diff mbox

drm/i915: Look for an active kernel context before switching

Message ID 20180523080439.20697-1-chris@chris-wilson.co.uk (mailing list archive)
State New, archived
Headers show

Commit Message

Chris Wilson May 23, 2018, 8:04 a.m. UTC
From: Mika Kuoppala <mika.kuoppala@intel.com>

We were not very carefully checking to see if an older request on the
engine was an earlier switch-to-kernel-context before deciding to emit a
new switch. The end result would be that we could get into a permanent
loop of trying to emit a new request to perform the switch simply to
flush the existing switch.

What we need is a means of tracking the completion of each timeline
versus the kernel context, that is to detect if a more recent request
has been submitted that would result in a switch away from the kernel
context. To realise this, we need only to look in our syncmap on the
kernel context and check that we have synchronized against all active
rings.

Reported-by: Mika Kuoppala <mika.kuoppala@intel.com>
Fixes: a89d1f921c15 ("drm/i915: Split i915_gem_timeline into individual timelines")
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Mika Kuoppala <mika.kuoppala@intel.com>
---
 drivers/gpu/drm/i915/i915_gem_context.c | 42 +++++++++++++++++++++----
 1 file changed, 36 insertions(+), 6 deletions(-)
diff mbox

Patch

diff --git a/drivers/gpu/drm/i915/i915_gem_context.c b/drivers/gpu/drm/i915/i915_gem_context.c
index b69b18ef8120..eb6df93a96e6 100644
--- a/drivers/gpu/drm/i915/i915_gem_context.c
+++ b/drivers/gpu/drm/i915/i915_gem_context.c
@@ -587,18 +587,45 @@  last_request_on_engine(struct i915_timeline *timeline,
 	return NULL;
 }
 
-static bool engine_has_idle_kernel_context(struct intel_engine_cs *engine)
+static bool engine_has_kernel_context_barrier(struct intel_engine_cs *engine)
 {
-	struct list_head * const active_rings = &engine->i915->gt.active_rings;
+	struct drm_i915_private *i915 = engine->i915;
+	struct i915_timeline *barrier =
+		to_intel_context(i915->kernel_context, engine)->ring->timeline;
 	struct intel_ring *ring;
+	bool any_active = false;
 
-	lockdep_assert_held(&engine->i915->drm.struct_mutex);
+	lockdep_assert_held(&i915->drm.struct_mutex);
+	list_for_each_entry(ring, &i915->gt.active_rings, active_link) {
+		struct i915_request *rq;
+
+		if (ring->timeline == barrier)
+			continue;
 
-	list_for_each_entry(ring, active_rings, active_link) {
-		if (last_request_on_engine(ring->timeline, engine))
+		rq = last_request_on_engine(ring->timeline, engine);
+		if (!rq)
+			continue;
+
+		/*
+		 * Was this request submitted after the previous
+		 * switch-to-kernel-context?
+		 */
+		if (!i915_timeline_sync_is_later(barrier, &rq->fence))
 			return false;
+
+		any_active = true;
 	}
 
+	/*
+	 * If any other timeline was still active and behind the last barrier,
+	 * then our last switch-to-kernel-context must still be queued and
+	 * will run last (leaving the engine in the kernel context when it
+	 * eventually idles).
+	 */
+	if (any_active)
+		return true;
+
+	/* The engine is idle; check that it is idling in the kernel context. */
 	return intel_engine_has_kernel_context(engine);
 }
 
@@ -615,7 +642,7 @@  int i915_gem_switch_to_kernel_context(struct drm_i915_private *i915)
 		struct intel_ring *ring;
 		struct i915_request *rq;
 
-		if (engine_has_idle_kernel_context(engine))
+		if (engine_has_kernel_context_barrier(engine))
 			continue;
 
 		rq = i915_request_alloc(engine, i915->kernel_context);
@@ -626,6 +653,9 @@  int i915_gem_switch_to_kernel_context(struct drm_i915_private *i915)
 		list_for_each_entry(ring, &i915->gt.active_rings, active_link) {
 			struct i915_request *prev;
 
+			if (ring == rq->ring)
+				continue;
+
 			prev = last_request_on_engine(ring->timeline, engine);
 			if (prev)
 				i915_sw_fence_await_sw_fence_gfp(&rq->submit,