On Wed, Jan 10, 2024 at 11:46:35AM +0100, Christian König wrote:
> Am 10.01.24 um 11:22 schrieb Daniel Vetter:
> > On Wed, Jan 10, 2024 at 11:19:37AM +0100, Christian König wrote:
> > > Am 10.01.24 um 10:56 schrieb Julia Zhang:
> > > > drm_gem_map_attach() requires drm_gem_object_funcs.get_sg_table to be
> > > > implemented, or else return ENOSYS. Virtio has no get_sg_table
> > > > implemented for vram object. To fix this, add a new device_attach to
> > > > call drm_gem_map_attach() for shmem object and return 0 for vram object
> > > > instead of calling drm_gem_map_attach for both of these two kinds of
> > > > object.
> > > Well as far as I can see this is nonsense from the DMA-buf side of things.
> > >
> > > SG tables are always needed as long as you don't re-import the same object
> > > into your driver and then you shouldn't end up in this function in the first
> > > place.
> > >
> > > So that drm_gem_map_attach() requires get_sg_table to be implemented is
> > > intentional and should never be overridden like this.
> > See my reply, tldr; you're allowed to reject ->attach with -EBUSY to
> > handle exactly this case of non-shareable buffer types. But definitely
> > don't silently fail, that's a "we'll oops on map_attachment" kind of bug
> > :-)
>
> Ah, yes that makes much more sense!
>
> So basically just the "return 0;" needs to be "return -EBUSY;".
Well plus 2nd patch to polish the virtio_dma_buf docs a bit, that would be
nice :-D
-Sima
>
> Regards,
> Christian.
>
> > -Sima
> >
> > > Regards,
> > > Christian.
> > >
> > > > Signed-off-by: Julia Zhang <julia.zhang@xxxxxxx>
> > > > ---
> > > > drivers/gpu/drm/virtio/virtgpu_prime.c | 14 +++++++++++++-
> > > > 1 file changed, 13 insertions(+), 1 deletion(-)
> > > >
> > > > diff --git a/drivers/gpu/drm/virtio/virtgpu_prime.c b/drivers/gpu/drm/virtio/virtgpu_prime.c
> > > > index 44425f20d91a..f0b0ff6f3813 100644
> > > > --- a/drivers/gpu/drm/virtio/virtgpu_prime.c
> > > > +++ b/drivers/gpu/drm/virtio/virtgpu_prime.c
> > > > @@ -71,6 +71,18 @@ static void virtgpu_gem_unmap_dma_buf(struct dma_buf_attachment *attach,
> > > > drm_gem_unmap_dma_buf(attach, sgt, dir);
> > > > }
> > > > +static int virtgpu_gem_device_attach(struct dma_buf *dma_buf,
> > > > + struct dma_buf_attachment *attach)
> > > > +{
> > > > + struct drm_gem_object *obj = attach->dmabuf->priv;
> > > > + struct virtio_gpu_object *bo = gem_to_virtio_gpu_obj(obj);
> > > > +
> > > > + if (virtio_gpu_is_vram(bo))
> > > > + return 0;
> > > > +
> > > > + return drm_gem_map_attach(dma_buf, attach);
> > > > +}
> > > > +
> > > > static const struct virtio_dma_buf_ops virtgpu_dmabuf_ops = {
> > > > .ops = {
> > > > .cache_sgt_mapping = true,
> > > > @@ -83,7 +95,7 @@ static const struct virtio_dma_buf_ops virtgpu_dmabuf_ops = {
> > > > .vmap = drm_gem_dmabuf_vmap,
> > > > .vunmap = drm_gem_dmabuf_vunmap,
> > > > },
> > > > - .device_attach = drm_gem_map_attach,
> > > > + .device_attach = virtgpu_gem_device_attach,
> > > > .get_uuid = virtgpu_virtio_get_uuid,
> > > > };
>
--
Daniel Vetter
Software Engineer, Intel Corporation
http://blog.ffwll.ch
