drm/scheduler: Add drm_sched_suspend/resume_timeout()

This patch adds two new functions to help client drivers suspend and
resume the scheduler job timeout. This can be useful in cases where the
hardware has preemption support enabled. Using this, it is possible to have
the timeout active only for the ring which is active on the ringbuffer.
This patch also makes the job_list_lock IRQ safe.

Suggested-by: Christian Koenig <Christian.Koenig@amd.com>
Signed-off-by: Sharat Masetty <smasetty@codeaurora.org>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

1 files changed, 73 insertions, 12 deletions

diff --git a/drivers/gpu/drm/scheduler/sched_main.c b/drivers/gpu/drm/scheduler/sched_main.c
index 05b803d1248d..dbb69063b3d5 100644
--- a/drivers/gpu/drm/scheduler/sched_main.c
+++ b/drivers/gpu/drm/scheduler/sched_main.c
@@ -211,6 +211,62 @@ void drm_sched_fault(struct drm_gpu_scheduler *sched)
 }
 EXPORT_SYMBOL(drm_sched_fault);
 
+/**
+ * drm_sched_suspend_timeout - Suspend scheduler job timeout
+ *
+ * @sched: scheduler instance for which to suspend the timeout
+ *
+ * Suspend the delayed work timeout for the scheduler. This is done by
+ * modifying the delayed work timeout to an arbitrary large value,
+ * MAX_SCHEDULE_TIMEOUT in this case. Note that this function can be
+ * called from an IRQ context.
+ *
+ * Returns the timeout remaining
+ *
+ */
+unsigned long drm_sched_suspend_timeout(struct drm_gpu_scheduler *sched)
+{
+	unsigned long sched_timeout, now = jiffies;
+
+	sched_timeout = sched->work_tdr.timer.expires;
+
+	/*
+	 * Modify the timeout to an arbitrarily large value. This also prevents
+	 * the timeout to be restarted when new submissions arrive
+	 */
+	if (mod_delayed_work(system_wq, &sched->work_tdr, MAX_SCHEDULE_TIMEOUT)
+			&& time_after(sched_timeout, now))
+		return sched_timeout - now;
+	else
+		return sched->timeout;
+}
+EXPORT_SYMBOL(drm_sched_suspend_timeout);
+
+/**
+ * drm_sched_resume_timeout - Resume scheduler job timeout
+ *
+ * @sched: scheduler instance for which to resume the timeout
+ * @remaining: remaining timeout
+ *
+ * Resume the delayed work timeout for the scheduler. Note that
+ * this function can be called from an IRQ context.
+ */
+void drm_sched_resume_timeout(struct drm_gpu_scheduler *sched,
+		unsigned long remaining)
+{
+	unsigned long flags;
+
+	spin_lock_irqsave(&sched->job_list_lock, flags);
+
+	if (list_empty(&sched->ring_mirror_list))
+		cancel_delayed_work(&sched->work_tdr);
+	else
+		mod_delayed_work(system_wq, &sched->work_tdr, remaining);
+
+	spin_unlock_irqrestore(&sched->job_list_lock, flags);
+}
+EXPORT_SYMBOL(drm_sched_resume_timeout);
+
 /* job_finish is called after hw fence signaled
  */
 static void drm_sched_job_finish(struct work_struct *work)
@@ -218,6 +274,7 @@ static void drm_sched_job_finish(struct work_struct *work)
 	struct drm_sched_job *s_job = container_of(work, struct drm_sched_job,
 						   finish_work);
 	struct drm_gpu_scheduler *sched = s_job->sched;
+	unsigned long flags;
 
 	/*
 	 * Canceling the timeout without removing our job from the ring mirror
@@ -228,12 +285,12 @@ static void drm_sched_job_finish(struct work_struct *work)
 	 */
 	cancel_delayed_work_sync(&sched->work_tdr);
 
-	spin_lock(&sched->job_list_lock);
+	spin_lock_irqsave(&sched->job_list_lock, flags);
 	/* remove job from ring_mirror_list */
 	list_del_init(&s_job->node);
 	/* queue TDR for next job */
 	drm_sched_start_timeout(sched);
-	spin_unlock(&sched->job_list_lock);
+	spin_unlock_irqrestore(&sched->job_list_lock, flags);
 
 	sched->ops->free_job(s_job);
 }
@@ -249,20 +306,22 @@ static void drm_sched_job_finish_cb(struct dma_fence *f,
 static void drm_sched_job_begin(struct drm_sched_job *s_job)
 {
 	struct drm_gpu_scheduler *sched = s_job->sched;
+	unsigned long flags;
 
 	dma_fence_add_callback(&s_job->s_fence->finished, &s_job->finish_cb,
 			       drm_sched_job_finish_cb);
 
-	spin_lock(&sched->job_list_lock);
+	spin_lock_irqsave(&sched->job_list_lock, flags);
 	list_add_tail(&s_job->node, &sched->ring_mirror_list);
 	drm_sched_start_timeout(sched);
-	spin_unlock(&sched->job_list_lock);
+	spin_unlock_irqrestore(&sched->job_list_lock, flags);
 }
 
 static void drm_sched_job_timedout(struct work_struct *work)
 {
 	struct drm_gpu_scheduler *sched;
 	struct drm_sched_job *job;
+	unsigned long flags;
 
 	sched = container_of(work, struct drm_gpu_scheduler, work_tdr.work);
 	job = list_first_entry_or_null(&sched->ring_mirror_list,
@@ -271,9 +330,9 @@ static void drm_sched_job_timedout(struct work_struct *work)
 	if (job)
 		job->sched->ops->timedout_job(job);
 
-	spin_lock(&sched->job_list_lock);
+	spin_lock_irqsave(&sched->job_list_lock, flags);
 	drm_sched_start_timeout(sched);
-	spin_unlock(&sched->job_list_lock);
+	spin_unlock_irqrestore(&sched->job_list_lock, flags);
 }
 
 /**
@@ -287,9 +346,10 @@ void drm_sched_hw_job_reset(struct drm_gpu_scheduler *sched, struct drm_sched_jo
 {
 	struct drm_sched_job *s_job;
 	struct drm_sched_entity *entity, *tmp;
+	unsigned long flags;
 	int i;
 
-	spin_lock(&sched->job_list_lock);
+	spin_lock_irqsave(&sched->job_list_lock, flags);
 	list_for_each_entry_reverse(s_job, &sched->ring_mirror_list, node) {
 		if (s_job->s_fence->parent &&
 		    dma_fence_remove_callback(s_job->s_fence->parent,
@@ -299,7 +359,7 @@ void drm_sched_hw_job_reset(struct drm_gpu_scheduler *sched, struct drm_sched_jo
 			atomic_dec(&sched->hw_rq_count);
 		}
 	}
-	spin_unlock(&sched->job_list_lock);
+	spin_unlock_irqrestore(&sched->job_list_lock, flags);
 
 	if (bad && bad->s_priority != DRM_SCHED_PRIORITY_KERNEL) {
 		atomic_inc(&bad->karma);
@@ -337,9 +397,10 @@ void drm_sched_job_recovery(struct drm_gpu_scheduler *sched)
 {
 	struct drm_sched_job *s_job, *tmp;
 	bool found_guilty = false;
+	unsigned long flags;
 	int r;
 
-	spin_lock(&sched->job_list_lock);
+	spin_lock_irqsave(&sched->job_list_lock, flags);
 	list_for_each_entry_safe(s_job, tmp, &sched->ring_mirror_list, node) {
 		struct drm_sched_fence *s_fence = s_job->s_fence;
 		struct dma_fence *fence;
@@ -353,7 +414,7 @@ void drm_sched_job_recovery(struct drm_gpu_scheduler *sched)
 		if (found_guilty && s_job->s_fence->scheduled.context == guilty_context)
 			dma_fence_set_error(&s_fence->finished, -ECANCELED);
 
-		spin_unlock(&sched->job_list_lock);
+		spin_unlock_irqrestore(&sched->job_list_lock, flags);
 		fence = sched->ops->run_job(s_job);
 		atomic_inc(&sched->hw_rq_count);
 
@@ -372,10 +433,10 @@ void drm_sched_job_recovery(struct drm_gpu_scheduler *sched)
 				drm_sched_expel_job_unlocked(s_job);
 			drm_sched_process_job(NULL, &s_fence->cb);
 		}
-		spin_lock(&sched->job_list_lock);
+		spin_lock_irqsave(&sched->job_list_lock, flags);
 	}
 	drm_sched_start_timeout(sched);
-	spin_unlock(&sched->job_list_lock);
+	spin_unlock_irqrestore(&sched->job_list_lock, flags);
 }
 EXPORT_SYMBOL(drm_sched_job_recovery);