drm/radeon/kms: Add support for multi-ring sync in CS ioctl (v2)

Use semaphores to sync buffers across rings in the CS
ioctl.  Add a reloc flag to allow userspace to skip
sync for buffers.

agd5f: port to latest CS ioctl changes.

v2: add ring lock/unlock to make sure changes hit the ring.
Signed-off-by: Christian König <deathsimple@vodafone.de>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

drivers/gpu/drm/radeon/radeon.h

@@ -230,6 +230,7 @@ struct radeon_fence {
 	bool				signaled;
 	/* RB, DMA, etc. */
 	int				ring;
+	struct radeon_semaphore		*semaphore;
 };
 
 int radeon_fence_driver_start_ring(struct radeon_device *rdev, int ring);
@@ -789,6 +790,7 @@ struct radeon_cs_parser {
 	struct radeon_cs_reloc	*relocs;
 	struct radeon_cs_reloc	**relocs_ptr;
 	struct list_head	validated;
+	bool			sync_to_ring[RADEON_NUM_RINGS];
 	/* indices of various chunks */
 	int			chunk_ib_idx;
 	int			chunk_relocs_idx;

drivers/gpu/drm/radeon/radeon_cs.c

@@ -84,6 +84,13 @@ int radeon_cs_parser_relocs(struct radeon_cs_parser *p)
 			p->relocs[i].flags = r->flags;
 			radeon_bo_list_add_object(&p->relocs[i].lobj,
 						  &p->validated);
+
+			if (p->relocs[i].robj->tbo.sync_obj && !(r->flags & RADEON_RELOC_DONT_SYNC)) {
+				struct radeon_fence *fence = p->relocs[i].robj->tbo.sync_obj;
+				if (!radeon_fence_signaled(fence)) {
+					p->sync_to_ring[fence->ring] = true;
+				}
+			}
 		} else
 			p->relocs[i].handle = 0;
 	}
@@ -109,6 +116,36 @@ 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)
+{
+	int i, r;
+
+	for (i = 0; i < RADEON_NUM_RINGS; ++i) {
+		/* no need to sync to our own or unused rings */
+		if (i == p->ring || !p->sync_to_ring[i] || !p->rdev->ring[i].ready)
+			continue;
+
+		if (!p->ib->fence->semaphore) {
+			r = radeon_semaphore_create(p->rdev, &p->ib->fence->semaphore);
+			if (r)
+				return r;
+		}
+
+		r = radeon_ring_lock(p->rdev, &p->rdev->ring[i], 3);
+		if (r)
+			return r;
+		radeon_semaphore_emit_signal(p->rdev, i, p->ib->fence->semaphore);
+		radeon_ring_unlock_commit(p->rdev, &p->rdev->ring[i]);
+
+		r = radeon_ring_lock(p->rdev, &p->rdev->ring[p->ring], 3);
+		if (r)
+			return r;
+		radeon_semaphore_emit_wait(p->rdev, p->ring, p->ib->fence->semaphore);
+		radeon_ring_unlock_commit(p->rdev, &p->rdev->ring[p->ring]);
+	}
+	return 0;
+}
+
 int radeon_cs_parser_init(struct radeon_cs_parser *p, void *data)
 {
 	struct drm_radeon_cs *cs = data;
@@ -314,6 +351,10 @@ 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");
+	}
 	parser->ib->vm_id = 0;
 	r = radeon_ib_schedule(rdev, parser->ib);
 	if (r) {
@@ -384,6 +425,10 @@ 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");
+	}
 	parser->ib->vm_id = vm->id;
 	/* ib pool is bind at 0 in virtual address space to gpu_addr is the
 	 * offset inside the pool bo

drivers/gpu/drm/radeon/radeon_fence.c

@@ -153,6 +153,8 @@ static void radeon_fence_destroy(struct kref *kref)
 	list_del(&fence->list);
 	fence->emitted = false;
 	write_unlock_irqrestore(&fence->rdev->fence_lock, irq_flags);
+	if (fence->semaphore)
+		radeon_semaphore_free(fence->rdev, fence->semaphore);
 	kfree(fence);
 }
 
@@ -172,6 +174,7 @@ int radeon_fence_create(struct radeon_device *rdev,
 	(*fence)->signaled = false;
 	(*fence)->seq = 0;
 	(*fence)->ring = ring;
+	(*fence)->semaphore = NULL;
 	INIT_LIST_HEAD(&(*fence)->list);
 
 	write_lock_irqsave(&rdev->fence_lock, irq_flags);

include/drm/radeon_drm.h

@@ -917,6 +917,9 @@ struct drm_radeon_cs_chunk {
 	uint64_t		chunk_data;
 };
 
+/* drm_radeon_cs_reloc.flags */
+#define RADEON_RELOC_DONT_SYNC		0x01
+
 struct drm_radeon_cs_reloc {
 	uint32_t		handle;
 	uint32_t		read_domains;