[PATCH 12/22] drm/i915/gem: Re-evaluate execbuf throttling after wait

[Chris Wilson]

Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
---
 .../gpu/drm/i915/gem/i915_gem_execbuffer.c    | 36 +++++++++++--------
 drivers/gpu/drm/i915/i915_request.c           |  2 ++
 drivers/gpu/drm/i915/i915_request.h           | 12 +++++++
 3 files changed, 35 insertions(+), 15 deletions(-)

diff --git a/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c b/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c
index 2da2b0589856..36923fc67bca 100644
--- a/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c
+++ b/drivers/gpu/drm/i915/gem/i915_gem_execbuffer.c
@@ -2247,7 +2247,6 @@ static struct i915_request *eb_throttle(struct intel_context *ce)
 static int __eb_pin_engine(struct i915_execbuffer *eb, struct intel_context *ce)
 {
 	struct intel_timeline *tl;
-	struct i915_request *rq;
 	int err;
 
 	/*
@@ -2270,6 +2269,12 @@ static int __eb_pin_engine(struct i915_execbuffer *eb, struct intel_context *ce)
 	if (err)
 		return err;
 
+	tl = intel_context_timeline_lock(ce);
+	if (IS_ERR(tl)) {
+		err = PTR_ERR(tl);
+		goto err_unpin;
+	}
+
 	/*
 	 * Take a local wakeref for preparing to dispatch the execbuf as
 	 * we expect to access the hardware fairly frequently in the
@@ -2278,21 +2283,18 @@ static int __eb_pin_engine(struct i915_execbuffer *eb, struct intel_context *ce)
 	 * until the timeline is idle, which in turn releases the wakeref
 	 * taken on the engine, and the parent device.
 	 */
-	tl = intel_context_timeline_lock(ce);
-	if (IS_ERR(tl)) {
-		err = PTR_ERR(tl);
-		goto err_unpin;
-	}
-
 	intel_context_enter(ce);
-	rq = eb_throttle(ce);
 
-	intel_context_timeline_unlock(tl);
-
-	if (rq) {
-		bool nonblock = eb->file->filp->f_flags & O_NONBLOCK;
+	do {
+		const bool nonblock = eb->file->filp->f_flags & O_NONBLOCK;
+		struct i915_request *rq;
 		long timeout;
 
+		rq = eb_throttle(ce);
+		intel_context_timeline_unlock(tl);
+		if (!rq)
+			break;
+
 		timeout = MAX_SCHEDULE_TIMEOUT;
 		if (nonblock)
 			timeout = 0;
@@ -2300,13 +2302,17 @@ static int __eb_pin_engine(struct i915_execbuffer *eb, struct intel_context *ce)
 		timeout = i915_request_wait(rq,
 					    I915_WAIT_INTERRUPTIBLE,
 					    timeout);
-		i915_request_put(rq);
-
 		if (timeout < 0) {
+			i915_request_put(rq);
 			err = nonblock ? -EWOULDBLOCK : timeout;
 			goto err_exit;
 		}
-	}
+
+		mutex_lock(&tl->mutex);
+		if (!i915_request_is_dead(rq))
+			i915_request_retire_upto(rq);
+		i915_request_put(rq);
+	} while (1);
 
 	eb->engine = ce->engine;
 	eb->context = ce;
diff --git a/drivers/gpu/drm/i915/i915_request.c b/drivers/gpu/drm/i915/i915_request.c
index 14a5a99284fa..7376767d79af 100644
--- a/drivers/gpu/drm/i915/i915_request.c
+++ b/drivers/gpu/drm/i915/i915_request.c
@@ -269,6 +269,8 @@ bool i915_request_retire(struct i915_request *rq)
 		set_bit(I915_FENCE_FLAG_ACTIVE, &rq->fence.flags);
 		__notify_execute_cb(rq);
 	}
+	GEM_BUG_ON(i915_request_is_dead(rq));
+	set_bit(I915_FENCE_FLAG_RETIRED, &rq->fence.flags);
 	GEM_BUG_ON(!list_empty(&rq->execute_cb));
 	spin_unlock_irq(&rq->lock);
 
diff --git a/drivers/gpu/drm/i915/i915_request.h b/drivers/gpu/drm/i915/i915_request.h
index 0e4fe3205ce7..9e90c2dc9262 100644
--- a/drivers/gpu/drm/i915/i915_request.h
+++ b/drivers/gpu/drm/i915/i915_request.h
@@ -76,6 +76,13 @@ enum {
 	 * a request is on the various signal_list.
 	 */
 	I915_FENCE_FLAG_SIGNAL,
+
+	/*
+	 * I915_FENCE_FLAG_RETIRED - this request has been retired
+	 *
+	 * Mark the request as having been retired; don't retire again!
+	 */
+	I915_FENCE_FLAG_RETIRED,
 };
 
 /**
@@ -333,6 +340,11 @@ static inline bool i915_request_is_active(const struct i915_request *rq)
 	return test_bit(I915_FENCE_FLAG_ACTIVE, &rq->fence.flags);
 }
 
+static inline bool i915_request_is_dead(const struct i915_request *rq)
+{
+	return test_bit(I915_FENCE_FLAG_RETIRED, &rq->fence.flags);
+}
+
 /**
  * Returns true if seq1 is later than seq2.
  */
-- 
2.24.1