enable using gpu semaphores for ring sync

4 files changed, 30 insertions, 19 deletions

diff --git a/sys/dev/pci/drm/i915_drv.c b/sys/dev/pci/drm/i915_drv.c
index c720b3869a6..c0b13ffce2e 100644
--- a/sys/dev/pci/drm/i915_drv.c
+++ b/sys/dev/pci/drm/i915_drv.c
@@ -71,6 +71,9 @@ int i915_panel_ignore_lid = 1;
 /* Enable powersavings, fbc, downclocking, etc. (default: true) */
 unsigned int i915_powersave = 1;
 
+/* Use semaphores for inter-ring sync (default: -1 (use per-chip defaults)) */
+int i915_semaphores = -1;
+
 /*
  * Enable frame buffer compression for power savings
  * (default: -1 (use per-chip default))
diff --git a/sys/dev/pci/drm/i915_drv.h b/sys/dev/pci/drm/i915_drv.h
index 40665eab1a6..60efe11849d 100644
--- a/sys/dev/pci/drm/i915_drv.h
+++ b/sys/dev/pci/drm/i915_drv.h
@@ -1501,6 +1501,7 @@ read64(struct inteldrm_softc *dev_priv, bus_size_t off)
 extern unsigned int i915_lvds_downclock;
 extern int i915_panel_ignore_lid;
 extern unsigned int i915_powersave;
+extern int i915_semaphores;
 extern int i915_enable_rc6;
 extern int i915_enable_fbc;
 
diff --git a/sys/dev/pci/drm/i915_gem.c b/sys/dev/pci/drm/i915_gem.c
index ef0db7eb69f..3f78da803c6 100644
--- a/sys/dev/pci/drm/i915_gem.c
+++ b/sys/dev/pci/drm/i915_gem.c
@@ -1195,12 +1195,17 @@ void
 i915_gem_retire_requests_ring(struct intel_ring_buffer *ring)
 {
 	uint32_t seqno;
+	int i;
 
 	if (list_empty(&ring->request_list))
 		return;
 
 	seqno = ring->get_seqno(ring, true);
 
+	for (i = 0; i < ARRAY_SIZE(ring->sync_seqno); i++)
+		if (seqno >= ring->sync_seqno[i])
+			ring->sync_seqno[i] = 0;
+
 	while (!list_empty(&ring->request_list)) {
 		struct drm_i915_gem_request *request;
 
diff --git a/sys/dev/pci/drm/i915_gem_execbuffer.c b/sys/dev/pci/drm/i915_gem_execbuffer.c
index 6cc802de500..823aed23018 100644
--- a/sys/dev/pci/drm/i915_gem_execbuffer.c
+++ b/sys/dev/pci/drm/i915_gem_execbuffer.c
@@ -297,8 +297,6 @@ i915_gem_execbuffer_flush(struct drm_device *dev,
 bool
 intel_enable_semaphores(struct drm_device *dev)
 {
-	return 0;
-#ifdef notyet
 	if (INTEL_INFO(dev)->gen < 6)
 		return 0;
 
@@ -310,7 +308,6 @@ intel_enable_semaphores(struct drm_device *dev)
 		return 0;
 
 	return 1;
-#endif
 }
 
 int
@@ -318,42 +315,31 @@ i915_gem_execbuffer_sync_rings(struct drm_i915_gem_object *obj,
 			       struct intel_ring_buffer *to)
 {
 	struct intel_ring_buffer *from = obj->ring;
-//	u32 seqno;
-//	int ret, idx;
+	u32 seqno;
+	int ret, idx;
 
 	if (from == NULL || to == from)
 		return 0;
 
 	/* XXX gpu semaphores are implicated in various hard hangs on SNB */
-//	if (!intel_enable_semaphores(obj->base.dev))
+	if (!intel_enable_semaphores(obj->base.dev))
 		return i915_gem_object_wait_rendering(obj, false);
-#ifdef notyet
 	idx = intel_ring_sync_index(from, to);
 
-	seqno = obj->last_rendering_seqno;
+	seqno = obj->last_read_seqno;
 	if (seqno <= from->sync_seqno[idx])
 		return 0;
 
 	if (seqno == from->outstanding_lazy_request) {
-		struct drm_i915_gem_request *request;
-
-		request = kzalloc(sizeof(*request), GFP_KERNEL);
-		if (request == NULL)
-			return -ENOMEM;
-
-		ret = i915_add_request(from, NULL, request);
+		ret = i915_add_request(from, NULL, &seqno);
 		if (ret) {
-			kfree(request);
 			return ret;
 		}
-
-		seqno = request->seqno;
 	}
 
 	from->sync_seqno[idx] = seqno;
 
 	return to->sync_to(to, from, seqno - 1);
-#endif
 }
 
 int
@@ -532,6 +518,7 @@ i915_gem_execbuffer2(struct drm_device *dev, void *data,
 	int					 ret, ret2, i;
 	int					 pinned = 0, pin_tries;
 	uint32_t				 reloc_index;
+	uint32_t				 seqno;
 
 	/*
 	 * Check for valid execbuffer offset. We can do this early because
@@ -691,6 +678,21 @@ i915_gem_execbuffer2(struct drm_device *dev, void *data,
 	if (ret)
 		goto err;
 
+	seqno = i915_gem_next_request_seqno(ring);
+	for (i = 0; i < ARRAY_SIZE(ring->sync_seqno); i++) {
+		if (seqno < ring->sync_seqno[i]) {
+			/* The GPU can not handle its semaphore value wrapping,
+			 * so every billion or so execbuffers, we need to stall
+			 * the GPU in order to reset the counters.
+			 */
+			ret = i915_gpu_idle(dev);
+			if (ret)
+				goto err;
+
+			BUG_ON(ring->sync_seqno[i]);
+                }
+	}
+
 	if (args->flags & I915_EXEC_GEN7_SOL_RESET) {
 		ret = i915_reset_gen7_sol_offsets(dev, ring);
 		if (ret)