[PATCH v5 0/5] drm/virtio: Import scanout buffers from other devices

[Dmitry Osipenko]

Hello, Vivek

All patches applied to misc-next with a small modification, thanks!

Note: While verifying move_notify(), I noticed that AMD/TTM driver moves
same shared FB GEMs after each framebuffer update when it renders into
FB, despite of the 32GB BAR. This should be rather inefficient. I'd
expect dma-buf staying static if there is no need to evict it. Something
to check how it works with DG2.

Fix: I made this change to the "Import prime buffers" patch after
spotting possibility of having race condition between move_notify() and
freeing GEM:

diff --git a/drivers/gpu/drm/virtio/virtgpu_prime.c
b/drivers/gpu/drm/virtio/virtgpu_prime.c
index 8644b87d473d..688810d1b611 100644
--- a/drivers/gpu/drm/virtio/virtgpu_prime.c
+++ b/drivers/gpu/drm/virtio/virtgpu_prime.c
@@ -189,13 +189,18 @@ static void virtgpu_dma_buf_free_obj(struct
drm_gem_object *obj)
 	struct virtio_gpu_object *bo = gem_to_virtio_gpu_obj(obj);
 	struct virtio_gpu_device *vgdev = obj->dev->dev_private;
 	struct dma_buf_attachment *attach = obj->import_attach;
+	struct dma_resv *resv = attach->dmabuf->resv;

 	if (attach) {
+		dma_resv_lock(resv, NULL);
+
 		virtio_gpu_detach_object_fenced(bo);

 		if (bo->sgt)
-			dma_buf_unmap_attachment_unlocked(attach, bo->sgt,
-							  DMA_BIDIRECTIONAL);
+			dma_buf_unmap_attachment(attach, bo->sgt,
+						 DMA_BIDIRECTIONAL);
+
+		dma_resv_unlock(resv);

 		dma_buf_detach(attach->dmabuf, attach);
 		dma_buf_put(attach->dmabuf);
@@ -268,7 +273,7 @@ static void virtgpu_dma_buf_move_notify(struct
dma_buf_attachment *attach)
 	struct drm_gem_object *obj = attach->importer_priv;
 	struct virtio_gpu_object *bo = gem_to_virtio_gpu_obj(obj);

-	if (bo->created) {
+	if (bo->created && kref_read(&obj->refcount)) {
 		virtio_gpu_detach_object_fenced(bo);

 		if (bo->sgt)


-- 
Best regards,
Dmitry