| Message ID | 1466099078-3919-1-git-send-email-will.deacon@arm.com (mailing list archive) |
|---|---|
| State | New, archived |
| Headers | show |
On 16/06/16 18:44, Will Deacon wrote: > 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. With this, I now see VFIO unmapping at the same granularity as the initial mapping. To think of all the cumulative hours we've spent watching it split the blocks and go 4K at a time... *sigh* Tested-by: Robin Murphy <robin.murphy@arm.com> > Reported-by: Robin Murphy <robin.murphy@arm.com> > Signed-off-by: Will Deacon <will.deacon@arm.com> > --- > drivers/iommu/io-pgtable-arm.c | 2 +- > 1 file changed, 1 insertion(+), 1 deletion(-) > > diff --git a/drivers/iommu/io-pgtable-arm.c b/drivers/iommu/io-pgtable-arm.c > index a1ed1b73fed4..f5c90e1366ce 100644 > --- a/drivers/iommu/io-pgtable-arm.c > +++ b/drivers/iommu/io-pgtable-arm.c > @@ -576,7 +576,7 @@ static phys_addr_t arm_lpae_iova_to_phys(struct io_pgtable_ops *ops, > return 0; > > found_translation: > - iova &= (ARM_LPAE_GRANULE(data) - 1); > + iova &= (ARM_LPAE_BLOCK_SIZE(lvl, data) - 1); > return ((phys_addr_t)iopte_to_pfn(pte,data) << data->pg_shift) | iova; > } > >
diff --git a/drivers/iommu/io-pgtable-arm.c b/drivers/iommu/io-pgtable-arm.c index a1ed1b73fed4..f5c90e1366ce 100644 --- a/drivers/iommu/io-pgtable-arm.c +++ b/drivers/iommu/io-pgtable-arm.c @@ -576,7 +576,7 @@ static phys_addr_t arm_lpae_iova_to_phys(struct io_pgtable_ops *ops, return 0; found_translation: - iova &= (ARM_LPAE_GRANULE(data) - 1); + iova &= (ARM_LPAE_BLOCK_SIZE(lvl, data) - 1); return ((phys_addr_t)iopte_to_pfn(pte,data) << data->pg_shift) | iova; }
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. Reported-by: Robin Murphy <robin.murphy@arm.com> Signed-off-by: Will Deacon <will.deacon@arm.com> --- drivers/iommu/io-pgtable-arm.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-)