Re: [PATCH v2 4/4] vfio: Add pinned interface to perform revoke semantics

[Leon Romanovsky <leon@kernel.org>]

On Mon, Jan 19, 2026 at 01:12:45PM +0100, Christian König wrote:
> On 1/18/26 13:08, Leon Romanovsky wrote:
> > From: Leon Romanovsky <leonro@nvidia.com>
> > 
> > DMABUF ->pin() interface is called when the DMABUF importer perform
> > its DMA mapping, so let's use this opportunity to check if DMABUF
> > exporter revoked its buffer or not.
> > 
> > Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
> > ---
> >  drivers/vfio/pci/vfio_pci_dmabuf.c | 16 ++++++++++++++++
> >  1 file changed, 16 insertions(+)
> > 
> > diff --git a/drivers/vfio/pci/vfio_pci_dmabuf.c b/drivers/vfio/pci/vfio_pci_dmabuf.c
> > index d4d0f7d08c53..af9c315ddf71 100644
> > --- a/drivers/vfio/pci/vfio_pci_dmabuf.c
> > +++ b/drivers/vfio/pci/vfio_pci_dmabuf.c
> > @@ -20,6 +20,20 @@ struct vfio_pci_dma_buf {
> >  	u8 revoked : 1;
> >  };
> >  
> > +static int vfio_pci_dma_buf_pin(struct dma_buf_attachment *attachment)
> > +{
> > +	struct vfio_pci_dma_buf *priv = attachment->dmabuf->priv;
> > +
> > +	dma_resv_assert_held(priv->dmabuf->resv);
> > +
> > +	return dma_buf_attachment_is_revoke(attachment) ? 0 : -EOPNOTSUPP;
> 
> It's probably better to do that check in vfio_pci_dma_buf_attach.

I assume you are proposing to add this check in both
vfio_pci_dma_buf_attach() and vfio_pci_dma_buf_pin(). Otherwise,
importers that lack .invalidate_mapping() will invoke dma_buf_pin()
and will not fail.

> 
> And BTW the function vfio_pci_dma_buf_move() seems to be broken:
> 
> void vfio_pci_dma_buf_move(struct vfio_pci_core_device *vdev, bool revoked)
> {
>         struct vfio_pci_dma_buf *priv;
>         struct vfio_pci_dma_buf *tmp;
> 
>         lockdep_assert_held_write(&vdev->memory_lock);
> 
>         list_for_each_entry_safe(priv, tmp, &vdev->dmabufs, dmabufs_elm) {
>                 if (!get_file_active(&priv->dmabuf->file))
>                         continue;
> 
>                 if (priv->revoked != revoked) {
>                         dma_resv_lock(priv->dmabuf->resv, NULL);
>                         priv->revoked = revoked;
>                         dma_buf_move_notify(priv->dmabuf);
> 
> A dma_buf_move_notify() just triggers asynchronous invalidation of the mapping!
> 
> You need to use dma_resv_wait() to wait for that to finish.

We (VFIO and IOMMUFD) followed the same pattern used in  
amdgpu_bo_move_notify(), which also does not wait.

I'll add wait here.

Thanks

> 
>                         dma_resv_unlock(priv->dmabuf->resv);
>                 }
>                 fput(priv->dmabuf->file);
>         }
> }
> 
> Regards,
> Christian.
> 
> 
> > +}
> > +
> > +static void vfio_pci_dma_buf_unpin(struct dma_buf_attachment *attachment)
> > +{
> > +	/* Do nothing */
> > +}
> > +
> >  static int vfio_pci_dma_buf_attach(struct dma_buf *dmabuf,
> >  				   struct dma_buf_attachment *attachment)
> >  {
> > @@ -76,6 +90,8 @@ static void vfio_pci_dma_buf_release(struct dma_buf *dmabuf)
> >  }
> >  
> >  static const struct dma_buf_ops vfio_pci_dmabuf_ops = {
> > +	.pin = vfio_pci_dma_buf_pin,
> > +	.unpin = vfio_pci_dma_buf_unpin,
> >  	.attach = vfio_pci_dma_buf_attach,
> >  	.map_dma_buf = vfio_pci_dma_buf_map,
> >  	.unmap_dma_buf = vfio_pci_dma_buf_unmap,
> > 
>