Re: [PATCH] vfio: Support VFIO_NOIOMMU with iommufd

[Jason Gunthorpe <jgg@nvidia.com>]

On Mon, Jan 09, 2023 at 04:34:34PM -0700, Alex Williamson wrote:
> On Mon,  9 Jan 2023 10:22:59 -0400
> Jason Gunthorpe <jgg@nvidia.com> wrote:
> 
> > Add a small amount of emulation to vfio_compat to accept the SET_IOMMU
> > to VFIO_NOIOMMU_IOMMU and have vfio just ignore iommufd if it is working
> > on a no-iommu enabled device.
> >
> > Signed-off-by: Jason Gunthorpe <jgg@nvidia.com>
> > ---
> >  drivers/iommu/iommufd/Kconfig       |  2 +-
> >  drivers/iommu/iommufd/vfio_compat.c | 46 ++++++++++++++++++++++++-----
> >  drivers/vfio/group.c                | 13 ++++----
> >  drivers/vfio/iommufd.c              | 21 ++++++++++++-
> >  include/linux/iommufd.h             |  6 ++--
> >  5 files changed, 70 insertions(+), 18 deletions(-)
> > 
> > This needs a testing confirmation with dpdk to go forward, thanks
> 
> How do we create a noiommu group w/o the vfio_noiommu flag that's
> provided by container.c?

Ah, the module option is now in the wrong place, I'll move it to
vfio_main.c

> Even without dpdk, you should be able to turn off the system IOMMU
> and get something bound to vfio-pci that still taints the kernel and
> provides a noiommu-%d group under /dev/vfio/.  There's a rudimentary
> unit test for noiommu here[1].  Thanks,

Thanks, I'll check it

Jason