[Intel-gfx] [PATCH 01/11] drm/i915/execlists: Avoid kicking the submission too early for rescheduling

Chris Wilson chris at chris-wilson.co.uk
Mon Mar 26 11:50:34 UTC 2018

If the request is still waiting on external fences, it has not yet been
submitted to the HW queue and so we can forgo kicking the submission
tasklet when re-evaluating its priority.

This should have no impact other than reducing the number of tasklet
wakeups under signal heavy workloads (e.g. switching between engines).

v2: Use prebaked container_of()

References: f6322eddaff7 ("drm/i915/preemption: Allow preemption between submission ports")
Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
Cc: Michał Winiarski <michal.winiarski at intel.com>
Cc: Michel Thierry <michel.thierry at intel.com>
Cc: Mika Kuoppala <mika.kuoppala at linux.intel.com>
Cc: Tvrtko Ursulin <tvrtko.ursulin at intel.com>
 drivers/gpu/drm/i915/intel_lrc.c | 15 +++++++++++----
 1 file changed, 11 insertions(+), 4 deletions(-)

diff --git a/drivers/gpu/drm/i915/intel_lrc.c b/drivers/gpu/drm/i915/intel_lrc.c
index b4ab06b05e58..104b69e0494f 100644
--- a/drivers/gpu/drm/i915/intel_lrc.c
+++ b/drivers/gpu/drm/i915/intel_lrc.c
@@ -1051,12 +1051,16 @@ static void queue_request(struct intel_engine_cs *engine,
 	list_add_tail(&pt->link, &lookup_priolist(engine, pt, prio)->requests);
-static void submit_queue(struct intel_engine_cs *engine, int prio)
+static void __submit_queue(struct intel_engine_cs *engine, int prio)
-	if (prio > engine->execlists.queue_priority) {
 		engine->execlists.queue_priority = prio;
-	}
+static void submit_queue(struct intel_engine_cs *engine, int prio)
+	if (prio > engine->execlists.queue_priority)
+		__submit_queue(engine, prio);
 static void execlists_submit_request(struct i915_request *request)
@@ -1189,7 +1193,10 @@ static void execlists_schedule(struct i915_request *request, int prio)
 			queue_request(engine, pt, prio);
-		submit_queue(engine, prio);
+		if (prio > engine->execlists.queue_priority &&
+		    i915_sw_fence_done(&pt_to_request(pt)->submit))
+			__submit_queue(engine, prio);

More information about the Intel-gfx mailing list