drm/amdgpu: get rid of incorrect TDR

original time out detect routine is incorrect, cuz it measures
the gap from job scheduled, but we should only measure the
gap from processed by hw.

Signed-off-by: Monk Liu <Monk.Liu@amd.com>
Reviewed-by: Chunming Zhou <david1.zhou@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drivers/gpu/drm/amd/scheduler/gpu_scheduler.c

@@ -418,46 +418,18 @@ static void amd_sched_process_job(struct fence *f, struct fence_cb *cb)
 	spin_unlock_irqrestore(&sched->job_list_lock, flags);
 
 	amd_sched_fence_signal(s_fence);
-	if (sched->timeout != MAX_SCHEDULE_TIMEOUT) {
-		cancel_delayed_work(&s_fence->dwork);
-		spin_lock_irqsave(&sched->fence_list_lock, flags);
-		list_del_init(&s_fence->list);
-		spin_unlock_irqrestore(&sched->fence_list_lock, flags);
-	}
+
 	trace_amd_sched_process_job(s_fence);
 	fence_put(&s_fence->base);
 	wake_up_interruptible(&sched->wake_up_worker);
 }
 
-static void amd_sched_fence_work_func(struct work_struct *work)
-{
-	struct amd_sched_fence *s_fence =
-		container_of(work, struct amd_sched_fence, dwork.work);
-	struct amd_gpu_scheduler *sched = s_fence->sched;
-	struct amd_sched_fence *entity, *tmp;
-	unsigned long flags;
-
-	DRM_ERROR("[%s] scheduler is timeout!\n", sched->name);
-
-	/* Clean all pending fences */
-	spin_lock_irqsave(&sched->fence_list_lock, flags);
-	list_for_each_entry_safe(entity, tmp, &sched->fence_list, list) {
-		DRM_ERROR("  fence no %d\n", entity->base.seqno);
-		cancel_delayed_work(&entity->dwork);
-		list_del_init(&entity->list);
-		fence_put(&entity->base);
-	}
-	spin_unlock_irqrestore(&sched->fence_list_lock, flags);
-}
-
 static int amd_sched_main(void *param)
 {
 	struct sched_param sparam = {.sched_priority = 1};
 	struct amd_gpu_scheduler *sched = (struct amd_gpu_scheduler *)param;
 	int r, count;
 
-	spin_lock_init(&sched->fence_list_lock);
-	INIT_LIST_HEAD(&sched->fence_list);
 	sched_setscheduler(current, SCHED_FIFO, &sparam);
 
 	while (!kthread_should_stop()) {
@@ -465,7 +437,6 @@ static int amd_sched_main(void *param)
 		struct amd_sched_fence *s_fence;
 		struct amd_sched_job *sched_job;
 		struct fence *fence;
-		unsigned long flags;
 
 		wait_event_interruptible(sched->wake_up_worker,
 			(entity = amd_sched_select_entity(sched)) ||
@@ -480,14 +451,6 @@ static int amd_sched_main(void *param)
 
 		s_fence = sched_job->s_fence;
 
-		if (sched->timeout != MAX_SCHEDULE_TIMEOUT) {
-			INIT_DELAYED_WORK(&s_fence->dwork, amd_sched_fence_work_func);
-			schedule_delayed_work(&s_fence->dwork, sched->timeout);
-			spin_lock_irqsave(&sched->fence_list_lock, flags);
-			list_add_tail(&s_fence->list, &sched->fence_list);
-			spin_unlock_irqrestore(&sched->fence_list_lock, flags);
-		}
-
 		atomic_inc(&sched->hw_rq_count);
 		amd_sched_job_pre_schedule(sched, sched_job);
 		fence = sched->ops->run_job(sched_job);

drivers/gpu/drm/amd/scheduler/gpu_scheduler.h

@@ -74,8 +74,6 @@ struct amd_sched_fence {
 	struct amd_gpu_scheduler	*sched;
 	spinlock_t			lock;
 	void                            *owner;
-	struct delayed_work		dwork;
-	struct list_head		list;
 	struct amd_sched_job 	*s_job;
 };
 
@@ -127,8 +125,6 @@ struct amd_gpu_scheduler {
 	wait_queue_head_t		wake_up_worker;
 	wait_queue_head_t		job_scheduled;
 	atomic_t			hw_rq_count;
-	struct list_head		fence_list;
-	spinlock_t			fence_list_lock;
 	struct task_struct		*thread;
 	struct list_head	ring_mirror_list;
 	spinlock_t			job_list_lock;