iommu/io-pgtable-arm: Fix iova_to_phys for block entries
authorWill Deacon <will.deacon@arm.com>
Thu, 16 Jun 2016 17:21:19 +0000 (18:21 +0100)
committerWill Deacon <will.deacon@arm.com>
Fri, 1 Jul 2016 13:04:37 +0000 (14:04 +0100)
commit7c6d90e2bb1a98b86d73b9e8ab4d97ed5507e37c
tree22bf6738ea995d27bf8033f9e6e6e078a192df16
parent112c898b59dd5cfd95ee30dfe7cc4fc11a6d484e
iommu/io-pgtable-arm: Fix iova_to_phys for block entries

The implementation of iova_to_phys for the long-descriptor ARM
io-pgtable code always masks with the granule size when inserting the
low virtual address bits into the physical address determined from the
page tables. In cases where the leaf entry is found before the final
level of table (i.e. due to a block mapping), this results in rounding
down to the bottom page of the block mapping. Consequently, the physical
address range batching in the vfio_unmap_unpin is defeated and we end
up taking the long way home.

This patch fixes the problem by masking the virtual address with the
appropriate mask for the level at which the leaf descriptor is located.
The short-descriptor code already gets this right, so no change is
needed there.

Cc: <stable@vger.kernel.org>
Reported-by: Robin Murphy <robin.murphy@arm.com>
Tested-by: Robin Murphy <robin.murphy@arm.com>
Signed-off-by: Will Deacon <will.deacon@arm.com>
drivers/iommu/io-pgtable-arm.c