On Tue, Jun 28, 2022 at 02:35:51PM +0530, Vasant Hegde wrote: > Sorry. I didn't get last statement ("device identity maps DMA requests > without PASID"). > Can you please elaborate?
When using v1 page-tables, each device supporting ATS/PRI/PASID needs to be direct-mapped, because the v1 page-tables basically act as a stage-2 page table for the PASID ones. But when the non-pasid case moves to the pasid==0 page-table, then there is not stage-2 anymore and a device can be used with ATS/PRI/PASID while non-PASID requests are translated too, no? I didn't get how this is handled in the current patch-set. Regards, Joerg _______________________________________________ iommu mailing list iommu@lists.linux-foundation.org https://lists.linuxfoundation.org/mailman/listinfo/iommu