> From: Jason Gunthorpe <jgg@xxxxxxxxxx> > Sent: Saturday, April 16, 2022 5:54 AM > > On Fri, Apr 15, 2022 at 03:57:14AM +0000, Tian, Kevin wrote: > > > From: Jason Gunthorpe <jgg@xxxxxxxxxx> > > > Sent: Friday, April 15, 2022 2:46 AM > > > > > > kvm and VFIO need to be coupled together however neither is willing to > > > tolerate a direct module dependency. Instead when kvm is given a VFIO > FD > > > it uses many symbol_get()'s to access VFIO. > > > > > > Provide a single VFIO function vfio_file_get_ops() which validates the > > > given struct file * is a VFIO file and then returns a struct of ops. > > > > VFIO has multiple files (container, group, and device). Here and other > > places seems to assume a VFIO file is just a group file. While it is correct > > in this external facing context, probably calling it 'VFIO group file' is > > clearer in various code comments and patch descriptions. > > > > > > > > Following patches will redo each of the symbol_get() calls into an > > > indirection through this ops struct. > > > > > > Signed-off-by: Jason Gunthorpe <jgg@xxxxxxxxxx> > > > > > > Reviewed-by: Kevin Tian <kevin.tian@xxxxxxxxx> > > > > Out of curiosity, how do you envision when iommufd is introduced? > > Will we need a generic ops abstraction so both vfio and iommufd > > register their own ops to keep kvm side generic or a new protocol > > will be introduced between iommufd and kvm? > > I imagine using the vfio_device in all these context where the vfio > group is used, not iommufd. This keeps everything internal to vfio. > In this case although the uAPI is called KVM_DEV_VFIO_GROUP_ADD Qemu will pass in a device fd and with this series KVM doesn't care whether it's actually a device or group and just use struct file to call vfio_file_ops. correct? You probably remember there is one additional requirement when adding ENQCMD virtualization on Intel platform. KVM is required to setup a guest PASID to host PASID translation table in CPU vmcs structure to support ENQCMD in the guest. Following above direction I suppose KVM will provide a new interface to allow user pass in [devfd, iommufd, guest_pasid] and then call a new vfio ops e.g. vfio_file_translate_guest_pasid(dev_file, iommufd, gpasid) to retrieve the host pasid. This sounds correct in concept as iommufd only knows host pasid and any g->h information is managed by vfio device driver. Does it also make sense to you? Just want to think forward a bit to make the whole picture clearer. Thanks Kevin
