drm/i915/gt: Track timeline activeness in enter/exit

Lift moving the timeline to/from the active_list on enter/exit in order
to shorten the active tracking span in comparison to the existing
pin/unpin.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Matthew Auld <matthew.auld@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190815205709.24285-1-chris@chris-wilson.co.uk

drivers/gpu/drm/i915/gem/i915_gem_pm.c

@@ -37,7 +37,6 @@ static void i915_gem_park(struct drm_i915_private *i915)
 	for_each_engine(engine, i915, id)
 		call_idle_barriers(engine); /* cleanup after wedging */
 
-	intel_timelines_park(i915);
 	i915_vma_parked(i915);
 
 	i915_globals_park();

drivers/gpu/drm/i915/gt/intel_context.c

@@ -280,10 +280,12 @@ int __init i915_global_context_init(void)
 void intel_context_enter_engine(struct intel_context *ce)
 {
 	intel_engine_pm_get(ce->engine);
+	intel_timeline_enter(ce->timeline);
 }
 
 void intel_context_exit_engine(struct intel_context *ce)
 {
+	intel_timeline_exit(ce->timeline);
 	intel_engine_pm_put(ce->engine);
 }
 

drivers/gpu/drm/i915/gt/intel_engine_pm.c

@@ -66,6 +66,8 @@ static bool switch_to_kernel_context(struct intel_engine_cs *engine)
 		/* Context switch failed, hope for the best! Maybe reset? */
 		return true;
 
+	intel_timeline_enter(rq->timeline);
+
 	/* Check again on the next retirement. */
 	engine->wakeref_serial = engine->serial + 1;
 	i915_request_add_active_barriers(rq);

drivers/gpu/drm/i915/gt/intel_lrc.c

@@ -3306,13 +3306,17 @@ static void virtual_context_enter(struct intel_context *ce)
 
 	for (n = 0; n < ve->num_siblings; n++)
 		intel_engine_pm_get(ve->siblings[n]);
+
+	intel_timeline_enter(ce->timeline);
 }
 
 static void virtual_context_exit(struct intel_context *ce)
 {
 	struct virtual_engine *ve = container_of(ce, typeof(*ve), context);
 	unsigned int n;
 
+	intel_timeline_exit(ce->timeline);
+
 	for (n = 0; n < ve->num_siblings; n++)
 		intel_engine_pm_put(ve->siblings[n]);
 }

drivers/gpu/drm/i915/gt/intel_timeline.c

@@ -278,64 +278,11 @@ void intel_timelines_init(struct drm_i915_private *i915)
 	timelines_init(&i915->gt);
 }
 
-static void timeline_add_to_active(struct intel_timeline *tl)
-{
-	struct intel_gt_timelines *gt = &tl->gt->timelines;
-
-	mutex_lock(&gt->mutex);
-	list_add(&tl->link, &gt->active_list);
-	mutex_unlock(&gt->mutex);
-}
-
-static void timeline_remove_from_active(struct intel_timeline *tl)
-{
-	struct intel_gt_timelines *gt = &tl->gt->timelines;
-
-	mutex_lock(&gt->mutex);
-	list_del(&tl->link);
-	mutex_unlock(&gt->mutex);
-}
-
-static void timelines_park(struct intel_gt *gt)
-{
-	struct intel_gt_timelines *timelines = &gt->timelines;
-	struct intel_timeline *timeline;
-
-	mutex_lock(&timelines->mutex);
-	list_for_each_entry(timeline, &timelines->active_list, link) {
-		/*
-		 * All known fences are completed so we can scrap
-		 * the current sync point tracking and start afresh,
-		 * any attempt to wait upon a previous sync point
-		 * will be skipped as the fence was signaled.
-		 */
-		i915_syncmap_free(&timeline->sync);
-	}
-	mutex_unlock(&timelines->mutex);
-}
-
-/**
- * intel_timelines_park - called when the driver idles
- * @i915: the drm_i915_private device
- *
- * When the driver is completely idle, we know that all of our sync points
- * have been signaled and our tracking is then entirely redundant. Any request
- * to wait upon an older sync point will be completed instantly as we know
- * the fence is signaled and therefore we will not even look them up in the
- * sync point map.
- */
-void intel_timelines_park(struct drm_i915_private *i915)
-{
-	timelines_park(&i915->gt);
-}
-
 void intel_timeline_fini(struct intel_timeline *timeline)
 {
 	GEM_BUG_ON(timeline->pin_count);
 	GEM_BUG_ON(!list_empty(&timeline->requests));
 
-	i915_syncmap_free(&timeline->sync);
-
 	if (timeline->hwsp_cacheline)
 		cacheline_free(timeline->hwsp_cacheline);
 	else
@@ -370,6 +317,7 @@ int intel_timeline_pin(struct intel_timeline *tl)
 	if (tl->pin_count++)
 		return 0;
 	GEM_BUG_ON(!tl->pin_count);
+	GEM_BUG_ON(tl->active_count);
 
 	err = i915_vma_pin(tl->hwsp_ggtt, 0, 0, PIN_GLOBAL | PIN_HIGH);
 	if (err)
@@ -380,7 +328,6 @@ int intel_timeline_pin(struct intel_timeline *tl)
 		offset_in_page(tl->hwsp_offset);
 
 	cacheline_acquire(tl->hwsp_cacheline);
-	timeline_add_to_active(tl);
 
 	return 0;
 
@@ -389,6 +336,40 @@ int intel_timeline_pin(struct intel_timeline *tl)
 	return err;
 }
 
+void intel_timeline_enter(struct intel_timeline *tl)
+{
+	struct intel_gt_timelines *timelines = &tl->gt->timelines;
+
+	GEM_BUG_ON(!tl->pin_count);
+	if (tl->active_count++)
+		return;
+	GEM_BUG_ON(!tl->active_count); /* overflow? */
+
+	mutex_lock(&timelines->mutex);
+	list_add(&tl->link, &timelines->active_list);
+	mutex_unlock(&timelines->mutex);
+}
+
+void intel_timeline_exit(struct intel_timeline *tl)
+{
+	struct intel_gt_timelines *timelines = &tl->gt->timelines;
+
+	GEM_BUG_ON(!tl->active_count);
+	if (--tl->active_count)
+		return;
+
+	mutex_lock(&timelines->mutex);
+	list_del(&tl->link);
+	mutex_unlock(&timelines->mutex);
+
+	/*
+	 * Since this timeline is idle, all bariers upon which we were waiting
+	 * must also be complete and so we can discard the last used barriers
+	 * without loss of information.
+	 */
+	i915_syncmap_free(&tl->sync);
+}
+
 static u32 timeline_advance(struct intel_timeline *tl)
 {
 	GEM_BUG_ON(!tl->pin_count);
@@ -546,16 +527,9 @@ void intel_timeline_unpin(struct intel_timeline *tl)
 	if (--tl->pin_count)
 		return;
 
-	timeline_remove_from_active(tl);
+	GEM_BUG_ON(tl->active_count);
 	cacheline_release(tl->hwsp_cacheline);
 
-	/*
-	 * Since this timeline is idle, all bariers upon which we were waiting
-	 * must also be complete and so we can discard the last used barriers
-	 * without loss of information.
-	 */
-	i915_syncmap_free(&tl->sync);
-
 	__i915_vma_unpin(tl->hwsp_ggtt);
 }
 

drivers/gpu/drm/i915/gt/intel_timeline.h

@@ -77,17 +77,18 @@ static inline bool intel_timeline_sync_is_later(struct intel_timeline *tl,
 }
 
 int intel_timeline_pin(struct intel_timeline *tl);
+void intel_timeline_enter(struct intel_timeline *tl);
 int intel_timeline_get_seqno(struct intel_timeline *tl,
 			     struct i915_request *rq,
 			     u32 *seqno);
+void intel_timeline_exit(struct intel_timeline *tl);
 void intel_timeline_unpin(struct intel_timeline *tl);
 
 int intel_timeline_read_hwsp(struct i915_request *from,
 			     struct i915_request *until,
 			     u32 *hwsp_offset);
 
 void intel_timelines_init(struct drm_i915_private *i915);
-void intel_timelines_park(struct drm_i915_private *i915);
 void intel_timelines_fini(struct drm_i915_private *i915);
 
 #endif

drivers/gpu/drm/i915/gt/intel_timeline_types.h

@@ -25,7 +25,25 @@ struct intel_timeline {
 
 	struct mutex mutex; /* protects the flow of requests */
 
+	/*
+	 * pin_count and active_count track essentially the same thing:
+	 * How many requests are in flight or may be under construction.
+	 *
+	 * We need two distinct counters so that we can assign different
+	 * lifetimes to the events for different use-cases. For example,
+	 * we want to permanently keep the timeline pinned for the kernel
+	 * context so that we can issue requests at any time without having
+	 * to acquire space in the GGTT. However, we want to keep tracking
+	 * the activity (to be able to detect when we become idle) along that
+	 * permanently pinned timeline and so end up requiring two counters.
+	 *
+	 * Note that the active_count is protected by the intel_timeline.mutex,
+	 * but the pin_count is protected by a combination of serialisation
+	 * from the intel_context caller plus internal atomicity.
+	 */
 	unsigned int pin_count;
+	unsigned int active_count;
+
 	const u32 *hwsp_seqno;
 	struct i915_vma *hwsp_ggtt;
 	u32 hwsp_offset;

drivers/gpu/drm/i915/gt/selftest_timeline.c

@@ -816,8 +816,6 @@ static int live_hwsp_recycle(void *arg)
 
 			if (err)
 				goto out;
-
-			intel_timelines_park(i915); /* Encourage recycling! */
 		} while (!__igt_timeout(end_time, NULL));
 	}