Package: src:linux
Version: 6.1.192-1
Severity: critical
Tags: regression

Dear Maintainer,

After upgrading a bookworm server from linux-image-6.1.0-54-amd64
(6.1.190-1)
to linux-image-6.1.0-55-amd64 (6.1.192-1) on 2026-11-16, the machine crashed
three times within about two and a half hours, each time with the same oops
in the ext4 block allocator, called from the writeback worker of the root
filesystem. The first crash left the system hung for 1.5 hours (all writes
to / blocked); after setting kernel.panic_on_oops=1 the next two became
automatic reboots. Uptime at crash: 1224 s, 915 s and 722 s.

Rolling back to 6.1.0-54 resolves it: we pinned GRUB to 6.1.0-54 and the
machine has been running stable on it since the third crash. 6.1.0-54 had
also run for 8 days on the same filesystem and workload without any oops.

Setup:
- Debian 12 (bookworm), amd64, MSI MSIS366/S3661, BIOS ES366AOC.10R
02/06/2026
- AMD EPYC 4345P, 64 GB RAM
- root filesystem: ext4 (ordered) on mdadm RAID1 (md3) over two Samsung
NVMe drives (MZWL6960HFJA)
- write-heavy workload: PostgreSQL 18, MariaDB 11.4, Docker, nginx/PHP
- the oops is in "flush-9:3", the writeback worker for md3 (/)

The filesystem reports no errors afterwards: dumpe2fs "state: clean",
/sys/fs/ext4/md3/errors_count = 0, md3 [UU]. Offline e2fsck not yet run.

Oops (first crash, from pstore/efi; the second and third are identical apart
from CPU/PID/worker: CPU 0 PID 3319 kworker/u32:16, and CPU 14 PID 101
kworker/u32:7):

BUG: kernel NULL pointer dereference, address: 0000000000000000
#PF: supervisor read access in kernel mode
#PF: error_code(0x0000) - not-present page
Oops: 0000 [#1] PREEMPT SMP NOPTI
CPU: 13 PID: 66042 Comm: kworker/u32:5 Not tainted 6.1.0-55-amd64 #1 Debian
6.1.192-1
Hardware name: MSI MSIS366/S3661, BIOS ES366AOC.10R 02/06/2026
Workqueue: writeback wb_workfn (flush-9:3)
RIP: 0010:ext4_mb_load_buddy_gfp+0xbc/0x380 [ext4]
Code: 89 44 24 08 f7 3c 24 48 63 d8 89 54 24 0c ba 01 00 00 00 48 89 de e8
d3 7e 54 fa 49 89 c7 48 3d 00 f0 ff ff 0f 87 28 01 00 00 <48> 8b 00 a8 04
0f 84 12 01 00 00 49 8b 07 a8 01 0f 85 07 01 00 00
RSP: 0018:ffffcd58a11a3550 EFLAGS: 00010207
CR2: 0000000000000000 CR3: 0000000464010000 CR4: 0000000000750ee0
Call Trace:
<TASK>
ext4_mb_regular_allocator+0x29e/0xce0 [ext4]
ext4_mb_new_blocks+0x820/0x1240 [ext4]
? jbd2_journal_dirty_metadata+0x166/0x230 [jbd2]
ext4_new_meta_blocks+0xd0/0x120 [ext4]
ext4_ext_insert_extent+0x56f/0x13c0 [ext4]
ext4_ext_map_blocks+0x76e/0x1990 [ext4]
ext4_map_blocks+0x1b6/0x5f0 [ext4]
ext4_writepages+0x88a/0x1190 [ext4]
do_writepages+0xbd/0x1c0
__writeback_single_inode+0x3d/0x360
writeback_sb_inodes+0x1f5/0x4d0
__writeback_inodes_wb+0x4c/0xf0
wb_writeback+0x204/0x2f0
wb_workfn+0x2aa/0x4f0
process_one_work+0x1c4/0x380
worker_thread+0x4d/0x380
kthread+0xd7/0x100
ret_from_fork+0x1f/0x30
</TASK>

Complete pstore dumps available on request. The 6.1.192-1 changelog lists
several ext4 changes, none in mballoc as far as I can tell. A second machine
on the same kernel (lighter write load) has not crashed so far.

Staying on 6.1.0-54 re-opens the issues fixed in DLA-4906-1, so a fixed 6.1
kernel would be very welcome; happy to test a candidate.

Kind regards,
Jeroen

Reply via email to