drm/i915: implement fastpath for overlay flip waiting

As long as the gpu can keep up, neither the cpu (waiting for gpu)
nore the gpu (waiting for vblank to do an overlay flip) stalls.
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Eric Anholt <eric@anholt.net>

drivers/gpu/drm/i915/i915_drv.h

@@ -808,6 +808,9 @@ void i915_gem_cleanup_ringbuffer(struct drm_device *dev);
 int i915_gem_do_init(struct drm_device *dev, unsigned long start,
 		     unsigned long end);
 int i915_gem_idle(struct drm_device *dev);
+uint32_t i915_add_request(struct drm_device *dev, struct drm_file *file_priv,
+			  uint32_t flush_domains);
+int i915_do_wait_request(struct drm_device *dev, uint32_t seqno, int interruptible);
 int i915_lp_ring_sync(struct drm_device *dev);
 int i915_gem_fault(struct vm_area_struct *vma, struct vm_fault *vmf);
 int i915_gem_object_set_to_gtt_domain(struct drm_gem_object *obj,

drivers/gpu/drm/i915/i915_gem.c

@@ -1583,7 +1583,7 @@ i915_gem_object_move_to_inactive(struct drm_gem_object *obj)
  *
  * Returned sequence numbers are nonzero on success.
  */
-static uint32_t
+uint32_t
 i915_add_request(struct drm_device *dev, struct drm_file *file_priv,
 		 uint32_t flush_domains)
 {
@@ -1820,7 +1820,7 @@ i915_gem_retire_work_handler(struct work_struct *work)
 	mutex_unlock(&dev->struct_mutex);
 }
 
-static int
+int
 i915_do_wait_request(struct drm_device *dev, uint32_t seqno, int interruptible)
 {
 	drm_i915_private_t *dev_priv = dev->dev_private;

drivers/gpu/drm/i915/intel_drv.h

@@ -126,7 +126,9 @@ struct intel_overlay {
 	u32 flip_addr;
 	struct drm_i915_gem_object *reg_bo;
 	void *virt_addr;
+	/* flip handling */
 	int hw_wedged;
+	uint32_t last_flip_req;
 };
 
 struct intel_crtc {

drivers/gpu/drm/i915/intel_overlay.c

@@ -251,7 +251,6 @@ static void intel_overlay_continue(struct intel_overlay *overlay,
         drm_i915_private_t *dev_priv = dev->dev_private;
 	u32 flip_addr = overlay->flip_addr;
 	u32 tmp;
-	int ret;
 	RING_LOCALS;
 
 	BUG_ON(!overlay->active);
@@ -264,11 +263,40 @@ static void intel_overlay_continue(struct intel_overlay *overlay,
 	if (tmp & (1 << 17))
 		DRM_DEBUG("overlay underrun, DOVSTA: %x\n", tmp);
 
-	BEGIN_LP_RING(6);
+	BEGIN_LP_RING(4);
 	OUT_RING(MI_FLUSH);
 	OUT_RING(MI_NOOP);
 	OUT_RING(MI_OVERLAY_FLIP | MI_OVERLAY_CONTINUE);
 	OUT_RING(flip_addr);
+        ADVANCE_LP_RING();
+
+	overlay->last_flip_req = i915_add_request(dev, NULL, 0);
+}
+
+static int intel_overlay_wait_flip(struct intel_overlay *overlay)
+{
+	struct drm_device *dev = overlay->dev;
+        drm_i915_private_t *dev_priv = dev->dev_private;
+	int ret;
+	u32 tmp;
+	RING_LOCALS;
+
+	if (overlay->last_flip_req != 0) {
+		ret = i915_do_wait_request(dev, overlay->last_flip_req, 0);
+
+		if (ret != 0)
+			return ret;
+
+		overlay->last_flip_req = 0;
+
+		tmp = I915_READ(ISR);
+
+		if (!(tmp & I915_OVERLAY_PLANE_FLIP_PENDING_INTERRUPT))
+			return 0;
+	}
+
+	/* synchronous slowpath */
+	BEGIN_LP_RING(2);
         OUT_RING(MI_WAIT_FOR_EVENT | MI_WAIT_FOR_OVERLAY_FLIP);
         OUT_RING(MI_NOOP);
         ADVANCE_LP_RING();
@@ -279,13 +307,8 @@ static void intel_overlay_continue(struct intel_overlay *overlay,
 		DRM_ERROR("intel overlay: ring sync failed, hw likely wedged\n");
 		overlay->hw_wedged = 1;
 	}
-}
 
-static int intel_overlay_wait_flip(struct intel_overlay *overlay)
-{
-	/* don't overcomplicate things for now with asynchronous operations
-	 * see comment above */
-	return 0;
+	return ret;
 }
 
 /* overlay needs to be disabled in OCMD reg */
@@ -344,7 +367,9 @@ static int intel_overlay_off(struct intel_overlay *overlay)
 	return ret;
 }
 
-/* wait for pending overlay flip and release old frame */
+/* Wait for pending overlay flip and release old frame.
+ * Needs to be called before the overlay register are changed
+ * via intel_overlay_(un)map_regs_atomic */
 static int intel_overlay_release_old_vid(struct intel_overlay *overlay)
 {
 	int ret;