On Wed, Oct 14, 2026 at 11:16:07PM +0200, Andrea Parri wrote: > On Wed, Oct 14, 2026 at 02:52:45PM -0300, Jason Gunthorpe wrote: > > > The mock page table has no memory type bits, so a mapping made with > > > IOMMU_MMIO cannot be told apart from one made with IOMMU_CACHE. Wrap the > > > mock's generic_pt map_range and unmap_range ops to record, per domain, > > > the pages mapped with IOMMU_MMIO, and add IOMMU_TEST_OP_MD_CHECK_MMIO to > > > check that a range is recorded. The check rejects ranges that wrap or > > > do not fit in unsigned long. > > > > I'm not keen on this, the special mock amd domain is just for mock > > usage, we should probably just add a mmio bit if we want to do this > > test. > > Makes sense, thanks. For v3 I have: > > - AMDV1PT_FMT_MOCK_MMIO (bit 52) in amdv1.h under AMDV1_IOMMUFD_SELFTEST, > set by amdv1pt_iommu_set_prot() for IOMMU_MMIO and kept by > amdv1pt_attr_from_entry(); > > - pt_iommu_amdv1_mock_iova_is_mmio() in fmt/iommu_mock.c, a walk like > iova_to_phys() that returns the bit, exported in GENERIC_PT_IOMMU for > IOMMU_TEST_OP_MD_CHECK_MMIO to use. > > The op wrappers and the xarray in mock_iommu_domain are gone. Is the > extra export from iommu_mock.c OK with you, or would you rather the > selftest read the PTE some other way?
Too much AI. This sounds better, but I don't know if we need all this code just for a selftest. Jason
