logged
Confirmed fixed on 7.0.0-38 — no hangs.
I've been running Ubuntu-7.0.0-38.38~24.04.4 since 2026-09-28 and have not
reproduced the poweroff deadlock once.
What I tested:
- 3 completed poweroffs, after sessions of ~8 hours, ~1h40m, and ~6 minutes.
Session length matters here: the deadlock was a race on pending ISM delayed
work, so it never fired from an idle fresh boot — only after real desktop and
gaming use. The long sessions are the meaningful ones.
- Zero hangs. All three powered off cleanly and stayed off.
How I verified rather than just eyeballing it:
- I had kernel.hung_task_panic=1 armed for the shutdown window only (via a
systemd unit ExecStop), with kernel.panic=20, so any hang becomes a panic
that efi_pstore records and systemd-pstore archives on the next boot. This is
exactly what captured the original deadlock trace. On 7.0.0-38 it has fired
zero times — /sys/fs/pstore and /var/lib/systemd/pstore contain no new
records.
- Confirmed the fix is actually present in the shipped binary, not just the
version number:
zstd -dc
/lib/modules/7.0.0-38-generic/kernel/drivers/gpu/drm/amd/amdgpu/amdgpu.ko.zst \
| strings | grep -c '^amdgpu_dm_ism_force_full_power$'
1
That symbol only exists in builds carrying the fix, since the fix split the
disable path into amdgpu_dm_ism_disable() (must not hold dc_lock) and the new
amdgpu_dm_ism_force_full_power() (holds it). For reference, 7.0.0-34 in
noble-updates does NOT have it and still deadlocks.
Nothing else regressed across the jump from 7.0.0-14: Display Core v3.2.369 on
DCN 4.0.1, gfx1200 on Mesa 25.2.8 / RADV, no amdgpu errors, no failed units.
One clarification on availability: 7.0.0-38.38~24.04.4 is still published to
noble-proposed only — linux-hwe-7.0 in noble-updates is currently 7.0.0-34.
I installed the proposed build deliberately to test the fix, with the pocket
pinned at priority 100 and removed immediately afterwards, so nothing else came
from proposed.
Caveat I want to be straight about: the original failure rate was roughly 1
shutdown in 3, so three clean shutdowns is encouraging but not statistically
conclusive on its own. I'll keep running it and will watch for consistency once
-38 reaches noble-updates. If it recurs I'll reopen with a fresh pstore trace.
Thanks for turning this around quickly.
--
You received this bug notification because you are a member of Ubuntu
Bugs, which is subscribed to Ubuntu.
https://bugs.launchpad.net/bugs/2166940
Title:
[regression] poweroff deadlocks in amdgpu_dm_ism_disable() — ISM
backport is missing upstream fix 3714fe2
To manage notifications about this bug go to:
https://bugs.launchpad.net/ubuntu/+source/linux-hwe-7.0/+bug/2166940/+subscriptions
--
ubuntu-bugs mailing list
[email protected]
https://lists.ubuntu.com/mailman/listinfo/ubuntu-bugs
