On Tue, Apr 29, 2025 at 06:58:32AM -0700, Rob Clark wrote: > On Tue, Apr 29, 2025 at 5:28 AM Jason Gunthorpe <j...@ziepe.ca> wrote: > > > > On Mon, Apr 28, 2025 at 01:54:10PM -0700, Rob Clark wrote: > > > From: Rob Clark <robdcl...@chromium.org> > > > > > > In situations where mapping/unmapping squence can be controlled by > > > userspace, attempting to map over a region that has not yet been > > > unmapped is an error. But not something that should spam dmesg. > > > > I think if you want to do something like that using the iommu API the > > expectation is for the caller to do a iova_to_phys to check what is > > mapped first? That seems kind of lame.. > > > > Maybe page table driver should not not be doing these WARNs at all. If > > we want to check for that the core iommu code should have the WARN_ON? > > > > eg iommufd already has a WARN_ON around iommu_unmap failures so having > > one in the ARM page table is a double WARN. > > > > Don't really like using a quirk to change the API contract. > > I'd also be ok to have the WARN_ON instead in the iommu code. In the > case where this quirk is needed, I'm using the io_pgtable helpers > directly, not going via the iommu layer.
Yes, that was my thought Then all the iommu_map/unmap() calls behave consistently here.. Jason
