diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
index 3596f3415624..6a98c0659324 100644
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -1568,6 +1568,21 @@ i915_add_request(struct intel_ring_buffer *ring,
 	int was_empty;
 	int ret;
 
+	/*
+	 * Emit any outstanding flushes - execbuf can fail to emit the flush
+	 * after having emitted the batchbuffer command. Hence we need to fix
+	 * things up similar to emitting the lazy request. The difference here
+	 * is that the flush _must_ happen before the next request, no matter
+	 * what.
+	 */
+	if (ring->gpu_caches_dirty) {
+		ret = i915_gem_flush_ring(ring, 0, I915_GEM_GPU_DOMAINS);
+		if (ret)
+			return ret;
+
+		ring->gpu_caches_dirty = false;
+	}
+
 	BUG_ON(request == NULL);
 	seqno = i915_gem_next_request_seqno(ring);
 
@@ -1613,6 +1628,9 @@ i915_add_request(struct intel_ring_buffer *ring,
 			queue_delayed_work(dev_priv->wq,
 					   &dev_priv->mm.retire_work, HZ);
 	}
+
+	WARN_ON(!list_empty(&ring->gpu_write_list));
+
 	return 0;
 }
 
@@ -1827,14 +1845,11 @@ i915_gem_retire_work_handler(struct work_struct *work)
 	 */
 	idle = true;
 	for_each_ring(ring, dev_priv, i) {
-		if (!list_empty(&ring->gpu_write_list)) {
+		if (ring->gpu_caches_dirty) {
 			struct drm_i915_gem_request *request;
-			int ret;
 
-			ret = i915_gem_flush_ring(ring,
-						  0, I915_GEM_GPU_DOMAINS);
 			request = kzalloc(sizeof(*request), GFP_KERNEL);
-			if (ret || request == NULL ||
+			if (request == NULL ||
 			    i915_add_request(ring, NULL, request))
 			    kfree(request);
 		}
diff --git a/drivers/gpu/drm/i915/i915_gem_execbuffer.c b/drivers/gpu/drm/i915/i915_gem_execbuffer.c
index f32d02464bce..88e2e114189c 100644
--- a/drivers/gpu/drm/i915/i915_gem_execbuffer.c
+++ b/drivers/gpu/drm/i915/i915_gem_execbuffer.c
@@ -810,33 +810,16 @@ i915_gem_execbuffer_relocate_slow(struct drm_device *dev,
 	return ret;
 }
 
-static int
+static void
 i915_gem_execbuffer_flush(struct drm_device *dev,
 			  uint32_t invalidate_domains,
-			  uint32_t flush_domains,
-			  uint32_t flush_rings)
+			  uint32_t flush_domains)
 {
-	drm_i915_private_t *dev_priv = dev->dev_private;
-	int i, ret;
-
 	if (flush_domains & I915_GEM_DOMAIN_CPU)
 		intel_gtt_chipset_flush();
 
 	if (flush_domains & I915_GEM_DOMAIN_GTT)
 		wmb();
-
-	if ((flush_domains | invalidate_domains) & I915_GEM_GPU_DOMAINS) {
-		for (i = 0; i < I915_NUM_RINGS; i++)
-			if (flush_rings & (1 << i)) {
-				ret = i915_gem_flush_ring(&dev_priv->ring[i],
-							  invalidate_domains,
-							  flush_domains);
-				if (ret)
-					return ret;
-			}
-	}
-
-	return 0;
 }
 
 static int
@@ -885,12 +868,9 @@ i915_gem_execbuffer_move_to_gpu(struct intel_ring_buffer *ring,
 		i915_gem_object_set_to_gpu_domain(obj, ring, &cd);
 
 	if (cd.invalidate_domains | cd.flush_domains) {
-		ret = i915_gem_execbuffer_flush(ring->dev,
-						cd.invalidate_domains,
-						cd.flush_domains,
-						cd.flush_rings);
-		if (ret)
-			return ret;
+		i915_gem_execbuffer_flush(ring->dev,
+					  cd.invalidate_domains,
+					  cd.flush_domains);
 	}
 
 	if (cd.flips) {
@@ -905,6 +885,11 @@ i915_gem_execbuffer_move_to_gpu(struct intel_ring_buffer *ring,
 			return ret;
 	}
 
+	/* Unconditionally invalidate gpu caches. */
+	ret = i915_gem_flush_ring(ring, I915_GEM_GPU_DOMAINS, 0);
+	if (ret)
+		return ret;
+
 	return 0;
 }
 
@@ -983,26 +968,13 @@ i915_gem_execbuffer_retire_commands(struct drm_device *dev,
 				    struct intel_ring_buffer *ring)
 {
 	struct drm_i915_gem_request *request;
-	u32 invalidate;
 
-	/*
-	 * Ensure that the commands in the batch buffer are
-	 * finished before the interrupt fires.
-	 *
-	 * The sampler always gets flushed on i965 (sigh).
-	 */
-	invalidate = I915_GEM_DOMAIN_COMMAND;
-	if (INTEL_INFO(dev)->gen >= 4)
-		invalidate |= I915_GEM_DOMAIN_SAMPLER;
-	if (ring->flush(ring, invalidate, 0)) {
-		i915_gem_next_request_seqno(ring);
-		return;
-	}
+	/* Unconditionally force add_request to emit a full flush. */
+	ring->gpu_caches_dirty = true;
 
 	/* Add a breadcrumb for the completion of the batch buffer */
 	request = kzalloc(sizeof(*request), GFP_KERNEL);
 	if (request == NULL || i915_add_request(ring, file, request)) {
-		i915_gem_next_request_seqno(ring);
 		kfree(request);
 	}
 }
diff --git a/drivers/gpu/drm/i915/intel_ringbuffer.h b/drivers/gpu/drm/i915/intel_ringbuffer.h
index 594c9c4ad396..1d3c81fdad92 100644
--- a/drivers/gpu/drm/i915/intel_ringbuffer.h
+++ b/drivers/gpu/drm/i915/intel_ringbuffer.h
@@ -113,6 +113,7 @@ struct  intel_ring_buffer {
 	 * Do we have some not yet emitted requests outstanding?
 	 */
 	u32 outstanding_lazy_request;
+	bool gpu_caches_dirty;
 
 	wait_queue_head_t irq_queue;