On Wed, Apr 09, 2025 at 04:30:17PM -0700, Dan Williams wrote: > Like Christoph said, a driver really has no business opting itself into > different DMA addressing domains. For TDISP we are also being careful to > make sure that flipping a device from shared to private is a suitably > violent event. This is because the Linux DMA layer does not have a > concept of allowing a device to have mappings from two different > addressing domains simultaneously. And this is a very important point, several of the architectures have two completely independent iommu tables, and maybe even completely different IOMMU instances for trusted and non-trusted DMA traffic. I expect configurations where trusted traffic is translated through the vIOMMU while non-trusted traffic is locked to an identity translation. There are more issue here than just swiotlb :\ > A "private_capable" flag might also make sense, but that is really a > property of a bus that need not be carried necessarily in 'struct > device'. However it works, it should be done before the driver is probed and remain stable for the duration of the driver attachment. From the iommu side the correct iommu domain, on the correct IOMMU instance to handle the expected traffic should be setup as the DMA API's iommu domain. Jason
