Re: [PATCH V3 0/9] iommu_ioas_map_file

[Jason Gunthorpe <jgg@nvidia.com>]

On Fri, Oct 04, 2024 at 11:48:11AM -0700, Steve Sistare wrote:
> Provide the IOMMU_IOAS_MAP_FILE ioctl, which allows a user to register
> memory by passing a memfd plus offset and length.  Implement it using
> the memfd_map_folios KAPI, and the proposed folio_add_pins KAPI.
> See the individual patches for details.
> 
> Changes in V2:
>  * changed names and commit message in "rename uptr in iopt_alloc_iova"
>  * normalized comments describing the iopt_map_user_pages interface
>  * submitted folio_split_user_page_pin (fka folio_repin_unhugely) separately
>  * replaced nupages[] optimization with folio-to-batch optimization
>  * added selftests for map file
> 
> Changes in V3:
>   * fixed bug setting user->locked
>   * fixed bug in pages->file refs
>   * replaced lockdep_off with down_write_nest_lock
>   * added ufolios_next to track folio consumption in reader
>   * combined IOMMU_IOAS_MAP_FILE interface and implementation
>   * added patch folio_add_pins (fka folio_split_user_page_pin )
>   * reformatted patches using clang-format
>   * misc cosmetic changes in response to review comments
> 
> Steve Sistare (9):
>   mm/gup: folio_add_pins
>   iommufd: rename uptr in iopt_alloc_iova
>   iommufd: generalize iopt_pages address
>   iommufd: pfn reader for file mappings
>   iommufd: IOMMU_IOAS_MAP_FILE
>   iommufd: file mappings for mdev
>   iommufd: pfn reader local variables
>   iommufd: optimize file mapping
>   iommufd: map file selftest

This doesn't apply, you need to base it on a clean rc release before
sending it..

Jason