Package: src:linux
Version: 6.1.192-1
Severity: important
X-Debbugs-CC: [email protected]
SUMMARY
-------
Since upgrading from linux-image-6.1.190-1 (linux-image-6.1.0-54-amd64) to
6.1.192-1 (linux-image-6.1.0-55-amd64), this machine produces a storm of
kernel BUGs from bluetoothd:
list_del corruption, <addr>->prev is LIST_POISON2 (dead000000000122)
kernel BUG at lib/list_debug.c:56!
...
hci_chan_del+0x18/0x90 [bluetooth]
l2cap_chan_put+0x106/0x130 [bluetooth]
l2cap_sock_release+0xca/0x100 [bluetooth]
at a rate of roughly one oops every 8 seconds (149 oops in a 20-minute
boot, 98 oops in a 47-second boot, 241 in another). After 20 minutes to
3 hours the kernel dies with
BUG: kernel NULL pointer dereference, address: 0000000000000000
and the machine hard-freezes: no further log output, no response to
SysRq, power cycle required. Reproduced 4 times on 2026-11-17 on this
machine.
This is a regression: 8 boots between 2026-11-06 and 2026-11-16 on
6.1.0-53 (6.1.187-1) and 6.1.0-54 (6.1.190-1) show 0 occurrences of
"list_del corruption" and no freeze, with the same Bluetooth headset
actively used.
ENVIRONMENT
-----------
Debian GNU/Linux 12 (bookworm), amd64
linux-image-6.1.0-55-amd64 6.1.192-1
Linux 6.1.0-55-amd64 #1 SMP PREEMPT_DYNAMIC Debian 6.1.192-1 (2026-11-14)
bluez 5.66-1+deb12u2
Hardware: Dell Inc. Precision 3581 / 0MFGJ1, BIOS 1.17.0 (09/09/2024)
CPU: 13th Gen Intel Core i7-13700H
GPU: Intel Raptor Lake-P Iris Xe (i915) + NVIDIA GA107GLM
Storage: Samsung PM9B1 NVMe, LUKS2 + LVM (rootfs on dm-crypt)
Bluetooth: Intel AX211, USB 8087:0033
Bluetooth audio device: "OpenRun Pro 2 by Shokz" (A8:F5:E1:CF:4B:4B),
paired + trusted, AVRCP profile
Other paired devices: Pixel 8a, Pixel Buds A-Series
OBSERVED TRACE (verbatim, oops #121 of a 149-oops boot)
-------------------------------------------------------
list_del corruption, ffff8ccd67100c00->prev is LIST_POISON2 (dead000000000122)
------------[ cut here ]------------
kernel BUG at lib/list_debug.c:56!
invalid opcode: 0000 [#121] PREEMPT SMP NOPTI
CPU: 8 PID: 35248 Comm: bluetoothd Tainted: G D 6.1.0-55-amd64 #1 Debian 
6.1.192-1
Hardware name: Dell Inc. Precision 3581/0MFGJ1, BIOS 1.17.0 09/09/2024
RIP: 0010:__list_del_entry_valid.cold+0x4b/0x6f
Call Trace:
<TASK>
hci_chan_del+0x18/0x90 [bluetooth]
l2cap_chan_put+0x106/0x130 [bluetooth]
l2cap_sock_release+0xca/0x100 [bluetooth]
__sock_release+0x3a/0xb0
sock_close+0x11/0x20
__fput+0x90/0x250
task_work_run+0x56/0x90
exit_to_user_mode_prepare+0x23d/0x240
syscall_exit_to_user_mode+0x17/0x40
do_syscall_64+0x69/0xb0
entry_SYSCALL_64_after_hwframe+0x6e/0xd8
</TASK>
---[ end trace 0000000000000000 ]---
(the same BUG repeats every ~8 s, with varying addresses and frames,
all ending in hci_chan_del via l2cap_chan_put)
...
and the last thing in the journal before the hang:
kernel: BUG: kernel NULL pointer dereference, address: 0000000000000000
TIMELINE ON 2026-11-17 (CET, all times from the persistent journal)
-------------------------------------------------------------------
08:58:47 - 09:11:45 6.1.0-55 (1st boot of this kernel) oops: ?
09:13:09 - 09:14:02 6.1.0-55 98 oops in 47 s
09:16:59 - 09:36:41 6.1.0-55 149 oops, froze at 09:36:41 (NULL deref)
09:53:xx - 10:33:xx 6.1.0-55 241 oops, froze
10:34:06 - 10:5x:xx 6.1.0-55 47 oops (Bluetooth headset off in that boot)
REGRESSION DATA (oopses of "list_del corruption", per boot)
-----------------------------------------------------------
6.1.0-53 (6.1.187-1) 2 boots 0 oops headset used (3-4 AVRCP regs)
6.1.0-54 (6.1.190-1) 6 boots 0 oops headset used (0-3 AVRCP regs)
6.1.0-55 (6.1.192-1) 4 boots 98 / 149 / 241 / 47 oops
The headset was already in use on 6.1.0-53 and 6.1.0-54, without a single
oops. The oopses appear only with 6.1.192.
CONTEXT DURING THE STORM
------------------------
During the failing boots, bluetoothd re-registers the AVRCP input device
every ~15.6 s:
input: OpenRun Pro 2 by Shokz (AVRCP) as /devices/virtual/input/input142
input: OpenRun Pro 2 by Shokz (AVRCP) as /devices/virtual/input/input143
... up to input156 and beyond, one every ~15.6 s
i.e. an AVRCP connect/disconnect loop; each L2CAP socket close in that loop
hits the corruption above. Note however that in one boot 47 oopses were
logged with no "OpenRun Pro" AVRCP registration at all, so the race is
probably reachable by any L2CAP socket close, not only by the AVRCP loop.
ANALYSIS
--------
The upstream fix for a closely related bug, commit ab4eedb790ca
("Bluetooth: L2CAP: Fix corrupted list in hci_chan_del"), IS present in
6.1.192-1. Verified on this host: the symbol introduced by that commit,
l2cap_conn_hold_unless_zero, is present in
/lib/modules/6.1.0-55-amd64/kernel/net/bluetooth/bluetooth.ko
(and 5 times in /proc/kallsyms), while it is absent from 6.1.0-54.
The trace fixed by ab4eedb790ca goes through l2cap_conn_put:
hci_chan_del <- l2cap_conn_put <- l2cap_sock_shutdown <- l2cap_sock_release
The trace observed here goes through a different path:
hci_chan_del <- l2cap_chan_put <- l2cap_sock_release
so this looks like a remaining/separate hole in the same locking area,
still present in 6.1.192. The list node being deleted has already been
poisoned (LIST_POISON2 in ->prev), i.e. hci_chan_del() is called twice on
the same hci_chan (or concurrently), which suggests l2cap_chan_put() can
drop the last reference while the connection is still tearing down.
IMPACT
------
Kernel list corruption (SLUB object reuse) in the Bluetooth core, leading
to repeated oopses and eventually a NULL pointer dereference and a full
machine hang requiring a power cycle. Work in progress is lost.
No kdump/crashkernel is configured on this host and /sys/fs/pstore is
empty, so no vmcore is available.
REPRODUCER
----------

  1.
Pair + trust "OpenRun Pro 2 by Shokz" and keep it powered/in range.
  2.
Boot linux-image-6.1.0-55-amd64 (6.1.192-1).
  3.
First oops about 9 s after boot; then ~1 oops / 8 s, all in bluetoothd.
  4.
After 20 min - 3 h: NULL deref and hard freeze.

WORKAROUND
----------
Disabling Bluetooth (systemctl disable --now bluetooth) stops the oops
storm and the freezes. 6.1.0-54 (6.1.190-1) is clean on this machine in
practice, so it is the practical workaround until a fixed kernel is
available.
ADDITIONAL INFORMATION
----------------------
linux-image-6.1.0-56-amd64 (6.1.192-2) is the same 6.1.192 source with an
ABI bump, so it presumably carries the same defect; I did not boot it yet.
I can provide, on request:

     *
the full journal (system + kernel, short-precise) of the failing
boots, and the complete 149-oops log;
     *
the per-boot oops/headset table quoted above;
     *
the output of any kernel debug you need (slub_debug, lockdep, KASAN
build, etc.);
     *
testing of any patched kernel (e.g. 6.1.193+ when it lands, or a
bookworm-backports build) on this exact hardware, since the failure
reproduces within seconds.

Thanks.

Reply via email to