drm/i915: Add flag to i915_add_request() to skip the cache flush
authorJohn Harrison <John.C.Harrison@Intel.com>
Fri, 29 May 2015 16:43:34 +0000 (17:43 +0100)
committerDaniel Vetter <daniel.vetter@ffwll.ch>
Tue, 23 Jun 2015 12:02:04 +0000 (14:02 +0200)
In order to explcitly track all GPU work (and completely remove the outstanding
lazy request), it is necessary to add extra i915_add_request() calls to various
places. Some of these do not need the implicit cache flush done as part of the
standard batch buffer submission process.

This patch adds a flag to _add_request() to specify whether the flush is
required or not.

For: VIZ-5115
Signed-off-by: John Harrison <John.C.Harrison@Intel.com>
Reviewed-by: Tomas Elf <tomas.elf@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
drivers/gpu/drm/i915/i915_drv.h
drivers/gpu/drm/i915/i915_gem.c
drivers/gpu/drm/i915/i915_gem_execbuffer.c
drivers/gpu/drm/i915/i915_gem_render_state.c
drivers/gpu/drm/i915/intel_lrc.c

index 14154c46076239f5a9b9a8eda61b44508ab4d9c8..104893bea2f1f95e4e6368878d310b3f52436131 100644 (file)
@@ -2890,9 +2890,12 @@ int __must_check i915_gpu_idle(struct drm_device *dev);
 int __must_check i915_gem_suspend(struct drm_device *dev);
 void __i915_add_request(struct intel_engine_cs *ring,
                        struct drm_file *file,
-                       struct drm_i915_gem_object *batch_obj);
+                       struct drm_i915_gem_object *batch_obj,
+                       bool flush_caches);
 #define i915_add_request(ring) \
-       __i915_add_request(ring, NULL, NULL)
+       __i915_add_request(ring, NULL, NULL, true)
+#define i915_add_request_no_flush(ring) \
+       __i915_add_request(ring, NULL, NULL, false)
 int __i915_wait_request(struct drm_i915_gem_request *req,
                        unsigned reset_counter,
                        bool interruptible,
index a0f51478581f0537e1289fb1c85bb57d8403a5ba..74c319350876ff57be0994b9c51a55fc6d28d903 100644 (file)
@@ -2470,7 +2470,8 @@ i915_gem_get_seqno(struct drm_device *dev, u32 *seqno)
  */
 void __i915_add_request(struct intel_engine_cs *ring,
                        struct drm_file *file,
-                       struct drm_i915_gem_object *obj)
+                       struct drm_i915_gem_object *obj,
+                       bool flush_caches)
 {
        struct drm_i915_private *dev_priv = ring->dev->dev_private;
        struct drm_i915_gem_request *request;
@@ -2502,12 +2503,14 @@ void __i915_add_request(struct intel_engine_cs *ring,
         * is that the flush _must_ happen before the next request, no matter
         * what.
         */
-       if (i915.enable_execlists)
-               ret = logical_ring_flush_all_caches(ringbuf, request->ctx);
-       else
-               ret = intel_ring_flush_all_caches(ring);
-       /* Not allowed to fail! */
-       WARN(ret, "*_ring_flush_all_caches failed: %d!\n", ret);
+       if (flush_caches) {
+               if (i915.enable_execlists)
+                       ret = logical_ring_flush_all_caches(ringbuf, request->ctx);
+               else
+                       ret = intel_ring_flush_all_caches(ring);
+               /* Not allowed to fail! */
+               WARN(ret, "*_ring_flush_all_caches failed: %d!\n", ret);
+       }
 
        /* Record the position of the start of the request so that
         * should we detect the updated seqno part-way through the
index 76bfc68d1a881a6d621ed9900ad94b4adaf43a08..a15517249bb9926376840d85cc5178661778f010 100644 (file)
@@ -1066,7 +1066,7 @@ i915_gem_execbuffer_retire_commands(struct i915_execbuffer_params *params)
        params->ring->gpu_caches_dirty = true;
 
        /* Add a breadcrumb for the completion of the batch buffer */
-       __i915_add_request(params->ring, params->file, params->batch_obj);
+       __i915_add_request(params->ring, params->file, params->batch_obj, true);
 }
 
 static int
index ce4788ff3df56aa9a41416a867a5f73b8d038d8c..4418616301e755efa54a59e03ea06c2d10a59f01 100644 (file)
@@ -173,7 +173,7 @@ int i915_gem_render_state_init(struct intel_engine_cs *ring)
 
        i915_vma_move_to_active(i915_gem_obj_to_ggtt(so.obj), ring);
 
-       __i915_add_request(ring, NULL, so.obj);
+       __i915_add_request(ring, NULL, so.obj, true);
        /* __i915_add_request moves object to inactive if it fails */
 out:
        i915_gem_render_state_fini(&so);
index 754aa39eb12b6517312d75451581eb37bd1eab47..47443b9aa17216f9ab824224c1ebb1677b4e5617 100644 (file)
@@ -1599,7 +1599,7 @@ static int intel_lr_context_render_state_init(struct intel_engine_cs *ring,
 
        i915_vma_move_to_active(i915_gem_obj_to_ggtt(so.obj), ring);
 
-       __i915_add_request(ring, file, so.obj);
+       __i915_add_request(ring, file, so.obj, true);
        /* intel_logical_ring_add_request moves object to inactive if it
         * fails */
 out: