[Intel-gfx] [PATCH v4 20/38] drm/i915: Added scheduler flush calls to ring throttle and idle functions

John.C.Harrison at Intel.com John.C.Harrison at Intel.com
Mon Jan 11 10:42:49 PST 2016


From: John Harrison <John.C.Harrison at Intel.com>

When requesting that all GPU work is completed, it is now necessary to
get the scheduler involved in order to flush out work that queued and
not yet submitted.

v2: Updated to add support for flushing the scheduler queue by time
stamp rather than just doing a blanket flush.

v3: Moved submit_max_priority() to this patch from an earlier patch
is it is no longer required in the other.

v4: Corrected the format of a comment to keep the style checker happy.
Downgraded a BUG_ON to a WARN_ON as the latter is preferred.

For: VIZ-1587
Signed-off-by: John Harrison <John.C.Harrison at Intel.com>
---
 drivers/gpu/drm/i915/i915_gem.c       |  24 +++++-
 drivers/gpu/drm/i915/i915_scheduler.c | 134 ++++++++++++++++++++++++++++++++++
 drivers/gpu/drm/i915/i915_scheduler.h |   3 +
 3 files changed, 160 insertions(+), 1 deletion(-)

diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
index 7716462..6ea0896 100644
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -3733,6 +3733,10 @@ int i915_gpu_idle(struct drm_device *dev)
 
 	/* Flush everything onto the inactive list. */
 	for_each_ring(ring, dev_priv, i) {
+		ret = i915_scheduler_flush(ring, true);
+		if (ret < 0)
+			return ret;
+
 		if (!i915.enable_execlists) {
 			struct drm_i915_gem_request *req;
 
@@ -4446,7 +4450,8 @@ i915_gem_ring_throttle(struct drm_device *dev, struct drm_file *file)
 	unsigned long recent_enough = jiffies - DRM_I915_THROTTLE_JIFFIES;
 	struct drm_i915_gem_request *request, *target = NULL;
 	unsigned reset_counter;
-	int ret;
+	int i, ret;
+	struct intel_engine_cs *ring;
 
 	ret = i915_gem_wait_for_error(&dev_priv->gpu_error);
 	if (ret)
@@ -4456,6 +4461,23 @@ i915_gem_ring_throttle(struct drm_device *dev, struct drm_file *file)
 	if (ret)
 		return ret;
 
+	for_each_ring(ring, dev_priv, i) {
+		/*
+		 * Flush out scheduler entries that are getting 'stale'. Note
+		 * that the following recent_enough test will only check
+		 * against the time at which the request was submitted to the
+		 * hardware (i.e. when it left the scheduler) not the time it
+		 * was submitted to the driver.
+		 *
+		 * Also, there is not much point worring about busy return
+		 * codes from the scheduler flush call. Even if more work
+		 * cannot be submitted right now for whatever reason, we
+		 * still want to throttle against stale work that has already
+		 * been submitted.
+		 */
+		i915_scheduler_flush_stamp(ring, recent_enough, false);
+	}
+
 	spin_lock(&file_priv->mm.lock);
 	list_for_each_entry(request, &file_priv->mm.request_list, client_list) {
 		if (time_after_eq(request->emitted_jiffies, recent_enough))
diff --git a/drivers/gpu/drm/i915/i915_scheduler.c b/drivers/gpu/drm/i915/i915_scheduler.c
index 8c05dd0..bd52752 100644
--- a/drivers/gpu/drm/i915/i915_scheduler.c
+++ b/drivers/gpu/drm/i915/i915_scheduler.c
@@ -31,6 +31,8 @@ static int         i915_scheduler_remove_dependent(struct i915_scheduler *schedu
 						   struct i915_scheduler_queue_entry *remove);
 static int         i915_scheduler_submit(struct intel_engine_cs *ring,
 					 bool is_locked);
+static int         i915_scheduler_submit_max_priority(struct intel_engine_cs *ring,
+						      bool is_locked);
 static uint32_t    i915_scheduler_count_flying(struct i915_scheduler *scheduler,
 					       struct intel_engine_cs *ring);
 static void        i915_scheduler_priority_bump_clear(struct i915_scheduler *scheduler);
@@ -589,6 +591,100 @@ void i915_gem_scheduler_work_handler(struct work_struct *work)
 	}
 }
 
+int i915_scheduler_flush_stamp(struct intel_engine_cs *ring,
+			       unsigned long target,
+			       bool is_locked)
+{
+	struct i915_scheduler_queue_entry *node;
+	struct drm_i915_private           *dev_priv;
+	struct i915_scheduler             *scheduler;
+	unsigned long       flags;
+	int                 flush_count = 0;
+
+	if (!ring)
+		return -EINVAL;
+
+	dev_priv  = ring->dev->dev_private;
+	scheduler = dev_priv->scheduler;
+
+	if (!scheduler)
+		return 0;
+
+	if (is_locked && (scheduler->flags[ring->id] & i915_sf_submitting)) {
+		/*
+		 * Scheduler is busy already submitting another batch,
+		 * come back later rather than going recursive...
+		 */
+		return -EAGAIN;
+	}
+
+	spin_lock_irqsave(&scheduler->lock, flags);
+	i915_scheduler_priority_bump_clear(scheduler);
+	list_for_each_entry(node, &scheduler->node_queue[ring->id], link) {
+		if (!I915_SQS_IS_QUEUED(node))
+			continue;
+
+		if (node->stamp > target)
+			continue;
+
+		flush_count = i915_scheduler_priority_bump(scheduler,
+					node, scheduler->priority_level_max);
+	}
+	spin_unlock_irqrestore(&scheduler->lock, flags);
+
+	if (flush_count) {
+		DRM_DEBUG_DRIVER("<%s> Bumped %d entries\n", ring->name, flush_count);
+		flush_count = i915_scheduler_submit_max_priority(ring, is_locked);
+	}
+
+	return flush_count;
+}
+
+int i915_scheduler_flush(struct intel_engine_cs *ring, bool is_locked)
+{
+	struct i915_scheduler_queue_entry *node;
+	struct drm_i915_private           *dev_priv;
+	struct i915_scheduler             *scheduler;
+	unsigned long       flags;
+	bool        found;
+	int         ret;
+	uint32_t    count = 0;
+
+	if (!ring)
+		return -EINVAL;
+
+	dev_priv  = ring->dev->dev_private;
+	scheduler = dev_priv->scheduler;
+
+	if (!scheduler)
+		return 0;
+
+	WARN_ON(is_locked && (scheduler->flags[ring->id] & i915_sf_submitting));
+
+	do {
+		found = false;
+		spin_lock_irqsave(&scheduler->lock, flags);
+		list_for_each_entry(node, &scheduler->node_queue[ring->id], link) {
+			if (!I915_SQS_IS_QUEUED(node))
+				continue;
+
+			found = true;
+			break;
+		}
+		spin_unlock_irqrestore(&scheduler->lock, flags);
+
+		if (found) {
+			ret = i915_scheduler_submit(ring, is_locked);
+			if (ret < 0)
+				return ret;
+
+			count += ret;
+		}
+	} while (found);
+
+	return count;
+}
+
 static void i915_scheduler_priority_bump_clear(struct i915_scheduler *scheduler)
 {
 	struct i915_scheduler_queue_entry *node;
@@ -642,6 +738,44 @@ static int i915_scheduler_priority_bump(struct i915_scheduler *scheduler,
 	return count;
 }
 
+static int i915_scheduler_submit_max_priority(struct intel_engine_cs *ring,
+					      bool is_locked)
+{
+	struct i915_scheduler_queue_entry  *node;
+	struct drm_i915_private            *dev_priv = ring->dev->dev_private;
+	struct i915_scheduler              *scheduler = dev_priv->scheduler;
+	unsigned long	flags;
+	int             ret, count = 0;
+	bool            found;
+
+	do {
+		found = false;
+		spin_lock_irqsave(&scheduler->lock, flags);
+		list_for_each_entry(node, &scheduler->node_queue[ring->id], link) {
+			if (!I915_SQS_IS_QUEUED(node))
+				continue;
+
+			if (node->priority < scheduler->priority_level_max)
+				continue;
+
+			found = true;
+			break;
+		}
+		spin_unlock_irqrestore(&scheduler->lock, flags);
+
+		if (!found)
+			break;
+
+		ret = i915_scheduler_submit(ring, is_locked);
+		if (ret < 0)
+			return ret;
+
+		count += ret;
+	} while (found);
+
+	return count;
+}
+
 static int i915_scheduler_pop_from_queue_locked(struct intel_engine_cs *ring,
 				struct i915_scheduler_queue_entry **pop_node,
 				unsigned long *flags)
diff --git a/drivers/gpu/drm/i915/i915_scheduler.h b/drivers/gpu/drm/i915/i915_scheduler.h
index c3e7ac6..7654013 100644
--- a/drivers/gpu/drm/i915/i915_scheduler.h
+++ b/drivers/gpu/drm/i915/i915_scheduler.h
@@ -93,6 +93,9 @@ int         i915_scheduler_queue_execbuffer(struct i915_scheduler_queue_entry *q
 bool        i915_scheduler_notify_request(struct drm_i915_gem_request *req);
 void        i915_scheduler_wakeup(struct drm_device *dev);
 void        i915_gem_scheduler_work_handler(struct work_struct *work);
+int         i915_scheduler_flush(struct intel_engine_cs *ring, bool is_locked);
+int         i915_scheduler_flush_stamp(struct intel_engine_cs *ring,
+				       unsigned long stamp, bool is_locked);
 bool        i915_scheduler_is_request_tracked(struct drm_i915_gem_request *req,
 					      bool *completed, bool *busy);
 
-- 
1.9.1



More information about the Intel-gfx mailing list