KVM: Ensure lockdep knows about kvm->lock vs. vcpu->mutex ordering rule

Documentation/virt/kvm/locking.rst tells us that kvm->lock is taken outside
vcpu->mutex. But that doesn't actually happen very often; it's only in
some esoteric cases like migration with AMD SEV. This means that lockdep
usually doesn't notice, and doesn't do its job of keeping us honest.

Ensure that lockdep *always* knows about the ordering of these two locks,
by briefly taking vcpu->mutex in kvm_vm_ioctl_create_vcpu() while kvm->lock
is held.

Signed-off-by: David Woodhouse <[email protected]>
Message-Id: <20230111180651[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>

KVM: x86/xen: Fix potential deadlock in kvm_xen_update_runstate_guest()

The kvm_xen_update_runstate_guest() function can be called when the vCPU
is being scheduled out, from a preempt notifier. It *opportunistically*
updates the runstate area in the guest memory, if the gfn_to_pfn_cache
which caches the appropriate address is still valid.

If there is *contention* when it attempts to obtain gpc->lock, then
locking inside the priority inheritance checks may cause a deadlock.
Lockdep reports:

[13890.148997] Chain exists of:
                 &gpc->lock --> &p->pi_lock --> &rq->__lock

[13890.149002]  Possible unsafe locking scenario:

[13890.149003]        CPU0                    CPU1
[13890.149004]        ----                    ----
[13890.149005]   lock(&rq->__lock);
[13890.149007]                                lock(&p->pi_lock);
[13890.149009]                                lock(&rq->__lock);
[13890.149011]   lock(&gpc->lock);
[13890.149013]
                *** DEADLOCK ***

In the general case, if there's contention for a read lock on gpc->lock,
that's going to be because something else is either invalidating or
revalidating the cache. Either way, we've raced with seeing it in an
invalid state, in which case we would have aborted the opportunistic
update anyway.

So in the 'atomic' case when called from the preempt notifier, just
switch to using read_trylock() and avoid the PI handling altogether.

Signed-off-by: David Woodhouse <[email protected]>
Message-Id: <20230111180651[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>

KVM: x86/xen: Fix lockdep warning on "recursive" gpc locking

In commit 5ec3289b31 ("KVM: x86/xen: Compatibility fixes for shared runstate
area") we declared it safe to obtain two gfn_to_pfn_cache locks at the same
time:
/*
 * The guest's runstate_info is split across two pages and we
 * need to hold and validate both GPCs simultaneously. We can
 * declare a lock ordering GPC1 > GPC2 because nothing else
 * takes them more than one at a time.
 */

However, we forgot to tell lockdep. Do so, by setting a subclass on the
first lock before taking the second.

Fixes: 5ec3289b31 ("KVM: x86/xen: Compatibility fixes for shared runstate area")
Suggested-by: Peter Zijlstra <[email protected]>
Signed-off-by: David Woodhouse <[email protected]>
Message-Id: <20230111180651[email protected]>
Signed-off-by: Paolo Bonzini <[email protected]>

Merge tag 'kvmarm-fixes-6.2-1' of git://git.kernel.org/pub/scm/linux/kernel/git/kvmarm/kvmarm into kvm-master

KVM/arm64 fixes for 6.2, take #1

- Fix the PMCR_EL0 reset value after the PMU rework

- Correctly handle S2 fault triggered by a S1 page table walk
  by not always classifying it as a write, as this breaks on
  R/O memslots

- Document why we cannot exit with KVM_EXIT_MMIO when taking
  a write fault from a S1 PTW on a R/O memslot

- Put the Apple M2 on the naughty step for not being able to
  correctly implement the vgic SEIS feature, just liek the M1
  before it

- Reviewer updates: Alex is stepping down, replaced by Zenghui

Documentation: kvm: fix SRCU locking order docs

kvm->srcu is taken in KVM_RUN and several other vCPU ioctls, therefore
vcpu->mutex is susceptible to the same deadlock that is documented
for kvm->slots_lock.  The same holds for kvm->lock, since kvm->lock
is held outside vcpu->mutex.  Fix the documentation and rearrange it
to highlight the difference between these locks and kvm->slots_arch_lock,
and how kvm->slots_arch_lock can be useful while processing a vmexit.

Signed-off-by: Paolo Bonzini <[email protected]>

perf auxtrace: Fix address filter duplicate symbol selection

When a match has been made to the nth duplicate symbol, return
success not error.

Example:

  Before:

    $ cat file.c
    cat: file.c: No such file or directory
    $ cat file1.c
    #include <stdio.h>

    static void func(void)
    {
            printf("First func\n");
    }

    void other(void);

    int main()
    {
            func();
            other();
            return 0;
    }
    $ cat file2.c
    #include <stdio.h>

    static void func(void)
    {
            printf("Second func\n");
    }

    void other(void)
    {
            func();
    }

    $ gcc -Wall -Wextra -o test file1.c file2.c
    $ perf record -e intel_pt//u --filter 'filter func @ ./test' -- ./test
    Multiple symbols with name 'func'
    #1      0x1149  l       func
                    which is near           main
    #2      0x1179  l       func
                    which is near           other
    Disambiguate symbol name by inserting #n after the name e.g. func #2
    Or select a global symbol by inserting #0 or #g or #G
    Failed to parse address filter: 'filter func @ ./test'
    Filter format is: filter|start|stop|tracestop <start symbol or address> [/ <end symbol or size>] [@<file name>]
    Where multiple filters are separated by space or comma.
    $ perf record -e intel_pt//u --filter 'filter func #2 @ ./test' -- ./test
    Failed to parse address filter: 'filter func #2 @ ./test'
    Filter format is: filter|start|stop|tracestop <start symbol or address> [/ <end symbol or size>] [@<file name>]
    Where multiple filters are separated by space or comma.

  After:

    $ perf record -e intel_pt//u --filter 'filter func #2 @ ./test' -- ./test
    First func
    Second func
    [ perf record: Woken up 1 times to write data ]
    [ perf record: Captured and wrote 0.016 MB perf.data ]
    $ perf script --itrace=b -Ftime,flags,ip,sym,addr --ns
    1231062.526977619:   tr strt                               0 [unknown] =>     558495708179 func
    1231062.526977619:   tr end  call               558495708188 func =>     558495708050 _init
    1231062.526979286:   tr strt                               0 [unknown] =>     55849570818d func
    1231062.526979286:   tr end  return             55849570818f func =>     55849570819d other

Fixes: 1b36c03e356936d6 ("perf record: Add support for using symbols in address filters")
Reported-by: Dmitrii Dolgov <[email protected]>
Signed-off-by: Adrian Hunter <[email protected]>
Tested-by: Dmitry Dolgov <[email protected]>
Cc: Adrian Hunter <[email protected]>
Cc: Ian Rogers <[email protected]>
Cc: Jiri Olsa <[email protected]>
Cc: Namhyung Kim <[email protected]>
Cc: [email protected]
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Arnaldo Carvalho de Melo <[email protected]>

drm/i915/gt: Cover rest of SVG unit MCR registers

CHICKEN_RASTER_{1,2} got overlooked with the move done in commit
a9e69428b1b4 ("drm/i915: Define MCR registers explicitly"). Registers
from the SVG unit became multicast as of Xe_HP graphics.

BSpec: 66534
Fixes: a9e69428b1b4 ("drm/i915: Define MCR registers explicitly")
Signed-off-by: Gustavo Sousa <[email protected]>
Cc: Matt Roper <[email protected]>
Reviewed-by: Matt Roper <[email protected]>
Signed-off-by: Matt Roper <[email protected]>
Link: https://patchwork.freedesktop.org/patch/msgid/[email protected]
(cherry picked from commit 10903b0a0f4d4964b352fa3df12d3d2ef5fb7a3b)
Signed-off-by: Rodrigo Vivi <[email protected]>

KVM: s390: interrupt: use READ_ONCE() before cmpxchg()

Use READ_ONCE() before cmpxchg() to prevent that the compiler generates
code that fetches the to be compared old value several times from memory.

Reviewed-by: Christian Borntraeger <[email protected]>
Acked-by: Christian Borntraeger <[email protected]>
Reviewed-by: Claudio Imbrenda <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Heiko Carstens <[email protected]>

s390/percpu: add READ_ONCE() to arch_this_cpu_to_op_simple()

Make sure that *ptr__ within arch_this_cpu_to_op_simple() is only
dereferenced once by using READ_ONCE(). Otherwise the compiler could
generate incorrect code.

Cc: <[email protected]>
Reviewed-by: Alexander Gordeev <[email protected]>
Signed-off-by: Heiko Carstens <[email protected]>

s390/cpum_sf: add READ_ONCE() semantics to compare and swap loops

The current cmpxchg_double() loops within the perf hw sampling code do not
have READ_ONCE() semantics to read the old value from memory. This allows
the compiler to generate code which reads the "old" value several times
from memory, which again allows for inconsistencies.

For example:

        /* Reset trailer (using compare-double-and-swap) */
        do {
                te_flags = te->flags & ~SDB_TE_BUFFER_FULL_MASK;
                te_flags |= SDB_TE_ALERT_REQ_MASK;
        } while (!cmpxchg_double(&te->flags, &te->overflow,
                 te->flags, te->overflow,
                 te_flags, 0ULL));

The compiler could generate code where te->flags used within the
cmpxchg_double() call may be refetched from memory and which is not
necessarily identical to the previous read version which was used to
generate te_flags. Which in turn means that an incorrect update could
happen.

Fix this by adding READ_ONCE() semantics to all cmpxchg_double()
loops. Given that READ_ONCE() cannot generate code on s390 which atomically
reads 16 bytes, use a private compare-and-swap-double implementation to
achieve that.

Also replace cmpxchg_double() with the private implementation to be able to
re-use the old value within the loops.

As a side effect this converts the whole code to only use bit fields
to read and modify bits within the hws trailer header.

Reported-by: Alexander Gordeev <[email protected]>
Acked-by: Alexander Gordeev <[email protected]>
Acked-by: Hendrik Brueckner <[email protected]>
Reviewed-by: Thomas Richter <[email protected]>
Cc: <[email protected]>
Link: https://lore.kernel.org/linux-s390/Y71QJBhNTIatvxUT@osiris/T/#ma14e2a5f7aa8ed4b94b6f9576799b3ad9c60f333
Signed-off-by: Heiko Carstens <[email protected]>

spi: Merge rename of spi-cs-setup-ns DT property

The newly added spi-cs-setup-ns doesn't really fit with the existing
property names for delays, rename it so that it does before it makes it
into a release and becomes ABI.

spi: spidev: remove debug messages that access spidev->spi without locking

The two debug messages in spidev_open() dereference spidev->spi without
taking the lock and without checking if it's not null. This can lead to
a crash. Drop the messages as they're not needed - the user-space will
get informed about ENOMEM with the syscall return value.

Signed-off-by: Bartosz Golaszewski <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Mark Brown <[email protected]>

spi: spidev: fix a race condition when accessing spidev->spi

There's a spinlock in place that is taken in file_operations callbacks
whenever we check if spidev->spi is still alive (not null). It's also
taken when spidev->spi is set to NULL in remove().

This however doesn't protect the code against driver unbind event while
one of the syscalls is still in progress. To that end we need a lock taken
continuously as long as we may still access spidev->spi. As both the file
ops and the remove callback are never called from interrupt context, we
can replace the spinlock with a mutex.

Signed-off-by: Bartosz Golaszewski <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Mark Brown <[email protected]>

Merge tag 'mlx5-fixes-2023-01-09' of git://git.kernel.org/pub/scm/linux/kernel/git/saeed/linux

mlx5-fixes-2023-01-09

ipv6: raw: Deduct extension header length in rawv6_push_pending_frames

The total cork length created by ip6_append_data includes extension
headers, so we must exclude them when comparing them against the
IPV6_CHECKSUM offset which does not include extension headers.

Reported-by: Kyle Zeng <[email protected]>
Fixes: 357b40a18b04 ("[IPV6]: IPV6_CHECKSUM socket option can corrupt kernel memory")
Signed-off-by: Herbert Xu <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

drm/nouveau: Remove file nouveau_fbcon.c

Commit 4a16dd9d18a0 ("drm/nouveau/kms: switch to drm fbdev helpers")
converted nouveau to generic fbdev emulation. The driver's internal
implementation later got accidentally restored during a merge commit.
Remove the file from the driver. No functional changes.

v2:
* point Fixes tag to merge commit (Alex)

Signed-off-by: Thomas Zimmermann <[email protected]>
Reviewed-by: Alex Deucher <[email protected]>
Fixes: 4e291f2f5853 ("Merge tag 'drm-misc-next-2022-11-10-1' of git://anongit.freedesktop.org/drm/drm-misc into drm-next")
Cc: Ben Skeggs <[email protected]>
Cc: Karol Herbst <[email protected]>
Cc: Lyude Paul <[email protected]>
Cc: Thomas Zimmermann <[email protected]>
Cc: Javier Martinez Canillas <[email protected]>
Cc: Sam Ravnborg <[email protected]>
Cc: Jani Nikula <[email protected]>
Cc: Dave Airlie <[email protected]>
Cc: [email protected]
Cc: [email protected]
Link: https://patchwork.freedesktop.org/patch/msgid/[email protected]

net: lan966x: check for ptp to be enabled in lan966x_ptp_deinit()

If ptp was not enabled due to missing IRQ for instance,
lan966x_ptp_deinit() will dereference NULL pointers.

Fixes: d096459494a8 ("net: lan966x: Add support for ptp clocks")
Signed-off-by: Clément Léger <[email protected]>
Reviewed-by: Horatiu Vultur <[email protected]>
Reviewed-by: Jiri Pirko <[email protected]>
Signed-off-by: David S. Miller <[email protected]>

powerpc/imc-pmu: Fix use of mutex in IRQs disabled section

Current imc-pmu code triggers a WARNING with CONFIG_DEBUG_ATOMIC_SLEEP
and CONFIG_PROVE_LOCKING enabled, while running a thread_imc event.

Command to trigger the warning:
  # perf stat -e thread_imc/CPM_CS_FROM_L4_MEM_X_DPTEG/ sleep 5

   Performance counter stats for 'sleep 5':

                   0      thread_imc/CPM_CS_FROM_L4_MEM_X_DPTEG/

         5.002117947 seconds time elapsed

         0.000131000 seconds user
         0.001063000 seconds sys

Below is snippet of the warning in dmesg:

  BUG: sleeping function called from invalid context at kernel/locking/mutex.c:580
  in_atomic(): 1, irqs_disabled(): 1, non_block: 0, pid: 2869, name: perf-exec
  preempt_count: 2, expected: 0
  4 locks held by perf-exec/2869:
   #0: c00000004325c540 (&sig->cred_guard_mutex){+.+.}-{3:3}, at: bprm_execve+0x64/0xa90
   #1: c00000004325c5d8 (&sig->exec_update_lock){++++}-{3:3}, at: begin_new_exec+0x460/0xef0
   #2: c0000003fa99d4e0 (&cpuctx_lock){-...}-{2:2}, at: perf_event_exec+0x290/0x510
   #3: c000000017ab8418 (&ctx->lock){....}-{2:2}, at: perf_event_exec+0x29c/0x510
  irq event stamp: 4806
  hardirqs last  enabled at (4805): [<c000000000f65b94>] _raw_spin_unlock_irqrestore+0x94/0xd0
  hardirqs last disabled at (4806): [<c0000000003fae44>] perf_event_exec+0x394/0x510
  softirqs last  enabled at (0): [<c00000000013c404>] copy_process+0xc34/0x1ff0
  softirqs last disabled at (0): [<0000000000000000>] 0x0
  CPU: 36 PID: 2869 Comm: perf-exec Not tainted 6.2.0-rc2-00011-g1247637727f2 #61
  Hardware name: 8375-42A POWER9 0x4e1202 opal:v7.0-16-g9b85f7d961 PowerNV
  Call Trace:
    dump_stack_lvl+0x98/0xe0 (unreliable)
    __might_resched+0x2f8/0x310
    __mutex_lock+0x6c/0x13f0
    thread_imc_event_add+0xf4/0x1b0
    event_sched_in+0xe0/0x210
    merge_sched_in+0x1f0/0x600
    visit_groups_merge.isra.92.constprop.166+0x2bc/0x6c0
    ctx_flexible_sched_in+0xcc/0x140
    ctx_sched_in+0x20c/0x2a0
    ctx_resched+0x104/0x1c0
    perf_event_exec+0x340/0x510
    begin_new_exec+0x730/0xef0
    load_elf_binary+0x3f8/0x1e10
  ...
  do not call blocking ops when !TASK_RUNNING; state=2001 set at [<00000000fd63e7cf>] do_nanosleep+0x60/0x1a0
  WARNING: CPU: 36 PID: 2869 at kernel/sched/core.c:9912 __might_sleep+0x9c/0xb0
  CPU: 36 PID: 2869 Comm: sleep Tainted: G        W          6.2.0-rc2-00011-g1247637727f2 #61
  Hardware name: 8375-42A POWER9 0x4e1202 opal:v7.0-16-g9b85f7d961 PowerNV
  NIP:  c000000000194a1c LR: c000000000194a18 CTR: c000000000a78670
  REGS: c00000004d2134e0 TRAP: 0700   Tainted: G        W           (6.2.0-rc2-00011-g1247637727f2)
  MSR:  9000000000021033 <SF,HV,ME,IR,DR,RI,LE>  CR: 48002824  XER: 00000000
  CFAR: c00000000013fb64 IRQMASK: 1

The above warning triggered because the current imc-pmu code uses mutex
lock in interrupt disabled sections. The function mutex_lock()
internally calls __might_resched(), which will check if IRQs are
disabled and in case IRQs are disabled, it will trigger the warning.

Fix the issue by changing the mutex lock to spinlock.

Fixes: 8f95faaac56c ("powerpc/powernv: Detect and create IMC device")
Reported-by: Michael Petlan <[email protected]>
Reported-by: Peter Zijlstra <[email protected]>
Signed-off-by: Kajol Jain <[email protected]>
[mpe: Fix comments, trim oops in change log, add reported-by tags]
Signed-off-by: Michael Ellerman <[email protected]>
Link: https://lore.kernel.org/r/[email protected]

powerpc/boot: Fix incorrect version calculation issue in ld_version

The ld_version() function computes the wrong version value for certain
ld versions such as the following:

  $ ld --version
  GNU ld (GNU Binutils; SUSE Linux Enterprise 15)
  2.37.20211103-150100.7.37

For input 2.37.20211103, the value computed is 202348030000 which is
higher than the value for a later version like 2.39.0, which is
23900000.

This issue was highlighted because with the above ld version, the
powerpc kernel build started failing with ld error: "unrecognized option
--no-warn-rwx-segments". This was caused due to the recent commit
579aee9fc594 ("powerpc: suppress some linker warnings in recent linker
versions") which added the --no-warn-rwx-segments linker flag if the ld
version is greater than 2.39.

Due to the bug in ld_version(), ld version 2.37.20111103 is wrongly
calculated to be greater than 2.39 and the unsupported flag is added.

To fix it, if version is of the form x.y.z and length(z) == 8, then most
probably it is a date [yyyymmdd] commonly used for release snapshots and
not an actual new version. Hence, ignore the date part replacing it with
0.

Fixes: 579aee9fc594 ("powerpc: suppress some linker warnings in recent linker versions")
Signed-off-by: Ojaswin Mujoo <[email protected]>
[mpe: Tweak change log wording/formatting, add Fixes tag]
Signed-off-by: Michael Ellerman <[email protected]>
Link: https://lore.kernel.org/r/[email protected]

cifs: fix potential memory leaks in session setup

Make sure to free cifs_ses::auth_key.response before allocating it as
we might end up leaking memory in reconnect or mounting.

Signed-off-by: Paulo Alcantara (SUSE) <[email protected]>
Signed-off-by: Steve French <[email protected]>

cifs: do not query ifaces on smb1 mounts

Users have reported the following error on every 600 seconds
(SMB_INTERFACE_POLL_INTERVAL) when mounting SMB1 shares:

CIFS: VFS: \\srv\share error -5 on ioctl to get interface list

It's supported only by SMB2+, so do not query network interfaces on
SMB1 mounts.

Fixes: 6e1c1c08cdf3 ("cifs: periodically query network interfaces from server")
Reviewed-by: Shyam Prasad N <[email protected]>
Reviewed-by: Tom Talpey <[email protected]>
Signed-off-by: Paulo Alcantara (SUSE) <[email protected]>
Signed-off-by: Steve French <[email protected]>

Merge branch '100GbE' of git://git.kernel.org/pub/scm/linux/kernel/git/tnguy/net-queue

Tony Nguyen says:

====================
Intel Wired LAN Driver Updates 2023-01-09 (ice)

This series contains updates to ice driver only.

Jiasheng Jiang frees allocated cmd_buf if write_buf allocation failed to
prevent memory leak.

Yuan Can adds check, and proper cleanup, of gnss_tty_port allocation call
to avoid memory leaks.

* '100GbE' of git://git.kernel.org/pub/scm/linux/kernel/git/tnguy/net-queue:
  ice: Add check for kzalloc
  ice: Fix potential memory leak in ice_gnss_tty_write()
====================

Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Jakub Kicinski <[email protected]>

net: sched: disallow noqueue for qdisc classes

While experimenting with applying noqueue to a classful queue discipline,
we discovered a NULL pointer dereference in the __dev_queue_xmit()
path that generates a kernel OOPS:

    # dev=enp0s5
    # tc qdisc replace dev $dev root handle 1: htb default 1
    # tc class add dev $dev parent 1: classid 1:1 htb rate 10mbit
    # tc qdisc add dev $dev parent 1:1 handle 10: noqueue
    # ping -I $dev -w 1 -c 1 1.1.1.1

[    2.172856] BUG: kernel NULL pointer dereference, address: 0000000000000000
[    2.173217] #PF: supervisor instruction fetch in kernel mode
...
[    2.178451] Call Trace:
[    2.178577]  <TASK>
[    2.178686]  htb_enqueue+0x1c8/0x370
[    2.178880]  dev_qdisc_enqueue+0x15/0x90
[    2.179093]  __dev_queue_xmit+0x798/0xd00
[    2.179305]  ? _raw_write_lock_bh+0xe/0x30
[    2.179522]  ? __local_bh_enable_ip+0x32/0x70
[    2.179759]  ? ___neigh_create+0x610/0x840
[    2.179968]  ? eth_header+0x21/0xc0
[    2.180144]  ip_finish_output2+0x15e/0x4f0
[    2.180348]  ? dst_output+0x30/0x30
[    2.180525]  ip_push_pending_frames+0x9d/0xb0
[    2.180739]  raw_sendmsg+0x601/0xcb0
[    2.180916]  ? _raw_spin_trylock+0xe/0x50
[    2.181112]  ? _raw_spin_unlock_irqrestore+0x16/0x30
[    2.181354]  ? get_page_from_freelist+0xcd6/0xdf0
[    2.181594]  ? sock_sendmsg+0x56/0x60
[    2.181781]  sock_sendmsg+0x56/0x60
[    2.181958]  __sys_sendto+0xf7/0x160
[    2.182139]  ? handle_mm_fault+0x6e/0x1d0
[    2.182366]  ? do_user_addr_fault+0x1e1/0x660
[    2.182627]  __x64_sys_sendto+0x1b/0x30
[    2.182881]  do_syscall_64+0x38/0x90
[    2.183085]  entry_SYSCALL_64_after_hwframe+0x63/0xcd
...
[    2.187402]  </TASK>

Previously in commit d66d6c3152e8 ("net: sched: register noqueue
qdisc"), NULL was set for the noqueue discipline on noqueue init
so that __dev_queue_xmit() falls through for the noqueue case. This
also sets a bypass of the enqueue NULL check in the
register_qdisc() function for the struct noqueue_disc_ops.

Classful queue disciplines make it past the NULL check in
__dev_queue_xmit() because the discipline is set to htb (in this case),
and then in the call to __dev_xmit_skb(), it calls into htb_enqueue()
which grabs a leaf node for a class and then calls qdisc_enqueue() by
passing in a queue discipline which assumes ->enqueue() is not set to NULL.

Fix this by not allowing classes to be assigned to the noqueue
discipline. Linux TC Notes states that classes cannot be set to
the noqueue discipline. [1] Let's enforce that here.

Links:
1. https://linux-tc-notes.sourceforge.net/tc/doc/sch_noqueue.txt

Fixes: d66d6c3152e8 ("net: sched: register noqueue qdisc")
Cc: [email protected]
Signed-off-by: Frederick Lawler <[email protected]>
Reviewed-by: Jakub Sitnicki <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Jakub Kicinski <[email protected]>

drm/amdkfd: Fix NULL pointer error for GC 11.0.1 on mGPU

The point bo->kfd_bo is NULL for queue's write pointer BO
when creating queue on mGPU. To avoid using the pointer
fixes the error.

Signed-off-by: Eric Huang <[email protected]>
Reviewed-by: Felix Kuehling <[email protected]>
Signed-off-by: Alex Deucher <[email protected]>

drm/amd/pm/smu13: BACO is supported when it's in BACO state

This leverages the logic in smu11. No need to talk to SMU to
check BACO enablement as it's in BACO state already.

Signed-off-by: Guchun Chen <[email protected]>
Reviewed-by: Kenneth Feng <[email protected]>
Reviewed-by: Lijo Lazar <[email protected]>
Signed-off-by: Alex Deucher <[email protected]>
Cc: [email protected] # 6.0, 6.1

drm/amdkfd: Add sync after creating vram bo

There will be data corruption on vram allocated by svm
if the initialization is not complete and application is
writting on the memory. Adding sync to wait for the
initialization completion is to resolve this issue.

Signed-off-by: Eric Huang <[email protected]>
Reviewed-by: Felix Kuehling <[email protected]>
Signed-off-by: Alex Deucher <[email protected]>

cifs: fix double free on failed kerberos auth

If session setup failed with kerberos auth, we ended up freeing
cifs_ses::auth_key.response twice in SMB2_auth_kerberos() and
sesInfoFree().

Fix this by zeroing out cifs_ses::auth_key.response after freeing it
in SMB2_auth_kerberos().

Fixes: a4e430c8c8ba ("cifs: replace kfree() with kfree_sensitive() for sensitive data")
Signed-off-by: Paulo Alcantara (SUSE) <[email protected]>
Acked-by: Ronnie Sahlberg <[email protected]>
Signed-off-by: Steve French <[email protected]>

cifs: remove redundant assignment to the variable match

The variable match is being assigned a value that is never read, it
is being re-assigned a new value later on. The assignment is redundant
and can be removed.

Cleans up clang scan-build warning:
fs/cifs/dfs_cache.c:1302:2: warning: Value stored to 'match' is never read

Signed-off-by: Colin Ian King <[email protected]>
Reviewed-by: Paulo Alcantara (SUSE) <[email protected]>
Signed-off-by: Steve French <[email protected]>

Merge tag 'nfsd-6.2-3' of git://git.kernel.org/pub/scm/linux/kernel/git/cel/linux

Pull nfsd fixes from Chuck Lever:

 - Fix a race when creating NFSv4 files

 - Revert the use of relaxed bitops

* tag 'nfsd-6.2-3' of git://git.kernel.org/pub/scm/linux/kernel/git/cel/linux:
  NFSD: Use set_bit(RQ_DROPME)
  Revert "SUNRPC: Use RMW bitops in single-threaded hot paths"
  nfsd: fix handling of cached open files in nfsd4_open codepath

Merge tag 'xtensa-20230110' of https://github.com/jcmvbkbc/linux-xtensa

Pull xtensa fixes from Max Filippov:

 - fix xtensa allmodconfig build broken by the kcsan test

 - drop unused members of struct thread_struct

* tag 'xtensa-20230110' of https://github.com/jcmvbkbc/linux-xtensa:
  xtensa: drop unused members of struct thread_struct
  kcsan: test: don't put the expect array on the stack

iavf/iavf_main: actually log ->src mask when talking about it

This fixes a copy-paste issue where dev_err would log the dst mask even
though it is clearly talking about src.

Found by Linux Verification Center (linuxtesting.org) with the SVACE
static analysis tool.

Fixes: 0075fa0fadd0 ("i40evf: Add support to apply cloud filters")
Signed-off-by: Daniil Tatianin <[email protected]>
Reviewed-by: Michal Swiatkowski <[email protected]>
Signed-off-by: Tony Nguyen <[email protected]>

igc: Fix PPS delta between two synchronized end-points

This patch fix the pulse per second output delta between
two synchronized end-points.

Based on Intel Discrete I225 Software User Manual Section
4.2.15 TimeSync Auxiliary Control Register, ST0[Bit 4] and
ST1[Bit 7] must be set to ensure that clock output will be
toggles based on frequency value defined. This is to ensure
that output of the PPS is aligned with the clock.

How to test:

1) Running time synchronization on both end points.
Ex: ptp4l --step_threshold=1 -m -f gPTP.cfg -i <interface name>

2) Configure PPS output using below command for both end-points
Ex: SDP0 on I225 REV4 SKU variant

./testptp -d /dev/ptp0 -L 0,2
./testptp -d /dev/ptp0 -p 1000000000

3) Measure the output using analyzer for both end-points

Fixes: 87938851b6ef ("igc: enable auxiliary PHC functions for the i225")
Signed-off-by: Christopher S Hall <[email protected]>
Signed-off-by: Muhammad Husaini Zulkifli <[email protected]>
Acked-by: Sasha Neftin <[email protected]>
Tested-by: Naama Meir <[email protected]>
Signed-off-by: Tony Nguyen <[email protected]>

ixgbe: fix pci device refcount leak

As the comment of pci_get_domain_bus_and_slot() says, it
returns a PCI device with refcount incremented, when finish
using it, the caller must decrement the reference count by
calling pci_dev_put().

In ixgbe_get_first_secondary_devfn() and ixgbe_x550em_a_has_mii(),
pci_dev_put() is called to avoid leak.

Fixes: 8fa10ef01260 ("ixgbe: register a mdiobus")
Signed-off-by: Yang Yingliang <[email protected]>
Tested-by: Gurucharan G <[email protected]> (A Contingent worker at Intel)
Signed-off-by: Tony Nguyen <[email protected]>

cpufreq: amd-pstate: fix kernel hang issue while amd-pstate unregistering

In the amd_pstate_adjust_perf(), there is one cpufreq_cpu_get() call to
increase increments the kobject reference count of policy and make it as
busy. Therefore, a corresponding call to cpufreq_cpu_put() is needed to
decrement the kobject reference count back, it will resolve the kernel
hang issue when unregistering the amd-pstate driver and register the
`amd_pstate_epp` driver instance.

Fixes: 1d215f0319 ("cpufreq: amd-pstate: Add fast switch function for AMD P-State")
Acked-by: Huang Rui <[email protected]>
Reviewed-by: Mario Limonciello <[email protected]>
Tested-by: Wyes Karny <[email protected]>
Signed-off-by: Perry Yuan <[email protected]>
Cc: 5.17+ <[email protected]> # 5.17+
Signed-off-by: Rafael J. Wysocki <[email protected]>

Merge tag 'cpufreq/arm/fixes-6.2-rc4' of git://git.kernel.org/pub/scm/linux/kernel/git/vireshk/pm

Pull cpufreq ARM fixes for 6.2-rc4 from Viresh Kumar:

"- Fix double initialization and set suspend-freq for Apple's cpufreq
   driver (Arnd Bergmann and Hector Martin).

 - Fix reading of "reg" property, update cpufreq-dt's blocklist and
   update DT documentation for Qualcomm's cpufreq driver (Konrad Dybcio
   and Krzysztof Kozlowski).

 - Replace 0 with NULL for Armada driver (Miles Chen).

 - Fix potential overflows in CPPC driver (Pierre Gondois).

 - Update blocklist for Tegra234 Soc (Sumit Gupta)."

ACPI: Fix selecting wrong ACPI fwnode for the iGPU on some Dell laptops

The Dell Latitude E6430 both with and without the optional NVidia dGPU
has a bug in its ACPI tables which is causing Linux to assign the wrong
ACPI fwnode / companion to the pci_device for the i915 iGPU.

Specifically under the PCI root bridge there are these 2 ACPI Device()s :

 Scope (_SB.PCI0)
 {
     Device (GFX0)
     {
         Name (_ADR, 0x00020000)  // _ADR: Address
     }

     ...

     Device (VID)
     {
         Name (_ADR, 0x00020000)  // _ADR: Address
         ...

         Method (_DOS, 1, NotSerialized)  // _DOS: Disable Output Switching
         {
             VDP8 = Arg0
             VDP1 (One, VDP8)
         }

         Method (_DOD, 0, NotSerialized)  // _DOD: Display Output Devices
         {
             ...
         }
         ...
     }
 }

The non-functional GFX0 ACPI device is a problem, because this gets
returned as ACPI companion-device by acpi_find_child_device() for the iGPU.

This is a long standing problem and the i915 driver does use the ACPI
companion for some things, but works fine without it.

However since commit 63f534b8bad9 ("ACPI: PCI: Rework acpi_get_pci_dev()")
acpi_get_pci_dev() relies on the physical-node pointer in the acpi_device
and that is set on the wrong acpi_device because of the wrong
acpi_find_child_device() return. This breaks the ACPI video code,
leading to non working backlight control in some cases.

Add a type.backlight flag, mark ACPI video bus devices with this and make
find_child_checks() return a higher score for children with this flag set,
so that it picks the right companion-device.

Fixes: 63f534b8bad9 ("ACPI: PCI: Rework acpi_get_pci_dev()")
Co-developed-by: Rafael J. Wysocki <[email protected]>
Signed-off-by: Hans de Goede <[email protected]>
Cc: 6.1+ <[email protected]> # 6.1+
Signed-off-by: Rafael J. Wysocki <[email protected]>

ACPI: video: Allow selecting NVidia-WMI-EC or Apple GMUX backlight from the cmdline

The patches adding NVidia-WMI-EC and Apple GMUX backlight detection
support to acpi_video_get_backlight_type(), forgot to update
acpi_video_parse_cmdline() to allow manually selecting these from
the commandline.

Add support for these to acpi_video_parse_cmdline().

Fixes: fe7aebb40d42 ("ACPI: video: Add Nvidia WMI EC brightness control detection (v3)")
Fixes: 21245df307cb ("ACPI: video: Add Apple GMUX brightness control detection")
Signed-off-by: Hans de Goede <[email protected]>
Signed-off-by: Rafael J. Wysocki <[email protected]>

ACPI: resource: Skip IRQ override on Asus Expertbook B2402CBA

Like the Asus Expertbook B2502CBA and various Asus Vivobook laptops,
the Asus Expertbook B2402CBA has an ACPI DSDT table that describes IRQ 1
as ActiveLow while the kernel overrides it to Edge_High. This prevents the
keyboard from working. To fix this issue, add this laptop to the
skip_override_table so that the kernel does not override IRQ 1.

Link: https://bugzilla.kernel.org/show_bug.cgi?id=216864
Tested-by: zelenat <[email protected]>
Signed-off-by: Tamim Khan <[email protected]>
Signed-off-by: Rafael J. Wysocki <[email protected]>

x86/resctrl: Fix event counts regression in reused RMIDs

When creating a new monitoring group, the RMID allocated for it may have
been used by a group which was previously removed. In this case, the
hardware counters will have non-zero values which should be deducted
from what is reported in the new group's counts.

resctrl_arch_reset_rmid() initializes the prev_msr value for counters to
0, causing the initial count to be charged to the new group. Resurrect
__rmid_read() and use it to initialize prev_msr correctly.

Unlike before, __rmid_read() checks for error bits in the MSR read so
that callers don't need to.

Fixes: 1d81d15db39c ("x86/resctrl: Move mbm_overflow_count() into resctrl_arch_rmid_read()")
Signed-off-by: Peter Newman <[email protected]>
Signed-off-by: Borislav Petkov (AMD) <[email protected]>
Reviewed-by: Reinette Chatre <[email protected]>
Tested-by: Babu Moger <[email protected]>
Cc: [email protected]
Link: https://lore.kernel.org/r/[email protected]

x86/resctrl: Fix task CLOSID/RMID update race

When the user moves a running task to a new rdtgroup using the task's
file interface or by deleting its rdtgroup, the resulting change in
CLOSID/RMID must be immediately propagated to the PQR_ASSOC MSR on the
task(s) CPUs.

x86 allows reordering loads with prior stores, so if the task starts
running between a task_curr() check that the CPU hoisted before the
stores in the CLOSID/RMID update then it can start running with the old
CLOSID/RMID until it is switched again because __rdtgroup_move_task()
failed to determine that it needs to be interrupted to obtain the new
CLOSID/RMID.

Refer to the diagram below:

CPU 0                                   CPU 1
-----                                   -----
__rdtgroup_move_task():
  curr <- t1->cpu->rq->curr
                                        __schedule():
                                          rq->curr <- t1
                                        resctrl_sched_in():
                                          t1->{closid,rmid} -> {1,1}
  t1->{closid,rmid} <- {2,2}
  if (curr == t1) // false
   IPI(t1->cpu)

A similar race impacts rdt_move_group_tasks(), which updates tasks in a
deleted rdtgroup.

In both cases, use smp_mb() to order the task_struct::{closid,rmid}
stores before the loads in task_curr().  In particular, in the
rdt_move_group_tasks() case, simply execute an smp_mb() on every
iteration with a matching task.

It is possible to use a single smp_mb() in rdt_move_group_tasks(), but
this would require two passes and a means of remembering which
task_structs were updated in the first loop. However, benchmarking
results below showed too little performance impact in the simple
approach to justify implementing the two-pass approach.

Times below were collected using `perf stat` to measure the time to
remove a group containing a 1600-task, parallel workload.

CPU: Intel(R) Xeon(R) Platinum P-8136 CPU @ 2.00GHz (112 threads)

  # mkdir /sys/fs/resctrl/test
  # echo $$ > /sys/fs/resctrl/test/tasks
  # perf bench sched messaging -g 40 -l 100000

task-clock time ranges collected using:

  # perf stat rmdir /sys/fs/resctrl/test

Baseline:                     1.54 - 1.60 ms
smp_mb() every matching task: 1.57 - 1.67 ms

  [ bp: Massage commit message. ]

Fixes: ae28d1aae48a ("x86/resctrl: Use an IPI instead of task_work_add() to update PQR_ASSOC MSR")
Fixes: 0efc89be9471 ("x86/intel_rdt: Update task closid immediately on CPU in rmdir and unmount")
Signed-off-by: Peter Newman <[email protected]>
Signed-off-by: Borislav Petkov (AMD) <[email protected]>
Reviewed-by: Reinette Chatre <[email protected]>
Reviewed-by: Babu Moger <[email protected]>
Cc: <[email protected]>
Link: https://lore.kernel.org/r/[email protected]

io_uring/fdinfo: include locked hash table in fdinfo output

A previous commit split the hash table for polled requests into two
parts, but didn't get the fdinfo output updated. This means that it's
less useful for debugging, as we may think a given request is not pending
poll.

Fix this up by dumping the locked hash table contents too.

Fixes: 9ca9fb24d5fe ("io_uring: mutex locked poll hashing")
Signed-off-by: Jens Axboe <[email protected]>

x86/pat: Fix pat_x_mtrr_type() for MTRR disabled case

Since

  72cbc8f04fe2 ("x86/PAT: Have pat_enabled() properly reflect state when running on Xen")

PAT can be enabled without MTRR.

This has resulted in problems e.g. for a SEV-SNP guest running under Hyper-V,
when trying to establish a new mapping via memremap() with WB caching mode, as
pat_x_mtrr_type() will call mtrr_type_lookup(), which in turn is returning
MTRR_TYPE_INVALID due to MTRR being disabled in this configuration.

The result is a mapping with UC- caching, leading to severe performance
degradation.

Fix that by handling MTRR_TYPE_INVALID the same way as MTRR_TYPE_WRBACK
in pat_x_mtrr_type() because MTRR_TYPE_INVALID means MTRRs are disabled.

  [ bp: Massage commit message. ]

Fixes: 72cbc8f04fe2 ("x86/PAT: Have pat_enabled() properly reflect state when running on Xen")
Reported-by: Michael Kelley (LINUX) <[email protected]>
Signed-off-by: Juergen Gross <[email protected]>
Signed-off-by: Borislav Petkov (AMD) <[email protected]>
Reviewed-by: Michael Kelley <[email protected]>
Tested-by: Michael Kelley <[email protected]>
Cc: <[email protected]>
Link: https://lore.kernel.org/r/[email protected]

drm/i915/gt: Reset twice

After applying an engine reset, on some platforms like Jasperlake, we
occasionally detect that the engine state is not cleared until shortly
after the resume. As we try to resume the engine with volatile internal
state, the first request fails with a spurious CS event (it looks like
it reports a lite-restore to the hung context, instead of the expected
idle->active context switch).

Signed-off-by: Chris Wilson <[email protected]>
Cc: [email protected]
Cc: Mika Kuoppala <[email protected]>
Signed-off-by: Andi Shyti <[email protected]>
Reviewed-by: Gwan-gyeong Mun <[email protected]>
Link: https://patchwork.freedesktop.org/patch/msgid/[email protected]
(cherry picked from commit 3db9d590557da3aa2c952f2fecd3e9b703dad790)
Signed-off-by: Rodrigo Vivi <[email protected]>

perf bpf: Avoid build breakage with libbpf < 0.8.0 + LIBBPF_DYNAMIC=1

In 746bd29e348f99b4 ("perf build: Use tools/lib headers from install
path") we stopped having the tools/lib/ directory from the kernel
sources in the header include path unconditionally, which breaks the
build on systems with older versions of libbpf-devel, in this case 0.7.0
as some of the structures and function declarations present in the newer
version of libbpf included in the kernel sources (tools/lib/bpf) are not
anymore used, just the ones in the system libbpf.

So instead of trying to provide alternative functions when the
libbpf-bpf_program__set_insns feature test fails, fail a
LIBBPF_DYNAMIC=1 build (requesting the use of the system's libbpf) and
emit this build error message:

  $ make LIBBPF_DYNAMIC=1 -C tools/perf
  Makefile.config:593: *** Error: libbpf devel library needs to be >= 0.8.0 to build with LIBBPF_DYNAMIC, update or build statically with the version that comes with the kernel sources.  Stop.
  $

For v6.3 these tests will be revamped and we'll require libbpf 1.0 as a
minimal version for using LIBBPF_DYNAMIC=1, most distros should have it
by now or at v6.3 time.

Fixes: 746bd29e348f99b4 ("perf build: Use tools/lib headers from install path")
Cc: Alexander Shishkin <[email protected]>
Cc: Ian Rogers <[email protected]>
Cc: Ingo Molnar <[email protected]>
Cc: Jiri Olsa <[email protected]>
Cc: Mark Rutland <[email protected]>
Cc: Mike Leach <[email protected]>
Cc: Namhyung Kim <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Link: http://lore.kernel.org/lkml/CAP-5=fVa51_URGsdDFVTzpyGmdDRj_Dj2EKPuDHNQ0BYgMSzUA@mail.gmail.com
Signed-off-by: Arnaldo Carvalho de Melo <[email protected]>

perf build: Fix build error when NO_LIBBPF=1

The $(LIBBPF) target should only be a dependency of prepare if the
static version of libbpf is needed. Add a new LIBBPF_STATIC variable
that is set by Makefile.config. Use LIBBPF_STATIC to determine whether
the CFLAGS, etc. need updating and for adding $(LIBBPF) as a prepare
dependency.

As Makefile.config isn't loaded for "clean" as a target, always set
LIBBPF_OUTPUT regardless of whether it is needed for $(LIBBPF). This
is done to minimize conditional logic for $(LIBBPF)-clean.

This issue and an original fix was reported by Mike Leach in:
https://lore.kernel.org/lkml/20230105172243[email protected]/

Fixes: 746bd29e348f99b4 ("perf build: Use tools/lib headers from install path")
Reported-by: Mike Leach <[email protected]>
Signed-off-by: Ian Rogers <[email protected]>
Tested-by: Jiri Olsa <[email protected]>
Cc: Alexander Shishkin <[email protected]>
Cc: [email protected]
Cc: Ian Rogers <[email protected]>
Cc: Mark Rutland <[email protected]>
Cc: Namhyung Kim <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Link: http://lore.kernel.org/lkml/[email protected]
Signed-off-by: Arnaldo Carvalho de Melo <[email protected]>

perf tools: Don't install libtraceevent plugins as its not anymore in the kernel sources

While doing 'make -C tools/perf build-test' one can notice error
messages while trying to install libtraceevent plugins, stop doing that
as libtraceevent isn't anymore a homie.

These are the warnings dealt with:

   make_install_prefix_slash_O: make install prefix=/tmp/krava/
    failed to find: /tmp/krava/etc/bash_completion.d/perf
    failed to find: /tmp/krava/lib64/traceevent/plugins/plugin_cfg80211.so
    failed to find: /tmp/krava/lib64/traceevent/plugins/plugin_scsi.so
    failed to find: /tmp/krava/lib64/traceevent/plugins/plugin_xen.so
    failed to find: /tmp/krava/lib64/traceevent/plugins/plugin_function.so
    failed to find: /tmp/krava/lib64/traceevent/plugins/plugin_sched_switch.so
    failed to find: /tmp/krava/lib64/traceevent/plugins/plugin_mac80211.so
    failed to find: /tmp/krava/lib64/traceevent/plugins/plugin_kvm.so
    failed to find: /tmp/krava/lib64/traceevent/plugins/plugin_kmem.so
    failed to find: /tmp/krava/lib64/traceevent/plugins/plugin_hrtimer.so
    failed to find: /tmp/krava/lib64/traceevent/plugins/plugin_jbd2.so

Fixes: 4171925aa9f3f7bf ("tools lib traceevent: Remove libtraceevent")
Acked-by: Ian Rogers <[email protected]>
Cc: Adrian Hunter <[email protected]>
Cc: Jiri Olsa <[email protected]>
Cc: Namhyung Kim <[email protected]>
Cc: Steven Rostedt <[email protected]>
Link: http://lore.kernel.org/lkml/[email protected]
Signed-off-by: Arnaldo Carvalho de Melo <[email protected]>

perf kmem: Support field "node" in evsel__process_alloc_event() coping with recent tracepoint restructuring

Commit 11e9734bcb6a7361 ("mm/slab_common: unify NUMA and UMA version of
tracepoints") adds the field "node" into the tracepoints 'kmalloc' and
'kmem_cache_alloc', so this patch modifies the event process function to
support the field "node".

If field "node" is detected by checking function evsel__field(), it
stats the cross allocation.

When the "node" value is NUMA_NO_NODE (-1), it means the memory can be
allocated from any memory node, in this case, we don't account it as a
cross allocation.

Fixes: 11e9734bcb6a7361 ("mm/slab_common: unify NUMA and UMA version of tracepoints")
Reported-by: Ravi Bangoria <[email protected]>
Reviewed-by: James Clark <[email protected]>
Signed-off-by: Leo Yan <[email protected]>
Cc: Alexander Shishkin <[email protected]>
Cc: Hyeonggon Yoo <[email protected]>
Cc: Ingo Molnar <[email protected]>
Cc: Jiri Olsa <[email protected]>
Cc: Mark Rutland <[email protected]>
Cc: Namhyung Kim <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Vlastimil Babka <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Arnaldo Carvalho de Melo <[email protected]>

perf kmem: Support legacy tracepoints

Commit 11e9734bcb6a7361 ("mm/slab_common: unify NUMA and UMA version of
tracepoints") removed tracepoints 'kmalloc_node' and
'kmem_cache_alloc_node', we need to consider the tool should be backward
compatible.

If it detect the tracepoint "kmem:kmalloc_node", this patch enables the
legacy tracepoints, otherwise, it will ignore them.

Fixes: 11e9734bcb6a7361 ("mm/slab_common: unify NUMA and UMA version of tracepoints")
Reported-by: Ravi Bangoria <[email protected]>
Reviewed-by: James Clark <[email protected]>
Signed-off-by: Leo Yan <[email protected]>
Cc: Alexander Shishkin <[email protected]>
Cc: Hyeonggon Yoo <[email protected]>
Cc: Ingo Molnar <[email protected]>
Cc: Jiri Olsa <[email protected]>
Cc: Mark Rutland <[email protected]>
Cc: Namhyung Kim <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Cc: Vlastimil Babka <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Arnaldo Carvalho de Melo <[email protected]>

perf build: Properly guard libbpf includes

Including libbpf header files should be guarded by HAVE_LIBBPF_SUPPORT.
In bpf_counter.h, move the skeleton utilities under HAVE_BPF_SKEL.

Fixes: d6a735ef3277c45f ("perf bpf_counter: Move common functions to bpf_counter.h")
Reported-by: Mike Leach <[email protected]>
Signed-off-by: Ian Rogers <[email protected]>
Tested-by: Arnaldo Carvalho de Melo <[email protected]>
Tested-by: Jiri Olsa <[email protected]>
Tested-by: Mike Leach <[email protected]>
Cc: Alexander Shishkin <[email protected]>
Cc: Mark Rutland <[email protected]>
Cc: Namhyung Kim <[email protected]>
Cc: Peter Zijlstra <[email protected]>
Link: http://lore.kernel.org/lkml/[email protected]
Signed-off-by: Arnaldo Carvalho de Melo <[email protected]>

s390/kexec: fix ipl report address for kdump

This commit addresses the following erroneous situation with file-based
kdump executed on a system with a valid IPL report.

On s390, a kdump kernel, its initrd and IPL report if present are loaded
into a special and reserved on boot memory region - crashkernel. When
a system crashes and kdump was activated before, the purgatory code
is entered first which swaps the crashkernel and [0 - crashkernel size]
memory regions. Only after that the kdump kernel is entered. For this
reason, the pointer to an IPL report in lowcore must point to the IPL report
after the swap and not to the address of the IPL report that was located in
crashkernel memory region before the swap. Failing to do so, makes the
kdump's decompressor try to read memory from the crashkernel memory region
which already contains the production's kernel memory.

The situation described above caused spontaneous kdump failures/hangs
on systems where the Secure IPL is activated because on such systems
an IPL report is always present. In that case kdump's decompressor tried
to parse an IPL report which frequently lead to illegal memory accesses
because an IPL report contains addresses to various data.

Cc: <[email protected]>
Fixes: 99feaa717e55 ("s390/kexec_file: Create ipl report and pass to next kernel")
Reviewed-by: Vasily Gorbik <[email protected]>
Signed-off-by: Alexander Egorenkov <[email protected]>
Signed-off-by: Heiko Carstens <[email protected]>

ASoC: fsl-asoc-card: Fix naming of AC'97 CODEC widgets

The fsl-asoc-card AC'97 support currently tries to route to Playback and
Capture widgets provided by the AC'97 CODEC. This doesn't work since the
generic AC'97 driver registers with an "AC97" at the front of the stream
and hence widget names, update to reflect reality. It's not clear to me
if or how this ever worked.

Acked-by: Shengjiu Wang <[email protected]>
Signed-off-by: Mark Brown <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Mark Brown <[email protected]>

ASoC: fsl_ssi: Rename AC'97 streams to avoid collisions with AC'97 CODEC

The SSI driver calls the AC'97 playback and transmit streams "AC97 Playback"
and "AC97 Capture" respectively. This is the same name used by the generic
AC'97 CODEC driver in ASoC, creating confusion for the Freescale ASoC card
when it attempts to use these widgets in routing. Add a "CPU" in the name
like the regular DAIs registered by the driver to disambiguate.

Acked-by: Shengjiu Wang <[email protected]>
Signed-off-by: Mark Brown <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Mark Brown <[email protected]>

x86/boot: Avoid using Intel mnemonics in AT&T syntax asm

With 'GNU assembler (GNU Binutils for Debian) 2.39.90.20221231' the
build now reports:

  arch/x86/realmode/rm/../../boot/bioscall.S: Assembler messages:
  arch/x86/realmode/rm/../../boot/bioscall.S:35: Warning: found `movsd'; assuming `movsl' was meant
  arch/x86/realmode/rm/../../boot/bioscall.S:70: Warning: found `movsd'; assuming `movsl' was meant

  arch/x86/boot/bioscall.S: Assembler messages:
  arch/x86/boot/bioscall.S:35: Warning: found `movsd'; assuming `movsl' was meant
  arch/x86/boot/bioscall.S:70: Warning: found `movsd'; assuming `movsl' was meant

Which is due to:

  PR gas/29525

  Note that with the dropped CMPSD and MOVSD Intel Syntax string insn
  templates taking operands, mixed IsString/non-IsString template groups
  (with memory operands) cannot occur anymore. With that
  maybe_adjust_templates() becomes unnecessary (and is hence being
  removed).

More details: https://sourceware.org/bugzilla/show_bug.cgi?id=29525

Borislav Petkov further explains:

  " the particular problem here is is that the 'd' suffix is
    "conflicting" in the sense that you can have SSE mnemonics like movsD %xmm...
    and the same thing also for string ops (which is the case here) so apparently
    the agreement in binutils land is to use the always accepted suffixes 'l' or 'q'
    and phase out 'd' slowly... "

Fixes: 7a734e7dd93b ("x86, setup: "glove box" BIOS calls -- infrastructure")
Signed-off-by: Peter Zijlstra (Intel) <[email protected]>
Signed-off-by: Ingo Molnar <[email protected]>
Acked-by: Borislav Petkov (AMD) <[email protected]>
Link: https://lore.kernel.org/r/[email protected]

Merge tag '6.2-rc3-ksmbd-server-fixes' of git://git.samba.org/ksmbd

Pull ksmb server fixes from Steve French:

 - fix possible infinite loop in socket handler

 - fix possible panic in ntlmv2 authentication

 - fix error handling on tree connect

* tag '6.2-rc3-ksmbd-server-fixes' of git://git.samba.org/ksmbd:
  ksmbd: fix infinite loop in ksmbd_conn_handler_loop()
  ksmbd: check nt_len to be at least CIFS_ENCPWD_SIZE in ksmbd_decode_ntlmssp_auth_blob
  ksmbd: send proper error response in smb2_tree_connect()

sh/mm: Fix pmd_t for real

Because typing is hard...

Fixes: 0862ff059c9e ("sh/mm: Make pmd_t similar to pte_t")
Reported-by: Guenter Roeck <[email protected]>
Signed-off-by: Peter Zijlstra (Intel) <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

drm/amdgpu: fix pipeline sync v2

This fixes a potential memory leak of dma_fence objects in the CS code
as well as glitches in firefox because of missing pipeline sync.

v2: use the scheduler instead of the fence context

Signed-off-by: Christian König <[email protected]>
Bug: https://gitlab.freedesktop.org/drm/amd/-/issues/2323
Tested-by: Michal Kubecek [email protected]
Tested-by: Vlastimil Babka <[email protected]>
Acked-by: Alex Deucher <[email protected]>
Link: https://patchwork.freedesktop.org/patch/msgid/[email protected]

octeontx2-pf: Fix resource leakage in VF driver unbind

resources allocated like mcam entries to support the Ntuple feature
and hash tables for the tc feature are not getting freed in driver
unbind. This patch fixes the issue.

Fixes: 2da489432747 ("octeontx2-pf: devlink params support to set mcam entry count")
Signed-off-by: Hariprasad Kelam <[email protected]>
Signed-off-by: Sunil Kovvuri Goutham <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Paolo Abeni <[email protected]>

Merge branch 'selftests-net-isolate-l2_tos_ttl_inherit-sh-in-its-own-netns'

Guillaume Nault says:

====================
selftests/net: Isolate l2_tos_ttl_inherit.sh in its own netns.

l2_tos_ttl_inherit.sh uses a veth pair to run its tests, but only one
of the veth interfaces runs in a dedicated netns. The other one remains
in the initial namespace where the existing network configuration can
interfere with the setup used for the tests.

Isolate both veth devices in their own netns and ensure everything gets
cleaned up when the script exits.

Link: https://lore.kernel.org/netdev/[email protected]/
====================

Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Paolo Abeni <[email protected]>

selftests/net: l2_tos_ttl_inherit.sh: Ensure environment cleanup on failure.

Use 'set -e' and an exit handler to stop the script if a command fails
and ensure the test environment is cleaned up in any case. Also, handle
the case where the script is interrupted by SIGINT.

The only command that's expected to fail is 'wait $ping_pid', since
it's killed by the script. Handle this case with '|| true' to make it
play well with 'set -e'.

Finally, return the Kselftest SKIP code (4) when the script breaks
because of an environment problem or a command line failure. The 0 and
1 return codes should now reliably indicate that all tests have been
run (0: all tests run and passed, 1: all tests run but at least one
failed, 4: test script didn't run completely).

Fixes: b690842d12fd ("selftests/net: test l2 tunnel TOS/TTL inheriting")
Reported-by: Mirsad Goran Todorovac <[email protected]>
Tested-by: Mirsad Goran Todorovac <[email protected]>
Signed-off-by: Guillaume Nault <[email protected]>
Signed-off-by: Paolo Abeni <[email protected]>

selftests/net: l2_tos_ttl_inherit.sh: Run tests in their own netns.

This selftest currently runs half in the current namespace and half in
a netns of its own. Therefore, the test can fail if the current
namespace is already configured with incompatible parameters (for
example if it already has a veth0 interface).

Adapt the script to put both ends of the veth pair in their own netns.
Now veth0 is created in NS0 instead of the current namespace, while
veth1 is set up in NS1 (instead of the 'testing' netns).

The user visible netns names are randomised to minimise the risk of
conflicts with already existing namespaces. The cleanup() function
doesn't need to remove the virtual interface anymore: deleting NS0 and
NS1 automatically removes the virtual interfaces they contained.

We can remove $ns, which was only used to run ip commands in the
'testing' netns (let's use the builtin "-netns" option instead).
However, we still need a similar functionality as ping and tcpdump
now need to run in NS0. So we now have $RUN_NS0 for that.

Fixes: b690842d12fd ("selftests/net: test l2 tunnel TOS/TTL inheriting")
Reported-by: Mirsad Goran Todorovac <[email protected]>
Tested-by: Mirsad Goran Todorovac <[email protected]>
Signed-off-by: Guillaume Nault <[email protected]>
Signed-off-by: Paolo Abeni <[email protected]>

selftests/net: l2_tos_ttl_inherit.sh: Set IPv6 addresses with "nodad".

The ping command can run before DAD completes. In that case, ping may
fail and break the selftest.

We don't need DAD here since we're working on isolated device pairs.

Fixes: b690842d12fd ("selftests/net: test l2 tunnel TOS/TTL inheriting")
Signed-off-by: Guillaume Nault <[email protected]>
Signed-off-by: Paolo Abeni <[email protected]>

ALSA: hda/hdmi: Add a HP device 0x8715 to force connect list

Add the 'HP Engage Flex Mini' device to the force connect list to
enable audio through HDMI.

Signed-off-by: Adrian Chan <[email protected]>
Cc: <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Takashi Iwai <[email protected]>

MAINTAINERS: stop nvme matching for nvmem files

The nvme patterns detect all include files starting with nvme, which
also picks up the nvmem subsystem header files. Fix this by using
a more specific pattern.

Signed-off-by: Russell King (Oracle) <[email protected]>
[hch: switched to a purely inclusive pattern instead of excluding nvmem*]
Signed-off-by: Christoph Hellwig <[email protected]>

nvme: don't allow unprivileged passthrough on partitions

Passthrough commands can always access the entire device, and thus
submitting them on partitions is an privelege escalation.

In hindsight we should have never allowed any passthrough commands on
partitions, but it's probably too late to change that decision now.

Fixes: e4fbcf32c860 ("nvme: identify-namespace without CAP_SYS_ADMIN")
Signed-off-by: Christoph Hellwig <[email protected]>
Reviewed-by: Keith Busch <[email protected]>
Reviewed-by: Kanchan Joshi <[email protected]>
Reviewed-by: Chaitanya Kulkarni <[email protected]>

nvme: replace the "bool vec" arguments with flags in the ioctl path

To prepare for passing down more information, replace the boolean
vec argument with a more extensible flags one.

Signed-off-by: Christoph Hellwig <[email protected]>
Reviewed-by: Keith Busch <[email protected]>
Reviewed-by: Kanchan Joshi <[email protected]>
Reviewed-by: Chaitanya Kulkarni <[email protected]>

nvme: remove __nvme_ioctl

Open code __nvme_ioctl in the two callers to make future changes that
pass down additional paramters in the ioctl path easier.

Signed-off-by: Christoph Hellwig <[email protected]>
Reviewed-by: Keith Busch <[email protected]>
Reviewed-by: Kanchan Joshi <[email protected]>
Reviewed-by: Chaitanya Kulkarni <[email protected]>

nvme-pci: fix error handling in nvme_pci_enable()

There are two issues in nvme_pci_enable():

 1) If pci_alloc_irq_vectors() fails, device is left enabled. Fix this by
    adding a goto disable statement.
 2) nvme_pci_configure_admin_queue could return -ENODEV, in this case,
    we will need to free IRQ properly.  Otherwise the following warning
    could be triggered:

[    5.286752] WARNING: CPU: 0 PID: 33 at kernel/irq/irqdomain.c:253 irq_domain_remove+0x12d/0x140
[    5.290547] Call Trace:
[    5.290626]  <TASK>
[    5.290695]  msi_remove_device_irq_domain+0xc9/0xf0
[    5.290843]  msi_device_data_release+0x15/0x80
[    5.290978]  release_nodes+0x58/0x90
[    5.293788] WARNING: CPU: 0 PID: 33 at kernel/irq/msi.c:276 msi_device_data_release+0x76/0x80
[    5.297573] Call Trace:
[    5.297651]  <TASK>
[    5.297719]  release_nodes+0x58/0x90
[    5.297831]  devres_release_all+0xef/0x140
[    5.298339]  device_unbind_cleanup+0x11/0xc0
[    5.298479]  really_probe+0x296/0x320

Fixes: a6ee7f19ebfd ("nvme-pci: call nvme_pci_configure_admin_queue from nvme_pci_enable")
Co-developed-by: Keith Busch <[email protected]>
Signed-off-by: Tong Zhang <[email protected]>
Reviewed-by: Keith Busch <[email protected]>
Signed-off-by: Christoph Hellwig <[email protected]>

nvme-pci: add NVME_QUIRK_IDENTIFY_CNS quirk to Apple T2 controllers

This mirrors the quirk added to Apple Silicon controllers in apple.c.
These controllers do not support the Active NS ID List command and
behave identically to the SoC version judging by existing user
reports/syslogs, so will need the same fix. This quirk reverts
back to NVMe 1.0 behavior and disables the broken commands.

Fixes: 811f4de0344d ("nvme: avoid fallback to sequential scan due to transient issues")
Signed-off-by: Hector Martin <[email protected]>
Tested-by: Orlando Chamberlain <[email protected]>
Signed-off-by: Christoph Hellwig <[email protected]>

nvme-apple: add NVME_QUIRK_IDENTIFY_CNS quirk to fix regression

From the get-go, this driver and the ANS syslog have been complaining
about namespace identification. In 6.2-rc1, commit 811f4de0344d ("nvme:
avoid fallback to sequential scan due to transient issues") regressed
the driver by no longer allowing fallback to sequential namespace scans,
leaving us with no namespaces.

It turns out that the real problem is that this controller claiming
NVMe 1.1 compat is treating the CNS field as a binary field, as in NVMe
1.0. This already has a quirk, NVME_QUIRK_IDENTIFY_CNS, so set it for
the controller to fix all this nonsense (including other errors
triggered by other CNS commands).

Fixes: 811f4de0344d ("nvme: avoid fallback to sequential scan due to transient issues")
Fixes: 5bd2927aceba ("nvme-apple: Add initial Apple SoC NVMe driver")
Signed-off-by: Hector Martin <[email protected]>
Reviewed-by: Sven Peter <[email protected]>
Signed-off-by: Christoph Hellwig <[email protected]>

net/mlx5e: Fix macsec possible null dereference when updating MAC security entity (SecY)

Upon updating MAC security entity (SecY) in hw offload path, the macsec
security association (SA) initialization routine is called. In case of
extended packet number (epn) is enabled the salt and ssci attributes are
retrieved using the MACsec driver rx_sa context which is unavailable when
updating a SecY property such as encoding-sa hence the null dereference.
Fix by using the provided SA to set those attributes.

Fixes: 4411a6c0abd3 ("net/mlx5e: Support MACsec offload extended packet number (EPN)")
Signed-off-by: Emeel Hakim <[email protected]>
Reviewed-by: Raed Salem <[email protected]>
Signed-off-by: Saeed Mahameed <[email protected]>

net/mlx5e: Fix macsec ssci attribute handling in offload path

Currently when macsec offload is set with extended packet number (epn)
enabled, the driver wrongly deduce the short secure channel identifier
(ssci) from the salt instead of the stand alone ssci attribute as it
should, consequently creating a mismatch between the kernel and driver's
ssci values.
Fix by using the ssci value from the relevant attribute.

Fixes: 4411a6c0abd3 ("net/mlx5e: Support MACsec offload extended packet number (EPN)")
Signed-off-by: Emeel Hakim <[email protected]>
Reviewed-by: Raed Salem <[email protected]>
Signed-off-by: Saeed Mahameed <[email protected]>

net/mlx5: E-switch, Coverity: overlapping copy

When a capability is set via port function caps callbacks, a memcpy() is
performed in which the source and the target are the same address, e.g.:
the copy is redundant. Hence, Remove it.
Discovered by Coverity.

Fixes: 7db98396ef45 ("net/mlx5: E-Switch, Implement devlink port function cmds to control RoCE")
Fixes: e5b9642a33be ("net/mlx5: E-Switch, Implement devlink port function cmds to control migratable")
Signed-off-by: Shay Drory <[email protected]>
Reviewed-by: Mark Bloch <[email protected]>
Signed-off-by: Saeed Mahameed <[email protected]>

net/mlx5e: Don't support encap rules with gbp option

Previously, encap rules with gbp option would be offloaded by mistake but
driver does not support gbp option offload.

To fix this issue, check if the encap rule has gbp option and don't
offload the rule

Fixes: d8f9dfae49ce ("net: sched: allow flower to match vxlan options")
Signed-off-by: Gavin Li <[email protected]>
Reviewed-by: Maor Dickman <[email protected]>
Signed-off-by: Saeed Mahameed <[email protected]>

net/mlx5: Fix ptp max frequency adjustment range

.max_adj of ptp_clock_info acts as an absolute value for the amount in ppb
that can be set for a single call of .adjfine. This means that a single
call to .getfine cannot be greater than .max_adj or less than -(.max_adj).
Provides correct value for max frequency adjustment value supported by
devices.

Fixes: 3d8c38af1493 ("net/mlx5e: Add PTP Hardware Clock (PHC) support")
Signed-off-by: Rahul Rameshbabu <[email protected]>
Reviewed-by: Gal Pressman <[email protected]>
Reviewed-by: Tariq Toukan <[email protected]>
Signed-off-by: Saeed Mahameed <[email protected]>

net/mlx5e: Fix memory leak on updating vport counters

When updating statistics driver queries the vport's counters. On fail,
add error path releasing the allocated buffer avoiding memory leak.

Fixes: 64b68e369649 ("net/mlx5: Refactor and expand rep vport stat group")
Signed-off-by: Aya Levin <[email protected]>
Reviewed-by: Maor Dickman <[email protected]>
Signed-off-by: Saeed Mahameed <[email protected]>

net/mlx5e: TC, Restore pkt rate policing support

The offending commit removed the support for all packet rate metering.
Restore the pkt rate metering support by removing the restriction.

Fixes: 3603f26633e7 ("net/mlx5e: TC, allow meter jump control action")
Signed-off-by: Oz Shlomo <[email protected]>
Reviewed-by: Roi Dayan <[email protected]>
Signed-off-by: Saeed Mahameed <[email protected]>

net/mlx5e: TC, ignore match level for post meter rules

The post meter table only matches on reg_c5. As such, the inner/outer
match levels are irrelevant for the match critieria. The cited patch only
sets the outer criteria to none, thus setting the inner match level for
encapsulated packets. This caused rules with police action on tunnel
devices to not find an existing flow group for the match criteria, thus
failing to offload the rule.

Set both the inner and outer match levels to none for post_meter rules.

Fixes: 0d8c38d44f33 ("net/mlx5e: TC, init post meter rules with branching attributes")
Signed-off-by: Oz Shlomo <[email protected]>
Reviewed-by: Roi Dayan <[email protected]>
Signed-off-by: Saeed Mahameed <[email protected]>

net/mlx5e: IPoIB, Fix child PKEY interface stats on rx path

The current code always does the accounting using the
stats from the parent interface (linked in the rq). This
doesn't work when there are child interfaces configured.

Fix this behavior by always using the stats from the child
interface priv. This will also work for parent only
interfaces: the child (netdev) and parent netdev (rq->netdev)
will point to the same thing.

Fixes: be98737a4faa ("net/mlx5e: Use dynamic per-channel allocations in stats")
Signed-off-by: Dragos Tatulea <[email protected]>
Reviewed-by: Tariq Toukan <[email protected]>
Signed-off-by: Saeed Mahameed <[email protected]>

net/mlx5e: IPoIB, Block PKEY interfaces with less rx queues than parent

A user is able to configure an arbitrary number of rx queues when
creating an interface via netlink. This doesn't work for child PKEY
interfaces because the child interface uses the parent receive channels.

Although the child shares the parent's receive channels, the number of
rx queues is important for the channel_stats array: the parent's rx
channel index is used to access the child's channel_stats. So the array
has to be at least as large as the parent's rx queue size for the
counting to work correctly and to prevent out of bound accesses.

This patch checks for the mentioned scenario and returns an error when
trying to create the interface. The error is propagated to the user.

Fixes: be98737a4faa ("net/mlx5e: Use dynamic per-channel allocations in stats")
Signed-off-by: Dragos Tatulea <[email protected]>
Reviewed-by: Tariq Toukan <[email protected]>
Signed-off-by: Saeed Mahameed <[email protected]>

net/mlx5e: IPoIB, Block queue count configuration when sub interfaces are present

PKEY sub interfaces share the receive queues with the parent interface.
While setting the sub interface queue count is not supported, it is
currently possible to change the number of queues of the parent interface.
Thus we can end up with inconsistent queue sizes between the parent and its
sub interfaces.

This change disallows setting the queue count on the parent interface when
sub interfaces are present.

This is achieved by introducing an explicit reference to the parent netdev
in the mlx5i_priv of the child interface. An additional counter is also
required on the parent side to detect when sub interfaces are attached and
for proper cleanup.

The rtnl lock is taken during the ethtool op and the sub interface
ndo_init/uninit ops. There is no race here around counting the sub
interfaces, reading the sub interfaces and setting the number of
channels. The ASSERT_RTNL was added to document that.

Fixes: be98737a4faa ("net/mlx5e: Use dynamic per-channel allocations in stats")
Signed-off-by: Dragos Tatulea <[email protected]>
Reviewed-by: Tariq Toukan <[email protected]>
Signed-off-by: Saeed Mahameed <[email protected]>

net/mlx5e: Verify dev is present for fix features ndo

The native NIC port net device instance is being used as Uplink
representor.  While changing profiles private resources are not
available, fix features ndo does not check if the netdev is present.
Add driver protection to verify private resources are ready.

Fixes: 7a9fb35e8c3a ("net/mlx5e: Do not reload ethernet ports when changing eswitch mode")
Signed-off-by: Roy Novich <[email protected]>
Reviewed-by: Roi Dayan <[email protected]>
Signed-off-by: Saeed Mahameed <[email protected]>

net/mlx5: Fix command stats access after free

Command may fail while driver is reloading and can't accept FW commands
till command interface is reinitialized. Such command failure is being
logged to command stats. This results in NULL pointer access as command
stats structure is being freed and reallocated during mlx5 devlink
reload (see kernel log below).

Fix it by making command stats statically allocated on driver probe.

Kernel log:
[ 2394.808802] BUG: unable to handle kernel paging request at 000000000002a9c0
[ 2394.810610] PGD 0 P4D 0
[ 2394.811811] Oops: 0002 [#1] SMP NOPTI
...
[ 2394.815482] RIP: 0010:native_queued_spin_lock_slowpath+0x183/0x1d0
...
[ 2394.829505] Call Trace:
[ 2394.830667]  _raw_spin_lock_irq+0x23/0x26
[ 2394.831858]  cmd_status_err+0x55/0x110 [mlx5_core]
[ 2394.833020]  mlx5_access_reg+0xe7/0x150 [mlx5_core]
[ 2394.834175]  mlx5_query_port_ptys+0x78/0xa0 [mlx5_core]
[ 2394.835337]  mlx5e_ethtool_get_link_ksettings+0x74/0x590 [mlx5_core]
[ 2394.836454]  ? kmem_cache_alloc_trace+0x140/0x1c0
[ 2394.837562]  __rh_call_get_link_ksettings+0x33/0x100
[ 2394.838663]  ? __rtnl_unlock+0x25/0x50
[ 2394.839755]  __ethtool_get_link_ksettings+0x72/0x150
[ 2394.840862]  duplex_show+0x6e/0xc0
[ 2394.841963]  dev_attr_show+0x1c/0x40
[ 2394.843048]  sysfs_kf_seq_show+0x9b/0x100
[ 2394.844123]  seq_read+0x153/0x410
[ 2394.845187]  vfs_read+0x91/0x140
[ 2394.846226]  ksys_read+0x4f/0xb0
[ 2394.847234]  do_syscall_64+0x5b/0x1a0
[ 2394.848228]  entry_SYSCALL_64_after_hwframe+0x65/0xca

Fixes: 34f46ae0d4b3 ("net/mlx5: Add command failures data to debugfs")
Signed-off-by: Moshe Shemesh <[email protected]>
Reviewed-by: Shay Drory <[email protected]>
Signed-off-by: Saeed Mahameed <[email protected]>

net/mlx5e: TC, Keep mod hdr actions after mod hdr alloc

When offloading TC NIC rule which has mod_hdr action, the
mod_hdr actions list is freed upon mod_hdr allocation.

In the new format of handling multi table actions and CT in
particular, the mod_hdr actions list is still relevant when
setting the pre and post rules and therefore, freeing the list
may cause adding rules which don't set the FTE_ID.

Therefore, the mod_hdr actions list needs to be kept for the
pre/post flows as well and should be left for these handler to
be freed.

Fixes: 8300f225268b ("net/mlx5e: Create new flow attr for multi table actions")
Signed-off-by: Ariel Levkovich <[email protected]>
Signed-off-by: Saeed Mahameed <[email protected]>

net/mlx5: check attr pointer validity before dereferencing it

Fix attr pointer validity checks after it was already
dereferenced.

Fixes: cb0d54cbf948 ("net/mlx5e: Fix wrong source vport matching on tunnel rule")
Signed-off-by: Ariel Levkovich <[email protected]>
Reported-by: kernel test robot <[email protected]>
Reported-by: Dan Carpenter <[email protected]>
Signed-off-by: Saeed Mahameed <[email protected]>

net/mlx5: DR, Fix 'stack frame size exceeds limit' error in dr_rule

If the kernel configuration asks the compiler to check frame limit of 1K,
dr_rule_create_rule_nic exceed this limit:
    "stack frame size (1184) exceeds limit (1024)"

Fixing this issue by checking configured frame limit and using the
optimization STE array only for cases with the usual 2K (or larger)
stack size warning.

Fixes: b9b81e1e9382 ("net/mlx5: DR, For short chains of STEs, avoid allocating ste_arr dynamically")
Reported-by: kernel test robot <[email protected]>
Signed-off-by: Yevgeny Kliteynik <[email protected]>

Revert "r8169: disable detection of chip version 36"

This reverts commit 42666b2c452ce87894786aae05e3fad3cfc6cb59.

This chip version seems to be very rare, but it exits in consumer
devices, see linked report.

Link: https://stackoverflow.com/questions/75049473/cant-setup-a-wired-network-in-archlinux-fresh-install
Fixes: 42666b2c452c ("r8169: disable detection of chip version 36")
Signed-off-by: Heiner Kallweit <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Jakub Kicinski <[email protected]>

cpufreq: armada-37xx: stop using 0 as NULL pointer

Use NULL for NULL pointer to fix the following sparse warning:
drivers/cpufreq/armada-37xx-cpufreq.c:448:32: sparse: warning: Using plain integer as NULL pointer

Signed-off-by: Miles Chen <[email protected]>
Signed-off-by: Viresh Kumar <[email protected]>

net/sched: act_mpls: Fix warning during failed attribute validation

The 'TCA_MPLS_LABEL' attribute is of 'NLA_U32' type, but has a
validation type of 'NLA_VALIDATE_FUNCTION'. This is an invalid
combination according to the comment above 'struct nla_policy':

"
Meaning of `validate' field, use via NLA_POLICY_VALIDATE_FN:
   NLA_BINARY           Validation function called for the attribute.
   All other            Unused - but note that it's a union
"

This can trigger the warning [1] in nla_get_range_unsigned() when
validation of the attribute fails. Despite being of 'NLA_U32' type, the
associated 'min'/'max' fields in the policy are negative as they are
aliased by the 'validate' field.

Fix by changing the attribute type to 'NLA_BINARY' which is consistent
with the above comment and all other users of NLA_POLICY_VALIDATE_FN().
As a result, move the length validation to the validation function.

No regressions in MPLS tests:

 # ./tdc.py -f tc-tests/actions/mpls.json
 [...]
 # echo $?
 0

[1]
WARNING: CPU: 0 PID: 17743 at lib/nlattr.c:118
nla_get_range_unsigned+0x1d8/0x1e0 lib/nlattr.c:117
Modules linked in:
CPU: 0 PID: 17743 Comm: syz-executor.0 Not tainted 6.1.0-rc8 #3
Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS
rel-1.13.0-48-gd9c812dda519-prebuilt.qemu.org 04/01/2014
RIP: 0010:nla_get_range_unsigned+0x1d8/0x1e0 lib/nlattr.c:117
[...]
Call Trace:
 <TASK>
 __netlink_policy_dump_write_attr+0x23d/0x990 net/netlink/policy.c:310
 netlink_policy_dump_write_attr+0x22/0x30 net/netlink/policy.c:411
 netlink_ack_tlv_fill net/netlink/af_netlink.c:2454 [inline]
 netlink_ack+0x546/0x760 net/netlink/af_netlink.c:2506
 netlink_rcv_skb+0x1b7/0x240 net/netlink/af_netlink.c:2546
 rtnetlink_rcv+0x18/0x20 net/core/rtnetlink.c:6109
 netlink_unicast_kernel net/netlink/af_netlink.c:1319 [inline]
 netlink_unicast+0x5e9/0x6b0 net/netlink/af_netlink.c:1345
 netlink_sendmsg+0x739/0x860 net/netlink/af_netlink.c:1921
 sock_sendmsg_nosec net/socket.c:714 [inline]
 sock_sendmsg net/socket.c:734 [inline]
 ____sys_sendmsg+0x38f/0x500 net/socket.c:2482
 ___sys_sendmsg net/socket.c:2536 [inline]
 __sys_sendmsg+0x197/0x230 net/socket.c:2565
 __do_sys_sendmsg net/socket.c:2574 [inline]
 __se_sys_sendmsg net/socket.c:2572 [inline]
 __x64_sys_sendmsg+0x42/0x50 net/socket.c:2572
 do_syscall_x64 arch/x86/entry/common.c:50 [inline]
 do_syscall_64+0x2b/0x70 arch/x86/entry/common.c:80
 entry_SYSCALL_64_after_hwframe+0x63/0xcd

Link: https://lore.kernel.org/netdev/CAO4mrfdmjvRUNbDyP0R03_DrD_eFCLCguz6OxZ2TYRSv0K9gxA@mail.gmail.com/
Fixes: 2a2ea50870ba ("net: sched: add mpls manipulation actions to TC")
Reported-by: Wei Chen <[email protected]>
Tested-by: Wei Chen <[email protected]>
Signed-off-by: Ido Schimmel <[email protected]>
Reviewed-by: Alexander Duyck <[email protected]>
Link: https://lore.kernel.org/r/[email protected]
Signed-off-by: Jakub Kicinski <[email protected]>

drm/vmwgfx: Remove rcu locks from user resources

User resource lookups used rcu to avoid two extra atomics. Unfortunately
the rcu paths were buggy and it was easy to make the driver crash by
submitting command buffers from two different threads. Because the
lookups never show up in performance profiles replace them with a
regular spin lock which fixes the races in accesses to those shared
resources.

Fixes kernel oops'es in IGT's vmwgfx execution_buffer stress test and
seen crashes with apps using shared resources.

Fixes: e14c02e6b699 ("drm/vmwgfx: Look up objects without taking a reference")
Signed-off-by: Zack Rusin <[email protected]>
Reviewed-by: Martin Krastev <[email protected]>
Reviewed-by: Maaz Mombasawala <[email protected]>
Link: https://patchwork.freedesktop.org/patch/msgid/[email protected]

drm/virtio: Fix GEM handle creation UAF

Userspace can guess the handle value and try to race GEM object creation
with handle close, resulting in a use-after-free if we dereference the
object after dropping the handle's reference.  For that reason, dropping
the handle's reference must be done *after* we are done dereferencing
the object.

Signed-off-by: Rob Clark <[email protected]>
Reviewed-by: Chia-I Wu <[email protected]>
Fixes: 62fb7a5e1096 ("virtio-gpu: add 3d/virgl support")
Cc: [email protected]
Signed-off-by: Dmitry Osipenko <[email protected]>
Link: https://patchwork.freedesktop.org/patch/msgid/[email protected]

gcc: disable -Warray-bounds for gcc-11 too

We had already disabled this warning for gcc-12 due to bugs in the value
range analysis, but it turns out we end up having some similar problems
with gcc-11.3 too, so let's disable it there too.

Older gcc versions end up being increasingly less relevant, and
hopefully clang and newer version of gcc (ie gcc-13) end up working
reliably enough that we still get the build coverage even when we
disable this for some versions.

Link: https://lore.kernel.org/all/[email protected]/
Link: https://lore.kernel.org/all/[email protected]/
Cc: Kees Cook <[email protected]>
Cc: Vlastimil Babka <[email protected]>
Cc: Guenter Roeck <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

io_uring/poll: add hash if ready poll request can't complete inline

If we don't, then we may lose access to it completely, leading to a
request leak. This will eventually stall the ring exit process as
well.

Cc: [email protected]
Fixes: 49f1c68e048f ("io_uring: optimise submission side poll_refs")
Reported-and-tested-by: [email protected]
Link: https://lore.kernel.org/io-uring/[email protected]/
Suggested-by: Pavel Begunkov <[email protected]>
Signed-off-by: Jens Axboe <[email protected]>

ice: Add check for kzalloc

Add the check for the return value of kzalloc in order to avoid
NULL pointer dereference.
Moreover, use the goto-label to share the clean code.

Fixes: d6b98c8d242a ("ice: add write functionality for GNSS TTY")
Signed-off-by: Jiasheng Jiang <[email protected]>
Reviewed-by: Jiri Pirko <[email protected]>
Tested-by: Gurucharan G <[email protected]> (A Contingent worker at Intel)
Signed-off-by: Tony Nguyen <[email protected]>

ice: Fix potential memory leak in ice_gnss_tty_write()

The ice_gnss_tty_write() return directly if the write_buf alloc failed,
leaking the cmd_buf.

Fix by free cmd_buf if write_buf alloc failed.

Fixes: d6b98c8d242a ("ice: add write functionality for GNSS TTY")
Signed-off-by: Yuan Can <[email protected]>
Reviewed-by: Leon Romanovsky <[email protected]>
Tested-by: Gurucharan G <[email protected]> (A Contingent worker at Intel)
Signed-off-by: Tony Nguyen <[email protected]>

drm/amdgpu: Fixed bug on error when unloading amdgpu

Fixed bug on error when unloading amdgpu.

The error message is as follows:
[  377.706202] kernel BUG at drivers/gpu/drm/drm_buddy.c:278!
[  377.706215] invalid opcode: 0000 [#1] PREEMPT SMP NOPTI
[  377.706222] CPU: 4 PID: 8610 Comm: modprobe Tainted: G          IOE      6.0.0-thomas #1
[  377.706231] Hardware name: ASUS System Product Name/PRIME Z390-A, BIOS 2004 11/02/2021
[  377.706238] RIP: 0010:drm_buddy_free_block+0x26/0x30 [drm_buddy]
[  377.706264] Code: 00 00 00 90 0f 1f 44 00 00 48 8b 0e 89 c8 25 00 0c 00 00 3d 00 04 00 00 75 10 48 8b 47 18 48 d3 e0 48 01 47 28 e9 fa fe ff ff <0f> 0b 0f 1f 84 00 00 00 00 00 0f 1f 44 00 00 41 54 55 48 89 f5 53
[  377.706282] RSP: 0018:ffffad2dc4683cb8 EFLAGS: 00010287
[  377.706289] RAX: 0000000000000000 RBX: ffff8b1743bd5138 RCX: 0000000000000000
[  377.706297] RDX: ffff8b1743bd5160 RSI: ffff8b1743bd5c78 RDI: ffff8b16d1b25f70
[  377.706304] RBP: ffff8b1743bd59e0 R08: 0000000000000001 R09: 0000000000000001
[  377.706311] R10: ffff8b16c8572400 R11: ffffad2dc4683cf0 R12: ffff8b16d1b25f70
[  377.706318] R13: ffff8b16d1b25fd0 R14: ffff8b1743bd59c0 R15: ffff8b16d1b25f70
[  377.706325] FS:  00007fec56c72c40(0000) GS:ffff8b1836500000(0000) knlGS:0000000000000000
[  377.706334] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[  377.706340] CR2: 00007f9b88c1ba50 CR3: 0000000110450004 CR4: 00000000003706e0
[  377.706347] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
[  377.706354] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
[  377.706361] Call Trace:
[  377.706365]  <TASK>
[  377.706369]  drm_buddy_free_list+0x2a/0x60 [drm_buddy]
[  377.706376]  amdgpu_vram_mgr_fini+0xea/0x180 [amdgpu]
[  377.706572]  amdgpu_ttm_fini+0x12e/0x1a0 [amdgpu]
[  377.706650]  amdgpu_bo_fini+0x22/0x90 [amdgpu]
[  377.706727]  gmc_v11_0_sw_fini+0x26/0x30 [amdgpu]
[  377.706821]  amdgpu_device_fini_sw+0xa1/0x3c0 [amdgpu]
[  377.706897]  amdgpu_driver_release_kms+0x12/0x30 [amdgpu]
[  377.706975]  drm_dev_release+0x20/0x40 [drm]
[  377.707006]  release_nodes+0x35/0xb0
[  377.707014]  devres_release_all+0x8b/0xc0
[  377.707020]  device_unbind_cleanup+0xe/0x70
[  377.707027]  device_release_driver_internal+0xee/0x160
[  377.707033]  driver_detach+0x44/0x90
[  377.707039]  bus_remove_driver+0x55/0xe0
[  377.707045]  pci_unregister_driver+0x3b/0x90
[  377.707052]  amdgpu_exit+0x11/0x6c [amdgpu]
[  377.707194]  __x64_sys_delete_module+0x142/0x2b0
[  377.707201]  ? fpregs_assert_state_consistent+0x22/0x50
[  377.707208]  ? exit_to_user_mode_prepare+0x3e/0x190
[  377.707215]  do_syscall_64+0x38/0x90
[  377.707221]  entry_SYSCALL_64_after_hwframe+0x63/0xcd

Signed-off-by: YiPeng Chai <[email protected]>
Reviewed-by: Christian König <[email protected]>
Signed-off-by: Alex Deucher <[email protected]>
Cc: [email protected]

drm/amd: Delay removal of the firmware framebuffer

Removing the firmware framebuffer from the driver means that even
if the driver doesn't support the IP blocks in a GPU it will no
longer be functional after the driver fails to initialize.

This change will ensure that unsupported IP blocks at least cause
the driver to work with the EFI framebuffer.

Cc: [email protected]
Suggested-by: Alex Deucher <[email protected]>
Reviewed-by: Alex Deucher <[email protected]>
Reviewed-by: Lijo Lazar <[email protected]>
Signed-off-by: Mario Limonciello <[email protected]>
Signed-off-by: Alex Deucher <[email protected]>

drm/amdgpu: Fix potential NULL dereference

Fix potential NULL dereference, in the case when "man", the resource manager
might be NULL, when/if we print debug information.

Cc: Alex Deucher <[email protected]>
Cc: Christian König <[email protected]>
Cc: AMD Graphics <[email protected]>
Cc: Dan Carpenter <[email protected]>
Cc: kernel test robot <[email protected]>
Fixes: 7554886daa31ea ("drm/amdgpu: Fix size validation for non-exclusive domains (v4)")
Signed-off-by: Luben Tuikov <[email protected]>
Reviewed-by: Christian König <[email protected]>
Signed-off-by: Alex Deucher <[email protected]>

cifs: fix file info setting in cifs_open_file()

In cifs_open_file(), @buf must hold a pointer to a cifs_open_info_data
structure which is passed by cifs_nt_open(), so assigning @buf
directly to @fi was obviously wrong.

Fix this by passing a valid FILE_ALL_INFO structure to SMBLegacyOpen()
and CIFS_open(), and then copy the set structure to the corresponding
cifs_open_info_data::fi field with move_cifs_info_to_smb2() helper.

Link: https://bugzilla.kernel.org/show_bug.cgi?id=216889
Fixes: 76894f3e2f71 ("cifs: improve symlink handling for smb2+")
Signed-off-by: Paulo Alcantara (SUSE) <[email protected]>
Signed-off-by: Steve French <[email protected]>

cifs: fix file info setting in cifs_query_path_info()

We missed to set file info when CIFSSMBQPathInfo() returned 0, thus
leaving cifs_open_info_data::fi unset.

Fix this by setting cifs_open_info_data::fi when either
CIFSSMBQPathInfo() or SMBQueryInformation() succeed.

Link: https://bugzilla.kernel.org/show_bug.cgi?id=216881
Fixes: 76894f3e2f71 ("cifs: improve symlink handling for smb2+")
Signed-off-by: Paulo Alcantara (SUSE) <[email protected]>
Signed-off-by: Steve French <[email protected]>

tools/nolibc: fix the O_* fcntl/open macro definitions for riscv

When RISCV port was imported in 5.2, the O_* macros were taken with
their octal value and written as-is in hex, resulting in the getdents64()
to fail in nolibc-test.

Fixes: 582e84f7b779 ("tool headers nolibc: add RISCV support") #5.2
Signed-off-by: Willy Tarreau <[email protected]>
Signed-off-by: Paul E. McKenney <[email protected]>