On Mon, Feb 14, 2022 at 10:02:36AM -0400, Jason Gunthorpe wrote: > That works for VFIO, but it doesn't work for other in-kernel > drivers.. Is there something ensuring the group is only the GPU and > sound device? Is the GPU never an addin card? GPUs supporting this functionality are always iGPUs, AFAIK. > I'd say the right way to code this after Lu's series to have both the > GPU and sound driver call iommu_attach_device() during their probe()'s > and specify the identity domain as the attaching domain. > > That would be quite similar to how the Tegra drivers got arranged. > > And then maybe someone could better guess what the "sound driver" is > since it would be marked with an iommu_attach_device() call. Device drivers calling into iommu_attach_device() is seldom a good idea. In this case the sound device has some generic hardware interface so that an existing sound driver can be re-used. Making this driver call iommu-specific functions for some devices is something hard to justify. With sub-groups on the other hand it would be a no-brainer, because the sound device would be in a separate sub-group. Basically any device in the same group as the GPU would be in a separate sub-group. Regards, Joerg
