On 10/10/2019 08:14, Chris Wilson wrote:
Since writing the comment that the scheduler is entirely passive, we've
added minimal timeslicing which adds the most primitive of active
elements (a timeout and reschedule).
Signed-off-by: Chris Wilson <chris@xxxxxxxxxxxxxxxxxx>
Cc: Tvrtko Ursulin <tvrtko.ursulin@xxxxxxxxxxxxxxx>
Cc: Ramalingam C <ramalingam.c@xxxxxxxxx>
---
drivers/gpu/drm/i915/i915_scheduler_types.h | 9 +++++++++
1 file changed, 9 insertions(+)
diff --git a/drivers/gpu/drm/i915/i915_scheduler_types.h b/drivers/gpu/drm/i915/i915_scheduler_types.h
index aad81acba9dc..d18e70550054 100644
--- a/drivers/gpu/drm/i915/i915_scheduler_types.h
+++ b/drivers/gpu/drm/i915/i915_scheduler_types.h
@@ -49,6 +49,15 @@ struct i915_sched_attr {
* DAG of each request, we are able to insert it into a sorted queue when it
* is ready, and are able to reorder its portion of the graph to accommodate
* dynamic priority changes.
+ *
+ * Ok, there is now one active element to the "scheduler" in the backends.
+ * We let a new context run for a small amount of time before re-evaluating
+ * the run order. As we re-evaluate, we maintain the strict ordering of
+ * dependencies, but attempt to rotate the active contexts (the current context
+ * is put to the back of its priority queue, then reshuffling its dependents).
+ * This provides minimal timeslicing and prevents a userspace hog (e.g.
+ * something waiting on a user semaphore [VkEvent]) from denying service to
+ * others.
*/
struct i915_sched_node {
struct list_head signalers_list; /* those before us, we depend upon */
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@xxxxxxxxx>
Regards,
Tvrtko
_______________________________________________
Intel-gfx mailing list
Intel-gfx@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
