Lima Job Timeout on clang build
System information
- Hardware: PinePhone 3GB
- OS: Arch Linux ARM (DanctNIX)
- GPU: Mali 400-MP2
- Kernel version: Linux 5.16.8-1-danctnix
- Mesa version: Mesa 22.1.0 devel (git-d8d32cf7)
- Desktop manager and compositor: Phosh, Phoc
Describe the issue
It has been difficult to pin down a specific reproduction step of this crash, so I have usually left a journalctl -f
ssh session on another machine to see the output (As once it has crashes the compositor is dead and no screen output).
Important to note, the mesa-git package mentioned above has been built with clang, and as we have seen in #5959 (closed) the same issue continues to occur.
Log files as attachment
journalctl -f from ssh to view the log during crash:
Feb 14 08:59:19 sipiphone kernel: [drm:lima_sched_timedout_job] *ERROR* lima job timeout
Feb 14 08:59:19 sipiphone kernel: lima 1c40000.gpu: fail to save task state from phoc pid 2999: error task list is full
Feb 14 08:59:19 sipiphone kernel: lima 1c40000.gpu: pp task error 0 int_state=0 status=5
Feb 14 08:59:19 sipiphone kernel: lima 1c40000.gpu: pp task error 1 int_state=0 status=5
This output is nearly identical to the crash log found in the issue for clang built mesa issue (MR #5959 (closed)), though obviously a different compositor name and pid.
[ 64.766199] [drm:lima_sched_timedout_job] *ERROR* lima job timeout
[ 64.766609] lima 1c40000.gpu: fail to save task state from gnome-shell pid 665: error task list is full
[ 64.766656] lima 1c40000.gpu: pp task error 0 int_state=0 status=5
[ 64.766776] lima 1c40000.gpu: pp task error 1 int_state=0 status=5
Extra information
I'll try and get more visibility / information from the mesa driver, just need to tinker around first as this is my first diagnostic / bug report!
Related to #5959 (closed).