[Intel-gfx] [PATCH 6/7] drm/i915/gt: Only wait for GPU activity before unbinding a GGTT fence

[Chris Wilson]

Only GPU activity via the GGTT fence is asynchronous, we know that we
control the CPU access directly, so we only need to wait for the GPU to
stop using the fence before we relinquish it.

Signed-off-by: Chris Wilson <chris at chris-wilson.co.uk>
---
 drivers/gpu/drm/i915/gt/intel_ggtt_fencing.c | 12 ++++++++----
 drivers/gpu/drm/i915/gt/intel_ggtt_fencing.h |  3 +++
 drivers/gpu/drm/i915/i915_vma.c              |  4 ++++
 3 files changed, 15 insertions(+), 4 deletions(-)

diff --git a/drivers/gpu/drm/i915/gt/intel_ggtt_fencing.c b/drivers/gpu/drm/i915/gt/intel_ggtt_fencing.c
index 51984cee18b3..f1ded8fbda3a 100644
--- a/drivers/gpu/drm/i915/gt/intel_ggtt_fencing.c
+++ b/drivers/gpu/drm/i915/gt/intel_ggtt_fencing.c
@@ -235,9 +235,12 @@ static int fence_update(struct i915_fence_reg *fence,
 		GEM_BUG_ON(!i915_gem_object_get_stride(vma->obj) ||
 			   !i915_gem_object_get_tiling(vma->obj));
 
-		ret = i915_vma_sync(vma);
-		if (ret)
-			return ret;
+		if (INTEL_GEN(fence_to_i915(fence)) < 4) {
+			/* implicit 'unfenced' GPU blits */
+			ret = i915_vma_sync(vma);
+			if (ret)
+				return ret;
+		}
 
 		fence->stride = i915_gem_object_get_stride(vma->obj);
 		fence->tiling = i915_gem_object_get_tiling(vma->obj);
@@ -247,7 +250,7 @@ static int fence_update(struct i915_fence_reg *fence,
 	old = xchg(&fence->vma, NULL);
 	if (old) {
 		/* XXX Ideally we would move the waiting to outside the mutex */
-		ret = i915_vma_sync(old);
+		ret = i915_active_wait(&fence->active);
 		if (ret) {
 			fence->vma = old;
 			return ret;
@@ -869,6 +872,7 @@ void intel_ggtt_init_fences(struct i915_ggtt *ggtt)
 	for (i = 0; i < num_fences; i++) {
 		struct i915_fence_reg *fence = &ggtt->fence_regs[i];
 
+		i915_active_init(&fence->active, NULL, NULL);
 		fence->ggtt = ggtt;
 		fence->id = i;
 		list_add_tail(&fence->link, &ggtt->fence_list);
diff --git a/drivers/gpu/drm/i915/gt/intel_ggtt_fencing.h b/drivers/gpu/drm/i915/gt/intel_ggtt_fencing.h
index ad3acc9b7f37..1731cd039bb9 100644
--- a/drivers/gpu/drm/i915/gt/intel_ggtt_fencing.h
+++ b/drivers/gpu/drm/i915/gt/intel_ggtt_fencing.h
@@ -28,6 +28,8 @@
 #include <linux/list.h>
 #include <linux/types.h>
 
+#include "i915_active.h"
+
 struct drm_i915_gem_object;
 struct i915_ggtt;
 struct i915_vma;
@@ -41,6 +43,7 @@ struct i915_fence_reg {
 	struct i915_ggtt *ggtt;
 	struct i915_vma *vma;
 	atomic_t pin_count;
+	struct i915_active active;
 	int id;
 	/**
 	 * Whether the tiling parameters for the currently
diff --git a/drivers/gpu/drm/i915/i915_vma.c b/drivers/gpu/drm/i915/i915_vma.c
index 5b3efb43a8ef..aedbd056fd45 100644
--- a/drivers/gpu/drm/i915/i915_vma.c
+++ b/drivers/gpu/drm/i915/i915_vma.c
@@ -1214,6 +1214,10 @@ int i915_vma_move_to_active(struct i915_vma *vma,
 		dma_resv_add_shared_fence(vma->resv, &rq->fence);
 		obj->write_domain = 0;
 	}
+
+	if (flags & EXEC_OBJECT_NEEDS_FENCE && vma->fence)
+		i915_active_add_request(&vma->fence->active, rq);
+
 	obj->read_domains |= I915_GEM_GPU_DOMAINS;
 	obj->mm.dirty = true;
 
-- 
2.20.1