All of lore.kernel.org
 help / color / mirror / Atom feed
* [PATCH] drm/i915/execlists: Disable submission tasklets when rescheduling
@ 2018-05-01  9:49 Chris Wilson
  2018-05-01 10:01 ` Chris Wilson
                   ` (2 more replies)
  0 siblings, 3 replies; 4+ messages in thread
From: Chris Wilson @ 2018-05-01  9:49 UTC (permalink / raw)
  To: intel-gfx

As we reschedule the requests, we do not want the submission tasklet
running until we finish updating the priority chains. (We start
rewriting priorities from the oldest, but the dequeue looks at the most
recent in-flight, so there is a small race condition where dequeue may
decide that preemption is falsely required.) Combine the tasklet kicking
from adding a new request with the set-wedge protection so that we only
have to adjust the preempt-counter once to achieve both goals.

Reported-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
---
 drivers/gpu/drm/i915/i915_gem.c     | 4 ++--
 drivers/gpu/drm/i915/i915_request.c | 5 +----
 2 files changed, 3 insertions(+), 6 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
index 438a2fc5bba0..a278e4dc5e09 100644
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -577,10 +577,10 @@ static void __fence_set_priority(struct dma_fence *fence,
 	rq = to_request(fence);
 	engine = rq->engine;
 
-	rcu_read_lock();
+	local_bh_disable(); /* RCU serialisation for set-wedged protection */
 	if (engine->schedule)
 		engine->schedule(rq, attr);
-	rcu_read_unlock();
+	local_bh_enable(); /* kick the tasklets if queues were reprioritised */
 }
 
 static void fence_set_priority(struct dma_fence *fence,
diff --git a/drivers/gpu/drm/i915/i915_request.c b/drivers/gpu/drm/i915/i915_request.c
index 7bb613c00cc3..eb66579db823 100644
--- a/drivers/gpu/drm/i915/i915_request.c
+++ b/drivers/gpu/drm/i915/i915_request.c
@@ -1111,12 +1111,9 @@ void __i915_request_add(struct i915_request *request, bool flush_caches)
 	 * decide whether to preempt the entire chain so that it is ready to
 	 * run at the earliest possible convenience.
 	 */
-	rcu_read_lock();
+	local_bh_disable();
 	if (engine->schedule)
 		engine->schedule(request, &request->ctx->sched);
-	rcu_read_unlock();
-
-	local_bh_disable();
 	i915_sw_fence_commit(&request->submit);
 	local_bh_enable(); /* Kick the execlists tasklet if just scheduled */
 
-- 
2.17.0

_______________________________________________
Intel-gfx mailing list
Intel-gfx@lists.freedesktop.org
https://lists.freedesktop.org/mailman/listinfo/intel-gfx

^ permalink raw reply related	[flat|nested] 4+ messages in thread
end of thread, other threads:[~2018-05-01 11:29 UTC | newest]

Thread overview: 4+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2018-05-01  9:49 [PATCH] drm/i915/execlists: Disable submission tasklets when rescheduling Chris Wilson
2018-05-01 10:01 ` Chris Wilson
2018-05-01 10:24 ` ✓ Fi.CI.BAT: success for " Patchwork
2018-05-01 11:29 ` ✗ Fi.CI.IGT: failure " Patchwork

This is an external index of several public inboxes,
see mirroring instructions on how to clone and mirror
all data and code used by this external index.