[Intel-gfx] [PATCH 04/15] drm/i915: drop lock support for i915_wait_request

[Ben Widawsky]

Provide a way for callers to instruct the wait request to drop
struct_mutex before the actually wait. This gives an opportunity for GPU
clients to submit independent work.

While it's tempting to make i915_wait_request always drop the lock and
enforce the caller to check their specific criteria for a sane
environment in turns out to be quite difficult to implement in most
cases as almost any state can change out from underneath one the lock is
dropped, and things such as domains aren't easy to figure out whether
they've changed (they could have been modified, and then changed back to
the original flags).

Cc: Chris Wilson <chris at chris-wilson.co.uk>
Cc: Daniel Vetter <daniel.vetter at ffwll.ch>
Signed-off-by: Ben Widawsky <ben at bwidawsk.net>
---
 drivers/gpu/drm/i915/i915_drv.h      |    3 +-
 drivers/gpu/drm/i915/i915_gem.c      |   42 +++++++++++++++++++++++++---------
 drivers/gpu/drm/i915/intel_overlay.c |    4 +-
 3 files changed, 35 insertions(+), 14 deletions(-)

diff --git a/drivers/gpu/drm/i915/i915_drv.h b/drivers/gpu/drm/i915/i915_drv.h
index d2da91f..9d9d160 100644
--- a/drivers/gpu/drm/i915/i915_drv.h
+++ b/drivers/gpu/drm/i915/i915_drv.h
@@ -1187,7 +1187,8 @@ int __must_check i915_add_request(struct intel_ring_buffer *ring,
 				  struct drm_file *file,
 				  struct drm_i915_gem_request *request);
 int __must_check i915_wait_request(struct intel_ring_buffer *ring,
-				   uint32_t seqno);
+				   uint32_t seqno,
+				   bool drop_mutex);
 int i915_gem_fault(struct vm_area_struct *vma, struct vm_fault *vmf);
 int __must_check
 i915_gem_object_set_to_gtt_domain(struct drm_i915_gem_object *obj,
diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
index ed0b68f..9c743ae 100644
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -1943,13 +1943,15 @@ i915_gem_retire_work_handler(struct work_struct *work)
  */
 int
 i915_wait_request(struct intel_ring_buffer *ring,
-		  uint32_t seqno)
+		  uint32_t seqno,
+		  bool drop_mutex)
 {
 	drm_i915_private_t *dev_priv = ring->dev->dev_private;
 	u32 ier;
 	int ret = 0;
 
 	BUG_ON(seqno == 0);
+	BUG_ON(!mutex_is_locked(&ring->dev->struct_mutex));
 
 	if (atomic_read(&dev_priv->mm.wedged)) {
 		struct completion *x = &dev_priv->error_completion;
@@ -1994,8 +1996,12 @@ i915_wait_request(struct intel_ring_buffer *ring,
 
 		trace_i915_gem_request_wait_begin(ring, seqno);
 
-		ring->waiting_seqno = seqno;
+		if (!drop_mutex)
+			ring->waiting_seqno = seqno;
 		if (ring->irq_get(ring)) {
+			if (drop_mutex)
+				mutex_unlock(&ring->dev->struct_mutex);
+
 			if (dev_priv->mm.interruptible)
 				ret = wait_event_interruptible(ring->irq_queue,
 							       i915_seqno_passed(ring->get_seqno(ring), seqno)
@@ -2005,12 +2011,23 @@ i915_wait_request(struct intel_ring_buffer *ring,
 					   i915_seqno_passed(ring->get_seqno(ring), seqno)
 					   || atomic_read(&dev_priv->mm.wedged));
 
+			if (drop_mutex)
+				mutex_lock(&ring->dev->struct_mutex);
+
 			ring->irq_put(ring);
-		} else if (wait_for(i915_seqno_passed(ring->get_seqno(ring),
-						      seqno) ||
-				    atomic_read(&dev_priv->mm.wedged), 3000))
-			ret = -EBUSY;
-		ring->waiting_seqno = 0;
+		} else {
+			if (drop_mutex)
+				mutex_unlock(&ring->dev->struct_mutex);
+			if (wait_for(i915_seqno_passed(ring->get_seqno(ring),
+						       seqno) ||
+				     atomic_read(&dev_priv->mm.wedged), 3000))
+				ret = -EBUSY;
+			if (drop_mutex)
+				mutex_lock(&ring->dev->struct_mutex);
+		}
+
+		if (!drop_mutex)
+			ring->waiting_seqno = 0;
 
 		trace_i915_gem_request_wait_end(ring, seqno);
 	}
@@ -2051,7 +2068,8 @@ i915_gem_object_wait_rendering(struct drm_i915_gem_object *obj)
 	 * it.
 	 */
 	if (obj->active) {
-		ret = i915_wait_request(obj->ring, obj->last_rendering_seqno);
+		ret = i915_wait_request(obj->ring, obj->last_rendering_seqno,
+					false);
 		if (ret)
 			return ret;
 	}
@@ -2186,7 +2204,7 @@ static int i915_ring_idle(struct intel_ring_buffer *ring)
 			return ret;
 	}
 
-	return i915_wait_request(ring, i915_gem_next_request_seqno(ring));
+	return i915_wait_request(ring, i915_gem_next_request_seqno(ring), false);
 }
 
 int
@@ -2400,7 +2418,8 @@ i915_gem_object_flush_fence(struct drm_i915_gem_object *obj,
 		if (!ring_passed_seqno(obj->last_fenced_ring,
 				       obj->last_fenced_seqno)) {
 			ret = i915_wait_request(obj->last_fenced_ring,
-						obj->last_fenced_seqno);
+						obj->last_fenced_seqno,
+						false);
 			if (ret)
 				return ret;
 		}
@@ -2541,7 +2560,8 @@ i915_gem_object_get_fence(struct drm_i915_gem_object *obj,
 				if (!ring_passed_seqno(obj->last_fenced_ring,
 						       reg->setup_seqno)) {
 					ret = i915_wait_request(obj->last_fenced_ring,
-								reg->setup_seqno);
+								reg->setup_seqno,
+								false);
 					if (ret)
 						return ret;
 				}
diff --git a/drivers/gpu/drm/i915/intel_overlay.c b/drivers/gpu/drm/i915/intel_overlay.c
index cdf17d4..8f27f2b 100644
--- a/drivers/gpu/drm/i915/intel_overlay.c
+++ b/drivers/gpu/drm/i915/intel_overlay.c
@@ -227,7 +227,7 @@ static int intel_overlay_do_wait_request(struct intel_overlay *overlay,
 	}
 	overlay->last_flip_req = request->seqno;
 	overlay->flip_tail = tail;
-	ret = i915_wait_request(LP_RING(dev_priv), overlay->last_flip_req);
+	ret = i915_wait_request(LP_RING(dev_priv), overlay->last_flip_req, false);
 	if (ret)
 		return ret;
 
@@ -448,7 +448,7 @@ static int intel_overlay_recover_from_interrupt(struct intel_overlay *overlay)
 	if (overlay->last_flip_req == 0)
 		return 0;
 
-	ret = i915_wait_request(LP_RING(dev_priv), overlay->last_flip_req);
+	ret = i915_wait_request(LP_RING(dev_priv), overlay->last_flip_req, false);
 	if (ret)
 		return ret;
 
-- 
1.7.7.3