fuse: enable fuse-over-io-uring

All required parts are handled now, fuse-io-uring can
be enabled.

Signed-off-by: Bernd Schubert <bschubert@ddn.com>
Reviewed-by: Pavel Begunkov <asml.silence@gmail.com> # io_uring
Reviewed-by: Luis Henriques <luis@igalia.com>
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>

fuse: block request allocation until io-uring init is complete

Avoid races and block request allocation until io-uring
queues are ready.

This is a especially important for background requests,
as bg request completion might cause lock order inversion
of the typical queue->lock and then fc->bg_lock

    fuse_request_end
       spin_lock(&fc->bg_lock);
       flush_bg_queue
         fuse_send_one
           fuse_uring_queue_fuse_req
           spin_lock(&queue->lock);

Signed-off-by: Bernd Schubert <bernd@bsbernd.com>
Reviewed-by: Luis Henriques <luis@igalia.com>
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>

fuse: {io-uring} Prevent mount point hang on fuse-server termination

When the fuse-server terminates while the fuse-client or kernel
still has queued URING_CMDs, these commands retain references
to the struct file used by the fuse connection. This prevents
fuse_dev_release() from being invoked, resulting in a hung mount
point.

This patch addresses the issue by making queued URING_CMDs
cancelable, allowing fuse_dev_release() to proceed as expected
and preventing the mount point from hanging.

Signed-off-by: Bernd Schubert <bschubert@ddn.com>
Reviewed-by: Pavel Begunkov <asml.silence@gmail.com> # io_uring
Reviewed-by: Luis Henriques <luis@igalia.com>
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>

fuse: Allow to queue bg requests through io-uring

This prepares queueing and sending background requests through
io-uring.

Signed-off-by: Bernd Schubert <bschubert@ddn.com>
Reviewed-by: Pavel Begunkov <asml.silence@gmail.com> # io_uring
Reviewed-by: Luis Henriques <luis@igalia.com>
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>

fuse: Allow to queue fg requests through io-uring

This prepares queueing and sending foreground requests through
io-uring.

Signed-off-by: Bernd Schubert <bschubert@ddn.com>
Reviewed-by: Pavel Begunkov <asml.silence@gmail.com> # io_uring
Reviewed-by: Luis Henriques <luis@igalia.com>
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>

fuse: {io-uring} Make fuse_dev_queue_{interrupt,forget} non-static

These functions are also needed by fuse-over-io-uring.

Signed-off-by: Bernd Schubert <bschubert@ddn.com>
Reviewed-by: Luis Henriques <luis@igalia.com>
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>

fuse: {io-uring} Handle teardown of ring entries

On teardown struct file_operations::uring_cmd requests
need to be completed by calling io_uring_cmd_done().
Not completing all ring entries would result in busy io-uring
tasks giving warning messages in intervals and unreleased
struct file.

Additionally the fuse connection and with that the ring can
only get released when all io-uring commands are completed.

Completion is done with ring entries that are
a) in waiting state for new fuse requests - io_uring_cmd_done
is needed

b) already in userspace - io_uring_cmd_done through teardown
is not needed, the request can just get released. If fuse server
is still active and commits such a ring entry, fuse_uring_cmd()
already checks if the connection is active and then complete the
io-uring itself with -ENOTCONN. I.e. special handling is not
needed.

This scheme is basically represented by the ring entry state
FRRS_WAIT and FRRS_USERSPACE.

Entries in state:
- FRRS_INIT: No action needed, do not contribute to
  ring->queue_refs yet
- All other states: Are currently processed by other tasks,
  async teardown is needed and it has to wait for the two
  states above. It could be also solved without an async
  teardown task, but would require additional if conditions
  in hot code paths. Also in my personal opinion the code
  looks cleaner with async teardown.

Signed-off-by: Bernd Schubert <bschubert@ddn.com>
Reviewed-by: Pavel Begunkov <asml.silence@gmail.com> # io_uring
Reviewed-by: Luis Henriques <luis@igalia.com>
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>

fuse: Add io-uring sqe commit and fetch support

This adds support for fuse request completion through ring SQEs
(FUSE_URING_CMD_COMMIT_AND_FETCH handling). After committing
the ring entry it becomes available for new fuse requests.
Handling of requests through the ring (SQE/CQE handling)
is complete now.

Fuse request data are copied through the mmaped ring buffer,
there is no support for any zero copy yet.

Signed-off-by: Bernd Schubert <bschubert@ddn.com>
Reviewed-by: Pavel Begunkov <asml.silence@gmail.com> # io_uring
Reviewed-by: Luis Henriques <luis@igalia.com>
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>

fuse: {io-uring} Make hash-list req unique finding functions non-static

fuse-over-io-uring uses existing functions to find requests based
on their unique id - make these functions non-static.

Signed-off-by: Bernd Schubert <bschubert@ddn.com>
Reviewed-by: Joanne Koong <joannelkoong@gmail.com>
Reviewed-by: Luis Henriques <luis@igalia.com>
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>

fuse: Add fuse-io-uring handling into fuse_copy

Add special fuse-io-uring into the fuse argument
copy handler.

Signed-off-by: Bernd Schubert <bschubert@ddn.com>
Reviewed-by: Joanne Koong <joannelkoong@gmail.com>
Reviewed-by: Luis Henriques <luis@igalia.com>
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>

fuse: Make fuse_copy non static

Move 'struct fuse_copy_state' and fuse_copy_* functions
to fuse_dev_i.h to make it available for fuse-io-uring.
'copy_out_args()' is renamed to 'fuse_copy_out_args'.

Signed-off-by: Bernd Schubert <bschubert@ddn.com>
Reviewed-by: Joanne Koong <joannelkoong@gmail.com>
Reviewed-by: Luis Henriques <luis@igalia.com>
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>

fuse: {io-uring} Handle SQEs - register commands

This adds basic support for ring SQEs (with opcode=IORING_OP_URING_CMD).
For now only FUSE_IO_URING_CMD_REGISTER is handled to register queue
entries.

Signed-off-by: Bernd Schubert <bschubert@ddn.com>
Reviewed-by: Pavel Begunkov <asml.silence@gmail.com> # io_uring
Reviewed-by: Luis Henriques <luis@igalia.com>
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>

fuse: make args->in_args[0] to be always the header

This change sets up FUSE operations to always have headers in
args.in_args[0], even for opcodes without an actual header.
This step prepares for a clean separation of payload from headers,
initially it is used by fuse-over-io-uring.

For opcodes without a header, we use a zero-sized struct as a
placeholder. This approach:
- Keeps things consistent across all FUSE operations
- Will help with payload alignment later
- Avoids future issues when header sizes change

Op codes that already have an op code specific header do not
need modification.
Op codes that have neither payload nor op code headers
are not modified either (FUSE_READLINK and FUSE_DESTROY).
FUSE_BATCH_FORGET already has the header in the right place,
but is not using fuse_copy_args - as -over-uring is currently
not handling forgets it does not matter for now, but header
separation will later need special attention for that op code.

Correct the struct fuse_args->in_args array max size.

Signed-off-by: Bernd Schubert <bschubert@ddn.com>
Reviewed-by: Joanne Koong <joannelkoong@gmail.com>
Reviewed-by: Luis Henriques <luis@igalia.com>
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>

fuse: Add fuse-io-uring design documentation

[Add several documentation updates I had missed after
renaming functions and also fixes 'make htmldocs'.]

Signed-off-by: Bernd Schubert <bschubert@ddn.com>
Reviewed-by: Luis Henriques <luis@igalia.com>
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>

fuse: Move request bits

These are needed by fuse-over-io-uring.

Signed-off-by: Bernd Schubert <bschubert@ddn.com>
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Reviewed-by: Joanne Koong <joannelkoong@gmail.com>
Reviewed-by: Luis Henriques <luis@igalia.com>
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>

fuse: Move fuse_get_dev to header file

Another preparation patch, as this function will be needed by
fuse/dev.c and fuse/dev_uring.c.

Signed-off-by: Bernd Schubert <bschubert@ddn.com>
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Reviewed-by: Joanne Koong <joannelkoong@gmail.com>
Reviewed-by: Luis Henriques <luis@igalia.com>
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>

fuse: rename to fuse_dev_end_requests and make non-static

This function is needed by fuse_uring.c to clean ring queues,
so make it non static. Especially in non-static mode the function
name 'end_requests' should be prefixed with fuse_

Signed-off-by: Bernd Schubert <bschubert@ddn.com>
Reviewed-by: Josef Bacik <josef@toxicpanda.com>
Reviewed-by: Joanne Koong <joannelkoong@gmail.com>
Reviewed-by: Luis Henriques <luis@igalia.com>
Signed-off-by: Miklos Szeredi <mszeredi@redhat.com>

Linux 6.13

Merge tag 'x86_urgent_for_v6.13' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip

Pull x86 fixes from Borislav Petkov:

 - Mark serialize() noinstr so that it can be used from instrumentation-
   free code

 - Make sure FRED's RSP0 MSR is synchronized with its corresponding
   per-CPU value in order to avoid double faults in hotplug scenarios

 - Disable EXECMEM_ROX on x86 for now because it didn't receive proper
   x86 maintainers review, went in and broke a bunch of things

* tag 'x86_urgent_for_v6.13' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  x86/asm: Make serialize() always_inline
  x86/fred: Fix the FRED RSP0 MSR out of sync with its per-CPU cache
  x86: Disable EXECMEM_ROX support

Merge tag 'timers_urgent_for_v6.13' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip

Pull timer fixes from Borislav Petkov:

 - Reset hrtimers correctly when a CPU hotplug state traversal happens
   "half-ways" and leaves hrtimers not (re-)initialized properly

 - Annotate accesses to a timer group's ignore flag to prevent KCSAN
   from raising data_race warnings

 - Make sure timer group initialization is visible to timer tree walkers
   and avoid a hypothetical race

 - Fix another race between CPU hotplug and idle entry/exit where timers
   on a fully idle system are getting ignored

 - Fix a case where an ignored signal is still being handled which it
   shouldn't be

* tag 'timers_urgent_for_v6.13' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  hrtimers: Handle CPU state correctly on hotplug
  timers/migration: Annotate accesses to ignore flag
  timers/migration: Enforce group initialization visibility to tree walkers
  timers/migration: Fix another race between hotplug and idle entry/exit
  signal/posixtimers: Handle ignore/blocked sequences correctly

Merge tag 'irq_urgent_for_v6.13' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip

Pull irq fixes from Borislav Petkov:

 - Fix an OF node leak in irqchip init's error handling path

 - Fix sunxi systems to wake up from suspend with an NMI by
   pressing the power button

 - Do not spuriously enable interrupts in gic-v3 in a nested
   interrupts-off section

 - Make sure gic-v3 handles properly a failure to enter a
   low power state

* tag 'irq_urgent_for_v6.13' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  irqchip: Plug a OF node reference leak in platform_irqchip_probe()
  irqchip/sunxi-nmi: Add missing SKIP_WAKE flag
  irqchip/gic-v3-its: Don't enable interrupts in its_irq_set_vcpu_affinity()
  irqchip/gic-v3: Handle CPU_PM_ENTER_FAILED correctly

Merge tag 'sched_urgent_for_v6.13' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip

Pull scheduler fixes from Borislav Petkov:

 - Do not adjust the weight of empty group entities and avoid
   scheduling artifacts

 - Avoid scheduling lag by computing lag properly and thus address
   an EEVDF entity placement issue

* tag 'sched_urgent_for_v6.13' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip:
  sched/fair: Fix update_cfs_group() vs DELAY_DEQUEUE
  sched/fair: Fix EEVDF entity placement bug causing scheduling lag

Merge tag 'trace-v6.13-rc7-2' of git://git.kernel.org/pub/scm/linux/kernel/git/trace/linux-trace

Pull tracing fix from Steven Rostedt:
 "Fix regression in GFP output in trace events

  It was reported that the GFP flags in trace events went from human
  readable to just their hex values:

      gfp_flags=GFP_HIGHUSER_MOVABLE|__GFP_COMP to gfp_flags=0x140cca

  This was caused by a change that added the use of enums in calculating
  the GFP flags.

  As defines get translated into their values in the trace event format
  files, the user space tooling could easily convert the GFP flags into
  their symbols via the __print_flags() helper macro.

  The problem is that enums do not get converted, and the names of the
  enums show up in the format files and user space tooling cannot
  translate them.

  Add TRACE_DEFINE_ENUM() around the enums used for GFP flags which is
  the tracing infrastructure macro that informs the tracing subsystem
  what the values for enums and it can then expose that to user space"

* tag 'trace-v6.13-rc7-2' of git://git.kernel.org/pub/scm/linux/kernel/git/trace/linux-trace:
  tracing: gfp: Fix the GFP enum values shown for user space tracing tools

Merge tag 'devicetree-fixes-for-6.13-2' of git://git.kernel.org/pub/scm/linux/kernel/git/robh/linux

Pull devicetree fixes from Rob Herring:
 "Another fix and testcase to avoid the newly added WARN in the case of
  non-translatable addresses"

* tag 'devicetree-fixes-for-6.13-2' of git://git.kernel.org/pub/scm/linux/kernel/git/robh/linux:
  of/address: Fix WARN when attempting translating non-translatable addresses
  of/unittest: Add test that of_address_to_resource() fails on non-translatable address

Merge tag 'soc-fixes-6.13-4' of git://git.kernel.org/pub/scm/linux/kernel/git/soc/soc

Pull SoC fixes from Arnd Bergmann:
 "Two last minute fixes: one build issue on TI soc drivers, and a
  regression in the renesas reset controller driver"

* tag 'soc-fixes-6.13-4' of git://git.kernel.org/pub/scm/linux/kernel/git/soc/soc:
  soc: ti: pruss: Fix pruss APIs
  reset: rzg2l-usbphy-ctrl: Assign proper of node to the allocated device

Merge tag 'platform-drivers-x86-v6.13-6' of git://git.kernel.org/pub/scm/linux/kernel/git/pdx86/platform-drivers-x86

Pull x86 platform driver fixes from Ilpo Järvinen:

 - dell-uart-backlight: Fix serdev race

 - lenovo-yoga-tab2-pro-1380-fastcharger: Fix serdev race

* tag 'platform-drivers-x86-v6.13-6' of git://git.kernel.org/pub/scm/linux/kernel/git/pdx86/platform-drivers-x86:
  platform/x86: lenovo-yoga-tab2-pro-1380-fastcharger: fix serdev race
  platform/x86: dell-uart-backlight: fix serdev race

Merge tag 'mtd/fixes-for-6.13' of git://git.kernel.org/pub/scm/linux/kernel/git/mtd/linux

Pull mtd revert from Miquel Raynal:
 "Very late this cycle we identified a breakage that could potentially
  hit several spi controller drivers because of a change in the way the
  dummy cycles validity is checked.

  We do not know at the moment how to handle the situation properly, so
  we prefer to revert the faulty patch for the next release"

* tag 'mtd/fixes-for-6.13' of git://git.kernel.org/pub/scm/linux/kernel/git/mtd/linux:
  Revert "mtd: spi-nor: core: replace dummy buswidth from addr to data"

tracing: gfp: Fix the GFP enum values shown for user space tracing tools

Tracing tools like perf and trace-cmd read the /sys/kernel/tracing/events/*/*/format
files to know how to parse the data and also how to print it. For the
"print fmt" portion of that file, if anything uses an enum that is not
exported to the tracing system, user space will not be able to parse it.

The GFP flags use to be defines, and defines get translated in the print
fmt sections. But now they are converted to use enums, which is not.

The mm_page_alloc trace event format use to have:

  print fmt: "page=%p pfn=0x%lx order=%d migratetype=%d gfp_flags=%s",
    REC->pfn != -1UL ? (((struct page *)vmemmap_base) + (REC->pfn)) : ((void
    *)0), REC->pfn != -1UL ? REC->pfn : 0, REC->order, REC->migratetype,
    (REC->gfp_flags) ? __print_flags(REC->gfp_flags, "|", {( unsigned
    long)(((((((( gfp_t)(0x400u|0x800u)) | (( gfp_t)0x40u) | (( gfp_t)0x80u) |
    (( gfp_t)0x100000u)) | (( gfp_t)0x02u)) | (( gfp_t)0x08u) | (( gfp_t)0)) |
    (( gfp_t)0x40000u) | (( gfp_t)0x80000u) | (( gfp_t)0x2000u)) & ~((
    gfp_t)(0x400u|0x800u))) | (( gfp_t)0x400u)), "GFP_TRANSHUGE"}, {( unsigned
    long)((((((( gfp_t)(0x400u|0x800u)) | (( gfp_t)0x40u) | (( gfp_t)0x80u) |
    (( gfp_t)0x100000u)) | (( gfp_t)0x02u)) | (( gfp_t)0x08u) | (( gfp_t)0)) ...

Where the GFP values are shown and not their names. But after the GFP
flags were converted to use enums, it has:

  print fmt: "page=%p pfn=0x%lx order=%d migratetype=%d gfp_flags=%s",
    REC->pfn != -1UL ? (vmemmap + (REC->pfn)) : ((void *)0), REC->pfn != -1UL
    ? REC->pfn : 0, REC->order, REC->migratetype, (REC->gfp_flags) ?
    __print_flags(REC->gfp_flags, "|", {( unsigned long)((((((((
    gfp_t)(((((1UL))) << (___GFP_DIRECT_RECLAIM_BIT))|((((1UL))) <<
    (___GFP_KSWAPD_RECLAIM_BIT)))) | (( gfp_t)((((1UL))) << (___GFP_IO_BIT)))
    | (( gfp_t)((((1UL))) << (___GFP_FS_BIT))) | (( gfp_t)((((1UL))) <<
    (___GFP_HARDWALL_BIT)))) | (( gfp_t)((((1UL))) << (___GFP_HIGHMEM_BIT))))
    | (( gfp_t)((((1UL))) << (___GFP_MOVABLE_BIT))) | (( gfp_t)0)) | ((
    gfp_t)((((1UL))) << (___GFP_COMP_BIT))) ...

Where the enums names like ___GFP_KSWAPD_RECLAIM_BIT are shown and not their
values. User space has no way to convert these names to their values and
the output will fail to parse. What is shown is now:

  mm_page_alloc:  page=0xffffffff981685f3 pfn=0x1d1ac1 order=0 migratetype=1 gfp_flags=0x140cca

The TRACE_DEFINE_ENUM() macro was created to handle enums in the print fmt
files. This causes them to be replaced at boot up with the numbers, so
that user space tooling can parse it. By using this macro, the output is
back to the human readable:

  mm_page_alloc: page=0xffffffff981685f3 pfn=0x122233 order=0 migratetype=1 gfp_flags=GFP_HIGHUSER_MOVABLE|__GFP_COMP

Cc: stable@vger.kernel.org
Cc: Masami Hiramatsu <mhiramat@kernel.org>
Cc: Mark Rutland <mark.rutland@arm.com>
Cc: Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Veronika Molnarova <vmolnaro@redhat.com>
Cc: Suren Baghdasaryan <surenb@google.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Link: https://lore.kernel.org/20250116214438.749504792@goodmis.org
Reported-by: Michael Petlan <mpetlan@redhat.com>
Closes: https://lore.kernel.org/all/87be5f7c-1a0-dad-daa0-54e342efaea7@redhat.com/
Fixes: 772dd0342727c ("mm: enumerate all gfp flags")
Signed-off-by: Steven Rostedt (Google) <rostedt@goodmis.org>

Merge tag 'hwmon-for-v6.13-rc8' of git://git.kernel.org/pub/scm/linux/kernel/git/groeck/linux-staging

Pull hwmon fixes from Guenter Roeck:

 - ltc2991, tmp513: Fix problems seen when dividing negative numbers

 - drivetemp: Handle large timeouts observed on some drives

 - acpi_power_meter: Fix loading the driver on platforms without _PMD
   method

* tag 'hwmon-for-v6.13-rc8' of git://git.kernel.org/pub/scm/linux/kernel/git/groeck/linux-staging:
  hwmon: (ltc2991) Fix mixed signed/unsigned in DIV_ROUND_CLOSEST
  hwmon: (drivetemp) Set scsi command timeout to 10s
  hwmon: (acpi_power_meter) Fix a check for the return value of read_domain_devices().
  hwmon: (tmp513) Fix division of negative numbers

Merge tag 'gpio-fixes-for-v6.13' of git://git.kernel.org/pub/scm/linux/kernel/git/brgl/linux

Pull gpio fix from Bartosz Golaszewski:

 - convert regular spinlock to raw spinlock in gpio-xilinx to avoid a
   lockdep splat

* tag 'gpio-fixes-for-v6.13' of git://git.kernel.org/pub/scm/linux/kernel/git/brgl/linux:
  gpio: xilinx: Convert gpio_lock to raw spinlock

Merge tag 'i2c-for-6.13-rc8' of git://git.kernel.org/pub/scm/linux/kernel/git/wsa/linux

Pull i2c fixes from Wolfram Sang:

 - fix ref leak in the I2C core

 - fix remove notification in the address translator

 - missing error check in the pinctrl demuxer (plus a typo fix)

 - fix NAK handling when Linux is testunit target

 - fix NAK handling for the Renesas R-Car controller when it is a target

* tag 'i2c-for-6.13-rc8' of git://git.kernel.org/pub/scm/linux/kernel/git/wsa/linux:
  i2c: testunit: on errors, repeat NACK until STOP
  i2c: rcar: fix NACK handling when being a target
  i2c: mux: demux-pinctrl: correct comment
  i2c: mux: demux-pinctrl: check initial mux selection, too
  i2c: atr: Fix client detach
  i2c: core: fix reference leak in i2c_register_adapter()

Merge tag 'pmdomain-v6.13-rc2-2' of git://git.kernel.org/pub/scm/linux/kernel/git/ulfh/linux-pm

Pull pmdomain fix from Ulf Hansson:

 - imx8mp-blk-ctrl: Add missing loop break condition

* tag 'pmdomain-v6.13-rc2-2' of git://git.kernel.org/pub/scm/linux/kernel/git/ulfh/linux-pm:
  pmdomain: imx8mp-blk-ctrl: add missing loop break condition

Merge tag 'mm-hotfixes-stable-2025-01-16-21-11' of git://git.kernel.org/pub/scm/linux/kernel/git/akpm/mm

Pull misc fixes from Andrew Morton:
 "7 singleton hotfixes.  6 are MM.

  Two are cc:stable and the remainder address post-6.12 issues"

* tag 'mm-hotfixes-stable-2025-01-16-21-11' of git://git.kernel.org/pub/scm/linux/kernel/git/akpm/mm:
  ocfs2: check dir i_size in ocfs2_find_entry
  mailmap: update entry for Ethan Carter Edwards
  mm: zswap: move allocations during CPU init outside the lock
  mm: khugepaged: fix call hpage_collapse_scan_file() for anonymous vma
  mm: shmem: use signed int for version handling in casefold option
  alloc_tag: skip pgalloc_tag_swap if profiling is disabled
  mm: page_alloc: fix missed updates of lowmem_reserve in adjust_managed_page_count

Merge tag '6.13-rc7-SMB3-client-fixes' of git://git.samba.org/sfrench/cifs-2.6

Pull smb client fixes from Steve French:

 - fix double free when reconnect racing with closing session

 - fix SMB1 reconnect with password rotation

* tag '6.13-rc7-SMB3-client-fixes' of git://git.samba.org/sfrench/cifs-2.6:
  smb: client: fix double free of TCP_Server_Info::hostname
  cifs: support reconnect with alternate password for SMB1

Merge tag 'drm-fixes-2025-01-17' of https://gitlab.freedesktop.org/drm/kernel

Pull drm fixes from Dave Airlie:
 "Final(?) set of fixes for 6.13, I think the holidays finally caught up
  with everyone, the misc changes are 2 weeks worth, otherwise amdgpu
  and xe are most of it. The largest pieces is a new test so I'm not too
  worried about that.

  kunit:
   - Fix W=1 build for kunit tests

  bridge:
   - Handle YCbCr420 better in bridge code, with tests
   - itee-it6263 error handling fix

  amdgpu:
   - SMU 13 fix
   - DP MST fixes
   - DCN 3.5 fix
   - PSR fixes
   - eDP fix
   - VRR fix
   - Enforce isolation fixes
   - GFX 12 fix
   - PSP 14.x fix

  xe:
   - Add steering info support for GuC register lists
   - Add means to wait for reset and synchronous reset
   - Make changing ccs_mode a synchronous action
   - Add missing mux registers
   - Mark ComputeCS read mode as UC on iGPU, unblocking ULLS on iGPU

  i915:
   - Relax clear color alignment to 64 bytes [fb]

  v3d:
   - Fix warn when unloading v3d

  nouveau:
   - Fix cross-device fence handling in nouveau
   - Fix backlight regression for macbooks 5,1

  vmwgfx:
   - Fix BO reservation handling in vmwgfx"

* tag 'drm-fixes-2025-01-17' of https://gitlab.freedesktop.org/drm/kernel: (33 commits)
  drm/xe: Mark ComputeCS read mode as UC on iGPU
  drm/xe/oa: Add missing VISACTL mux registers
  drm/xe: make change ccs_mode a synchronous action
  drm/xe: introduce xe_gt_reset and xe_gt_wait_for_reset
  drm/xe/guc: Adding steering info support for GuC register lists
  drm/bridge: ite-it6263: Prevent error pointer dereference in probe()
  drm/v3d: Ensure job pointer is set to NULL after job completion
  drm/vmwgfx: Add new keep_resv BO param
  drm/vmwgfx: Remove busy_places
  drm/vmwgfx: Unreserve BO on error
  drm/amdgpu: fix fw attestation for MP0_14_0_{2/3}
  drm/amdgpu: always sync the GFX pipe on ctx switch
  drm/amdgpu: disable gfxoff with the compute workload on gfx12
  drm/amdgpu: Fix Circular Locking Dependency in AMDGPU GFX Isolation
  drm/i915/fb: Relax clear color alignment to 64 bytes
  drm/amd/display: Disable replay and psr while VRR is enabled
  drm/amd/display: Fix PSR-SU not support but still call the amdgpu_dm_psr_enable
  nouveau/fence: handle cross device fences properly
  drm/tests: connector: Add ycbcr_420_allowed tests
  drm/connector: hdmi: Validate supported_formats matches ycbcr_420_allowed
  ...

Merge tag 'io_uring-6.13-20250116' of git://git.kernel.dk/linux

Pull io_uring fixes from Jens Axboe:
 "One fix for the error handling in buffer cloning, and one fix for the
  ring resizing.

  Two minor followups for the latter as well.

  Both of these issues only affect 6.13, so not marked for stable"

* tag 'io_uring-6.13-20250116' of git://git.kernel.dk/linux:
  io_uring/register: cache old SQ/CQ head reading for copies
  io_uring/register: document io_register_resize_rings() shared mem usage
  io_uring/register: use stable SQ/CQ ring data during resize
  io_uring/rsrc: fixup io_clone_buffers() error handling

Merge tag 'drm-xe-fixes-2025-01-16' of https://gitlab.freedesktop.org/drm/xe/kernel into drm-fixes

Driver Changes:
- Add steering info support for GuC register lists (Jesus Narvaez)
- Add means to wait for reset and synchronous reset (Maciej)
- Make changing ccs_mode a synchronous action (Maciej)
- Add missing mux registers (Ashutosh)
- Mark ComputeCS read mode as UC on iGPU, unblocking ULLS on iGPU (Matt Brost)

Signed-off-by: Dave Airlie <airlied@redhat.com>
From: Thomas Hellstrom <thomas.hellstrom@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/Z4ll3F1anLEwCvrf@fedora

Merge tag 'trace-v6.13-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/trace/linux-trace

Pull tracing fixes from Steven Rostedt:

 - Fix a regression in the irqsoff and wakeup latency tracing

   The function graph tracer infrastructure has become generic so that
   fprobes and BPF can be based on it. As it use to only handle function
   graph tracing, it would always calculate the time the function
   entered so that it could then calculate the time it exits and give
   the length of time the function executed for. But this is not needed
   for the other users (fprobes and BPF) and reading the clock adds a
   non-negligible overhead, so the calculation was moved into the
   function graph tracer logic.

   But the irqsoff and wakeup latency tracers, when the "display-graph"
   option was set, would use the function graph tracer to calculate the
   times of functions during the latency. The movement of the calltime
   calculation made the value zero for these tracers, and the output no
   longer showed the length of time of each tracer, but instead the
   absolute timestamp of when the function returned (rettime - calltime
   where calltime is now zero).

   Have the irqsoff and wakeup latency tracers also do the calltime
   calculation as the function graph tracer does and report the proper
   length of the function timings.

 - Update the tracing display to reflect the new preempt lazy model

   When the system is configured with preempt lazy, the output of the
   trace data would state "unknown" for the current preemption model.
   Because the lazy preemption model was just added, make it known to
   the tracing subsystem too. This is just a one line change.

 - Document multiple function graph having slightly different timings

   Now that function graph tracer infrastructure is separate, this also
   allows the function graph tracer to run in multiple instances (it
   wasn't able to do so before). If two instances ran the function graph
   tracer and traced the same functions, the timings for them will be
   slightly different because each does their own timings and collects
   the timestamps differently. Document this to not have people be
   confused by it.

* tag 'trace-v6.13-rc7' of git://git.kernel.org/pub/scm/linux/kernel/git/trace/linux-trace:
  ftrace: Document that multiple function_graph tracing may have different times
  tracing: Print lazy preemption model
  tracing: Fix irqsoff and wakeup latency tracers when using function graph

Merge tag 'drm-intel-fixes-2025-01-15' of https://gitlab.freedesktop.org/drm/i915/kernel into drm-fixes

- Relax clear color alignment to 64 bytes [fb] (Ville Syrjälä)

Signed-off-by: Dave Airlie <airlied@redhat.com>
From: Tvrtko Ursulin <tursulin@igalia.com>
Link: https://patchwork.freedesktop.org/patch/msgid/Z4fdIVf68qsqIpiN@linux

drm/xe: Mark ComputeCS read mode as UC on iGPU

RING_CMD_CCTL read index should be UC on iGPU parts due to L3 caching
structure. Having this as WB blocks ULLS from being enabled. Change to
UC to unblock ULLS on iGPU.

v2:
 - Drop internal communications commnet, bspec is updated

Cc: Balasubramani Vivekanandan <balasubramani.vivekanandan@intel.com>
Cc: Michal Mrozek <michal.mrozek@intel.com>
Cc: Paulo Zanoni <paulo.r.zanoni@intel.com>
Cc: José Roberto de Souza <jose.souza@intel.com>
Cc: stable@vger.kernel.org
Fixes: 328e089bfb37 ("drm/xe: Leverage ComputeCS read L3 caching")
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Acked-by: Michal Mrozek <michal.mrozek@intel.com>
Reviewed-by: Stuart Summers <stuart.summers@intel.com>
Reviewed-by: Matt Roper <matthew.d.roper@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20250114002507.114087-1-matthew.brost@intel.com
(cherry picked from commit 758debf35b9cda5450e40996991a6e4b222899bd)
Signed-off-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>

Merge tag 'net-6.13-rc8' of git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net

Pull networking fixes from Paolo Abeni:
 "Notably this includes fixes for a few regressions spotted very
  recently. No known outstanding ones.

  Current release - regressions:

   - core: avoid CFI problems with sock priv helpers

   - xsk: bring back busy polling support

   - netpoll: ensure skb_pool list is always initialized

  Current release - new code bugs:

   - core: make page_pool_ref_netmem work with net iovs

   - ipv4: route: fix drop reason being overridden in
     ip_route_input_slow

   - udp: make rehash4 independent in udp_lib_rehash()

  Previous releases - regressions:

   - bpf: fix bpf_sk_select_reuseport() memory leak

   - openvswitch: fix lockup on tx to unregistering netdev with carrier

   - mptcp: be sure to send ack when mptcp-level window re-opens

   - eth:
      - bnxt: always recalculate features after XDP clearing, fix
        null-deref
      - mlx5: fix sub-function add port error handling
      - fec: handle page_pool_dev_alloc_pages error

  Previous releases - always broken:

   - vsock: some fixes due to transport de-assignment

   - eth:
      - ice: fix E825 initialization
      - mlx5e: fix inversion dependency warning while enabling IPsec
        tunnel
      - gtp: destroy device along with udp socket's netns dismantle.
      - xilinx: axienet: Fix IRQ coalescing packet count overflow"

* tag 'net-6.13-rc8' of git://git.kernel.org/pub/scm/linux/kernel/git/netdev/net: (44 commits)
  netdev: avoid CFI problems with sock priv helpers
  net/mlx5e: Always start IPsec sequence number from 1
  net/mlx5e: Rely on reqid in IPsec tunnel mode
  net/mlx5e: Fix inversion dependency warning while enabling IPsec tunnel
  net/mlx5: Clear port select structure when fail to create
  net/mlx5: SF, Fix add port error handling
  net/mlx5: Fix a lockdep warning as part of the write combining test
  net/mlx5: Fix RDMA TX steering prio
  net: make page_pool_ref_netmem work with net iovs
  net: ethernet: xgbe: re-add aneg to supported features in PHY quirks
  net: pcs: xpcs: actively unset DW_VR_MII_DIG_CTRL1_2G5_EN for 1G SGMII
  net: pcs: xpcs: fix DW_VR_MII_DIG_CTRL1_2G5_EN bit being set for 1G SGMII w/o inband
  selftests: net: Adapt ethtool mq tests to fix in qdisc graft
  net: fec: handle page_pool_dev_alloc_pages error
  net: netpoll: ensure skb_pool list is always initialized
  net: xilinx: axienet: Fix IRQ coalescing packet count overflow
  nfp: bpf: prevent integer overflow in nfp_bpf_event_output()
  selftests: mptcp: avoid spurious errors on disconnect
  mptcp: fix spurious wake-up on under memory pressure
  mptcp: be sure to send ack when mptcp-level window re-opens
  ...

Merge tag 'pm-6.13-rc8' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael/linux-pm

Pull power management fixes from Rafael Wysocki:
 "Update the documentation of cpuidle governors that does not match the
  code any more after previous functional changes (Rafael Wysocki) and
  fix up the cpufreq Kconfig file broken inadvertently by a previous
  update (Viresh Kumar)"

* tag 'pm-6.13-rc8' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael/linux-pm:
  cpufreq: Move endif to the end of Kconfig file
  cpuidle: teo: Update documentation after previous changes
  cpuidle: menu: Update documentation after previous changes

Merge tag 'acpi-6.13-rc8' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael/linux-pm

Pull ACPI fix from Rafael Wysocki:
 "Prevent acpi_video_device_EDID() from returning a pointer to a memory
  region that should not be passed to kfree() which causes one of its
  users to crash randomly on attempts to free it (Chris Bainbridge)"

* tag 'acpi-6.13-rc8' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael/linux-pm:
  ACPI: video: Fix random crashes due to bad kfree()

Merge tag 'for-6.13-rc7-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/kdave/linux

Pull btrfs fix from David Sterba:

 - handle d_path() errors when canonicalizing device mapper paths during
   device scan

* tag 'for-6.13-rc7-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/kdave/linux:
  btrfs: add the missing error handling inside get_canonical_dev_path

x86/asm: Make serialize() always_inline

In order to allow serialize() to be used from noinstr code, make it
__always_inline.

Fixes: 0ef8047b737d ("x86/static-call: provide a way to do very early static-call updates")
Closes: https://lore.kernel.org/oe-kbuild-all/202412181756.aJvzih2K-lkp@intel.com/
Reported-by: kernel test robot <lkp@intel.com>
Signed-off-by: Juergen Gross <jgross@suse.com>
Signed-off-by: Borislav Petkov (AMD) <bp@alien8.de>
Link: https://lore.kernel.org/r/20241218100918.22167-1-jgross@suse.com

pmdomain: imx8mp-blk-ctrl: add missing loop break condition

Currently imx8mp_blk_ctrl_remove() will continue the for loop
until an out-of-bounds exception occurs.

pstate: 60000005 (nZCv daif -PAN -UAO -TCO -DIT -SSBS BTYPE=--)
pc : dev_pm_domain_detach+0x8/0x48
lr : imx8mp_blk_ctrl_shutdown+0x58/0x90
sp : ffffffc084f8bbf0
x29: ffffffc084f8bbf0 x28: ffffff80daf32ac0 x27: 0000000000000000
x26: ffffffc081658d78 x25: 0000000000000001 x24: ffffffc08201b028
x23: ffffff80d0db9490 x22: ffffffc082340a78 x21: 00000000000005b0
x20: ffffff80d19bc180 x19: 000000000000000a x18: ffffffffffffffff
x17: ffffffc080a39e08 x16: ffffffc080a39c98 x15: 4f435f464f006c72
x14: 0000000000000004 x13: ffffff80d0172110 x12: 0000000000000000
x11: ffffff80d0537740 x10: ffffff80d05376c0 x9 : ffffffc0808ed2d8
x8 : ffffffc084f8bab0 x7 : 0000000000000000 x6 : 0000000000000000
x5 : ffffff80d19b9420 x4 : fffffffe03466e60 x3 : 0000000080800077
x2 : 0000000000000000 x1 : 0000000000000001 x0 : 0000000000000000
Call trace:
 dev_pm_domain_detach+0x8/0x48
 platform_shutdown+0x2c/0x48
 device_shutdown+0x158/0x268
 kernel_restart_prepare+0x40/0x58
 kernel_kexec+0x58/0xe8
 __do_sys_reboot+0x198/0x258
 __arm64_sys_reboot+0x2c/0x40
 invoke_syscall+0x5c/0x138
 el0_svc_common.constprop.0+0x48/0xf0
 do_el0_svc+0x24/0x38
 el0_svc+0x38/0xc8
 el0t_64_sync_handler+0x120/0x130
 el0t_64_sync+0x190/0x198
Code: 8128c2d0 ffffffc0 aa1e03e9 d503201f

Fixes: 556f5cf9568a ("soc: imx: add i.MX8MP HSIO blk-ctrl")
Cc: stable@vger.kernel.org
Signed-off-by: Xiaolei Wang <xiaolei.wang@windriver.com>
Reviewed-by: Lucas Stach <l.stach@pengutronix.de>
Reviewed-by: Fabio Estevam <festevam@gmail.com>
Reviewed-by: Frank Li <Frank.Li@nxp.com>
Link: https://lore.kernel.org/r/20250115014118.4086729-1-xiaolei.wang@windriver.com
Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>

Merge branch 'pm-cpufreq'

Merge a cpufreq fix for 6.13:

 - Fix cpufreq Kconfig breakage after previous changes (Viresh Kumar).

* pm-cpufreq:
  cpufreq: Move endif to the end of Kconfig file

netdev: avoid CFI problems with sock priv helpers

Li Li reports that casting away callback type may cause issues
for CFI. Let's generate a small wrapper for each callback,
to make sure compiler sees the anticipated types.

Reported-by: Li Li <dualli@chromium.org>
Link: https://lore.kernel.org/CANBPYPjQVqmzZ4J=rVQX87a9iuwmaetULwbK_5_3YWk2eGzkaA@mail.gmail.com
Fixes: 170aafe35cb9 ("netdev: support binding dma-buf to netdevice")
Signed-off-by: Jakub Kicinski <kuba@kernel.org>
Reviewed-by: Mina Almasry <almasrymina@google.com>
Link: https://patch.msgid.link/20250115161436.648646-1-kuba@kernel.org
Signed-off-by: Paolo Abeni <pabeni@redhat.com>

hrtimers: Handle CPU state correctly on hotplug

Consider a scenario where a CPU transitions from CPUHP_ONLINE to halfway
through a CPU hotunplug down to CPUHP_HRTIMERS_PREPARE, and then back to
CPUHP_ONLINE:

Since hrtimers_prepare_cpu() does not run, cpu_base.hres_active remains set
to 1 throughout. However, during a CPU unplug operation, the tick and the
clockevents are shut down at CPUHP_AP_TICK_DYING. On return to the online
state, for instance CFS incorrectly assumes that the hrtick is already
active, and the chance of the clockevent device to transition to oneshot
mode is also lost forever for the CPU, unless it goes back to a lower state
than CPUHP_HRTIMERS_PREPARE once.

This round-trip reveals another issue; cpu_base.online is not set to 1
after the transition, which appears as a WARN_ON_ONCE in enqueue_hrtimer().

Aside of that, the bulk of the per CPU state is not reset either, which
means there are dangling pointers in the worst case.

Address this by adding a corresponding startup() callback, which resets the
stale per CPU state and sets the online flag.

[ tglx: Make the new callback unconditionally available, remove the online
   modification in the prepare() callback and clear the remaining
   state in the starting callback instead of the prepare callback ]

Fixes: 5c0930ccaad5 ("hrtimers: Push pending hrtimers away from outgoing CPU earlier")
Signed-off-by: Koichiro Den <koichiro.den@canonical.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: stable@vger.kernel.org
Link: https://lore.kernel.org/all/20241220134421.3809834-1-koichiro.den@canonical.com

timers/migration: Annotate accesses to ignore flag

The group's ignore flag is:

_ read under the group's lock (idle entry, remote expiry)
_ turned on/off under the group's lock (idle entry, remote expiry)
_ turned on locklessly on idle exit

When idle entry or remote expiry clear the "ignore" flag of a group, the
operation must be synchronized against other concurrent idle entry or
remote expiry to make sure the related group timer is never missed. To
enforce this synchronization, both "ignore" clear and read are
performed under the group lock.

On the contrary, whether idle entry or remote expiry manage to observe
the "ignore" flag turned on by a CPU exiting idle is a matter of
optimization. If that flag set is missed or cleared concurrently, the
worst outcome is a migrator wasting time remotely handling a "ghost"
timer. This is why the ignore flag can be set locklessly.

Unfortunately, the related lockless accesses are bare and miss
appropriate annotations. KCSAN rightfully complains:

 BUG: KCSAN: data-race in __tmigr_cpu_activate / print_report

 write to 0xffff88842fc28004 of 1 bytes by task 0 on cpu 0:
 __tmigr_cpu_activate
 tmigr_cpu_activate
 timer_clear_idle
 tick_nohz_restart_sched_tick
 tick_nohz_idle_exit
 do_idle
 cpu_startup_entry
 kernel_init
 do_initcalls
 clear_bss
 reserve_bios_regions
 common_startup_64

 read to 0xffff88842fc28004 of 1 bytes by task 0 on cpu 1:
 print_report
 kcsan_report_known_origin
 kcsan_setup_watchpoint
 tmigr_next_groupevt
 tmigr_update_events
 tmigr_inactive_up
 __walk_groups+0x50/0x77
 walk_groups
 __tmigr_cpu_deactivate
 tmigr_cpu_deactivate
 __get_next_timer_interrupt
 timer_base_try_to_set_idle
 tick_nohz_stop_tick
 tick_nohz_idle_stop_tick
 cpuidle_idle_call
 do_idle

Although the relevant accesses could be marked as data_race(), the
"ignore" flag being read several times within the same
tmigr_update_events() function is confusing and error prone. Prefer
reading it once in that function and make use of similar/paired accesses
elsewhere with appropriate comments when necessary.

Reported-by: kernel test robot <oliver.sang@intel.com>
Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Link: https://lore.kernel.org/all/20250114231507.21672-4-frederic@kernel.org
Closes: https://lore.kernel.org/oe-lkp/202501031612.62e0c498-lkp@intel.com

timers/migration: Enforce group initialization visibility to tree walkers

Commit 2522c84db513 ("timers/migration: Fix another race between hotplug
and idle entry/exit") fixed yet another race between idle exit and CPU
hotplug up leading to a wrong "0" value migrator assigned to the top
level. However there is yet another situation that remains unhandled:

         [GRP0:0]
      migrator  = TMIGR_NONE
      active    = NONE
      groupmask = 1
      /     \      \
     0       1     2..7
   idle      idle   idle

0) The system is fully idle.

         [GRP0:0]
      migrator  = CPU 0
      active    = CPU 0
      groupmask = 1
      /     \      \
     0       1     2..7
   active   idle   idle

1) CPU 0 is activating. It has done the cmpxchg on the top's ->migr_state
but it hasn't yet returned to __walk_groups().

         [GRP0:0]
      migrator  = CPU 0
      active    = CPU 0, CPU 1
      groupmask = 1
      /     \      \
     0       1     2..7
   active  active  idle

2) CPU 1 is activating. CPU 0 stays the migrator (still stuck in
__walk_groups(), delayed by #VMEXIT for example).

                    [GRP1:0]
                migrator = TMIGR_NONE
                active   = NONE
                groupmask = 1
             /                   \
         [GRP0:0]                  [GRP0:1]
      migrator  = CPU 0           migrator = TMIGR_NONE
      active    = CPU 0, CPU1     active   = NONE
      groupmask = 1               groupmask = 2
      /     \      \
     0       1     2..7                   8
   active  active  idle                !online

3) CPU 8 is preparing to boot. CPUHP_TMIGR_PREPARE is being ran by CPU 1
which has created the GRP0:1 and the new top GRP1:0 connected to GRP0:1
and GRP0:0. CPU 1 hasn't yet propagated its activation up to GRP1:0.

                    [GRP1:0]
               migrator = GRP0:0
               active   = GRP0:0
               groupmask = 1
             /                   \
         [GRP0:0]                  [GRP0:1]
     migrator  = CPU 0           migrator = TMIGR_NONE
     active    = CPU 0, CPU1     active   = NONE
     groupmask = 1               groupmask = 2
     /     \      \
    0       1     2..7                   8
  active  active  idle                !online

4) CPU 0 finally resumed after its #VMEXIT. It's in __walk_groups()
returning from tmigr_cpu_active(). The new top GRP1:0 is visible and
fetched and the pre-initialized groupmask of GRP0:0 is also visible.
As a result tmigr_active_up() is called to GRP1:0 with GRP0:0 as active
and migrator. CPU 0 is returning to __walk_groups() but suffers again
a #VMEXIT.

                    [GRP1:0]
               migrator = GRP0:0
               active   = GRP0:0
               groupmask = 1
             /                   \
         [GRP0:0]                  [GRP0:1]
     migrator  = CPU 0           migrator = TMIGR_NONE
     active    = CPU 0, CPU1     active   = NONE
     groupmask = 1               groupmask = 2
     /     \      \
    0       1     2..7                   8
  active  active  idle                 !online

5) CPU 1 propagates its activation of GRP0:0 to GRP1:0. This has no
   effect since CPU 0 did it already.

                    [GRP1:0]
               migrator = GRP0:0
               active   = GRP0:0, GRP0:1
               groupmask = 1
             /                   \
         [GRP0:0]                  [GRP0:1]
     migrator  = CPU 0           migrator = CPU 8
     active    = CPU 0, CPU1     active   = CPU 8
     groupmask = 1               groupmask = 2
     /     \      \                     \
    0       1     2..7                   8
  active  active  idle                 active

6) CPU 1 links CPU 8 to its group. CPU 8 boots and goes through
   CPUHP_AP_TMIGR_ONLINE which propagates activation.

                                   [GRP2:0]
                              migrator = TMIGR_NONE
                              active   = NONE
                              groupmask = 1
                             /                \
                    [GRP1:0]                    [GRP1:1]
               migrator = GRP0:0              migrator = TMIGR_NONE
               active   = GRP0:0, GRP0:1      active   = NONE
               groupmask = 1                  groupmask = 2
             /                   \
         [GRP0:0]                  [GRP0:1]                [GRP0:2]
     migrator  = CPU 0           migrator = CPU 8        migrator = TMIGR_NONE
     active    = CPU 0, CPU1     active   = CPU 8        active   = NONE
     groupmask = 1               groupmask = 2           groupmask = 0
     /     \      \                     \
    0       1     2..7                   8                  64
  active  active  idle                 active             !online

7) CPU 64 is booting. CPUHP_TMIGR_PREPARE is being ran by CPU 1
which has created the GRP1:1, GRP0:2 and the new top GRP2:0 connected to
GRP1:1 and GRP1:0. CPU 1 hasn't yet propagated its activation up to
GRP2:0.

                                   [GRP2:0]
                              migrator = 0 (!!!)
                              active   = NONE
                              groupmask = 1
                             /                \
                    [GRP1:0]                    [GRP1:1]
               migrator = GRP0:0              migrator = TMIGR_NONE
               active   = GRP0:0, GRP0:1      active   = NONE
               groupmask = 1                  groupmask = 2
             /                   \
         [GRP0:0]                  [GRP0:1]                [GRP0:2]
     migrator  = CPU 0           migrator = CPU 8        migrator = TMIGR_NONE
     active    = CPU 0, CPU1     active   = CPU 8        active   = NONE
     groupmask = 1               groupmask = 2           groupmask = 0
     /     \      \                     \
    0       1     2..7                   8                  64
  active  active  idle                 active             !online

8) CPU 0 finally resumed after its #VMEXIT. It's in __walk_groups()
returning from tmigr_cpu_active(). The new top GRP2:0 is visible and
fetched but the pre-initialized groupmask of GRP1:0 is not because no
ordering made its initialization visible. As a result tmigr_active_up()
may be called to GRP2:0 with a "0" child's groumask. Leaving the timers
ignored for ever when the system is fully idle.

The race is highly theoretical and perhaps impossible in practice but
the groupmask of the child is not the only concern here as the whole
initialization of the child is not guaranteed to be visible to any
tree walker racing against hotplug (idle entry/exit, remote handling,
etc...). Although the current code layout seem to be resilient to such
hazards, this doesn't tell much about the future.

Fix this with enforcing address dependency between group initialization
and the write/read to the group's parent's pointer. Fortunately that
doesn't involve any barrier addition in the fast paths.

Fixes: 10a0e6f3d3db ("timers/migration: Move hierarchy setup into cpuhotplug prepare callback")
Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: stable@vger.kernel.org
Link: https://lore.kernel.org/all/20250114231507.21672-3-frederic@kernel.org

timers/migration: Fix another race between hotplug and idle entry/exit

Commit 10a0e6f3d3db ("timers/migration: Move hierarchy setup into
cpuhotplug prepare callback") fixed a race between idle exit and CPU
hotplug up leading to a wrong "0" value migrator assigned to the top
level. However there is still a situation that remains unhandled:

         [GRP0:0]
        migrator  = TMIGR_NONE
        active    = NONE
        groupmask = 0
        /     \      \
       0       1     2..7
     idle      idle   idle

0) The system is fully idle.

         [GRP0:0]
        migrator  = CPU 0
        active    = CPU 0
        groupmask = 0
        /     \      \
       0       1     2..7
     active   idle   idle

1) CPU 0 is activating. It has done the cmpxchg on the top's ->migr_state
but it hasn't yet returned to __walk_groups().

         [GRP0:0]
        migrator  = CPU 0
        active    = CPU 0, CPU 1
        groupmask = 0
        /     \      \
       0       1     2..7
     active  active  idle

2) CPU 1 is activating. CPU 0 stays the migrator (still stuck in
__walk_groups(), delayed by #VMEXIT for example).

                 [GRP1:0]
              migrator = TMIGR_NONE
              active   = NONE
              groupmask = 0
              /                  \
        [GRP0:0]                      [GRP0:1]
       migrator  = CPU 0           migrator = TMIGR_NONE
       active    = CPU 0, CPU1     active   = NONE
       groupmask = 2               groupmask = 1
       /     \      \
      0       1     2..7                   8
    active  active  idle              !online

3) CPU 8 is preparing to boot. CPUHP_TMIGR_PREPARE is being ran by CPU 1
which has created the GRP0:1 and the new top GRP1:0 connected to GRP0:1
and GRP0:0. The groupmask of GRP0:0 is now 2. CPU 1 hasn't yet
propagated its activation up to GRP1:0.

                 [GRP1:0]
              migrator = 0 (!!!)
              active   = NONE
              groupmask = 0
              /                  \
        [GRP0:0]                  [GRP0:1]
       migrator  = CPU 0           migrator = TMIGR_NONE
       active    = CPU 0, CPU1     active   = NONE
       groupmask = 2               groupmask = 1
       /     \      \
      0       1     2..7                   8
    active  active  idle                !online

4) CPU 0 finally resumed after its #VMEXIT. It's in __walk_groups()
returning from tmigr_cpu_active(). The new top GRP1:0 is visible and
fetched but the freshly updated groupmask of GRP0:0 may not be visible
due to lack of ordering! As a result tmigr_active_up() is called to
GRP0:0 with a child's groupmask of "0". This buggy "0" groupmask then
becomes the migrator for GRP1:0 forever. As a result, timers on a fully
idle system get ignored.

One possible fix would be to define TMIGR_NONE as "0" so that such a
race would have no effect. And after all TMIGR_NONE doesn't need to be
anything else. However this would leave an uncomfortable state machine
where gears happen not to break by chance but are vulnerable to future
modifications.

Keep TMIGR_NONE as is instead and pre-initialize to "1" the groupmask of
any newly created top level. This groupmask is guaranteed to be visible
upon fetching the corresponding group for the 1st time:

_ By the upcoming CPU thanks to CPU hotplug synchronization between the
  control CPU (BP) and the booting one (AP).

_ By the control CPU since the groupmask and parent pointers are
  initialized locally.

_ By all CPUs belonging to the same group than the control CPU because
  they must wait for it to ever become idle before needing to walk to
  the new top. The cmpcxhg() on ->migr_state then makes sure its
  groupmask is visible.

With this pre-initialization, it is guaranteed that if a future top level
is linked to an old one, it is walked through with a valid groupmask.

Fixes: 10a0e6f3d3db ("timers/migration: Move hierarchy setup into cpuhotplug prepare callback")
Signed-off-by: Frederic Weisbecker <frederic@kernel.org>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: stable@vger.kernel.org
Link: https://lore.kernel.org/all/20250114231507.21672-2-frederic@kernel.org

Merge branch 'mlx5-misc-fixes-2025-01-15'

Tariq Toukan says:

====================
mlx5 misc fixes 2025-01-15

This patchset provides misc bug fixes from the team to the mlx5 core and
Eth drivers.
====================

Link: https://patch.msgid.link/20250115113910.1990174-1-tariqt@nvidia.com
Signed-off-by: Paolo Abeni <pabeni@redhat.com>

net/mlx5e: Always start IPsec sequence number from 1

According to RFC4303, section "3.3.3. Sequence Number Generation",
the first packet sent using a given SA will contain a sequence
number of 1.

This is applicable to both ESN and non-ESN mode, which was not covered
in commit mentioned in Fixes line.

Fixes: 3d42c8cc67a8 ("net/mlx5e: Ensure that IPsec sequence packet number starts from 1")
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Reviewed-by: Jacob Keller <jacob.e.keller@intel.com>
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>

net/mlx5e: Rely on reqid in IPsec tunnel mode

All packet offloads SAs have reqid in it to make sure they have
corresponding policy. While it is not strictly needed for transparent
mode, it is extremely important in tunnel mode. In that mode, policy and
SAs have different match criteria.

Policy catches the whole subnet addresses, and SA catches the tunnel gateways
addresses. The source address of such tunnel is not known during egress packet
traversal in flow steering as it is added only after successful encryption.

As reqid is required for packet offload and it is unique for every SA,
we can safely rely on it only.

The output below shows the configured egress policy and SA by strongswan:

[leonro@vm ~]$ sudo ip x s
src 192.169.101.2 dst 192.169.101.1
        proto esp spi 0xc88b7652 reqid 1 mode tunnel
        replay-window 0 flag af-unspec esn
        aead rfc4106(gcm(aes)) 0xe406a01083986e14d116488549094710e9c57bc6 128
        anti-replay esn context:
         seq-hi 0x0, seq 0x0, oseq-hi 0x0, oseq 0x0
         replay_window 1, bitmap-length 1
         00000000
        crypto offload parameters: dev eth2 dir out mode packet

[leonro@064 ~]$ sudo ip x p
src 192.170.0.0/16 dst 192.170.0.0/16
        dir out priority 383615 ptype main
        tmpl src 192.169.101.2 dst 192.169.101.1
                proto esp spi 0xc88b7652 reqid 1 mode tunnel
        crypto offload parameters: dev eth2 mode packet

Fixes: b3beba1fb404 ("net/mlx5e: Allow policies with reqid 0, to support IKE policy holes")
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Reviewed-by: Jacob Keller <jacob.e.keller@intel.com>
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>

net/mlx5e: Fix inversion dependency warning while enabling IPsec tunnel

Attempt to enable IPsec packet offload in tunnel mode in debug kernel
generates the following kernel panic, which is happening due to two
issues:
1. In SA add section, the should be _bh() variant when marking SA mode.
2. There is not needed flush_workqueue in SA delete routine. It is not
needed as at this stage as it is removed from SADB and the running work
will be canceled later in SA free.

 =====================================================
 WARNING: SOFTIRQ-safe -> SOFTIRQ-unsafe lock order detected
 6.12.0+ #4 Not tainted
 -----------------------------------------------------
 charon/1337 [HC0[0]:SC0[4]:HE1:SE0] is trying to acquire:
 ffff88810f365020 (&xa->xa_lock#24){+.+.}-{3:3}, at: mlx5e_xfrm_del_state+0xca/0x1e0 [mlx5_core]

 and this task is already holding:
 ffff88813e0f0d48 (&x->lock){+.-.}-{3:3}, at: xfrm_state_delete+0x16/0x30
 which would create a new lock dependency:
  (&x->lock){+.-.}-{3:3} -> (&xa->xa_lock#24){+.+.}-{3:3}

 but this new dependency connects a SOFTIRQ-irq-safe lock:
  (&x->lock){+.-.}-{3:3}

 ... which became SOFTIRQ-irq-safe at:
   lock_acquire+0x1be/0x520
   _raw_spin_lock_bh+0x34/0x40
   xfrm_timer_handler+0x91/0xd70
   __hrtimer_run_queues+0x1dd/0xa60
   hrtimer_run_softirq+0x146/0x2e0
   handle_softirqs+0x266/0x860
   irq_exit_rcu+0x115/0x1a0
   sysvec_apic_timer_interrupt+0x6e/0x90
   asm_sysvec_apic_timer_interrupt+0x16/0x20
   default_idle+0x13/0x20
   default_idle_call+0x67/0xa0
   do_idle+0x2da/0x320
   cpu_startup_entry+0x50/0x60
   start_secondary+0x213/0x2a0
   common_startup_64+0x129/0x138

 to a SOFTIRQ-irq-unsafe lock:
  (&xa->xa_lock#24){+.+.}-{3:3}

 ... which became SOFTIRQ-irq-unsafe at:
 ...
   lock_acquire+0x1be/0x520
   _raw_spin_lock+0x2c/0x40
   xa_set_mark+0x70/0x110
   mlx5e_xfrm_add_state+0xe48/0x2290 [mlx5_core]
   xfrm_dev_state_add+0x3bb/0xd70
   xfrm_add_sa+0x2451/0x4a90
   xfrm_user_rcv_msg+0x493/0x880
   netlink_rcv_skb+0x12e/0x380
   xfrm_netlink_rcv+0x6d/0x90
   netlink_unicast+0x42f/0x740
   netlink_sendmsg+0x745/0xbe0
   __sock_sendmsg+0xc5/0x190
   __sys_sendto+0x1fe/0x2c0
   __x64_sys_sendto+0xdc/0x1b0
   do_syscall_64+0x6d/0x140
   entry_SYSCALL_64_after_hwframe+0x4b/0x53

 other info that might help us debug this:

  Possible interrupt unsafe locking scenario:

        CPU0                    CPU1
        ----                    ----
   lock(&xa->xa_lock#24);
                                local_irq_disable();
                                lock(&x->lock);
                                lock(&xa->xa_lock#24);
   <Interrupt>
     lock(&x->lock);

  *** DEADLOCK ***

 2 locks held by charon/1337:
  #0: ffffffff87f8f858 (&net->xfrm.xfrm_cfg_mutex){+.+.}-{4:4}, at: xfrm_netlink_rcv+0x5e/0x90
  #1: ffff88813e0f0d48 (&x->lock){+.-.}-{3:3}, at: xfrm_state_delete+0x16/0x30

 the dependencies between SOFTIRQ-irq-safe lock and the holding lock:
 -> (&x->lock){+.-.}-{3:3} ops: 29 {
    HARDIRQ-ON-W at:
                     lock_acquire+0x1be/0x520
                     _raw_spin_lock_bh+0x34/0x40
                     xfrm_alloc_spi+0xc0/0xe60
                     xfrm_alloc_userspi+0x5f6/0xbc0
                     xfrm_user_rcv_msg+0x493/0x880
                     netlink_rcv_skb+0x12e/0x380
                     xfrm_netlink_rcv+0x6d/0x90
                     netlink_unicast+0x42f/0x740
                     netlink_sendmsg+0x745/0xbe0
                     __sock_sendmsg+0xc5/0x190
                     __sys_sendto+0x1fe/0x2c0
                     __x64_sys_sendto+0xdc/0x1b0
                     do_syscall_64+0x6d/0x140
                     entry_SYSCALL_64_after_hwframe+0x4b/0x53
    IN-SOFTIRQ-W at:
                     lock_acquire+0x1be/0x520
                     _raw_spin_lock_bh+0x34/0x40
                     xfrm_timer_handler+0x91/0xd70
                     __hrtimer_run_queues+0x1dd/0xa60
                     hrtimer_run_softirq+0x146/0x2e0
                     handle_softirqs+0x266/0x860
                     irq_exit_rcu+0x115/0x1a0
                     sysvec_apic_timer_interrupt+0x6e/0x90
                     asm_sysvec_apic_timer_interrupt+0x16/0x20
                     default_idle+0x13/0x20
                     default_idle_call+0x67/0xa0
                     do_idle+0x2da/0x320
                     cpu_startup_entry+0x50/0x60
                     start_secondary+0x213/0x2a0
                     common_startup_64+0x129/0x138
    INITIAL USE at:
                    lock_acquire+0x1be/0x520
                    _raw_spin_lock_bh+0x34/0x40
                    xfrm_alloc_spi+0xc0/0xe60
                    xfrm_alloc_userspi+0x5f6/0xbc0
                    xfrm_user_rcv_msg+0x493/0x880
                    netlink_rcv_skb+0x12e/0x380
                    xfrm_netlink_rcv+0x6d/0x90
                    netlink_unicast+0x42f/0x740
                    netlink_sendmsg+0x745/0xbe0
                    __sock_sendmsg+0xc5/0x190
                    __sys_sendto+0x1fe/0x2c0
                    __x64_sys_sendto+0xdc/0x1b0
                    do_syscall_64+0x6d/0x140
                    entry_SYSCALL_64_after_hwframe+0x4b/0x53
  }
  ... key      at: [<ffffffff87f9cd20>] __key.18+0x0/0x40

 the dependencies between the lock to be acquired
  and SOFTIRQ-irq-unsafe lock:
 -> (&xa->xa_lock#24){+.+.}-{3:3} ops: 9 {
    HARDIRQ-ON-W at:
                     lock_acquire+0x1be/0x520
                     _raw_spin_lock_bh+0x34/0x40
                     mlx5e_xfrm_add_state+0xc5b/0x2290 [mlx5_core]
                     xfrm_dev_state_add+0x3bb/0xd70
                     xfrm_add_sa+0x2451/0x4a90
                     xfrm_user_rcv_msg+0x493/0x880
                     netlink_rcv_skb+0x12e/0x380
                     xfrm_netlink_rcv+0x6d/0x90
                     netlink_unicast+0x42f/0x740
                     netlink_sendmsg+0x745/0xbe0
                     __sock_sendmsg+0xc5/0x190
                     __sys_sendto+0x1fe/0x2c0
                     __x64_sys_sendto+0xdc/0x1b0
                     do_syscall_64+0x6d/0x140
                     entry_SYSCALL_64_after_hwframe+0x4b/0x53
    SOFTIRQ-ON-W at:
                     lock_acquire+0x1be/0x520
                     _raw_spin_lock+0x2c/0x40
                     xa_set_mark+0x70/0x110
                     mlx5e_xfrm_add_state+0xe48/0x2290 [mlx5_core]
                     xfrm_dev_state_add+0x3bb/0xd70
                     xfrm_add_sa+0x2451/0x4a90
                     xfrm_user_rcv_msg+0x493/0x880
                     netlink_rcv_skb+0x12e/0x380
                     xfrm_netlink_rcv+0x6d/0x90
                     netlink_unicast+0x42f/0x740
                     netlink_sendmsg+0x745/0xbe0
                     __sock_sendmsg+0xc5/0x190
                     __sys_sendto+0x1fe/0x2c0
                     __x64_sys_sendto+0xdc/0x1b0
                     do_syscall_64+0x6d/0x140
                     entry_SYSCALL_64_after_hwframe+0x4b/0x53
    INITIAL USE at:
                    lock_acquire+0x1be/0x520
                    _raw_spin_lock_bh+0x34/0x40
                    mlx5e_xfrm_add_state+0xc5b/0x2290 [mlx5_core]
                    xfrm_dev_state_add+0x3bb/0xd70
                    xfrm_add_sa+0x2451/0x4a90
                    xfrm_user_rcv_msg+0x493/0x880
                    netlink_rcv_skb+0x12e/0x380
                    xfrm_netlink_rcv+0x6d/0x90
                    netlink_unicast+0x42f/0x740
                    netlink_sendmsg+0x745/0xbe0
                    __sock_sendmsg+0xc5/0x190
                    __sys_sendto+0x1fe/0x2c0
                    __x64_sys_sendto+0xdc/0x1b0
                    do_syscall_64+0x6d/0x140
                    entry_SYSCALL_64_after_hwframe+0x4b/0x53
  }
  ... key      at: [<ffffffffa078ff60>] __key.48+0x0/0xfffffffffff210a0 [mlx5_core]
  ... acquired at:
    __lock_acquire+0x30a0/0x5040
    lock_acquire+0x1be/0x520
    _raw_spin_lock_bh+0x34/0x40
    mlx5e_xfrm_del_state+0xca/0x1e0 [mlx5_core]
    xfrm_dev_state_delete+0x90/0x160
    __xfrm_state_delete+0x662/0xae0
    xfrm_state_delete+0x1e/0x30
    xfrm_del_sa+0x1c2/0x340
    xfrm_user_rcv_msg+0x493/0x880
    netlink_rcv_skb+0x12e/0x380
    xfrm_netlink_rcv+0x6d/0x90
    netlink_unicast+0x42f/0x740
    netlink_sendmsg+0x745/0xbe0
    __sock_sendmsg+0xc5/0x190
    __sys_sendto+0x1fe/0x2c0
    __x64_sys_sendto+0xdc/0x1b0
    do_syscall_64+0x6d/0x140
    entry_SYSCALL_64_after_hwframe+0x4b/0x53

 stack backtrace:
 CPU: 7 UID: 0 PID: 1337 Comm: charon Not tainted 6.12.0+ #4
 Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS rel-1.13.0-0-gf21b5a4aeb02-prebuilt.qemu.org 04/01/2014
 Call Trace:
  <TASK>
  dump_stack_lvl+0x74/0xd0
  check_irq_usage+0x12e8/0x1d90
  ? print_shortest_lock_dependencies_backwards+0x1b0/0x1b0
  ? check_chain_key+0x1bb/0x4c0
  ? __lockdep_reset_lock+0x180/0x180
  ? check_path.constprop.0+0x24/0x50
  ? mark_lock+0x108/0x2fb0
  ? print_circular_bug+0x9b0/0x9b0
  ? mark_lock+0x108/0x2fb0
  ? print_usage_bug.part.0+0x670/0x670
  ? check_prev_add+0x1c4/0x2310
  check_prev_add+0x1c4/0x2310
  __lock_acquire+0x30a0/0x5040
  ? lockdep_set_lock_cmp_fn+0x190/0x190
  ? lockdep_set_lock_cmp_fn+0x190/0x190
  lock_acquire+0x1be/0x520
  ? mlx5e_xfrm_del_state+0xca/0x1e0 [mlx5_core]
  ? lockdep_hardirqs_on_prepare+0x400/0x400
  ? __xfrm_state_delete+0x5f0/0xae0
  ? lock_downgrade+0x6b0/0x6b0
  _raw_spin_lock_bh+0x34/0x40
  ? mlx5e_xfrm_del_state+0xca/0x1e0 [mlx5_core]
  mlx5e_xfrm_del_state+0xca/0x1e0 [mlx5_core]
  xfrm_dev_state_delete+0x90/0x160
  __xfrm_state_delete+0x662/0xae0
  xfrm_state_delete+0x1e/0x30
  xfrm_del_sa+0x1c2/0x340
  ? xfrm_get_sa+0x250/0x250
  ? check_chain_key+0x1bb/0x4c0
  xfrm_user_rcv_msg+0x493/0x880
  ? copy_sec_ctx+0x270/0x270
  ? check_chain_key+0x1bb/0x4c0
  ? lockdep_set_lock_cmp_fn+0x190/0x190
  ? lockdep_set_lock_cmp_fn+0x190/0x190
  netlink_rcv_skb+0x12e/0x380
  ? copy_sec_ctx+0x270/0x270
  ? netlink_ack+0xd90/0xd90
  ? netlink_deliver_tap+0xcd/0xb60
  xfrm_netlink_rcv+0x6d/0x90
  netlink_unicast+0x42f/0x740
  ? netlink_attachskb+0x730/0x730
  ? lock_acquire+0x1be/0x520
  netlink_sendmsg+0x745/0xbe0
  ? netlink_unicast+0x740/0x740
  ? __might_fault+0xbb/0x170
  ? netlink_unicast+0x740/0x740
  __sock_sendmsg+0xc5/0x190
  ? fdget+0x163/0x1d0
  __sys_sendto+0x1fe/0x2c0
  ? __x64_sys_getpeername+0xb0/0xb0
  ? do_user_addr_fault+0x856/0xe30
  ? lock_acquire+0x1be/0x520
  ? __task_pid_nr_ns+0x117/0x410
  ? lock_downgrade+0x6b0/0x6b0
  __x64_sys_sendto+0xdc/0x1b0
  ? lockdep_hardirqs_on_prepare+0x284/0x400
  do_syscall_64+0x6d/0x140
  entry_SYSCALL_64_after_hwframe+0x4b/0x53
 RIP: 0033:0x7f7d31291ba4
 Code: 7d e8 89 4d d4 e8 4c 42 f7 ff 44 8b 4d d0 4c 8b 45 c8 89 c3 44 8b 55 d4 8b 7d e8 b8 2c 00 00 00 48 8b 55 d8 48 8b 75 e0 0f 05 <48> 3d 00 f0 ff ff 77 34 89 df 48 89 45 e8 e8 99 42 f7 ff 48 8b 45
 RSP: 002b:00007f7d2ccd94f0 EFLAGS: 00000297 ORIG_RAX: 000000000000002c
 RAX: ffffffffffffffda RBX: 0000000000000001 RCX: 00007f7d31291ba4
 RDX: 0000000000000028 RSI: 00007f7d2ccd96a0 RDI: 000000000000000a
 RBP: 00007f7d2ccd9530 R08: 00007f7d2ccd9598 R09: 000000000000000c
 R10: 0000000000000000 R11: 0000000000000297 R12: 0000000000000028
 R13: 00007f7d2ccd9598 R14: 00007f7d2ccd96a0 R15: 00000000000000e1
  </TASK>

Fixes: 4c24272b4e2b ("net/mlx5e: Listen to ARP events to update IPsec L2 headers in tunnel mode")
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>

net/mlx5: Clear port select structure when fail to create

Clear the port select structure on error so no stale values left after
definers are destroyed. That's because the mlx5_lag_destroy_definers()
always try to destroy all lag definers in the tt_map, so in the flow
below lag definers get double-destroyed and cause kernel crash:

  mlx5_lag_port_sel_create()
    mlx5_lag_create_definers()
      mlx5_lag_create_definer()     <- Failed on tt 1
        mlx5_lag_destroy_definers() <- definers[tt=0] gets destroyed
  mlx5_lag_port_sel_create()
    mlx5_lag_create_definers()
      mlx5_lag_create_definer()     <- Failed on tt 0
        mlx5_lag_destroy_definers() <- definers[tt=0] gets double-destroyed

 Unable to handle kernel NULL pointer dereference at virtual address 0000000000000008
 Mem abort info:
   ESR = 0x0000000096000005
   EC = 0x25: DABT (current EL), IL = 32 bits
   SET = 0, FnV = 0
   EA = 0, S1PTW = 0
   FSC = 0x05: level 1 translation fault
 Data abort info:
   ISV = 0, ISS = 0x00000005, ISS2 = 0x00000000
   CM = 0, WnR = 0, TnD = 0, TagAccess = 0
   GCS = 0, Overlay = 0, DirtyBit = 0, Xs = 0
 user pgtable: 64k pages, 48-bit VAs, pgdp=0000000112ce2e00
 [0000000000000008] pgd=0000000000000000, p4d=0000000000000000, pud=0000000000000000
 Internal error: Oops: 0000000096000005 [#1] PREEMPT SMP
 Modules linked in: iptable_raw bonding ip_gre ip6_gre gre ip6_tunnel tunnel6 geneve ip6_udp_tunnel udp_tunnel ipip tunnel4 ip_tunnel rdma_ucm(OE) rdma_cm(OE) iw_cm(OE) ib_ipoib(OE) ib_cm(OE) ib_umad(OE) mlx5_ib(OE) ib_uverbs(OE) mlx5_fwctl(OE) fwctl(OE) mlx5_core(OE) mlxdevm(OE) ib_core(OE) mlxfw(OE) memtrack(OE) mlx_compat(OE) openvswitch nsh nf_conncount psample xt_conntrack xt_MASQUERADE nf_conntrack_netlink nfnetlink xfrm_user xfrm_algo xt_addrtype iptable_filter iptable_nat nf_nat nf_conntrack nf_defrag_ipv6 nf_defrag_ipv4 br_netfilter bridge stp llc netconsole overlay efi_pstore sch_fq_codel zram ip_tables crct10dif_ce qemu_fw_cfg fuse ipv6 crc_ccitt [last unloaded: mlx_compat(OE)]
  CPU: 3 UID: 0 PID: 217 Comm: kworker/u53:2 Tainted: G           OE      6.11.0+ #2
  Tainted: [O]=OOT_MODULE, [E]=UNSIGNED_MODULE
  Hardware name: QEMU KVM Virtual Machine, BIOS 0.0.0 02/06/2015
  Workqueue: mlx5_lag mlx5_do_bond_work [mlx5_core]
  pstate: 60400005 (nZCv daif +PAN -UAO -TCO -DIT -SSBS BTYPE=--)
  pc : mlx5_del_flow_rules+0x24/0x2c0 [mlx5_core]
  lr : mlx5_lag_destroy_definer+0x54/0x100 [mlx5_core]
  sp : ffff800085fafb00
  x29: ffff800085fafb00 x28: ffff0000da0c8000 x27: 0000000000000000
  x26: ffff0000da0c8000 x25: ffff0000da0c8000 x24: ffff0000da0c8000
  x23: ffff0000c31f81a0 x22: 0400000000000000 x21: ffff0000da0c8000
  x20: 0000000000000000 x19: 0000000000000001 x18: 0000000000000000
  x17: 0000000000000000 x16: 0000000000000000 x15: 0000ffff8b0c9350
  x14: 0000000000000000 x13: ffff800081390d18 x12: ffff800081dc3cc0
  x11: 0000000000000001 x10: 0000000000000b10 x9 : ffff80007ab7304c
  x8 : ffff0000d00711f0 x7 : 0000000000000004 x6 : 0000000000000190
  x5 : ffff00027edb3010 x4 : 0000000000000000 x3 : 0000000000000000
  x2 : ffff0000d39b8000 x1 : ffff0000d39b8000 x0 : 0400000000000000
  Call trace:
   mlx5_del_flow_rules+0x24/0x2c0 [mlx5_core]
   mlx5_lag_destroy_definer+0x54/0x100 [mlx5_core]
   mlx5_lag_destroy_definers+0xa0/0x108 [mlx5_core]
   mlx5_lag_port_sel_create+0x2d4/0x6f8 [mlx5_core]
   mlx5_activate_lag+0x60c/0x6f8 [mlx5_core]
   mlx5_do_bond_work+0x284/0x5c8 [mlx5_core]
   process_one_work+0x170/0x3e0
   worker_thread+0x2d8/0x3e0
   kthread+0x11c/0x128
   ret_from_fork+0x10/0x20
  Code: a9025bf5 aa0003f6 a90363f7 f90023f9 (f9400400)
  ---[ end trace 0000000000000000 ]---

Fixes: dc48516ec7d3 ("net/mlx5: Lag, add support to create definers for LAG")
Signed-off-by: Mark Zhang <markzhang@nvidia.com>
Reviewed-by: Leon Romanovsky <leonro@nvidia.com>
Reviewed-by: Mark Bloch <mbloch@nvidia.com>
Reviewed-by: Jacob Keller <jacob.e.keller@intel.com>
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>

net/mlx5: SF, Fix add port error handling

If failed to add SF, error handling doesn't delete the SF from the
SF table. But the hw resources are deleted. So when unload driver,
hw resources will be deleted again. Firmware will report syndrome
0x68def3 which means "SF is not allocated can not deallocate".

Fix it by delete SF from SF table if failed to add SF.

Fixes: 2597ee190b4e ("net/mlx5: Call mlx5_sf_id_erase() once in mlx5_sf_dealloc()")
Signed-off-by: Chris Mi <cmi@nvidia.com>
Reviewed-by: Shay Drori <shayd@nvidia.com>
Reviewed-by: Jacob Keller <jacob.e.keller@intel.com>
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>

net/mlx5: Fix a lockdep warning as part of the write combining test

Fix a lockdep warning [1] observed during the write combining test.

The warning indicates a potential nested lock scenario that could lead
to a deadlock.

However, this is a false positive alarm because the SF lock and its
parent lock are distinct ones.

The lockdep confusion arises because the locks belong to the same object
class (i.e., struct mlx5_core_dev).

To resolve this, the code has been refactored to avoid taking both
locks. Instead, only the parent lock is acquired.

[1]
raw_ethernet_bw/2118 is trying to acquire lock:
[  213.619032] ffff88811dd75e08 (&dev->wc_state_lock){+.+.}-{3:3}, at:
               mlx5_wc_support_get+0x18c/0x210 [mlx5_core]
[  213.620270]
[  213.620270] but task is already holding lock:
[  213.620943] ffff88810b585e08 (&dev->wc_state_lock){+.+.}-{3:3}, at:
               mlx5_wc_support_get+0x10c/0x210 [mlx5_core]
[  213.622045]
[  213.622045] other info that might help us debug this:
[  213.622778]  Possible unsafe locking scenario:
[  213.622778]
[  213.623465]        CPU0
[  213.623815]        ----
[  213.624148]   lock(&dev->wc_state_lock);
[  213.624615]   lock(&dev->wc_state_lock);
[  213.625071]
[  213.625071]  *** DEADLOCK ***
[  213.625071]
[  213.625805]  May be due to missing lock nesting notation
[  213.625805]
[  213.626522] 4 locks held by raw_ethernet_bw/2118:
[  213.627019]  #0: ffff88813f80d578 (&uverbs_dev->disassociate_srcu){.+.+}-{0:0},
                at: ib_uverbs_ioctl+0xc4/0x170 [ib_uverbs]
[  213.628088]  #1: ffff88810fb23930 (&file->hw_destroy_rwsem){.+.+}-{3:3},
                at: ib_init_ucontext+0x2d/0xf0 [ib_uverbs]
[  213.629094]  #2: ffff88810fb23878 (&file->ucontext_lock){+.+.}-{3:3},
                at: ib_init_ucontext+0x49/0xf0 [ib_uverbs]
[  213.630106]  #3: ffff88810b585e08 (&dev->wc_state_lock){+.+.}-{3:3},
                at: mlx5_wc_support_get+0x10c/0x210 [mlx5_core]
[  213.631185]
[  213.631185] stack backtrace:
[  213.631718] CPU: 1 UID: 0 PID: 2118 Comm: raw_ethernet_bw Not tainted
               6.12.0-rc7_internal_net_next_mlx5_89a0ad0 #1
[  213.632722] Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS
               rel-1.13.0-0-gf21b5a4aeb02-prebuilt.qemu.org 04/01/2014
[  213.633785] Call Trace:
[  213.634099]
[  213.634393]  dump_stack_lvl+0x7e/0xc0
[  213.634806]  print_deadlock_bug+0x278/0x3c0
[  213.635265]  __lock_acquire+0x15f4/0x2c40
[  213.635712]  lock_acquire+0xcd/0x2d0
[  213.636120]  ? mlx5_wc_support_get+0x18c/0x210 [mlx5_core]
[  213.636722]  ? mlx5_ib_enable_lb+0x24/0xa0 [mlx5_ib]
[  213.637277]  __mutex_lock+0x81/0xda0
[  213.637697]  ? mlx5_wc_support_get+0x18c/0x210 [mlx5_core]
[  213.638305]  ? mlx5_wc_support_get+0x18c/0x210 [mlx5_core]
[  213.638902]  ? rcu_read_lock_sched_held+0x3f/0x70
[  213.639400]  ? mlx5_wc_support_get+0x18c/0x210 [mlx5_core]
[  213.640016]  mlx5_wc_support_get+0x18c/0x210 [mlx5_core]
[  213.640615]  set_ucontext_resp+0x68/0x2b0 [mlx5_ib]
[  213.641144]  ? debug_mutex_init+0x33/0x40
[  213.641586]  mlx5_ib_alloc_ucontext+0x18e/0x7b0 [mlx5_ib]
[  213.642145]  ib_init_ucontext+0xa0/0xf0 [ib_uverbs]
[  213.642679]  ib_uverbs_handler_UVERBS_METHOD_GET_CONTEXT+0x95/0xc0
                [ib_uverbs]
[  213.643426]  ? _copy_from_user+0x46/0x80
[  213.643878]  ib_uverbs_cmd_verbs+0xa6b/0xc80 [ib_uverbs]
[  213.644426]  ? ib_uverbs_handler_UVERBS_METHOD_INVOKE_WRITE+0x130/0x130
               [ib_uverbs]
[  213.645213]  ? __lock_acquire+0xa99/0x2c40
[  213.645675]  ? lock_acquire+0xcd/0x2d0
[  213.646101]  ? ib_uverbs_ioctl+0xc4/0x170 [ib_uverbs]
[  213.646625]  ? reacquire_held_locks+0xcf/0x1f0
[  213.647102]  ? do_user_addr_fault+0x45d/0x770
[  213.647586]  ib_uverbs_ioctl+0xe0/0x170 [ib_uverbs]
[  213.648102]  ? ib_uverbs_ioctl+0xc4/0x170 [ib_uverbs]
[  213.648632]  __x64_sys_ioctl+0x4d3/0xaa0
[  213.649060]  ? do_user_addr_fault+0x4a8/0x770
[  213.649528]  do_syscall_64+0x6d/0x140
[  213.649947]  entry_SYSCALL_64_after_hwframe+0x4b/0x53
[  213.650478] RIP: 0033:0x7fa179b0737b
[  213.650893] Code: ff ff ff 85 c0 79 9b 49 c7 c4 ff ff ff ff 5b 5d 4c
               89 e0 41 5c c3 66 0f 1f 84 00 00 00 00 00 f3 0f 1e fa b8
               10 00 00 00 0f 05 <48> 3d 01 f0 ff ff 73 01 c3 48 8b 0d
               7d 2a 0f 00 f7 d8 64 89 01 48
[  213.652619] RSP: 002b:00007ffd2e6d46e8 EFLAGS: 00000246 ORIG_RAX:
               0000000000000010
[  213.653390] RAX: ffffffffffffffda RBX: 00007ffd2e6d47f8 RCX:
               00007fa179b0737b
[  213.654084] RDX: 00007ffd2e6d47e0 RSI: 00000000c0181b01 RDI:
               0000000000000003
[  213.654767] RBP: 00007ffd2e6d47c0 R08: 00007fa1799be010 R09:
               0000000000000002
[  213.655453] R10: 00007ffd2e6d4960 R11: 0000000000000246 R12:
               00007ffd2e6d487c
[  213.656170] R13: 0000000000000027 R14: 0000000000000001 R15:
               00007ffd2e6d4f70

Fixes: d98995b4bf98 ("net/mlx5: Reimplement write combining test")
Signed-off-by: Yishai Hadas <yishaih@nvidia.com>
Reviewed-by: Michael Guralnik <michaelgur@nvidia.com>
Reviewed-by: Larysa Zaremba <larysa.zaremba@intel.com>
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>

net/mlx5: Fix RDMA TX steering prio

User added steering rules at RDMA_TX were being added to the first prio,
which is the counters prio.
Fix that so that they are correctly added to the BYPASS_PRIO instead.

Fixes: 24670b1a3166 ("net/mlx5: Add support for RDMA TX steering")
Signed-off-by: Patrisious Haddad <phaddad@nvidia.com>
Reviewed-by: Mark Bloch <mbloch@nvidia.com>
Reviewed-by: Jacob Keller <jacob.e.keller@intel.com>
Signed-off-by: Tariq Toukan <tariqt@nvidia.com>
Signed-off-by: Paolo Abeni <pabeni@redhat.com>

Merge tag 'amd-drm-fixes-6.13-2025-01-15' of https://gitlab.freedesktop.org/agd5f/linux into drm-fixes

amd-drm-fixes-6.13-2025-01-15:

amdgpu:
- SMU 13 fix
- DP MST fixes
- DCN 3.5 fix
- PSR fixes
- eDP fix
- VRR fix
- Enforce isolation fixes
- GFX 12 fix
- PSP 14.x fix

Signed-off-by: Dave Airlie <airlied@redhat.com>
From: Alex Deucher <alexander.deucher@amd.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20250115151602.210704-1-alexander.deucher@amd.com

ocfs2: check dir i_size in ocfs2_find_entry

syz reports an out of bounds read:

==================================================================
BUG: KASAN: slab-out-of-bounds in ocfs2_match fs/ocfs2/dir.c:334
[inline]
BUG: KASAN: slab-out-of-bounds in ocfs2_search_dirblock+0x283/0x6e0
fs/ocfs2/dir.c:367
Read of size 1 at addr ffff88804d8b9982 by task syz-executor.2/14802

CPU: 0 UID: 0 PID: 14802 Comm: syz-executor.2 Not tainted 6.13.0-rc4 #2
Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.15.0-1
04/01/2014
Sched_ext: serialise (enabled+all), task: runnable_at=-10ms
Call Trace:
<TASK>
__dump_stack lib/dump_stack.c:94 [inline]
dump_stack_lvl+0x229/0x350 lib/dump_stack.c:120
print_address_description mm/kasan/report.c:378 [inline]
print_report+0x164/0x530 mm/kasan/report.c:489
kasan_report+0x147/0x180 mm/kasan/report.c:602
ocfs2_match fs/ocfs2/dir.c:334 [inline]
ocfs2_search_dirblock+0x283/0x6e0 fs/ocfs2/dir.c:367
ocfs2_find_entry_id fs/ocfs2/dir.c:414 [inline]
ocfs2_find_entry+0x1143/0x2db0 fs/ocfs2/dir.c:1078
ocfs2_find_files_on_disk+0x18e/0x530 fs/ocfs2/dir.c:1981
ocfs2_lookup_ino_from_name+0xb6/0x110 fs/ocfs2/dir.c:2003
ocfs2_lookup+0x30a/0xd40 fs/ocfs2/namei.c:122
lookup_open fs/namei.c:3627 [inline]
open_last_lookups fs/namei.c:3748 [inline]
path_openat+0x145a/0x3870 fs/namei.c:3984
do_filp_open+0xe9/0x1c0 fs/namei.c:4014
do_sys_openat2+0x135/0x1d0 fs/open.c:1402
do_sys_open fs/open.c:1417 [inline]
__do_sys_openat fs/open.c:1433 [inline]
__se_sys_openat fs/open.c:1428 [inline]
__x64_sys_openat+0x15d/0x1c0 fs/open.c:1428
do_syscall_x64 arch/x86/entry/common.c:52 [inline]
do_syscall_64+0xf6/0x210 arch/x86/entry/common.c:83
entry_SYSCALL_64_after_hwframe+0x77/0x7f
RIP: 0033:0x7f01076903ad
Code: c3 e8 a7 2b 00 00 0f 1f 80 00 00 00 00 f3 0f 1e fa 48 89 f8 48 89
f7 48 89 d6 48 89 ca 4d 89 c2 4d 89 c8 4c 8b 4c 24 08 0f 05 <48> 3d 01
f0 ff ff 73 01 c3 48 c7 c1 b0 ff ff ff f7 d8 64 89 01 48
RSP: 002b:00007f01084acfc8 EFLAGS: 00000246 ORIG_RAX: 0000000000000101
RAX: ffffffffffffffda RBX: 00007f01077cbf80 RCX: 00007f01076903ad
RDX: 0000000000105042 RSI: 0000000020000080 RDI: ffffffffffffff9c
RBP: 00007f01077cbf80 R08: 0000000000000000 R09: 0000000000000000
R10: 00000000000001ff R11: 0000000000000246 R12: 0000000000000000
R13: 00007f01077cbf80 R14: 00007f010764fc90 R15: 00007f010848d000
</TASK>
==================================================================

And a general protection fault in ocfs2_prepare_dir_for_insert:

==================================================================
loop0: detected capacity change from 0 to 32768
JBD2: Ignoring recovery information on journal
ocfs2: Mounting device (7,0) on (node local, slot 0) with ordered data
mode.
Oops: general protection fault, probably for non-canonical address
0xdffffc0000000001: 0000 [#1] PREEMPT SMP KASAN NOPTI
KASAN: null-ptr-deref in range [0x0000000000000008-0x000000000000000f]
CPU: 0 UID: 0 PID: 5096 Comm: syz-executor792 Not tainted
6.11.0-rc4-syzkaller-00002-gb0da640826ba #0
Hardware name: QEMU Standard PC (Q35 + ICH9, 2009), BIOS
1.16.3-debian-1.16.3-2~bpo12+1 04/01/2014
RIP: 0010:ocfs2_find_dir_space_id fs/ocfs2/dir.c:3406 [inline]
RIP: 0010:ocfs2_prepare_dir_for_insert+0x3309/0x5c70 fs/ocfs2/dir.c:4280
Code: 00 00 e8 2a 25 13 fe e9 ba 06 00 00 e8 20 25 13 fe e9 4f 01 00 00
e8 16 25 13 fe 49 8d 7f 08 49 8d 5f 09 48 89 f8 48 c1 e8 03 <42> 0f b6
04 20 84 c0 0f 85 bd 23 00 00 48 89 d8 48 c1 e8 03 42 0f
RSP: 0018:ffffc9000af9f020 EFLAGS: 00010202
RAX: 0000000000000001 RBX: 0000000000000009 RCX: ffff88801e27a440
RDX: 0000000000000000 RSI: 0000000000000400 RDI: 0000000000000008
RBP: ffffc9000af9f830 R08: ffffffff8380395b R09: ffffffff838090a7
R10: 0000000000000002 R11: ffff88801e27a440 R12: dffffc0000000000
R13: ffff88803c660878 R14: f700000000000088 R15: 0000000000000000
FS:  000055555a677380(0000) GS:ffff888020800000(0000)
knlGS:0000000000000000
CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
CR2: 0000560bce569178 CR3: 000000001de5a000 CR4: 0000000000350ef0
DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
Call Trace:
<TASK>
ocfs2_mknod+0xcaf/0x2b40 fs/ocfs2/namei.c:292
vfs_mknod+0x36d/0x3b0 fs/namei.c:4088
do_mknodat+0x3ec/0x5b0
__do_sys_mknodat fs/namei.c:4166 [inline]
__se_sys_mknodat fs/namei.c:4163 [inline]
__x64_sys_mknodat+0xa7/0xc0 fs/namei.c:4163
do_syscall_x64 arch/x86/entry/common.c:52 [inline]
do_syscall_64+0xf3/0x230 arch/x86/entry/common.c:83
entry_SYSCALL_64_after_hwframe+0x77/0x7f
RIP: 0033:0x7f2dafda3a99
Code: 28 00 00 00 75 05 48 83 c4 28 c3 e8 f1 17 00 00 90 48 89
f8 48 89 f7 48 89 d6 48 89 ca 4d 89 c2 4d 89 c8 4c 8b 4c 24 08
0f 05 <48> 3d 01 f0 ff ff 73 01 c3 48 c7 c1 b8 ff ff ff f7 d8
64 89 01 48
RSP: 002b:00007ffe336a6658 EFLAGS: 00000246 ORIG_RAX:
0000000000000103
RAX: ffffffffffffffda RBX: 0000000000000000 RCX:
00007f2dafda3a99
RDX: 00000000000021c0 RSI: 0000000020000040 RDI:
00000000ffffff9c
RBP: 00007f2dafe1b5f0 R08: 0000000000004480 R09:
000055555a6784c0
R10: 0000000000000103 R11: 0000000000000246 R12:
00007ffe336a6680
R13: 00007ffe336a68a8 R14: 431bde82d7b634db R15:
00007f2dafdec03b
</TASK>
==================================================================

The two reports are all caused invalid negative i_size of dir inode.  For
ocfs2, dir_inode can't be negative or zero.

Here add a check in which is called by ocfs2_check_dir_for_entry().  It
fixes the second report as ocfs2_check_dir_for_entry() must be called
before ocfs2_prepare_dir_for_insert().  Also set a up limit for dir with
OCFS2_INLINE_DATA_FL.  The i_size can't be great than blocksize.

Link: https://lkml.kernel.org/r/20250106140640.92260-1-glass.su@suse.com
Reported-by: Jiacheng Xu <stitch@zju.edu.cn>
Link: https://lore.kernel.org/ocfs2-devel/17a04f01.1ae74.19436d003fc.Coremail.stitch@zju.edu.cn/T/#u
Reported-by: syzbot+5a64828fcc4c2ad9b04f@syzkaller.appspotmail.com
Link: https://lore.kernel.org/all/0000000000005894f3062018caf1@google.com/T/
Signed-off-by: Su Yue <glass.su@suse.com>
Reviewed-by: Heming Zhao <heming.zhao@suse.com>
Reviewed-by: Joseph Qi <joseph.qi@linux.alibaba.com>
Cc: Mark Fasheh <mark@fasheh.com>
Cc: Joel Becker <jlbec@evilplan.org>
Cc: Junxiao Bi <junxiao.bi@oracle.com>
Cc: Changwei Ge <gechangwei@live.cn>
Cc: Jun Piao <piaojun@huawei.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

mailmap: update entry for Ethan Carter Edwards

Map old gmail + name to my current full name and email.

Link: https://lkml.kernel.org/r/xbfkmvmp4wyxrvlan57bjnul5icrwfyt67vnhhw2cyr5rzbnee@mfvihhd6s7l5
Signed-off-by: Ethan Carter Edwards <ethan@ethancedwards.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

mm: zswap: move allocations during CPU init outside the lock

In zswap_cpu_comp_prepare(), allocations are made and assigned to various
members of acomp_ctx under acomp_ctx->mutex.  However, allocations may
recurse into zswap through reclaim, trying to acquire the same mutex and
deadlocking.

Move the allocations before the mutex critical section.  Only the
initialization of acomp_ctx needs to be done with the mutex held.

Link: https://lkml.kernel.org/r/20250113214458.2123410-1-yosryahmed@google.com
Fixes: 12dcb0ef5406 ("mm: zswap: properly synchronize freeing resources during CPU hotunplug")
Signed-off-by: Yosry Ahmed <yosryahmed@google.com>
Reviewed-by: Chengming Zhou <chengming.zhou@linux.dev>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Nhat Pham <nphamcs@gmail.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

mm: khugepaged: fix call hpage_collapse_scan_file() for anonymous vma

syzkaller reported such a BUG_ON():

 ------------[ cut here ]------------
 kernel BUG at mm/khugepaged.c:1835!
 Internal error: Oops - BUG: 00000000f2000800 [#1] SMP
 ...
 CPU: 6 UID: 0 PID: 8009 Comm: syz.15.106 Kdump: loaded Tainted: G        W          6.13.0-rc6 #22
 Tainted: [W]=WARN
 Hardware name: QEMU KVM Virtual Machine, BIOS 0.0.0 02/06/2015
 pstate: 00400005 (nzcv daif +PAN -UAO -TCO -DIT -SSBS BTYPE=--)
 pc : collapse_file+0xa44/0x1400
 lr : collapse_file+0x88/0x1400
 sp : ffff80008afe3a60
 ...
 Call trace:
  collapse_file+0xa44/0x1400 (P)
  hpage_collapse_scan_file+0x278/0x400
  madvise_collapse+0x1bc/0x678
  madvise_vma_behavior+0x32c/0x448
  madvise_walk_vmas.constprop.0+0xbc/0x140
  do_madvise.part.0+0xdc/0x2c8
  __arm64_sys_madvise+0x68/0x88
  invoke_syscall+0x50/0x120
  el0_svc_common.constprop.0+0xc8/0xf0
  do_el0_svc+0x24/0x38
  el0_svc+0x34/0x128
  el0t_64_sync_handler+0xc8/0xd0
  el0t_64_sync+0x190/0x198

This indicates that the pgoff is unaligned.  After analysis, I confirm the
vma is mapped to /dev/zero.  Such a vma certainly has vm_file, but it is
set to anonymous by mmap_zero().  So even if it's mmapped by 2m-unaligned,
it can pass the check in thp_vma_allowable_order() as it is an
anonymous-mmap, but then be collapsed as a file-mmap.

It seems the problem has existed for a long time, but actually, since we
have khugepaged_max_ptes_none check before, we will skip collapse it as it
is /dev/zero and so has no present page.  But commit d8ea7cc8547c limit
the check for only khugepaged, so the BUG_ON() can be triggered by
madvise_collapse().

Add vma_is_anonymous() check to make such vma be processed by
hpage_collapse_scan_pmd().

Link: https://lkml.kernel.org/r/20250111034511.2223353-1-liushixin2@huawei.com
Fixes: d8ea7cc8547c ("mm/khugepaged: add flag to predicate khugepaged-only behavior")
Signed-off-by: Liu Shixin <liushixin2@huawei.com>
Reviewed-by: Yang Shi <yang@os.amperecomputing.com>
Acked-by: David Hildenbrand <david@redhat.com>
Cc: Chengming Zhou <chengming.zhou@linux.dev>
Cc: Johannes Weiner <hannes@cmpxchg.org>
Cc: Kefeng Wang <wangkefeng.wang@huawei.com>
Cc: Mattew Wilcox <willy@infradead.org>
Cc: Muchun Song <muchun.song@linux.dev>
Cc: Nanyong Sun <sunnanyong@huawei.com>
Cc: Qi Zheng <zhengqi.arch@bytedance.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

mm: shmem: use signed int for version handling in casefold option

Fixes an issue where the use of an unsigned data type in
`shmem_parse_opt_casefold()` caused incorrect evaluation of negative
conditions.

Link: https://lkml.kernel.org/r/20250111-unsignedcompare1601569-v3-1-c861b4221831@gmail.com
Fixes: 58e55efd6c72 ("tmpfs: Add casefold lookup support")
Reviewed-by: André Almeida <andrealmeid@igalia.com>
Reviewed-by: Gabriel Krisman Bertazi <gabriel@krisman.be>
Signed-off-by: Karan Sanghavi <karansanghvi98@gmail.com>
Cc: Christian Brauner <brauner@kernel.org>
Cc: Hugh Dickens <hughd@google.com>
Cc: Shuah khan <skhan@linuxfoundation.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

alloc_tag: skip pgalloc_tag_swap if profiling is disabled

When memory allocation profiling is disabled, there is no need to swap
allocation tags during migration.  Skip it to avoid unnecessary overhead.

Once I added these checks, the overhead of the mode when memory profiling
is enabled but turned off went down by about 50%.

Link: https://lkml.kernel.org/r/20241226211639.1357704-2-surenb@google.com
Fixes: e0a955bf7f61 ("mm/codetag: add pgalloc_tag_copy()")
Signed-off-by: Suren Baghdasaryan <surenb@google.com>
Cc: David Wang <00107082@163.com>
Cc: Kent Overstreet <kent.overstreet@linux.dev>
Cc: Yu Zhao <yuzhao@google.com>
Cc: Zhenhua Huang <quic_zhenhuah@quicinc.com>
Cc: <stable@vger.kernel.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

mm: page_alloc: fix missed updates of lowmem_reserve in adjust_managed_page_count

In the kernel, the zone's lowmem_reserve and _watermark, and the global
variable 'totalreserve_pages' depend on the value of managed_pages, but
after running adjust_managed_page_count, these values aren't updated,
which causes some problems.

For example, in a system with six 1GB large pages, we found that the value
of protection in zoneinfo (zone->lowmem_reserve), is not right.  Its value
seems to be calculated from the initial managed_pages, but after the
managed_pages changed, was not updated.  Only after reading the file
/proc/sys/vm/lowmem_reserve_ratio, updates happen.

read file /proc/sys/vm/lowmem_reserve_ratio:

lowmem_reserve_ratio_sysctl_handler
----setup_per_zone_lowmem_reserve
--------calculate_totalreserve_pages

protection changed after reading file:

[root@test ~]# cat /proc/zoneinfo | grep protection
        protection: (0, 2719, 57360, 0)
        protection: (0, 0, 54640, 0)
        protection: (0, 0, 0, 0)
        protection: (0, 0, 0, 0)
[root@test ~]# cat /proc/sys/vm/lowmem_reserve_ratio
256     256     32      0
[root@test ~]# cat /proc/zoneinfo | grep protection
        protection: (0, 2735, 63524, 0)
        protection: (0, 0, 60788, 0)
        protection: (0, 0, 0, 0)
        protection: (0, 0, 0, 0)

lowmem_reserve increased also makes the totalreserve_pages increased,
which causes a decrease in available memory.  The one above is just a test
machine, and the increase is not significant.  On our online machine, the
reserved memory will increase by several GB due to reading this file.  It
is clearly unreasonable to cause a sharp drop in available memory just by
reading a file.

In this patch, we update reserve memory when update managed_pages, The
size of reserved memory becomes stable.  But it seems that the _watermark
should also be updated along with the managed_pages.  We have not done it
because we are unsure if it is reasonable to set the watermark through the
initial managed_pages.  If it is not reasonable, we will propose new
patch.

Link: https://lkml.kernel.org/r/20241225021034.45693-1-15645113830zzh@gmail.com
Signed-off-by: zihan zhou <15645113830zzh@gmail.com>
Signed-off-by: yaowenchao <yaowenchao@jd.com>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>

net: make page_pool_ref_netmem work with net iovs

page_pool_ref_netmem() should work with either netmem representation, but
currently it casts to a page with netmem_to_page(), which will fail with
net iovs. Use netmem_get_pp_ref_count_ref() instead.

Fixes: 8ab79ed50cf1 ("page_pool: devmem support")
Signed-off-by: Pavel Begunkov <asml.silence@gmail.com>
Signed-off-by: David Wei <dw@davidwei.uk>
Link: https://lore.kernel.org/20250108220644.3528845-2-dw@davidwei.uk
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

Merge tag 'drm-misc-fixes-2025-01-15' of https://gitlab.freedesktop.org/drm/misc/kernel into drm-fixes

drm-misc-fixes for v6.13:

- itee-it6263 error handling fix.
- Fix warn when unloading v3d.
- Fix W=1 build for kunit tests.
- Fix backlight regression for macbooks 5,1 in nouveau.
- Handle YCbCr420 better in bridge code, with tests.
- Fix cross-device fence handling in nouveau.
- Fix BO reservation handling in vmwgfx.

Signed-off-by: Dave Airlie <airlied@redhat.com>
From: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/a89adcd5-2042-4e7f-93f4-2b299bb1ef17@linux.intel.com

smb: client: fix double free of TCP_Server_Info::hostname

When shutting down the server in cifs_put_tcp_session(), cifsd thread
might be reconnecting to multiple DFS targets before it realizes it
should exit the loop, so @server->hostname can't be freed as long as
cifsd thread isn't done.  Otherwise the following can happen:

  RIP: 0010:__slab_free+0x223/0x3c0
  Code: 5e 41 5f c3 cc cc cc cc 4c 89 de 4c 89 cf 44 89 44 24 08 4c 89
  1c 24 e8 fb cf 8e 00 44 8b 44 24 08 4c 8b 1c 24 e9 5f fe ff ff <0f>
  0b 41 f7 45 08 00 0d 21 00 0f 85 2d ff ff ff e9 1f ff ff ff 80
  RSP: 0018:ffffb26180dbfd08 EFLAGS: 00010246
  RAX: ffff8ea34728e510 RBX: ffff8ea34728e500 RCX: 0000000000800068
  RDX: 0000000000800068 RSI: 0000000000000000 RDI: ffff8ea340042400
  RBP: ffffe112041ca380 R08: 0000000000000001 R09: 0000000000000000
  R10: 6170732e31303000 R11: 70726f632e786563 R12: ffff8ea34728e500
  R13: ffff8ea340042400 R14: ffff8ea34728e500 R15: 0000000000800068
  FS: 0000000000000000(0000) GS:ffff8ea66fd80000(0000)
  000000
  CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
  CR2: 00007ffc25376080 CR3: 000000012a2ba001 CR4:
  PKRU: 55555554
  Call Trace:
   <TASK>
   ? show_trace_log_lvl+0x1c4/0x2df
   ? show_trace_log_lvl+0x1c4/0x2df
   ? __reconnect_target_unlocked+0x3e/0x160 [cifs]
   ? __die_body.cold+0x8/0xd
   ? die+0x2b/0x50
   ? do_trap+0xce/0x120
   ? __slab_free+0x223/0x3c0
   ? do_error_trap+0x65/0x80
   ? __slab_free+0x223/0x3c0
   ? exc_invalid_op+0x4e/0x70
   ? __slab_free+0x223/0x3c0
   ? asm_exc_invalid_op+0x16/0x20
   ? __slab_free+0x223/0x3c0
   ? extract_hostname+0x5c/0xa0 [cifs]
   ? extract_hostname+0x5c/0xa0 [cifs]
   ? __kmalloc+0x4b/0x140
   __reconnect_target_unlocked+0x3e/0x160 [cifs]
   reconnect_dfs_server+0x145/0x430 [cifs]
   cifs_handle_standard+0x1ad/0x1d0 [cifs]
   cifs_demultiplex_thread+0x592/0x730 [cifs]
   ? __pfx_cifs_demultiplex_thread+0x10/0x10 [cifs]
   kthread+0xdd/0x100
   ? __pfx_kthread+0x10/0x10
   ret_from_fork+0x29/0x50
   </TASK>

Fixes: 7be3248f3139 ("cifs: To match file servers, make sure the server hostname matches")
Reported-by: Jay Shin <jaeshin@redhat.com>
Signed-off-by: Paulo Alcantara (Red Hat) <pc@manguebit.com>
Signed-off-by: Steve French <stfrench@microsoft.com>

hwmon: (ltc2991) Fix mixed signed/unsigned in DIV_ROUND_CLOSEST

Fix use of DIV_ROUND_CLOSEST where a possibly negative value is divided
by an unsigned type by casting the unsigned type to the signed type of
the same size (st->r_sense_uohm[channel] has type of u32).

The docs on the DIV_ROUND_CLOSEST macro explain that dividing a negative
value by an unsigned type is undefined behavior. The actual behavior is
that it converts both values to unsigned before doing the division, for
example:

    int ret = DIV_ROUND_CLOSEST(-100, 3U);

results in ret == 1431655732 instead of -33.

Fixes: 2b9ea4262ae9 ("hwmon: Add driver for ltc2991")
Signed-off-by: David Lechner <dlechner@baylibre.com>
Link: https://lore.kernel.org/r/20250115-hwmon-ltc2991-fix-div-round-closest-v1-1-b4929667e457@baylibre.com
Signed-off-by: Guenter Roeck <linux@roeck-us.net>

net: ethernet: xgbe: re-add aneg to supported features in PHY quirks

In 4.19, before the switch to linkmode bitmaps, PHY_GBIT_FEATURES
included feature bits for aneg and TP/MII ports.

 SUPPORTED_TP | \
 SUPPORTED_MII)

 SUPPORTED_10baseT_Full)

 SUPPORTED_100baseT_Full)

 SUPPORTED_1000baseT_Full)

 PHY_100BT_FEATURES | \
 PHY_DEFAULT_FEATURES)

 PHY_1000BT_FEATURES)

Referenced commit expanded PHY_GBIT_FEATURES, silently removing
PHY_DEFAULT_FEATURES. The removed part can be re-added by using
the new PHY_GBIT_FEATURES definition.
Not clear to me is why nobody seems to have noticed this issue.

I stumbled across this when checking what it takes to make
phy_10_100_features_array et al private to phylib.

Fixes: d0939c26c53a ("net: ethernet: xgbe: expand PHY_GBIT_FEAUTRES")
Cc: stable@vger.kernel.org
Signed-off-by: Heiner Kallweit <hkallweit1@gmail.com>
Link: https://patch.msgid.link/46521973-7738-4157-9f5e-0bb6f694acba@gmail.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: pcs: xpcs: actively unset DW_VR_MII_DIG_CTRL1_2G5_EN for 1G SGMII

xpcs_config_2500basex() sets DW_VR_MII_DIG_CTRL1_2G5_EN, but
xpcs_config_aneg_c37_sgmii() never unsets it. So, on a protocol change
from 2500base-x to sgmii, the DW_VR_MII_DIG_CTRL1_2G5_EN bit will remain
set.

Fixes: f27abde3042a ("net: pcs: add 2500BASEX support for Intel mGbE controller")
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Maxime Chevallier <maxime.chevallier@bootlin.com>
Reviewed-by: Russell King (Oracle) <rmk+kernel@armlinux.org.uk>
Link: https://patch.msgid.link/20250114164721.2879380-2-vladimir.oltean@nxp.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: pcs: xpcs: fix DW_VR_MII_DIG_CTRL1_2G5_EN bit being set for 1G SGMII w/o inband

On a port with SGMII fixed-link at SPEED_1000, DW_VR_MII_DIG_CTRL1 gets
set to 0x2404. This is incorrect, because bit 2 (DW_VR_MII_DIG_CTRL1_2G5_EN)
is set.

It comes from the previous write to DW_VR_MII_AN_CTRL, because the "val"
variable is reused and is dirty. Actually, its value is 0x4, aka
FIELD_PREP(DW_VR_MII_PCS_MODE_MASK, DW_VR_MII_PCS_MODE_C37_SGMII).

Resolve the issue by clearing "val" to 0 when writing to a new register.
After the fix, the register value is 0x2400.

Prior to the blamed commit, when the read-modify-write was open-coded,
the code saved the content of the DW_VR_MII_DIG_CTRL1 register in the
"ret" variable.

Fixes: ce8d6081fcf4 ("net: pcs: xpcs: add _modify() accessors")
Signed-off-by: Vladimir Oltean <vladimir.oltean@nxp.com>
Reviewed-by: Maxime Chevallier <maxime.chevallier@bootlin.com>
Reviewed-by: Russell King (Oracle) <rmk+kernel@armlinux.org.uk>
Link: https://patch.msgid.link/20250114164721.2879380-1-vladimir.oltean@nxp.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

i2c: testunit: on errors, repeat NACK until STOP

This backend requests a NACK from the controller driver when it detects
an error. If that request gets ignored from some reason, subsequent
accesses will wrongly be handled OK. To fix this, an error now changes
the state machine, so the backend will report NACK until a STOP
condition has been detected. This make the driver more robust against
controllers which will sadly apply the NACK not to the current byte but
the next one.

Fixes: a8335c64c5f0 ("i2c: add slave testunit driver")
Signed-off-by: Wolfram Sang <wsa+renesas@sang-engineering.com>

i2c: rcar: fix NACK handling when being a target

When this controller is a target, the NACK handling had two issues.
First, the return value from the backend was not checked on the initial
WRITE_REQUESTED. So, the driver missed to send a NACK in this case.
Also, the NACK always arrives one byte late on the bus, even in the
WRITE_RECEIVED case. This seems to be a HW issue. We should then not
rely on the backend to correctly NACK the superfluous byte as well. Fix
both issues by introducing a flag which gets set whenever the backend
requests a NACK and keep sending it until we get a STOP condition.

Fixes: de20d1857dd6 ("i2c: rcar: add slave support")
Signed-off-by: Wolfram Sang <wsa+renesas@sang-engineering.com>

i2c: mux: demux-pinctrl: correct comment

Two characters flipped, fix them.

Signed-off-by: Wolfram Sang <wsa+renesas@sang-engineering.com>

i2c: mux: demux-pinctrl: check initial mux selection, too

When misconfigured, the initial setup of the current mux channel can
fail, too. It must be checked as well.

Fixes: 50a5ba876908 ("i2c: mux: demux-pinctrl: add driver")
Signed-off-by: Wolfram Sang <wsa+renesas@sang-engineering.com>

Revert "mtd: spi-nor: core: replace dummy buswidth from addr to data"

This reverts commit 98d1fb94ce75f39febd456d6d3cbbe58b6678795.

The commit uses data nbits instead of addr nbits for dummy phase. This
causes a regression for all boards where spi-tx-bus-width is smaller
than spi-rx-bus-width. It is a common pattern for boards to have
spi-tx-bus-width == 1 and spi-rx-bus-width > 1. The regression causes
all reads with a dummy phase to become unavailable for such boards,
leading to a usually slower 0-dummy-cycle read being selected.

Most controllers' supports_op hooks call spi_mem_default_supports_op().
In spi_mem_default_supports_op(), spi_mem_check_buswidth() is called to
check if the buswidths for the op can actually be supported by the
board's wiring. This wiring information comes from (among other things)
the spi-{tx,rx}-bus-width DT properties. Based on these properties,
SPI_TX_* or SPI_RX_* flags are set by of_spi_parse_dt().
spi_mem_check_buswidth() then uses these flags to make the decision
whether an op can be supported by the board's wiring (in a way,
indirectly checking against spi-{rx,tx}-bus-width).

Now the tricky bit here is that spi_mem_check_buswidth() does:

if (op->dummy.nbytes &&
    spi_check_buswidth_req(mem, op->dummy.buswidth, true))
return false;

The true argument to spi_check_buswidth_req() means the op is treated as
a TX op. For a board that has say 1-bit TX and 4-bit RX, a 4-bit dummy
TX is considered as unsupported, and the op gets rejected.

The commit being reverted uses the data buswidth for dummy buswidth. So
for reads, the RX buswidth gets used for the dummy phase, uncovering
this issue. In reality, a dummy phase is neither RX nor TX. As the name
suggests, these are just dummy cycles that send or receive no data, and
thus don't really need to have any buswidth at all.

Ideally, dummy phases should not be checked against the board's wiring
capabilities at all, and should only be sanity-checked for having a sane
buswidth value. Since we are now at rc7 and such a change might
introduce many unexpected bugs, revert the commit for now. It can be
sent out later along with the spi_mem_check_buswidth() fix.

Fixes: 98d1fb94ce75 ("mtd: spi-nor: core: replace dummy buswidth from addr to data")
Reported-by: Alexander Stein <alexander.stein@ew.tq-group.com>
Closes: https://lore.kernel.org/linux-mtd/3342163.44csPzL39Z@steina-w/
Tested-by: Alexander Stein <alexander.stein@ew.tq-group.com>
Reviewed-by: Tudor Ambarus <tudor.ambarus@linaro.org>
Signed-off-by: Pratyush Yadav <pratyush@kernel.org>
Signed-off-by: Miquel Raynal <miquel.raynal@bootlin.com>

signal/posixtimers: Handle ignore/blocked sequences correctly

syzbot triggered the warning in posixtimer_send_sigqueue(), which warns
about a non-ignored signal being already queued on the ignored list.

The warning is actually bogus, as the following sequence causes this:

    signal($SIG, SIGIGN);
    timer_settime(...); // arm periodic timer

      timer fires, signal is ignored and queued on ignored list

    sigprocmask(SIG_BLOCK, ...);        // block the signal
    timer_settime(...); // re-arm periodic timer

      timer fires, signal is not ignored because it is blocked
        ---> Warning triggers as signal is on the ignored list

Ideally timer_settime() could remove the signal, but that's racy and
incomplete vs. other scenarios and requires a full reevaluation of the
pending signal list.

Instead of adding more complexity, handle it gracefully by removing the
warning and requeueing the signal to the pending list. That's correct
versus:

  1) sig[timed]wait() as that does not check for SIGIGN and only relies on
     dequeue_signal() -> posixtimers_deliver_signal() to check whether the
     pending signal is still valid.

  2) Unblocking of the signal.

     - If the unblocking happens before SIGIGN is replaced by a signal
       handler, then the timer is rearmed in dequeue_signal(), but
       get_signal() will ignore it. The next timer expiry will move it back
       to the ignored list.

     - If SIGIGN was replaced before unblocking, then the signal will be
       delivered and a subsequent expiry will queue a signal on the pending
       list again.

There is a related scenario to trigger the complementary warning in the
signal ignored path, which does not expect the signal to be on the pending
list when it is ignored. That can be triggered even before the above change
via:

task1 task2

signal($SIG, SIGIGN);
sigprocmask(SIG_BLOCK, ...);

timer_create(); // Signal target is task2
timer_settime(...); // arm periodic timer

   timer fires, signal is not ignored because it is blocked
   and queued on the pending list of task2

                     syscall()
   // Sets the pending flag
   sigprocmask(SIG_UNBLOCK, ...);

-> preemption, task2 cannot dequeue the signal

timer_settime(...); // re-arm periodic timer

   timer fires, signal is ignored
        ---> Warning triggers as signal is on task2's pending list
     and the thread group is not exiting

Consequently, remove that warning too and just keep the signal on the
pending list.

The following attempt to deliver the signal on return to user space of
task2 will ignore the signal and a subsequent expiry will bring it back to
the ignored list, if it did not get blocked or un-ignored before that.

Fixes: df7a996b4dab ("signal: Queue ignored posixtimers on ignore list")
Reported-by: syzbot+3c2e3cc60665d71de2f7@syzkaller.appspotmail.com
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Frederic Weisbecker <frederic@kernel.org>
Link: https://lore.kernel.org/all/87ikqhcnjn.ffs@tglx

io_uring/register: cache old SQ/CQ head reading for copies

The SQ and CQ ring heads are read twice - once for verifying that it's
within bounds, and once inside the loops copying SQE and CQE entries.
This is technically incorrect, in case the values could get modified
in between verifying them and using them in the copy loop. While this
won't lead to anything truly nefarious, it may cause longer loop times
for the copies than expected.

Read the ring head values once, and use the verified value in the copy
loops.

Signed-off-by: Jens Axboe <axboe@kernel.dk>

io_uring/register: document io_register_resize_rings() shared mem usage

It can be a bit hard to tell which parts of io_register_resize_rings()
are operating on shared memory, and which ones are not. And anything
reading or writing to those regions should really use the read/write
once primitives.

Hence add those, ensuring sanity in how this memory is accessed, and
helping document the shared nature of it.

Signed-off-by: Jens Axboe <axboe@kernel.dk>

Merge tag 'ti-driver-soc-for-v6.14' of https://git.kernel.org/pub/scm/linux/kernel/git/ti/linux into arm/fixes

TI SoC driver updates for v6.14

- Build fixup when CONFIG_TI_PRUSS is disabled.

io_uring/register: use stable SQ/CQ ring data during resize

Normally the kernel would not expect an application to modify any of
the data shared with the kernel during a resize operation, but of
course the kernel cannot always assume good intent on behalf of the
application.

As part of resizing the rings, existing SQEs and CQEs are copied over
to the new storage. Resizing uses the masks in the newly allocated
shared storage to index the arrays, however it's possible that malicious
userspace could modify these after they have been sanity checked.

Use the validated and locally stored CQ and SQ ring sizing for masking
to ensure the values are both stable and valid.

Fixes: 79cfe9e59c2a ("io_uring/register: add IORING_REGISTER_RESIZE_RINGS")
Reported-by: Jann Horn <jannh@google.com>
Signed-off-by: Jens Axboe <axboe@kernel.dk>

hwmon: (drivetemp) Set scsi command timeout to 10s

There's at least one drive (MaxDigitalData OOS14000G) such that if it
receives a large amount of I/O while entering an idle power state will
first exit idle before responding, including causing SMART temperature
requests to be delayed.

This causes the drivetemp request to exceed its timeout of 1 second.

Signed-off-by: Russell Harmon <russ@har.mn>
Link: https://lore.kernel.org/r/20250115131340.3178988-1-russ@har.mn
Signed-off-by: Guenter Roeck <linux@roeck-us.net>

hwmon: (acpi_power_meter) Fix a check for the return value of read_domain_devices().

After commit fabb1f813ec0 ("hwmon: (acpi_power_meter) Fix fail to load
module on platform without _PMD method"),
the acpi_power_meter driver fails to load if the platform has _PMD method.

To address this, add a check for successful read_domain_devices().
Tested on Nvidia Grace machine.

Fixes: fabb1f813ec0 ("hwmon: (acpi_power_meter) Fix fail to load module on platform without _PMD method")
Signed-off-by: Kazuhiro Abe <fj1078ii@aa.jp.fujitsu.com>
Link: https://lore.kernel.org/r/20250115073532.3211000-1-fj1078ii@aa.jp.fujitsu.com
[groeck: Dropped unnecessary () from expression]
Signed-off-by: Guenter Roeck <linux@roeck-us.net>

Merge tag 'reset-fixes-for-v6.13' of git://git.pengutronix.de/pza/linux into arm/fixes

Reset controller fixes for v6.13

* Fix rzg2l-usb-vbus-regulator lookup by assigning the proper of node
  to the allocated platform device in the rzg2l-usbphy-ctrl driver.

* tag 'reset-fixes-for-v6.13' of git://git.pengutronix.de/pza/linux:
  reset: rzg2l-usbphy-ctrl: Assign proper of node to the allocated device

Link: https://lore.kernel.org/r/20250113163642.1757160-1-p.zabel@pengutronix.de
Signed-off-by: Arnd Bergmann <arnd@arndb.de>

drm/xe/oa: Add missing VISACTL mux registers

Add missing VISACTL mux registers required for some OA
config's (e.g. RenderPipeCtrl).

Fixes: cdf02fe1a94a ("drm/xe/oa/uapi: Add/remove OA config perf ops")
Cc: stable@vger.kernel.org
Signed-off-by: Ashutosh Dixit <ashutosh.dixit@intel.com>
Reviewed-by: Umesh Nerlige Ramappa <umesh.nerlige.ramappa@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20250111021539.2920346-1-ashutosh.dixit@intel.com
(cherry picked from commit c26f22dac3449d8a687237cdfc59a6445eb8f75a)
Signed-off-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>

drm/xe: make change ccs_mode a synchronous action

If ccs_mode is being modified via
   /sys/class/drm/cardX/device/tileY/gtY/ccs_mode
the asynchronous reset is triggered and the write returns immediately.

With that some test receive false information about number of CCS engines
or even fail if they proceed without delay after changing the ccs_mode.

Changing the ccs_mode change from async to sync to prevent failures in
tests.

Signed-off-by: Maciej Patelczyk <maciej.patelczyk@intel.com>
Reviewed-by: Lucas De Marchi <lucas.demarchi@intel.com>
Fixes: f3bc5bb4d53d ("drm/xe: Allow userspace to configure CCS mode")
Reviewed-by: Nirmoy Das <nirmoy.das@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20241211111727.1481476-3-maciej.patelczyk@intel.com
Signed-off-by: Nirmoy Das <nirmoy.das@intel.com>
(cherry picked from commit 480fb9806e2e073532f7786166287114c696b340)
Signed-off-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>

drm/xe: introduce xe_gt_reset and xe_gt_wait_for_reset

Add synchronous version gt reset as there are few places where it
is expected.
Also add a wait helper to wait until gt reset is done.

Signed-off-by: Maciej Patelczyk <maciej.patelczyk@intel.com>
Reviewed-by: Lucas De Marchi <lucas.demarchi@intel.com>
Fixes: f3bc5bb4d53d ("drm/xe: Allow userspace to configure CCS mode")
Reviewed-by: Nirmoy Das <nirmoy.das@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20241211111727.1481476-2-maciej.patelczyk@intel.com
Signed-off-by: Nirmoy Das <nirmoy.das@intel.com>
(cherry picked from commit 155c77f45f63dd58a37eeb0896b0b140ab785836)
Signed-off-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>

drm/xe/guc: Adding steering info support for GuC register lists

The guc_mmio_reg interface supports steering, but it is currently not
implemented. This will allow the GuC to control steering of MMIO
registers after save-restore and avoid reading from fused off MCR
register instances.

Fixes: 9c57bc08652a ("drm/xe/lnl: Drop force_probe requirement")
Signed-off-by: Jesus Narvaez <jesus.narvaez@intel.com>
Cc: Matt Roper <matthew.d.roper@intel.com>
Cc: Lucas De Marchi <lucas.demarchi@intel.com>
Cc: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Reviewed-by: Jonathan Cavitt <jonathan.cavitt@intel.com>
Signed-off-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20241212190100.3768068-1-jesus.narvaez@intel.com
(cherry picked from commit ee5a1321df90891d59d83b7c9d5b6c5b755d059d)
Signed-off-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>

irqchip: Plug a OF node reference leak in platform_irqchip_probe()

platform_irqchip_probe() leaks a OF node when irq_init_cb() fails. Fix it
by declaring par_np with the __free(device_node) cleanup construct.

This bug was found by an experimental static analysis tool that I am
developing.

Fixes: f8410e626569 ("irqchip: Add IRQCHIP_PLATFORM_DRIVER_BEGIN/END and IRQCHIP_MATCH helper macros")
Signed-off-by: Joe Hattori <joe@pf.is.s.u-tokyo.ac.jp>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Cc: stable@vger.kernel.org
Link: https://lore.kernel.org/all/20241215033945.3414223-1-joe@pf.is.s.u-tokyo.ac.jp

selftests: net: Adapt ethtool mq tests to fix in qdisc graft

Because of patch[1] the graft behaviour changed

So the command:

tcq replace parent 100:1 handle 204:

Is no longer valid and will not delete 100:4 added by command:

tcq replace parent 100:4 handle 204: pfifo_fast

So to maintain the original behaviour, this patch manually deletes 100:4
and grafts 100:1

Note: This change will also work fine without [1]

[1] https://lore.kernel.org/netdev/20250111151455.75480-1-jhs@mojatatu.com/T/#u

Signed-off-by: Victor Nogueira <victor@mojatatu.com>
Reviewed-by: Jamal Hadi Salim <jhs@mojatatu.com>
Signed-off-by: David S. Miller <davem@davemloft.net>

irqchip/sunxi-nmi: Add missing SKIP_WAKE flag

Some boards with Allwinner SoCs connect the PMIC's IRQ pin to the SoC's NMI
pin instead of a normal GPIO. Since the power key is connected to the PMIC,
and people expect to wake up a suspended system via this key, the NMI IRQ
controller must stay alive when the system goes into suspend.

Add the SKIP_WAKE flag to prevent the sunxi NMI controller from going to
sleep, so that the power key can wake up those systems.

[ tglx: Fixed up coding style ]

Signed-off-by: Philippe Simons <simons.philippe@gmail.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Link: https://lore.kernel.org/all/20250112123402.388520-1-simons.philippe@gmail.com

irqchip/gic-v3-its: Don't enable interrupts in its_irq_set_vcpu_affinity()

The following call-chain leads to enabling interrupts in a nested interrupt
disabled section:

irq_set_vcpu_affinity()
  irq_get_desc_lock()
     raw_spin_lock_irqsave()   <--- Disable interrupts
  its_irq_set_vcpu_affinity()
     guard(raw_spinlock_irq)   <--- Enables interrupts when leaving the guard()
  irq_put_desc_unlock()        <--- Warns because interrupts are enabled

This was broken in commit b97e8a2f7130, which replaced the original
raw_spin_[un]lock() pair with guard(raw_spinlock_irq).

Fix the issue by using guard(raw_spinlock).

[ tglx: Massaged change log ]

Fixes: b97e8a2f7130 ("irqchip/gic-v3-its: Fix potential race condition in its_vlpi_prop_update()")
Signed-off-by: Tomas Krcka <krckatom@amazon.de>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Reviewed-by: Marc Zyngier <maz@kernel.org>
Cc: stable@vger.kernel.org
Link: https://lore.kernel.org/all/20241230150825.62894-1-krckatom@amazon.de

irqchip/gic-v3: Handle CPU_PM_ENTER_FAILED correctly

When a CPU attempts to enter low power mode, it disables the redistributor
and Group 1 interrupts and reinitializes the system registers upon wakeup.

If the transition into low power mode fails, then the CPU_PM framework
invokes the PM notifier callback with CPU_PM_ENTER_FAILED to allow the
drivers to undo the state changes.

The GIC V3 driver ignores CPU_PM_ENTER_FAILED, which leaves the GIC in
disabled state.

Handle CPU_PM_ENTER_FAILED in the same way as CPU_PM_EXIT to restore normal
operation.

[ tglx: Massage change log, add Fixes tag ]

Fixes: 3708d52fc6bb ("irqchip: gic-v3: Implement CPU PM notifier")
Signed-off-by: Yogesh Lal <quic_ylal@quicinc.com>
Signed-off-by: Thomas Gleixner <tglx@linutronix.de>
Acked-by: Marc Zyngier <maz@kernel.org>
Cc: stable@vger.kernel.org
Link: https://lore.kernel.org/all/20241220093907.2747601-1-quic_ylal@quicinc.com

drm/bridge: ite-it6263: Prevent error pointer dereference in probe()

If devm_i2c_new_dummy_device() fails then we were supposed to return an
error code, but instead the function continues and will crash on the next
line.  Add the missing return statement.

Fixes: 049723628716 ("drm/bridge: Add ITE IT6263 LVDS to HDMI converter")
Signed-off-by: Dan Carpenter <dan.carpenter@linaro.org>
Reviewed-by: Biju Das <biju.das.jz@bp.renesas.com>
Reviewed-by: Liu Ying <victor.liu@nxp.com>
Signed-off-by: Liu Ying <victor.liu@nxp.com>
Link: https://patchwork.freedesktop.org/patch/msgid/804a758b-f2e7-4116-b72d-29bc8905beed@stanley.mountain

net: fec: handle page_pool_dev_alloc_pages error

The fec_enet_update_cbd function calls page_pool_dev_alloc_pages but did
not handle the case when it returned NULL. There was a WARN_ON(!new_page)
but it would still proceed to use the NULL pointer and then crash.

This case does seem somewhat rare but when the system is under memory
pressure it can happen. One case where I can duplicate this with some
frequency is when writing over a smbd share to a SATA HDD attached to an
imx6q.

Setting /proc/sys/vm/min_free_kbytes to higher values also seems to solve
the problem for my test case. But it still seems wrong that the fec driver
ignores the memory allocation error and can crash.

This commit handles the allocation error by dropping the current packet.

Fixes: 95698ff6177b5 ("net: fec: using page pool to manage RX buffers")
Signed-off-by: Kevin Groeneveld <kgroeneveld@lenbrook.com>
Reviewed-by: Jacob Keller <jacob.e.keller@intel.com>
Reviewed-by: Wei Fang <wei.fang@nxp.com>
Link: https://patch.msgid.link/20250113154846.1765414-1-kgroeneveld@lenbrook.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>

net: netpoll: ensure skb_pool list is always initialized

When __netpoll_setup() is called directly, instead of through
netpoll_setup(), the np->skb_pool list head isn't initialized.
If skb_pool_flush() is later called, then we hit a NULL pointer
in skb_queue_purge_reason().  This can be seen with this repro,
when CONFIG_NETCONSOLE is enabled as a module:

    ip tuntap add mode tap tap0
    ip link add name br0 type bridge
    ip link set dev tap0 master br0
    modprobe netconsole netconsole=4444@10.0.0.1/br0,9353@10.0.0.2/
    rmmod netconsole

The backtrace is:

    BUG: kernel NULL pointer dereference, address: 0000000000000008
    #PF: supervisor write access in kernel mode
    #PF: error_code(0x0002) - not-present page
    ... ... ...
    Call Trace:
     <TASK>
     __netpoll_free+0xa5/0xf0
     br_netpoll_cleanup+0x43/0x50 [bridge]
     do_netpoll_cleanup+0x43/0xc0
     netconsole_netdev_event+0x1e3/0x300 [netconsole]
     unregister_netdevice_notifier+0xd9/0x150
     cleanup_module+0x45/0x920 [netconsole]
     __se_sys_delete_module+0x205/0x290
     do_syscall_64+0x70/0x150
     entry_SYSCALL_64_after_hwframe+0x76/0x7e

Move the skb_pool list setup and initial skb fill into __netpoll_setup().

Fixes: 221a9c1df790 ("net: netpoll: Individualize the skb pool")
Signed-off-by: John Sperbeck <jsperbeck@google.com>
Reviewed-by: Breno Leitao <leitao@debian.org>
Link: https://patch.msgid.link/20250114011354.2096812-1-jsperbeck@google.com
Signed-off-by: Jakub Kicinski <kuba@kernel.org>