| CVE |
Vendors |
Products |
Updated |
CVSS v3.1 |
| In the Linux kernel, the following vulnerability has been resolved:
smb/client: flush dirty data before punching a hole
Punching a hole after a large buffered write may leave the range
reported as data. Reproduce it with:
xfs_io -f \
-c "pwrite -b 3m -S 0x61 0 3m" \
-c "fpunch 1m 1m" \
-c "seek -h 0" \
-c "seek -d 1m" \
/mnt/test/repro
Punching 1 MiB at offset 1 MiB should produce:
0 1 MiB 2 MiB 3 MiB
| DATA | HOLE | DATA | EOF
Instead, the entire file is reported as data. SEEK_HOLE(0) returns EOF,
and SEEK_DATA(1M) returns 1M.
This happens because a dirty folio spanning the punched range can be
written back after the punch and refill the hole.
Fix this by flushing and waiting for dirty data in the punched range
before invalidating the page cache and issuing FSCTL_SET_ZERO_DATA.
The xfstests generic/539 pass against Samba/ksmbd with this change. |
| In the Linux kernel, the following vulnerability has been resolved:
ALSA: seq: oss: Serialize readq reset state with q->lock
snd_seq_oss_readq_clear() resets qlen, head, and tail without
q->lock even though the normal reader and producer paths serialize the
same ring state under that spinlock. A reset can therefore race
snd_seq_oss_readq_free() or snd_seq_oss_readq_put_event() and leave
stale records in the queue, drop freshly queued ones, or report the
wrong readiness after wakeup. KCSAN reports a data race between
snd_seq_oss_readq_clear() and snd_seq_oss_readq_free().
Take q->lock while clearing the ring and resetting input_time. Factor
the enqueue logic into a caller-locked helper so
snd_seq_oss_readq_put_timestamp() updates its suppression state under
the same lock instead of racing the reset path.
The buggy scenario involves two paths, with each column showing the
order within that path:
reset path: locked readq updater:
1. snd_seq_oss_reset() or 1. A reader or callback producer
release reaches takes q->lock on the same queue.
snd_seq_oss_readq_clear().
2. snd_seq_oss_readq_clear() 2. The updater tests or modifies
resets qlen, head, tail, qlen, head, and tail.
and input_time.
3. snd_seq_oss_readq_clear() 3. The updater completes its
wakes sleepers on read-modify-write sequence.
q->midi_sleep.
4. Without q->lock, the reset 4. The resulting ring state drives
can overlap the locked later reads and readiness.
update.
KCSAN reports:
BUG: KCSAN: data-race in snd_seq_oss_readq_clear /
snd_seq_oss_readq_free
write to 0xffff8881069fe608 of 4 bytes by task 120516 on cpu 0:
snd_seq_oss_readq_free+0x6c/0x80
snd_seq_oss_read+0xcb/0x250
odev_read+0x38/0x60
vfs_read+0xff/0x600
ksys_read+0xb4/0x140
__x64_sys_read+0x46/0x60
do_syscall_64+0xbb/0x2f0
entry_SYSCALL_64_after_hwframe+0x77/0x7f
read to 0xffff8881069fe608 of 4 bytes by task 120517 on cpu 1:
snd_seq_oss_readq_clear+0x1f/0x90
snd_seq_oss_reset+0xa7/0xf0
snd_seq_oss_ioctl+0x6f6/0x7e0
odev_ioctl+0x56/0xc0
__x64_sys_ioctl+0xd1/0x120
do_syscall_64+0xbb/0x2f0
entry_SYSCALL_64_after_hwframe+0x77/0x7f
value changed: 0x00000001 -> 0x00000000 |
| In the Linux kernel, the following vulnerability has been resolved:
coresight: ete: Always save state on power down
System register ETMs and ETE are unlikely to be preserved on CPU power
down. The ETE DT binding also never documented
"arm,coresight-loses-context-with-cpu" so nobody would have legitimately
been able to use that binding to fix it and ACPI has no such binding at
all.
Fix it by hard coding the setting for sysreg ETMs (ETE is always sysreg)
or ACPI boots. Use a local variable when setting up save_state so that
it's immune to concurrent probing when devices have different
configurations which is an issue with modifying the global.
This fixes the following error when using Coresight with ACPI on the FVP
which supports CPU PM:
coresight ete0: External agent took claim tag
WARNING: drivers/hwtracing/coresight/coresight-core.c:248 at coresight_disclaim_device_unlocked+0xe0/0xe8, CPU#0: perf/117 |
| In the Linux kernel, the following vulnerability has been resolved:
ACPI: processor_idle: Mark LPI enter functions as __cpuidle
When function tracing or Kprobes is enabled, entering an ACPI Low
Power Idle (LPI) state triggers the following RCU splat:
RCU not on for: acpi_idle_lpi_enter+0x4/0xd8
WARNING: CPU: 8 PID: 0 at include/linux/trace_recursion.h:162 function_trace_call+0x1e8/0x228
The acpi_idle_lpi_enter() function is invoked within the cpuidle
path after RCU has already been disabled for the current local CPU.
Consequently, ftrace's function_trace_call() expects RCU to be
actively watching before recording trace data, emitting a warning
if it is not.
Fix this by annotating acpi_idle_lpi_enter(), the generic __weak
stub, and the RISC-V implementation of acpi_processor_ffh_lpi_enter()
with __cpuidle. This moves these functions into the '.cpuidle.text'
section, implicitly disabling ftrace instrumentation (notrace) along
this sensitive path and preventing trace-induced RCU warnings during
idle entry. |
| In the Linux kernel, the following vulnerability has been resolved:
tipc: serialize udp bearer replicast list updates
tipc_udp_rcast_add() and cleanup_bearer() both update ub->rcast.list with
list_add_rcu() / list_del_rcu(), but nothing serializes them. The add runs
from the encap receive softirq (via tipc_udp_rcast_disc()) without
rtnl_lock(), so it can race the cleanup delete and corrupt the list:
list_del corruption. prev->next should be ffff8880298d7ab8,
but was ffff88802449ad38. (prev=ffff888027e3ec98)
kernel BUG at lib/list_debug.c:62!
RIP: __list_del_entry_valid_or_report+0x17a/0x200
Workqueue: events cleanup_bearer
Call Trace:
cleanup_bearer (net/tipc/udp_media.c:811)
process_one_work (kernel/workqueue.c:3302)
worker_thread (kernel/workqueue.c:3466)
The bearer can be enabled from an unprivileged user namespace, as the
TIPCv2 generic-netlink ops carry no GENL_ADMIN_PERM.
Add a spinlock to struct udp_bearer and take it around the list_add_rcu()
in tipc_udp_rcast_add() and the list_del_rcu() loop in cleanup_bearer() so
the two writers can no longer corrupt the list.
Reject a duplicate peer under the same lock before allocating, and remove
tipc_udp_is_known_peer(). The old lockless pre-check in
tipc_udp_rcast_disc() was racy: two softirqs discovering the same peer
could both find it absent and add it twice.
cleanup_bearer() runs from a workqueue after tipc_udp_disable() clears the
bearer's up bit, so an encap softirq can still reach tipc_udp_rcast_add()
and add a peer after cleanup_bearer() has already emptied the list, leaking
that entry when the bearer is freed. Mark the bearer disabled under
rcast_lock once the list is emptied and refuse further additions. |
| In the Linux kernel, the following vulnerability has been resolved:
hwmon: (lm90) Add lock protection to lm90_alert
Sashiko reports:
lm90_alert() executes in the smbus alert context and calls
lm90_update_confreg() to disable the hardware alert line, without
acquiring hwmon_lock.
Concurrently, sysfs write operations (such as lm90_write_convrate) hold
the hwmon_lock, temporarily modify data->config, and then restore it.
If an alert interrupt occurs concurrently with a sysfs write, the sysfs
path will overwrite the alert handler's modifications to data->config
and the hardware register.
This unintentionally re-enables the hardware alert line while the alarm is
still active, causing an interrupt storm.
Add the missing lock to lm90_alert() to solve the problem. |
| In the Linux kernel, the following vulnerability has been resolved:
netfilter: bridge: eb_tables: close module init race
sashiko reports for unrelated patch:
Does the core ebtables initialization in ebtables.c suffer from a similar race?
Once nf_register_sockopt() completes, the sockopts are exposed globally.
sockopt has to be registered last, just like in ip/ip6/arptables. |
| In the Linux kernel, the following vulnerability has been resolved:
Bluetooth: ISO: Fix data-race on iso_pi fields in hci_get_route calls
iso_connect_bis(), iso_connect_cis(), iso_listen_bis(), and
iso_conn_big_sync() call hci_get_route() using iso_pi(sk)->dst,
iso_pi(sk)->src, and iso_pi(sk)->src_type without holding lock_sock().
These fields may be modified concurrently by connect() or setsockopt()
on the same socket, resulting in data-races reported by KCSAN.
Fix this by snapshotting the required fields under lock_sock() before
calling hci_get_route().
BUG: KCSAN: data-race in memcmp+0x45/0xb0
race at unknown origin, with read to 0xffff8880122135cf of 1 bytes
by task 333 on cpu 1:
memcmp+0x45/0xb0
hci_get_route+0x27e/0x490
iso_connect_cis+0x4c/0xa10
iso_sock_connect+0x60e/0xb30
__sys_connect_file+0xbd/0xe0
__sys_connect+0xe0/0x110
__x64_sys_connect+0x40/0x50
x64_sys_call+0xcad/0x1c60
do_syscall_64+0x133/0x590
entry_SYSCALL_64_after_hwframe+0x77/0x7f |
| In the Linux kernel, the following vulnerability has been resolved:
bpf: Fix RCU stall in bpf_fd_array_map_clear()
Add a missing cond_resched() in bpf_fd_array_map_clear() loop.
For PROG_ARRAY maps with many entries this loop calls
prog_array_map_poke_run() per entry which can be expensive, and
without yielding this can cause RCU stalls under load:
rcu: Stack dump where RCU GP kthread last ran:
CPU: 0 UID: 0 PID: 30932 Comm: kworker/0:2 Not tainted 6.14.0-13195-g967e8def1100 #2 PREEMPT(undef)
Workqueue: events prog_array_map_clear_deferred
RIP: 0010:write_comp_data+0x38/0x90 kernel/kcov.c:246
Call Trace:
<TASK>
prog_array_map_poke_run+0x77/0x380 kernel/bpf/arraymap.c:1096
__fd_array_map_delete_elem+0x197/0x310 kernel/bpf/arraymap.c:925
bpf_fd_array_map_clear kernel/bpf/arraymap.c:1000 [inline]
prog_array_map_clear_deferred+0x119/0x1b0 kernel/bpf/arraymap.c:1141
process_one_work+0x898/0x19d0 kernel/workqueue.c:3238
process_scheduled_works kernel/workqueue.c:3319 [inline]
worker_thread+0x770/0x10b0 kernel/workqueue.c:3400
kthread+0x465/0x880 kernel/kthread.c:464
ret_from_fork+0x4d/0x80 arch/x86/kernel/process.c:153
ret_from_fork_asm+0x19/0x30 arch/x86/entry/entry_64.S:245
</TASK> |
| In the Linux kernel, the following vulnerability has been resolved:
ALSA: pcm: oss: Fix data race at accessing runtime.oss.trigger
Currently the runtime.oss.trigger field may be accessed concurrently
without protection, which may lead to the data race. And, in this
case, it may lead to more severe problem because it's a bit field; as
writing the data, it may overwrite other bit fields as well, which
confuses the operation completely, as spotted by fuzzing.
Fix it by covering runtime.oss.trigger bit fled also with the existing
params_lock mutex in both snd_pcm_oss_get_trigger() and
snd_pcm_oss_poll(). |
| In the Linux kernel, the following vulnerability has been resolved:
i3c: mipi-i3c-hci: Fix race in DMA ring dequeue
The HCI DMA dequeue path (hci_dma_dequeue_xfer()) may be invoked for
multiple transfers that timeout around the same time. However, the
function is not serialized and can race with itself.
When a timeout occurs, hci_dma_dequeue_xfer() stops the ring, processes
incomplete transfers, and then restarts the ring. If another timeout
triggers a parallel call into the same function, the two instances may
interfere with each other - stopping or restarting the ring at unexpected
times.
Add a mutex so that hci_dma_dequeue_xfer() is serialized with respect to
itself. |
| In the Linux kernel, the following vulnerability has been resolved:
hwmon: (pmbus/core) Protect regulator operations with mutex
The regulator operations pmbus_regulator_get_voltage(),
pmbus_regulator_set_voltage(), and pmbus_regulator_list_voltage()
access PMBus registers and shared data but were not protected by
the update_lock mutex. This could lead to race conditions.
However, adding mutex protection directly to these functions causes
a deadlock because pmbus_regulator_notify() (which calls
regulator_notifier_call_chain()) is often called with the mutex
already held (e.g., from pmbus_fault_handler()). If a regulator
callback then calls one of the now-protected voltage functions,
it will attempt to acquire the same mutex.
Rework pmbus_regulator_notify() to utilize a worker function to
send notifications outside of the mutex protection. Events are
stored as atomics in a per-page bitmask and processed by the worker.
Initialize the worker and its associated data during regulator
registration, and ensure it is cancelled on device removal using
devm_add_action_or_reset().
While at it, remove the unnecessary include of linux/of.h. |
| In the Linux kernel, the following vulnerability has been resolved:
blktrace: fix __this_cpu_read/write in preemptible context
tracing_record_cmdline() internally uses __this_cpu_read() and
__this_cpu_write() on the per-CPU variable trace_cmdline_save, and
trace_save_cmdline() explicitly asserts preemption is disabled via
lockdep_assert_preemption_disabled(). These operations are only safe
when preemption is off, as they were designed to be called from the
scheduler context (probe_wakeup_sched_switch() / probe_wakeup()).
__blk_add_trace() was calling tracing_record_cmdline(current) early in
the blk_tracer path, before ring buffer reservation, from process
context where preemption is fully enabled. This triggers the following
using blktests/blktrace/002:
blktrace/002 (blktrace ftrace corruption with sysfs trace) [failed]
runtime 0.367s ... 0.437s
something found in dmesg:
[ 81.211018] run blktests blktrace/002 at 2026-02-25 22:24:33
[ 81.239580] null_blk: disk nullb1 created
[ 81.357294] BUG: using __this_cpu_read() in preemptible [00000000] code: dd/2516
[ 81.362842] caller is tracing_record_cmdline+0x10/0x40
[ 81.362872] CPU: 16 UID: 0 PID: 2516 Comm: dd Tainted: G N 7.0.0-rc1lblk+ #84 PREEMPT(full)
[ 81.362877] Tainted: [N]=TEST
[ 81.362878] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.17.0-0-gb52ca86e094d-prebuilt.qemu.org 04/01/2014
[ 81.362881] Call Trace:
[ 81.362884] <TASK>
[ 81.362886] dump_stack_lvl+0x8d/0xb0
...
(See '/mnt/sda/blktests/results/nodev/blktrace/002.dmesg' for the entire message)
[ 81.211018] run blktests blktrace/002 at 2026-02-25 22:24:33
[ 81.239580] null_blk: disk nullb1 created
[ 81.357294] BUG: using __this_cpu_read() in preemptible [00000000] code: dd/2516
[ 81.362842] caller is tracing_record_cmdline+0x10/0x40
[ 81.362872] CPU: 16 UID: 0 PID: 2516 Comm: dd Tainted: G N 7.0.0-rc1lblk+ #84 PREEMPT(full)
[ 81.362877] Tainted: [N]=TEST
[ 81.362878] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.17.0-0-gb52ca86e094d-prebuilt.qemu.org 04/01/2014
[ 81.362881] Call Trace:
[ 81.362884] <TASK>
[ 81.362886] dump_stack_lvl+0x8d/0xb0
[ 81.362895] check_preemption_disabled+0xce/0xe0
[ 81.362902] tracing_record_cmdline+0x10/0x40
[ 81.362923] __blk_add_trace+0x307/0x5d0
[ 81.362934] ? lock_acquire+0xe0/0x300
[ 81.362940] ? iov_iter_extract_pages+0x101/0xa30
[ 81.362959] blk_add_trace_bio+0x106/0x1e0
[ 81.362968] submit_bio_noacct_nocheck+0x24b/0x3a0
[ 81.362979] ? lockdep_init_map_type+0x58/0x260
[ 81.362988] submit_bio_wait+0x56/0x90
[ 81.363009] __blkdev_direct_IO_simple+0x16c/0x250
[ 81.363026] ? __pfx_submit_bio_wait_endio+0x10/0x10
[ 81.363038] ? rcu_read_lock_any_held+0x73/0xa0
[ 81.363051] blkdev_read_iter+0xc1/0x140
[ 81.363059] vfs_read+0x20b/0x330
[ 81.363083] ksys_read+0x67/0xe0
[ 81.363090] do_syscall_64+0xbf/0xf00
[ 81.363102] entry_SYSCALL_64_after_hwframe+0x76/0x7e
[ 81.363106] RIP: 0033:0x7f281906029d
[ 81.363111] Code: 31 c0 e9 c6 fe ff ff 50 48 8d 3d 66 63 0a 00 e8 59 ff 01 00 66 0f 1f 84 00 00 00 00 00 80 3d 41 33 0e 00 00 74 17 31 c0 0f 05 <48> 3d 00 f0 ff ff 77 5b c3 66 2e 0f 1f 84 00 00 00 00 00 48 83 ec
[ 81.363113] RSP: 002b:00007ffca127dd48 EFLAGS: 00000246 ORIG_RAX: 0000000000000000
[ 81.363120] RAX: ffffffffffffffda RBX: 0000000000000000 RCX: 00007f281906029d
[ 81.363122] RDX: 0000000000001000 RSI: 0000559f8bfae000 RDI: 0000000000000000
[ 81.363123] RBP: 0000000000001000 R08: 0000002863a10a81 R09: 00007f281915f000
[ 81.363124] R10: 00007f2818f77b60 R11: 0000000000000246 R12: 0000559f8bfae000
[ 81.363126] R13: 0000000000000000 R14: 0000000000000000 R15: 000000000000000a
[ 81.363142] </TASK>
The same BUG fires from blk_add_trace_plug(), blk_add_trace_unplug(),
and blk_add_trace_rq() paths as well.
The purpose of tracin
---truncated--- |
| In the Linux kernel, the following vulnerability has been resolved:
sched_ext: Build the cid tables privately and publish them with RCU
The cid tables are visible to the cid kfuncs while being modified: the
first enable publishes the global pointers before filling them,
ops.init_cids() overrides rewrite them in place, and re-enables rebuild
them in place. A racing TRACING or SYSCALL program can read unfilled
entries, including uninitialized memory in the kmalloc'd tables, or torn
topo updates.
Tie the tables' lifetimes to the root sched instead: each root enable
builds a fresh set privately and publishes the per-table __rcu globals once
the layout is final, and root disable unpublishes and RCU-frees the set. A
non-NULL global is now always a fully built table which stays valid for the
reader's RCU read section, and lookups stay two loads. Kfuncs treat NULL as
no-mapping, also after the scheduler exits instead of reporting the stale
last mapping.
The cid kfuncs are available whether the root scheduler is cid-form or
cpu-form, the latter to allow gradual migration to cids. Every root
therefore builds and publishes a default mapping.
Every reader must either be gated on scheduler liveness or NULL-check
inside an RCU read section. Fix the two kfuncs that were neither:
scx_bpf_this_cid() read the table with no RCU or preemption protection and
scx_bpf_task_cid() relied on KF_RCU, which doesn't put a sleepable program
in an RCU read section. The hotplug callbacks are instead serialized by
retiring the tables inside the cpus_read_lock() section that clears
scx_root.
v2: Document why every root builds the tables (desc + cid.c comment). |
| In the Linux kernel, the following vulnerability has been resolved:
seg6: ensure packet data is writable before modifying SRH and IPv6 DA
advance_nextseg() modifies the SRH Segments Left field and the IPv6
destination address without ensuring the packet data is writable.
seg6_next_csid_advance_arg() has the same problem when it advances the
NEXT-C-SID argument in the destination address.
The skb may be cloned, for example by an AF_PACKET socket receiving on
the ingress device. advance_nextseg() and seg6_next_csid_advance_arg()
then write into the packet data shared with the clone. A read from that
socket can return the modified packet instead of the received one. The
simplified path below shows this for advance_nextseg():
__netif_receive_skb_one_core
__netif_receive_skb_core
deliver_skb [orig: users=2, cloned=0]
packet_rcv
skb_clone clone queued to the AF_PACKET socket
consume_skb(orig) [orig: users=1, cloned=1]
ipv6_rcv
ip6_rcv_core skb_share_check: no-op [orig: users=1]
[...]
input_action_end_core
advance_nextseg writes into the data shared with the clone
Call skb_ensure_writable() in advance_nextseg() and in
seg6_next_csid_advance_arg() before they modify the packet data.
skb_ensure_writable() may reallocate skb->head, which invalidates the
pointers into the packet data taken before the call.
advance_nextseg() now returns a valid SRH pointer, or NULL if
skb_ensure_writable() fails.
seg6_next_csid_advance_arg() takes the pointer to the destination
address from the skb after skb_ensure_writable().
On failure, the callers drop the packet with SKB_DROP_REASON_NOMEM. |
| In the Linux kernel, the following vulnerability has been resolved:
KVM: s390: Add missing srcu in kvm_s390_set_irq_state()
Like kvm_s390_inject_vcpu(), kvm_s390_set_irq_state() also needs the
kvm->srcu or the slots lock when performing the Store status operation.
Fix by taking kvm->srcu in kvm_s390_set_irq_state(). |
| In the Linux kernel, the following vulnerability has been resolved:
smb: client: flush and commit data before querying allocated ranges
The mode 0 / FALLOC_FL_KEEP_SIZE fallocate emulation for small interior
ranges in smb3_simple_fallocate_range() issues
FSCTL_QUERY_ALLOCATED_RANGES and zero-fills any sub-range the server
reports as an unallocated hole, to force block allocation without
changing file contents.
This is unsafe against Windows servers. Per the Win32 documentation,
FSCTL_QUERY_ALLOCATED_RANGES only reports ranges that *may* contain
nonzero data and is explicitly not coherent with recently written data:
"a call to FSCTL_QUERY_ALLOCATED_RANGES [after writing to a network
file] would not necessarily return a correct list of allocated
regions. To ensure coherency [...] flush the data to the file"
fsx (generic/363) reproduces the resulting corruption against Windows
Server 2022: a range is written, an uncached read returns the data, yet
the next FSCTL_QUERY_ALLOCATED_RANGES on that range reports it as a whole
hole, so the emulation overwrites live data with zeroes. A network trace
confirmed the WRITE, the READ returning data and the query returning an
empty range list within microseconds. Samba does not exhibit this; its
query is coherent with the written data.
Write back the dirty pages, drain any outstanding I/O and issue an SMB2
FLUSH to force the server to commit the data before querying the
allocated ranges, so the query reflects the data actually present.
With this, generic/363 passes against both Windows Server 2022 and Samba
over 100000 fsx operations. |
| In the Linux kernel, the following vulnerability has been resolved:
KVM: arm64: Take the SRCU lock for page table walks in fault injection and AT emulation
walk_s1() and kvm_walk_nested_s2() expect to be called while holding
kvm->srcu to guard against memslot changes. While this is generally
the case, __kvm_at_s12() and __kvm_find_s1_desc_level() call into the
respective walkers without taking kvm->srcu.
Fix by acquiring kvm->srcu prior to the table walk in both instances. |
| In the Linux kernel, the following vulnerability has been resolved:
net: mana: Sync page pool RX frags for CPU
MANA allocates RX buffers from page pool fragments when frag_count is
greater than 1. In that case the buffers remain DMA mapped by page pool
and the RX completion path does not call dma_unmap_single(). As a result,
the implicit sync-for-CPU normally performed by dma_unmap_single() is
missing before the packet data is passed to the networking stack.
This breaks RX on configurations which require explicit DMA syncing, for
example when booted with swiotlb=force.
Fix this by recording the page pool page and DMA sync offset when the RX
buffer is allocated, and syncing the received packet range for CPU access
before handing the RX buffer to the stack. |
| In the Linux kernel, the following vulnerability has been resolved:
net/mlx5e: xsk: Fix unlocked writing to ICOSQ
During napi poll, when the affinity changes and there's still XSK work
to be done, we trigger an ICOSQ interrupt on the new CPU. However, this
triggering on the ICOSQ is done unprotected.
There are 2 such races:
A) mlx5e_trigger_irq() is called while mlx5e_xsk_alloc_rx_mpwqe() is
running from a different CPU due to affinity change. This can happen
because IRQ triggering is done after napi_complete_done(). At this point
the NAPI can be scheduled on a different CPU. Like this:
CPU A (old affinity, NAPI tail) CPU B (new affinity, fresh NAPI)
------------------------------- --------------------------------
napi_complete_done() clears SCHED
mlx5e_cq_arm(...)
napi_schedule_prep() sets SCHED
mlx5e_napi_poll()
mlx5e_xsk_alloc_rx_mpwqe()
mlx5e_icosq_sync_lock() // noop
memcpy 640 B UMR body
advance sq->pc by 10
mlx5e_trigger_irq(&c->icosq)
wqe_info[pi] = {NOP, 1}
mlx5e_post_nop() advances sq->pc
B) mlx5e_trigger_irq() is called on the ICOSQ when
mlx5e_trigger_napi_icosq() is running.
The obvious fix would be to lock the ICOSQ. But ICOSQ has an optimized
locking scheme that doesn't work for this scenario. Kick the async ICOSQ
instead which is always locked.
This issue was noticed in the wild with the following splat:
netdevice: ge-0-0-1: Bad OP in ICOSQ CQE: 0xd
WARNING: drivers/net/ethernet/mellanox/mlx5/core/en_rx.c:826 [...]
[...]
Call Trace:
<IRQ>
mlx5e_napi_poll+0x11d/0x7f0 [mlx5_core]
__napi_poll+0x30/0x200
? skb_defer_free_flush+0x9c/0xc0
net_rx_action+0x2fe/0x3f0
handle_softirqs+0xd8/0x340
__irq_exit_rcu+0xbc/0xe0
common_interrupt+0x85/0xa0
</IRQ>
<TASK>
asm_common_interrupt+0x26/0x40
[...]
---[ end trace 0000000000000000 ]---
mlx5_core 0000:08:00.0 ge-0-0-1: Error cqe on cqn 0x548, ci 0x2022, qn 0x8f4,
opcode 0xd, syndrome 0x2, vendor syndrome 0x68
00000000: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
00000010: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
00000020: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
00000030: 00 00 00 00 01 00 68 02 01 00 08 f4 de 14 59 d2
WQE DUMP: WQ size 16384 WQ cur size 0, WQE index 0x1e14, len: 64
00000000: 00 00 00 01 d9 ed 80 02 00 00 00 01 d9 ed 90 02
00000010: 00 00 00 01 d9 ed a0 02 00 00 00 01 d9 ed b0 02
00000020: 00 00 00 01 d9 ed c0 02 00 00 00 01 d9 ed d0 02
00000030: 00 00 00 01 d9 ed e0 02 00 00 00 01 d9 ed f0 02
mlx5_core 0000:08:00.0 ge-0-0-1: Error cqe on cqn 0x548, ci 0x2023, qn 0x8f4,
opcode 0xd, syndrome 0x5, vendor syndrome 0xf9
00000000: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
00000010: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
00000020: 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00 00
00000030: 00 00 00 00 01 00 f9 05 01 00 08 f4 de 15 cf d2 |
