[RFC PATCH v2 06/11] iommufd: Map memory provider files

[Jason Gunthorpe]

On Wed, Oct 21, 2026 at 08:17:05PM +0000, Samiullah Khawaja wrote:

> With this approach, silently skipping holes during IOAS_MAP_FILE means
> iommufd has no tracking of which IOVA ranges in the area are actually
> mapped, which complicates unmap and could also hide a bug if a hole in
> the range was not expected?

I think if we are going to do this it should only work with iommupt
and iommupt should provide an unmap that is compatible with the holes.

There are other places that would like such an operation, IIRC.

This will have to block an access reaching into a guest memfd mapping
as well, so that should be OK.

Jason