Update: a third occurrence of this crash happened a few hours after
filing (2026-06-14 19:14:25, signal 6 in gnome-shell, same as before).
This one provides much tighter evidence about the trigger:

Memory monitoring (~30s resolution) shows the same "Shmem spike" pattern
I'd previously seen 30 minutes before occurrence 1, but this time
happening twice within about 90 seconds and immediately preceding the
crash:

- 19:12:59 — Shmem jumped from ~550MB to ~12GB in 30s.
- 19:13:35 — kernel OOM killed a Spotify process (this is expected/intentional 
on my system due to a snap memory quota I set on Spotify to make it the 
preferred OOM victim).
- Shmem briefly collapsed, then at 19:14:29 it spiked again to ~24GB.
- 19:14:25 — kernel OOM killed two Firefox "Web Content" (content process) 
instances, and gnome-shell's SIGABRT crash timestamp is the *same second* 
(19:14:25) — versus a ~30 minute lag in occurrence 1.

I'd previously suspected Spotify's on-disk cache size (22GB) was related
to the Shmem spike size, and cleared it (down to 53MB). It has only
regrown to ~2.1GB since, yet the same multi-GB Shmem-spike pattern
recurred — so the disk cache size doesn't appear to be the actual
source. Spotify is just the OOM-preferred victim due to my quota config,
unrelated to the Shmem growth itself.

Given `/proc/meminfo` Shmem includes GPU/DRM GEM shmem-backed buffer
objects, my current working theory is that a burst of GPU buffer
allocations (multi-GB, within ~30s) is the proximate trigger for the
gallium driver's fence/buffer-swap abort (`dri_create_fence_fd` /
`cogl_onscreen_swap_buffers_with_damage`). I haven't yet identified
which process/subsystem is responsible for the buffer burst itself —
planning to add finer-grained (1-2s) monitoring of per-process DRM
memory via `/proc/*/fdinfo` to try to catch it in the act. Will update
if I find more.

-- 
You received this bug notification because you are a member of Ubuntu
Bugs, which is subscribed to Ubuntu.
https://bugs.launchpad.net/bugs/2156739

Title:
  gnome-shell crashes with SIGABRT in dri_create_fence_fd() (mesa-
  libgallium 25.2.8) on Arrow Lake-U / Xe2 iGPU during
  cogl_onscreen_swap_buffers_with_damage

To manage notifications about this bug go to:
https://bugs.launchpad.net/ubuntu/+source/mesa/+bug/2156739/+subscriptions


-- 
ubuntu-bugs mailing list
[email protected]
https://lists.ubuntu.com/mailman/listinfo/ubuntu-bugs

Reply via email to