On 9/9/2022 11:02 PM, Andrey Grodzovsky wrote:
What exactly is the scenario which this patch fixes in more detail please ?
GPU reset issue started after adding [PATCH 6/6].
Root cause -> In drm_sched_get_cleanup_job(), We use the finished fence status bit to check the job status dma_fence_is_signaled(). If a job is signaled (DMA_FENCE_FLAG_SIGNALED_BIT is set), then we cancel the reset worker thread.
After applying [patch 6] now we are checking enable signaling in dma_fence_is_signaled() by checking DMA_FENCE_FLAG_ENABLE_SIGNAL_BIT bit. but signaling is not enabled for the finished fence. As a result, dma_fence_is_signaled() always returns false, and drm_sched_get_cleanup_job() will not cancel the reset worker thread, resulting in the GPU reset.
To Fix the above issue Christian suggested that we can use parent(hardware) fence instead of finished fence because signaling enabled by the calling of dma_fence_add_callback() for parent fence. As a result, dma_fence_is_signaled() will return the correct fence status and reset worker thread can be cancelled in drm_sched_get_cleanup_job().
~arvind
Andrey
On 2022-09-09 13:08, Arvind Yadav wrote:
Using the parent fence instead of the finished fence to get the job status. This change is to avoid GPU scheduler timeout error which can cause GPU reset.
Signed-off-by: Arvind Yadav Arvind.Yadav@amd.com
changes in v1,v2 - Enable signaling for finished fence in sche_main() is removed
drivers/gpu/drm/scheduler/sched_main.c | 4 ++-- 1 file changed, 2 insertions(+), 2 deletions(-)
diff --git a/drivers/gpu/drm/scheduler/sched_main.c b/drivers/gpu/drm/scheduler/sched_main.c index e0ab14e0fb6b..2ac28ad11432 100644 --- a/drivers/gpu/drm/scheduler/sched_main.c +++ b/drivers/gpu/drm/scheduler/sched_main.c @@ -829,7 +829,7 @@ drm_sched_get_cleanup_job(struct drm_gpu_scheduler *sched) job = list_first_entry_or_null(&sched->pending_list, struct drm_sched_job, list); - if (job && dma_fence_is_signaled(&job->s_fence->finished)) { + if (job && dma_fence_is_signaled(job->s_fence->parent)) { /* remove job from pending_list */ list_del_init(&job->list); @@ -841,7 +841,7 @@ drm_sched_get_cleanup_job(struct drm_gpu_scheduler *sched) if (next) { next->s_fence->scheduled.timestamp = - job->s_fence->finished.timestamp; + job->s_fence->parent->timestamp; /* start TO timer for next job */ drm_sched_start_timeout(sched); }