[Intel-gfx] [PATCH v3] drm/i915/execlists: Avoid kicking priority on the current context

[Chris Wilson]

If the request is currently on the HW (in port 0), then we do not need
to kick the submission tasklet to evaluate whether we should be
preempting itself in order to execute it again.

In the case that was annoying me:

   execlists_schedule: rq(18:211173).prio=0 -> 2
   need_preempt: last(18:211174).prio=0, queue.prio=2

We are bumping the priority of the first of a pair of requests running
in the current context. Then when evaluating preempt, we would see that
that our priority request is higher than the last executing request in
ELSP0 and so trigger preemption, not realising that our intended request
was already executing.

v2: As we assume state of the execlists->port[] that is only valid while
we hold the timeline lock we have to repeat some earlier tests that on
the validity of the node.
v3: Wrap guc submission under the timeline.lock as is now the way of all
things.

Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin at intel.com>
---
 drivers/gpu/drm/i915/intel_guc_submission.c |  3 ++
 drivers/gpu/drm/i915/intel_lrc.c            | 36 +++++++++++++++++----
 2 files changed, 33 insertions(+), 6 deletions(-)

diff --git a/drivers/gpu/drm/i915/intel_guc_submission.c b/drivers/gpu/drm/i915/intel_guc_submission.c
index 195adbd0ebf7..aa539c9c653d 100644
--- a/drivers/gpu/drm/i915/intel_guc_submission.c
+++ b/drivers/gpu/drm/i915/intel_guc_submission.c
@@ -785,7 +785,9 @@ static void guc_submission_tasklet(unsigned long data)
 	struct intel_engine_execlists * const execlists = &engine->execlists;
 	struct execlist_port *port = execlists->port;
 	struct i915_request *rq;
+	unsigned long flags;
 
+	spin_lock_irqsave(&engine->timeline.lock, flags);
 	rq = port_request(port);
 	while (rq && i915_request_completed(rq)) {
 		trace_i915_request_out(rq);
@@ -800,6 +802,7 @@ static void guc_submission_tasklet(unsigned long data)
 			rq = NULL;
 		}
 	}
+	spin_unlock_irqrestore(&engine->timeline.lock, flags);
 
 	if (execlists_is_active(execlists, EXECLISTS_ACTIVE_PREEMPT) &&
 	    intel_read_status_page(engine, I915_GEM_HWS_PREEMPT_INDEX) ==
diff --git a/drivers/gpu/drm/i915/intel_lrc.c b/drivers/gpu/drm/i915/intel_lrc.c
index e5385dbfcdda..2257bc7b167f 100644
--- a/drivers/gpu/drm/i915/intel_lrc.c
+++ b/drivers/gpu/drm/i915/intel_lrc.c
@@ -1238,9 +1238,13 @@ static void execlists_schedule(struct i915_request *request,
 
 		engine = sched_lock_engine(node, engine);
 
+		/* Recheck after acquiring the engine->timeline.lock */
 		if (prio <= node->attr.priority)
 			continue;
 
+		if (i915_sched_node_signaled(node))
+			continue;
+
 		node->attr.priority = prio;
 		if (!list_empty(&node->link)) {
 			if (last != engine) {
@@ -1249,14 +1253,34 @@ static void execlists_schedule(struct i915_request *request,
 			}
 			GEM_BUG_ON(pl->priority != prio);
 			list_move_tail(&node->link, &pl->requests);
+		} else {
+			/*
+			 * If the request is not in the priolist queue because
+			 * it is not yet runnable, then it doesn't contribute
+			 * to our preemption decisions. On the other hand,
+			 * if the request is on the HW, it too is not in the
+			 * queue; but in that case we may still need to reorder
+			 * the inflight requests.
+			 */
+			if (!i915_sw_fence_done(&sched_to_request(node)->submit))
+				continue;
 		}
 
-		if (prio > engine->execlists.queue_priority &&
-		    i915_sw_fence_done(&sched_to_request(node)->submit)) {
-			/* defer submission until after all of our updates */
-			__update_queue(engine, prio);
-			tasklet_hi_schedule(&engine->execlists.tasklet);
-		}
+		if (prio <= engine->execlists.queue_priority)
+			continue;
+
+		/*
+		 * If we are already the currently executing context, don't
+		 * bother evaluating if we should preempt ourselves.
+		 */
+		if (sched_to_request(node)->global_seqno &&
+		    i915_seqno_passed(port_request(engine->execlists.port)->global_seqno,
+				      sched_to_request(node)->global_seqno))
+			continue;
+
+		/* Defer (tasklet) submission until after all of our updates. */
+		__update_queue(engine, prio);
+		tasklet_hi_schedule(&engine->execlists.tasklet);
 	}
 
 	spin_unlock_irq(&engine->timeline.lock);
-- 
2.18.0