vfio/type1: handle case where IOMMU does not support PAGE_SIZE size
authorEric Auger <eric.auger@linaro.org>
Thu, 29 Oct 2015 17:49:42 +0000 (17:49 +0000)
committerAlex Williamson <alex.williamson@redhat.com>
Tue, 3 Nov 2015 19:53:53 +0000 (12:53 -0700)
commit4644321fd3c119a819ab24fd2bc2d1f9bca4a695
tree3188e61fff67f7233ca755d1b967ac835c413fc4
parent1276ece32c5d18790e8bcff89e692fd3c1790bab
vfio/type1: handle case where IOMMU does not support PAGE_SIZE size

Current vfio_pgsize_bitmap code hides the supported IOMMU page
sizes smaller than PAGE_SIZE. As a result, in case the IOMMU
does not support PAGE_SIZE page, the alignment check on map/unmap
is done with larger page sizes, if any. This can fail although
mapping could be done with pages smaller than PAGE_SIZE.

This patch modifies vfio_pgsize_bitmap implementation so that,
in case the IOMMU supports page sizes smaller than PAGE_SIZE
we pretend PAGE_SIZE is supported and hide sub-PAGE_SIZE sizes.
That way the user will be able to map/unmap buffers whose size/
start address is aligned with PAGE_SIZE. Pinning code uses that
granularity while iommu driver can use the sub-PAGE_SIZE size
to map the buffer.

Signed-off-by: Eric Auger <eric.auger@linaro.org>
Acked-by: Will Deacon <will.deacon@arm.com>
Signed-off-by: Alex Williamson <alex.williamson@redhat.com>
drivers/vfio/vfio_iommu_type1.c