drm/radeon: rework ring syncing code

Move inter ring syncing with semaphores into the
existing ring allocations, with that we need to
lock the ring mutex only once.

Signed-off-by: Christian König <deathsimple@vodafone.de>
Reviewed-by: Jerome Glisse <jglisse@redhat.com>

drivers/gpu/drm/radeon/evergreen_blit_kms.c

@@ -622,7 +622,8 @@ int evergreen_blit_init(struct radeon_device *rdev)
 	rdev->r600_blit.primitives.draw_auto = draw_auto;
 	rdev->r600_blit.primitives.set_default_state = set_default_state;
 
-	rdev->r600_blit.ring_size_common = 55; /* shaders + def state */
+	rdev->r600_blit.ring_size_common = 8; /* sync semaphore */
+	rdev->r600_blit.ring_size_common += 55; /* shaders + def state */
 	rdev->r600_blit.ring_size_common += 16; /* fence emit for VB IB */
 	rdev->r600_blit.ring_size_common += 5; /* done copy */
 	rdev->r600_blit.ring_size_common += 16; /* fence emit for done copy */

drivers/gpu/drm/radeon/r600.c

@@ -2311,15 +2311,16 @@ int r600_copy_blit(struct radeon_device *rdev,
 		   unsigned num_gpu_pages,
 		   struct radeon_fence **fence)
 {
+	struct radeon_semaphore *sem = NULL;
 	struct radeon_sa_bo *vb = NULL;
 	int r;
 
-	r = r600_blit_prepare_copy(rdev, num_gpu_pages, &vb);
+	r = r600_blit_prepare_copy(rdev, num_gpu_pages, fence, &vb, &sem);
 	if (r) {
 		return r;
 	}
 	r600_kms_blit_copy(rdev, src_offset, dst_offset, num_gpu_pages, vb);
-	r600_blit_done_copy(rdev, fence, vb);
+	r600_blit_done_copy(rdev, fence, vb, sem);
 	return 0;
 }
 

drivers/gpu/drm/radeon/r600_blit_kms.c

@@ -512,7 +512,8 @@ int r600_blit_init(struct radeon_device *rdev)
 	rdev->r600_blit.primitives.draw_auto = draw_auto;
 	rdev->r600_blit.primitives.set_default_state = set_default_state;
 
-	rdev->r600_blit.ring_size_common = 40; /* shaders + def state */
+	rdev->r600_blit.ring_size_common = 8; /* sync semaphore */
+	rdev->r600_blit.ring_size_common += 40; /* shaders + def state */
 	rdev->r600_blit.ring_size_common += 5; /* done copy */
 	rdev->r600_blit.ring_size_common += 16; /* fence emit for done copy */
 
@@ -666,7 +667,8 @@ static unsigned r600_blit_create_rect(unsigned num_gpu_pages,
 
 
 int r600_blit_prepare_copy(struct radeon_device *rdev, unsigned num_gpu_pages,
-			   struct radeon_sa_bo **vb)
+			   struct radeon_fence **fence, struct radeon_sa_bo **vb,
+			   struct radeon_semaphore **sem)
 {
 	struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
 	int r;
@@ -689,22 +691,37 @@ int r600_blit_prepare_copy(struct radeon_device *rdev, unsigned num_gpu_pages,
 		return r;
 	}
 
+	r = radeon_semaphore_create(rdev, sem);
+	if (r) {
+		radeon_sa_bo_free(rdev, vb, NULL);
+		return r;
+	}
+
 	/* calculate number of loops correctly */
 	ring_size = num_loops * dwords_per_loop;
 	ring_size += rdev->r600_blit.ring_size_common;
 	r = radeon_ring_lock(rdev, ring, ring_size);
 	if (r) {
 		radeon_sa_bo_free(rdev, vb, NULL);
+		radeon_semaphore_free(rdev, sem, NULL);
 		return r;
 	}
 
+	if (radeon_fence_need_sync(*fence, RADEON_RING_TYPE_GFX_INDEX)) {
+		radeon_semaphore_sync_rings(rdev, *sem, (*fence)->ring,
+					    RADEON_RING_TYPE_GFX_INDEX);
+		radeon_fence_note_sync(*fence, RADEON_RING_TYPE_GFX_INDEX);
+	} else {
+		radeon_semaphore_free(rdev, sem, NULL);
+	}
+
 	rdev->r600_blit.primitives.set_default_state(rdev);
 	rdev->r600_blit.primitives.set_shaders(rdev);
 	return 0;
 }
 
 void r600_blit_done_copy(struct radeon_device *rdev, struct radeon_fence **fence,
-			 struct radeon_sa_bo *vb)
+			 struct radeon_sa_bo *vb, struct radeon_semaphore *sem)
 {
 	struct radeon_ring *ring = &rdev->ring[RADEON_RING_TYPE_GFX_INDEX];
 	int r;
@@ -717,6 +734,7 @@ void r600_blit_done_copy(struct radeon_device *rdev, struct radeon_fence **fence
 
 	radeon_ring_unlock_commit(rdev, ring);
 	radeon_sa_bo_free(rdev, &vb, *fence);
+	radeon_semaphore_free(rdev, &sem, *fence);
 }
 
 void r600_kms_blit_copy(struct radeon_device *rdev,

drivers/gpu/drm/radeon/radeon.h

@@ -470,10 +470,9 @@ void radeon_semaphore_emit_wait(struct radeon_device *rdev, int ring,
 				struct radeon_semaphore *semaphore);
 int radeon_semaphore_sync_rings(struct radeon_device *rdev,
 				struct radeon_semaphore *semaphore,
-				bool sync_to[RADEON_NUM_RINGS],
-				int dst_ring);
+				int signaler, int waiter);
 void radeon_semaphore_free(struct radeon_device *rdev,
-			   struct radeon_semaphore *semaphore,
+			   struct radeon_semaphore **semaphore,
 			   struct radeon_fence *fence);
 
 /*
@@ -653,6 +652,7 @@ struct radeon_ib {
 	struct radeon_fence		*fence;
 	unsigned			vm_id;
 	bool				is_const_ib;
+	struct radeon_fence		*sync_to[RADEON_NUM_RINGS];
 	struct radeon_semaphore		*semaphore;
 };
 

drivers/gpu/drm/radeon/radeon_asic.h

@@ -363,9 +363,10 @@ int r600_hdmi_buffer_status_changed(struct drm_encoder *encoder);
 void r600_hdmi_update_audio_settings(struct drm_encoder *encoder);
 /* r600 blit */
 int r600_blit_prepare_copy(struct radeon_device *rdev, unsigned num_gpu_pages,
-			   struct radeon_sa_bo **vb);
+			   struct radeon_fence **fence, struct radeon_sa_bo **vb,
+			   struct radeon_semaphore **sem);
 void r600_blit_done_copy(struct radeon_device *rdev, struct radeon_fence **fence,
-			 struct radeon_sa_bo *vb);
+			 struct radeon_sa_bo *vb, struct radeon_semaphore *sem);
 void r600_kms_blit_copy(struct radeon_device *rdev,
 			u64 src_gpu_addr, u64 dst_gpu_addr,
 			unsigned num_gpu_pages,

drivers/gpu/drm/radeon/radeon_cs.c

@@ -115,36 +115,20 @@ static int radeon_cs_get_ring(struct radeon_cs_parser *p, u32 ring, s32 priority
 	return 0;
 }
 
-static int radeon_cs_sync_rings(struct radeon_cs_parser *p)
+static void radeon_cs_sync_rings(struct radeon_cs_parser *p)
 {
-	bool sync_to_ring[RADEON_NUM_RINGS] = { };
-	bool need_sync = false;
-	int i, r;
+	int i;
 
 	for (i = 0; i < p->nrelocs; i++) {
-		struct radeon_fence *fence;
+		struct radeon_fence *a, *b;
 
 		if (!p->relocs[i].robj || !p->relocs[i].robj->tbo.sync_obj)
 			continue;
 
-		fence = p->relocs[i].robj->tbo.sync_obj;
-		if (fence->ring != p->ring && !radeon_fence_signaled(fence)) {
-			sync_to_ring[fence->ring] = true;
-			need_sync = true;
-		}
-	}
-
-	if (!need_sync) {
-		return 0;
-	}
-
-	r = radeon_semaphore_create(p->rdev, &p->ib.semaphore);
-	if (r) {
-		return r;
+		a = p->relocs[i].robj->tbo.sync_obj;
+		b = p->ib.sync_to[a->ring];
+		p->ib.sync_to[a->ring] = radeon_fence_later(a, b);
 	}
-
-	return radeon_semaphore_sync_rings(p->rdev, p->ib.semaphore,
-					   sync_to_ring, p->ring);
 }
 
 /* XXX: note that this is called from the legacy UMS CS ioctl as well */
@@ -368,10 +352,7 @@ static int radeon_cs_ib_chunk(struct radeon_device *rdev,
 		DRM_ERROR("Invalid command stream !\n");
 		return r;
 	}
-	r = radeon_cs_sync_rings(parser);
-	if (r) {
-		DRM_ERROR("Failed to synchronize rings !\n");
-	}
+	radeon_cs_sync_rings(parser);
 	parser->ib.vm_id = 0;
 	r = radeon_ib_schedule(rdev, &parser->ib);
 	if (r) {
@@ -468,10 +449,7 @@ static int radeon_cs_ib_vm_chunk(struct radeon_device *rdev,
 	if (r) {
 		goto out;
 	}
-	r = radeon_cs_sync_rings(parser);
-	if (r) {
-		DRM_ERROR("Failed to synchronize rings !\n");
-	}
+	radeon_cs_sync_rings(parser);
 
 	if ((rdev->family >= CHIP_TAHITI) &&
 	    (parser->chunk_const_ib_idx != -1)) {

drivers/gpu/drm/radeon/radeon_ring.c

@@ -42,7 +42,7 @@ int radeon_debugfs_sa_init(struct radeon_device *rdev);
 int radeon_ib_get(struct radeon_device *rdev, int ring,
 		  struct radeon_ib *ib, unsigned size)
 {
-	int r;
+	int i, r;
 
 	r = radeon_sa_bo_new(rdev, &rdev->ring_tmp_bo, &ib->sa_bo, size, 256, true);
 	if (r) {
@@ -50,20 +50,26 @@ int radeon_ib_get(struct radeon_device *rdev, int ring,
 		return r;
 	}
 
+	r = radeon_semaphore_create(rdev, &ib->semaphore);
+	if (r) {
+		return r;
+	}
+
 	ib->ring = ring;
 	ib->fence = NULL;
 	ib->ptr = radeon_sa_bo_cpu_addr(ib->sa_bo);
 	ib->gpu_addr = radeon_sa_bo_gpu_addr(ib->sa_bo);
 	ib->vm_id = 0;
 	ib->is_const_ib = false;
-	ib->semaphore = NULL;
+	for (i = 0; i < RADEON_NUM_RINGS; ++i)
+		ib->sync_to[i] = NULL;
 
 	return 0;
 }
 
 void radeon_ib_free(struct radeon_device *rdev, struct radeon_ib *ib)
 {
-	radeon_semaphore_free(rdev, ib->semaphore, ib->fence);
+	radeon_semaphore_free(rdev, &ib->semaphore, ib->fence);
 	radeon_sa_bo_free(rdev, &ib->sa_bo, ib->fence);
 	radeon_fence_unref(&ib->fence);
 }
@@ -71,7 +77,8 @@ void radeon_ib_free(struct radeon_device *rdev, struct radeon_ib *ib)
 int radeon_ib_schedule(struct radeon_device *rdev, struct radeon_ib *ib)
 {
 	struct radeon_ring *ring = &rdev->ring[ib->ring];
-	int r = 0;
+	bool need_sync = false;
+	int i, r = 0;
 
 	if (!ib->length_dw || !ring->ready) {
 		/* TODO: Nothings in the ib we should report. */
@@ -80,11 +87,24 @@ int radeon_ib_schedule(struct radeon_device *rdev, struct radeon_ib *ib)
 	}
 
 	/* 64 dwords should be enough for fence too */
-	r = radeon_ring_lock(rdev, ring, 64);
+	r = radeon_ring_lock(rdev, ring, 64 + RADEON_NUM_RINGS * 8);
 	if (r) {
 		dev_err(rdev->dev, "scheduling IB failed (%d).\n", r);
 		return r;
 	}
+	for (i = 0; i < RADEON_NUM_RINGS; ++i) {
+		struct radeon_fence *fence = ib->sync_to[i];
+		if (radeon_fence_need_sync(fence, ib->ring)) {
+			need_sync = true;
+			radeon_semaphore_sync_rings(rdev, ib->semaphore,
+						    fence->ring, ib->ring);
+			radeon_fence_note_sync(fence, ib->ring);
+		}
+	}
+	/* immediately free semaphore when we don't need to sync */
+	if (!need_sync) {
+		radeon_semaphore_free(rdev, &ib->semaphore, NULL);
+	}
 	radeon_ring_ib_execute(rdev, ib->ring, ib);
 	r = radeon_fence_emit(rdev, &ib->fence, ib->ring);
 	if (r) {

drivers/gpu/drm/radeon/radeon_semaphore.c

@@ -68,70 +68,49 @@ void radeon_semaphore_emit_wait(struct radeon_device *rdev, int ring,
 	radeon_semaphore_ring_emit(rdev, ring, &rdev->ring[ring], semaphore, true);
 }
 
+/* caller must hold ring lock */
 int radeon_semaphore_sync_rings(struct radeon_device *rdev,
 				struct radeon_semaphore *semaphore,
-				bool sync_to[RADEON_NUM_RINGS],
-				int dst_ring)
+				int signaler, int waiter)
 {
-	int i = 0, r;
+	int r;
 
-	mutex_lock(&rdev->ring_lock);
-	r = radeon_ring_alloc(rdev, &rdev->ring[dst_ring], RADEON_NUM_RINGS * 8);
-	if (r) {
-		goto error;
+	/* no need to signal and wait on the same ring */
+	if (signaler == waiter) {
+		return 0;
 	}
 
-	for (i = 0; i < RADEON_NUM_RINGS; ++i) {
-		/* no need to sync to our own or unused rings */
-		if (!sync_to[i] || i == dst_ring)
-			continue;
-
-		/* prevent GPU deadlocks */
-		if (!rdev->ring[i].ready) {
-			dev_err(rdev->dev, "Trying to sync to a disabled ring!");
-			r = -EINVAL;
-			goto error;
-		}
-
-		r = radeon_ring_alloc(rdev, &rdev->ring[i], 8);
-		if (r) {
-			goto error;
-		}
-
-		radeon_semaphore_emit_signal(rdev, i, semaphore);
-		radeon_semaphore_emit_wait(rdev, dst_ring, semaphore);
+	/* prevent GPU deadlocks */
+	if (!rdev->ring[signaler].ready) {
+		dev_err(rdev->dev, "Trying to sync to a disabled ring!");
+		return -EINVAL;
+	}
 
-		radeon_ring_commit(rdev, &rdev->ring[i]);
+	r = radeon_ring_alloc(rdev, &rdev->ring[signaler], 8);
+	if (r) {
+		return r;
 	}
+	radeon_semaphore_emit_signal(rdev, signaler, semaphore);
+	radeon_ring_commit(rdev, &rdev->ring[signaler]);
 
-	radeon_ring_commit(rdev, &rdev->ring[dst_ring]);
-	mutex_unlock(&rdev->ring_lock);
+	/* we assume caller has already allocated space on waiters ring */
+	radeon_semaphore_emit_wait(rdev, waiter, semaphore);
 
 	return 0;
-
-error:
-	/* unlock all locks taken so far */
-	for (--i; i >= 0; --i) {
-		if (sync_to[i] || i == dst_ring) {
-			radeon_ring_undo(&rdev->ring[i]);
-		}
-	}
-	radeon_ring_undo(&rdev->ring[dst_ring]);
-	mutex_unlock(&rdev->ring_lock);
-	return r;
 }
 
 void radeon_semaphore_free(struct radeon_device *rdev,
-			   struct radeon_semaphore *semaphore,
+			   struct radeon_semaphore **semaphore,
 			   struct radeon_fence *fence)
 {
-	if (semaphore == NULL) {
+	if (semaphore == NULL || *semaphore == NULL) {
 		return;
 	}
-	if (semaphore->waiters > 0) {
+	if ((*semaphore)->waiters > 0) {
 		dev_err(rdev->dev, "semaphore %p has more waiters than signalers,"
-			" hardware lockup imminent!\n", semaphore);
+			" hardware lockup imminent!\n", *semaphore);
 	}
-	radeon_sa_bo_free(rdev, &semaphore->sa_bo, fence);
-	kfree(semaphore);
+	radeon_sa_bo_free(rdev, &(*semaphore)->sa_bo, fence);
+	kfree(*semaphore);
+	*semaphore = NULL;
 }

drivers/gpu/drm/radeon/radeon_test.c

@@ -303,8 +303,7 @@ void radeon_test_ring_sync(struct radeon_device *rdev,
 	}
 
 out_cleanup:
-	if (semaphore)
-		radeon_semaphore_free(rdev, semaphore, NULL);
+	radeon_semaphore_free(rdev, &semaphore, NULL);
 
 	if (fence1)
 		radeon_fence_unref(&fence1);
@@ -422,8 +421,7 @@ void radeon_test_ring_sync2(struct radeon_device *rdev,
 	}
 
 out_cleanup:
-	if (semaphore)
-		radeon_semaphore_free(rdev, semaphore, NULL);
+	radeon_semaphore_free(rdev, &semaphore, NULL);
 
 	if (fenceA)
 		radeon_fence_unref(&fenceA);

drivers/gpu/drm/radeon/radeon_ttm.c

@@ -223,7 +223,6 @@ static int radeon_move_blit(struct ttm_buffer_object *bo,
 	struct radeon_device *rdev;
 	uint64_t old_start, new_start;
 	struct radeon_fence *fence;
-	struct radeon_semaphore *sem = NULL;
 	int r, ridx;
 
 	rdev = radeon_get_rdev(bo->bdev);
@@ -262,31 +261,12 @@ static int radeon_move_blit(struct ttm_buffer_object *bo,
 
 	/* sync other rings */
 	fence = bo->sync_obj;
-	if (fence && fence->ring != ridx
-	    && !radeon_fence_signaled(fence)) {
-		bool sync_to_ring[RADEON_NUM_RINGS] = { };
-		sync_to_ring[fence->ring] = true;
-
-		r = radeon_semaphore_create(rdev, &sem);
-		if (r) {
-			return r;
-		}
-
-		r = radeon_semaphore_sync_rings(rdev, sem, sync_to_ring, ridx);
-		if (r) {
-			radeon_semaphore_free(rdev, sem, NULL);
-			return r;
-		}
-	}
-
-	fence = NULL;
 	r = radeon_copy(rdev, old_start, new_start,
 			new_mem->num_pages * (PAGE_SIZE / RADEON_GPU_PAGE_SIZE), /* GPU pages */
 			&fence);
 	/* FIXME: handle copy error */
 	r = ttm_bo_move_accel_cleanup(bo, (void *)fence, NULL,
 				      evict, no_wait_reserve, no_wait_gpu, new_mem);
-	radeon_semaphore_free(rdev, sem, fence);
 	radeon_fence_unref(&fence);
 	return r;
 }