Hello,

I would like to request the backport of upstream commit

  dbd9d1cbf9700528c8595ab1fa7ef832e79821fe
  "ALSA: usb-audio: fix list_add double-add in push_back_to_ready_list"

to Debian's 6.12 kernel series (trixie).

== What happens ==

On Debian 13 with linux-image-6.12.116+deb13-amd64, the machine
panicked in interrupt context in snd_complete_urb() while idle:

  list_add corruption. prev->next should be next (ffff8f5041027590),
  but was ffff8f5041026348. (prev=ffff8f5041026348).
  kernel BUG at lib/list_debug.c:32!
  Oops: invalid opcode: 0000 [#1] PREEMPT SMP NOPTI
  RIP: 0010:__list_add_valid_or_report.cold+0x3a/0x5b
  Call Trace:
   <IRQ>
   snd_complete_urb+0xf4/0x540 [snd_usb_audio]
   __usb_hcd_giveback_urb+0x9d/0x120 [usbcore]
   usb_giveback_urb_bh+0xba/0x140 [usbcore]
   process_one_work+0x188/0x340
   bh_worker+0x17b/0x1a0
   tasklet_hi_action+0x13/0x30
   handle_softirqs+0xcc/0x280
   ? handle_edge_irq+0x8b/0x230
   __irq_exit_rcu+0x8d/0xb0
   common_interrupt+0x85/0xa0
   </IRQ>
   <TASK>
   asm_common_interrupt+0x26/0x40
   RIP: 0010:cpuidle_enter_state+0xc6/0x420
   cpuidle_enter+0x2d/0x40
   do_idle+0x142/0x2a0
   cpu_startup_entry+0x29/0x30
   start_secondary+0x11e/0x140
   common_startup_64+0x13e/0x141
   </TASK>
  Kernel panic - not syncing: Fatal exception in interrupt

The affected device is a Jabra Link 390 USB audio adapter
(0b0e:2e51), the only USB audio device on the system. The panic
happened after several suspend/resume cycles, while the CPU was
idle. The full EFI pstore dump is attached.

== Root cause ==

This is the bug fixed by dbd9d1cbf970. In sound/usb/endpoint.c,
stop_urbs() clears ep->ready_playback_urbs with INIT_LIST_HEAD()
instead of unlinking each queued snd_urb_ctx. If a URB survives
wait_clear_urbs()'s forced STOPPING->STOPPED timeout, its ctx is
left looking "linked" even though the list head no longer points to
it. When the endpoint is restarted and the same ctx is queued again,
and the old URB's completion then calls push_back_to_ready_list()
for it a second time, list_add() sees a double add and triggers
the BUG.

The fix is small and self-contained:

  - guard push_back_to_ready_list() with list_empty(&ctx->ready_list)
  - unlink each ctx with list_del_init() in stop_urbs()

7 insertions, 2 deletions in sound/usb/endpoint.c.

== Why 6.12 needs it ==

The commit is in mainline 7.3 (and 7.3.y), but it was not Cc'd to
stable. As of 2026-11-13 it is absent from 6.12.y, 6.17.y, 6.18.y,
6.19.y, 7.0.y and 7.2.y; I verified that the 6.12.y and 7.2.y
sources still have the unconditional list_add_tail() in
push_back_to_ready_list(). Debian's 6.12.116 therefore still has the
bug, and the 7.2.12 currently in trixie-backports does not contain
the fix either.

This is a local panic/DoS: any USB audio device using implicit
feedback or low-latency playback can hit it, and suspend/resume makes
it more likely. No special privileges are required beyond being able
to use the audio device.

== Request ==

Could you please consider carrying dbd9d1cbf970 in the Debian 6.12
series (or requesting its inclusion in 6.12.y stable), so Debian 13
users with USB audio devices get the fix?

I am willing to test a patched kernel on the affected hardware:
HP Elite x360 1040 14 inch G11, Debian 13 trixie,
6.12.116+deb13-amd64, Jabra Link 390.

References:
  syzbot: https://syzkaller.appspot.com/bug?extid=9fe3b8d9f5c64ff410a7
  fix:    
https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?id=dbd9d1cbf9700528c8595ab1fa7ef832e79821fe

Thanks,

Reply via email to