drm/v3d: Add job to pending list if the reset was skipped
authorMaíra Canal <mcanal@igalia.com>
Wed, 30 Apr 2025 20:51:52 +0000 (17:51 -0300)
committerMaíra Canal <mcanal@igalia.com>
Fri, 2 May 2025 19:25:14 +0000 (16:25 -0300)
commit35e4079bf1a2570abffce6ababa631afcf8ea0e5
treeb2f45986652e5c9b3e4d35e41e89d13803a5dc98
parentb662b162c3d06f120749eea0351ec9317d9dd905
drm/v3d: Add job to pending list if the reset was skipped

When a CL/CSD job times out, we check if the GPU has made any progress
since the last timeout. If so, instead of resetting the hardware, we skip
the reset and let the timer get rearmed. This gives long-running jobs a
chance to complete.

However, when `timedout_job()` is called, the job in question is removed
from the pending list, which means it won't be automatically freed through
`free_job()`. Consequently, when we skip the reset and keep the job
running, the job won't be freed when it finally completes.

This situation leads to a memory leak, as exposed in [1] and [2].

Similarly to commit 704d3d60fec4 ("drm/etnaviv: don't block scheduler when
GPU is still active"), this patch ensures the job is put back on the
pending list when extending the timeout.

Cc: stable@vger.kernel.org # 6.0
Reported-by: Daivik Bhatia <dtgs1208@gmail.com>
Closes: https://gitlab.freedesktop.org/mesa/mesa/-/issues/12227 [1]
Closes: https://github.com/raspberrypi/linux/issues/6817 [2]
Reviewed-by: Iago Toral Quiroga <itoral@igalia.com>
Acked-by: Tvrtko Ursulin <tvrtko.ursulin@igalia.com>
Link: https://lore.kernel.org/r/20250430210643.57924-1-mcanal@igalia.com
Signed-off-by: Maíra Canal <mcanal@igalia.com>
drivers/gpu/drm/v3d/v3d_sched.c