On Fri, Feb 03, 2023 at 08:31:52AM -0400, Jason Gunthorpe wrote:
On Fri, Feb 03, 2023 at 10:10:45AM +0000, Tian, Kevin wrote:
From: Nicolin Chen nicolinc@nvidia.com Sent: Thursday, February 2, 2023 3:05 PM
Support an access->ioas replacement in iommufd_access_set_ioas(), which sets the access->ioas to NULL provisionally so that any further incoming iommufd_access_pin_pages() callback can be blocked.
Then, call access->ops->unmap() to clean up the entire iopt. To allow an iommufd_access_unpin_pages() callback to happen via this unmap() call, add an ioas_unpin pointer so the unpin routine won't be affected by the "access->ioas = NULL" trick above.
Also, a vdev without an ops->dma_unmap implementation cannot replace its access->ioas pointer. So add an iommufd_access_ioas_is_attached() helper to sanity that.
Presumably a driver which doesn't implement ops->dma_unmap shouldn't be allowed to do pin/unpin. But it could use vfio_dma_rw() to access an iova range. In the latter case I don't see why replace cannot work.
Probably what's required here is to deny !ops->dma_unmap in vfio_pin/unpin_pages then making here replace always allowed?
That makes sense
Will change that in v2.
Thanks Nic