[PATCH 4/4] drm/radeon: add flag to NOT fence a BO on CS

[Christian König]

From: Christian König <christian.koenig at amd.com>

This is useful and only allowed with pure sync objects. E.g.
the CS depends on finishing a previous CS, but should not add
the new fence the handle.

Signed-off-by: Christian König <christian.koenig at amd.com>
---
 drivers/gpu/drm/radeon/radeon_cs.c | 29 ++++++++++++++++++++++++-----
 include/uapi/drm/radeon_drm.h      |  1 +
 2 files changed, 25 insertions(+), 5 deletions(-)

diff --git a/drivers/gpu/drm/radeon/radeon_cs.c b/drivers/gpu/drm/radeon/radeon_cs.c
index 11e4789..bdb8eda 100644
--- a/drivers/gpu/drm/radeon/radeon_cs.c
+++ b/drivers/gpu/drm/radeon/radeon_cs.c
@@ -225,7 +225,7 @@ static int radeon_cs_get_ring(struct radeon_cs_parser *p, u32 ring, s32 priority
 	return 0;
 }
 
-static void radeon_cs_sync_rings(struct radeon_cs_parser *p)
+static int radeon_cs_sync_rings(struct radeon_cs_parser *p)
 {
 	int i;
 
@@ -237,6 +237,18 @@ static void radeon_cs_sync_rings(struct radeon_cs_parser *p)
 		if (!bo)
 			continue;
 
+		fence = bo->tbo.sync_obj;
+		if (reloc->flags & RADEON_RELOC_DONT_FENCE) {
+
+			/* only allowed for sync objects */
+			if (radeon_bo_size(bo) != 0)
+				return -EINVAL;
+
+			list_del(&p->relocs[i].tv.head);
+			radeon_bo_unreserve(bo);
+			p->relocs[i].robj = NULL;
+		}
+
 		/* always sync to the last operation
 		   the clients doesn't know about */
 		radeon_semaphore_sync_to(p->ib.presync, bo->last_sync);
@@ -245,8 +257,6 @@ static void radeon_cs_sync_rings(struct radeon_cs_parser *p)
 		    reloc->flags & RADEON_RELOC_DONT_SYNC)
 			continue;
 
-		fence = bo->tbo.sync_obj;
-
 		if (bo->written && radeon_fence_signaled(bo->written))
 			radeon_fence_unref(&bo->written);
 
@@ -257,6 +267,8 @@ static void radeon_cs_sync_rings(struct radeon_cs_parser *p)
 		else
 			radeon_semaphore_sync_to(p->ib.postsync, fence);
 	}
+
+	return 0;
 }
 
 /* XXX: note that this is called from the legacy UMS CS ioctl as well */
@@ -498,7 +510,10 @@ static int radeon_cs_ib_chunk(struct radeon_device *rdev,
 		 (parser->ring == TN_RING_TYPE_VCE2_INDEX))
 		radeon_vce_note_usage(rdev);
 
-	radeon_cs_sync_rings(parser);
+	r = radeon_cs_sync_rings(parser);
+	if (r)
+		return r;
+
 	r = radeon_ib_schedule(rdev, &parser->ib, NULL);
 	if (r) {
 		DRM_ERROR("Failed to schedule IB !\n");
@@ -585,7 +600,11 @@ static int radeon_cs_ib_vm_chunk(struct radeon_device *rdev,
 	if (r) {
 		goto out;
 	}
-	radeon_cs_sync_rings(parser);
+
+	r = radeon_cs_sync_rings(parser);
+	if (r)
+		goto out;
+
 	radeon_semaphore_sync_to(parser->ib.presync, vm->fence);
 
 	if ((rdev->family >= CHIP_TAHITI) &&
diff --git a/include/uapi/drm/radeon_drm.h b/include/uapi/drm/radeon_drm.h
index 5bd3f68..289a0ca 100644
--- a/include/uapi/drm/radeon_drm.h
+++ b/include/uapi/drm/radeon_drm.h
@@ -946,6 +946,7 @@ struct drm_radeon_cs_chunk {
 /* drm_radeon_cs_reloc.flags */
 #define RADEON_RELOC_PRIO_MASK		(0xf << 0)
 #define RADEON_RELOC_DONT_SYNC		(1 << 4)
+#define RADEON_RELOC_DONT_FENCE		(1 << 5)
 
 struct drm_radeon_cs_reloc {
 	uint32_t		handle;
-- 
1.9.1