman@techsingularity.net>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Hao Luo <haoluo@google.com>
Cc: Michal Suchanek <msuchanek@suse.de>
Cc: Jiri Olsa <jolsa@kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2 years agomm/page_alloc: switch to pr_debug
Heiner Kallweit [Tue, 29 Jun 2021 02:41:31 +0000 (19:41 -0700)]
mm/page_alloc: switch to pr_debug

Having such debug messages in the dmesg log may confuse users.  Therefore
restrict debug output to cases where DEBUG is defined or dynamic debugging
is enabled for the respective code piece.

Link: https://lkml.kernel.org/r/976adb93-3041-ce63-48fc-55a6096a51c1@gmail.com
Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2 years agomm: optimise nth_page for contiguous memmap
Matthew Wilcox (Oracle) [Tue, 29 Jun 2021 02:41:28 +0000 (19:41 -0700)]
mm: optimise nth_page for contiguous memmap

If the memmap is virtually contiguous (either because we're using a
virtually mapped memmap or because we don't support a discontig memmap at
all), then we can implement nth_page() by simple addition.  Contrary to
popular belief, the compiler is not able to optimise this itself for a
vmemmap configuration.  This reduces one example user (sg.c) by four
instructions:

        struct page *page = nth_page(rsv_schp->pages[k], offset >> PAGE_SHIFT);

before:
   49 8b 45 70             mov    0x70(%r13),%rax
   48 63 c9                movslq %ecx,%rcx
   48 c1 eb 0c             shr    $0xc,%rbx
   48 8b 04 c8             mov    (%rax,%rcx,8),%rax
   48 2b 05 00 00 00 00    sub    0x0(%rip),%rax
           R_X86_64_PC32      vmemmap_base-0x4
   48 c1 f8 06             sar    $0x6,%rax
   48 01 d8                add    %rbx,%rax
   48 c1 e0 06             shl    $0x6,%rax
   48 03 05 00 00 00 00    add    0x0(%rip),%rax
           R_X86_64_PC32      vmemmap_base-0x4

after:
   49 8b 45 70             mov    0x70(%r13),%rax
   48 63 c9                movslq %ecx,%rcx
   48 c1 eb 0c             shr    $0xc,%rbx
   48 c1 e3 06             shl    $0x6,%rbx
   48 03 1c c8             add    (%rax,%rcx,8),%rbx

Link: https://lkml.kernel.org/r/20210413194625.1472345-1-willy@infradead.org
Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: David Hildenbrand <david@redhat.com>
Reviewed-by: Zi Yan <ziy@nvidia.com>
Reviewed-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Tejun Heo <tj@kernel.org>
Cc: FUJITA Tomonori <fujita.tomonori@lab.ntt.co.jp>
Cc: Douglas Gilbert <dougg@torque.net>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2 years agomm: constify page_count and page_ref_count
Matthew Wilcox (Oracle) [Tue, 29 Jun 2021 02:41:25 +0000 (19:41 -0700)]
mm: constify page_count and page_ref_count

Now that compound_head() accepts a const struct page pointer, these two
functions can be marked as not modifying the page pointer they are passed.

Link: https://lkml.kernel.org/r/20210416231531.2521383-7-willy@infradead.org
Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Reviewed-by: Vlastimil Babka <vbabka@suse.cz>
Reviewed-by: Anshuman Khandual <anshuman.khandual@arm.com>
Reviewed-by: William Kucharski <william.kucharski@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2 years agomm: constify get_pfnblock_flags_mask and get_pfnblock_migratetype
Matthew Wilcox (Oracle) [Tue, 29 Jun 2021 02:41:22 +0000 (19:41 -0700)]
mm: constify get_pfnblock_flags_mask and get_pfnblock_migratetype

The struct page is not modified by these routines, so it can be marked
const.

Link: https://lkml.kernel.org/r/20210416231531.2521383-6-willy@infradead.org
Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Reviewed-by: Vlastimil Babka <vbabka@suse.cz>
Reviewed-by: Anshuman Khandual <anshuman.khandual@arm.com>
Reviewed-by: William Kucharski <william.kucharski@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2 years agomm: make compound_head const-preserving
Matthew Wilcox (Oracle) [Tue, 29 Jun 2021 02:41:19 +0000 (19:41 -0700)]
mm: make compound_head const-preserving

If you pass a const pointer to compound_head(), you get a const pointer
back; if you pass a mutable pointer, you get a mutable pointer back.  Also
remove an unnecessary forward definition of struct page; we're about to
dereference page->compound_head, so it must already have been defined.

Link: https://lkml.kernel.org/r/20210416231531.2521383-5-willy@infradead.org
Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Reviewed-by: Vlastimil Babka <vbabka@suse.cz>
Reviewed-by: Anshuman Khandual <anshuman.khandual@arm.com>
Reviewed-by: William Kucharski <william.kucharski@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2 years agomm/page_owner: constify dump_page_owner
Matthew Wilcox (Oracle) [Tue, 29 Jun 2021 02:41:16 +0000 (19:41 -0700)]
mm/page_owner: constify dump_page_owner

dump_page_owner() only uses struct page to find the page_ext, and
lookup_page_ext() already takes a const argument.

Link: https://lkml.kernel.org/r/20210416231531.2521383-4-willy@infradead.org
Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Reviewed-by: Vlastimil Babka <vbabka@suse.cz>
Reviewed-by: Anshuman Khandual <anshuman.khandual@arm.com>
Reviewed-by: William Kucharski <william.kucharski@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2 years agomm/debug: factor PagePoisoned out of __dump_page
Matthew Wilcox (Oracle) [Tue, 29 Jun 2021 02:41:13 +0000 (19:41 -0700)]
mm/debug: factor PagePoisoned out of __dump_page

Move the PagePoisoned test into dump_page().  Skip the hex print for
poisoned pages -- we know they're full of ffffffff.  Move the reason
printing from __dump_page() to dump_page().

Link: https://lkml.kernel.org/r/20210416231531.2521383-3-willy@infradead.org
Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Reviewed-by: Vlastimil Babka <vbabka@suse.cz>
Reviewed-by: Anshuman Khandual <anshuman.khandual@arm.com>
Reviewed-by: William Kucharski <william.kucharski@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2 years agomm/page_alloc: bail out on fatal signal during reclaim/compaction retry attempt
Aaron Tomlin [Tue, 29 Jun 2021 02:41:10 +0000 (19:41 -0700)]
mm/page_alloc: bail out on fatal signal during reclaim/compaction retry attempt

A customer experienced a low-memory situation and decided to issue a
SIGKILL (i.e.  a fatal signal).  Instead of promptly terminating as one
would expect, the aforementioned task remained unresponsive.

Further investigation indicated that the task was "stuck" in the
reclaim/compaction retry loop.  Now, it does not make sense to retry
compaction when a fatal signal is pending.

In the context of try_to_compact_pages(), indeed COMPACT_SKIPPED can be
returned; albeit, not every zone, on the zone list, would be considered in
the case a fatal signal is found to be pending.  Yet, in
should_compact_retry(), given the last known compaction result, each zone,
on the zone list, can be considered/or checked (see
compaction_zonelist_suitable()).  For example, if a zone was found to
succeed, then reclaim/compaction would be tried again (notwithstanding the
above).

This patch ensures that compaction is not needlessly retried irrespective
of the last known compaction result e.g.  if it was skipped, in the
unlikely case a fatal signal is found pending.  So, OOM is at least
attempted.

Link: https://lkml.kernel.org/r/20210520142901.3371299-1-atomlin@redhat.com
Signed-off-by: Aaron Tomlin <atomlin@redhat.com>
Reviewed-by: Vlastimil Babka <vbabka@suse.cz>
Cc: Michal Hocko <mhocko@suse.com>
Cc: Matthew Wilcox <willy@infradead.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
2 years agomm: make __dump_page static
Matthew Wilcox (Oracle) [Tue, 29 Jun 2021 02:41:07 +0000 (19:41 -0700)]
mm: make __dump_page static

Patch series "Constify struct page arguments".

While working on various solutions to the 32-bit struct page size
regression, one of the problems I found was the networking stack expects
to be able to pass const struct page pointers around, and the mm doesn't
provide a lot of const-friendly functions to call.  The root tangle of
problems is that a lot of functions call VM_BUG_ON_PAGE(), which calls
dump_page(), which calls a lot of functions which don't take a const
struct page (but could be const).

This patch (of 6):

The only caller of __dump_page() now opencodes dump_page(), so remove it
as an externally visible symbol.

Link: https://lkml.kernel.org/r/20210416231531.2521383-1-willy@infradead.org
Link: https://lkml.kernel.org/r/20210416231531.2521383-2-willy@infradead.org
Signed-off-by: Matthew Wilcox (Oracle) <willy@infradead.org>
Reviewed-by: Vlastimil Babka <vbabka@suse.cz>
Reviewed-by: Anshuman Khandual <anshuman.khandual@arm.com>
Reviewed-by: William Kucharski <william.kucharski@oracle.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>