Am 08.09.23 um 20:24 schrieb Yunxiang Li:
Reading fence timestamp always need to check the corresponding flag bit before to make sure the write is actually visible, otherwise we can read garbage here.
Yeah, Teddy already stumbled over that problem as well I think. We are currently testing a more wider fix for this. Regards, Christian.
Fixes: 1774baa64f93 ("drm/scheduler: Change scheduled fence track v2") Signed-off-by: Yunxiang Li <yunxiang...@amd.com> --- drivers/gpu/drm/scheduler/sched_main.c | 3 +++ 1 file changed, 3 insertions(+) diff --git a/drivers/gpu/drm/scheduler/sched_main.c b/drivers/gpu/drm/scheduler/sched_main.c index 506371c42745..8ed00f54c59d 100644 --- a/drivers/gpu/drm/scheduler/sched_main.c +++ b/drivers/gpu/drm/scheduler/sched_main.c @@ -928,6 +928,9 @@ drm_sched_get_cleanup_job(struct drm_gpu_scheduler *sched) typeof(*next), list);if (next) {+ while (!test_bit(DMA_FENCE_FLAG_TIMESTAMP_BIT, + &job->s_fence->finished.flags)) + cpu_relax(); next->s_fence->scheduled.timestamp = job->s_fence->finished.timestamp; /* start TO timer for next job */
