drm/ttm: add a way to bo_wait for either the last read or last write

Sometimes we want to know whether a buffer is busy and wait for it (bo_wait).
However, sometimes it would be more useful to be able to query whether
a buffer is busy and being either read or written, and wait until it's stopped
being either read or written. The point of this is to be able to avoid
unnecessary waiting, e.g. if a GPU has written something to a buffer and is now
reading that buffer, and a CPU wants to map that buffer for read, it needs to
only wait for the last write. If there were no write, there wouldn't be any
waiting needed.

This, or course, requires user space drivers to send read/write flags
with each relocation (like we have read/write domains in radeon, so we can
actually use those for something useful now).

Now how this patch works:

The read/write flags should passed to ttm_validate_buffer. TTM maintains
separate sync objects of the last read and write for each buffer, in addition
to the sync object of the last use of a buffer. ttm_bo_wait then operates
with one the sync objects.

Signed-off-by: Marek Olšák <maraeo@gmail.com>
Reviewed-by: Jerome Glisse <jglisse@redhat.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

1 files changed, 21 insertions, 5 deletions

diff --git a/drivers/gpu/drm/ttm/ttm_bo_util.c b/drivers/gpu/drm/ttm/ttm_bo_util.c
index ae3c6f5dd2b7..6135f58169ce 100644
--- a/drivers/gpu/drm/ttm/ttm_bo_util.c
+++ b/drivers/gpu/drm/ttm/ttm_bo_util.c
@@ -436,6 +436,8 @@ static int ttm_buffer_object_transfer(struct ttm_buffer_object *bo,
 	atomic_set(&fbo->cpu_writers, 0);
 
 	fbo->sync_obj = driver->sync_obj_ref(bo->sync_obj);
+	fbo->sync_obj_read = driver->sync_obj_ref(bo->sync_obj_read);
+	fbo->sync_obj_write = driver->sync_obj_ref(bo->sync_obj_write);
 	kref_init(&fbo->list_kref);
 	kref_init(&fbo->kref);
 	fbo->destroy = &ttm_transfered_destroy;
@@ -618,20 +620,30 @@ int ttm_bo_move_accel_cleanup(struct ttm_buffer_object *bo,
 	struct ttm_mem_reg *old_mem = &bo->mem;
 	int ret;
 	struct ttm_buffer_object *ghost_obj;
-	void *tmp_obj = NULL;
+	void *tmp_obj = NULL, *tmp_obj_read = NULL, *tmp_obj_write = NULL;
 
 	spin_lock(&bdev->fence_lock);
-	if (bo->sync_obj) {
+	if (bo->sync_obj)
 		tmp_obj = bo->sync_obj;
-		bo->sync_obj = NULL;
-	}
+	if (bo->sync_obj_read)
+		tmp_obj_read = bo->sync_obj_read;
+	if (bo->sync_obj_write)
+		tmp_obj_write = bo->sync_obj_write;
+
 	bo->sync_obj = driver->sync_obj_ref(sync_obj);
+	bo->sync_obj_read = driver->sync_obj_ref(sync_obj);
+	bo->sync_obj_write = driver->sync_obj_ref(sync_obj);
 	bo->sync_obj_arg = sync_obj_arg;
 	if (evict) {
-		ret = ttm_bo_wait(bo, false, false, false);
+		ret = ttm_bo_wait(bo, false, false, false,
+				  TTM_USAGE_READWRITE);
 		spin_unlock(&bdev->fence_lock);
 		if (tmp_obj)
 			driver->sync_obj_unref(&tmp_obj);
+		if (tmp_obj_read)
+			driver->sync_obj_unref(&tmp_obj_read);
+		if (tmp_obj_write)
+			driver->sync_obj_unref(&tmp_obj_write);
 		if (ret)
 			return ret;
 
@@ -655,6 +667,10 @@ int ttm_bo_move_accel_cleanup(struct ttm_buffer_object *bo,
 		spin_unlock(&bdev->fence_lock);
 		if (tmp_obj)
 			driver->sync_obj_unref(&tmp_obj);
+		if (tmp_obj_read)
+			driver->sync_obj_unref(&tmp_obj_read);
+		if (tmp_obj_write)
+			driver->sync_obj_unref(&tmp_obj_write);
 
 		ret = ttm_buffer_object_transfer(bo, &ghost_obj);
 		if (ret)