IOMMUFD relies on a private protocol with VFIO, and this always operated
in pinned mode.
Now that VFIO can support pinned importers update IOMMUFD to invoke the
normal dma-buf flow to request pin.
This isn't enough to allow IOMMUFD to work with other exporters, it still
needs a way to get the physical address list which is another series.
IOMMUFD supports the defined revoke semantics. It immediately stops and
fences access to the memory inside it's invalidate_mappings() callback,
and it currently doesn't use scatterlists so doesn't call map/unmap at
all.
It is expected that a future revision can synchronously call unmap from
the move_notify callback as well.
Reviewed-by: Kevin Tian <kevin.tian@intel.com>
Reviewed-by: Jason Gunthorpe <jgg@nvidia.com>
Acked-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Leon Romanovsky <leonro@nvidia.com>
Signed-off-by: Christian König <christian.koenig@amd.com>
Link: https://lore.kernel.org/r/20260131-dmabuf-revoke-v7-8-463d956bd527@nvidia.com
mutex_unlock(&pages->mutex);
}
- rc = sym_vfio_pci_dma_buf_iommufd_map(attach, &pages->dmabuf.phys);
+ rc = dma_buf_pin(attach);
if (rc)
goto err_detach;
+ rc = sym_vfio_pci_dma_buf_iommufd_map(attach, &pages->dmabuf.phys);
+ if (rc)
+ goto err_unpin;
+
dma_resv_unlock(dmabuf->resv);
/* On success iopt_release_pages() will detach and put the dmabuf. */
pages->dmabuf.attach = attach;
return 0;
+err_unpin:
+ dma_buf_unpin(attach);
err_detach:
dma_resv_unlock(dmabuf->resv);
dma_buf_detach(dmabuf, attach);
if (iopt_is_dmabuf(pages) && pages->dmabuf.attach) {
struct dma_buf *dmabuf = pages->dmabuf.attach->dmabuf;
+ dma_buf_unpin(pages->dmabuf.attach);
dma_buf_detach(dmabuf, pages->dmabuf.attach);
dma_buf_put(dmabuf);
WARN_ON(!list_empty(&pages->dmabuf.tracker));