Re: [RFC PATCH 3/6] iommu/amd: Introduce Guest-ID struct amd_iommu_vminfo

[Suthikulpanit, Suravee]

On 1/5/2024 9:38 PM, Jason Gunthorpe wrote:
> On Fri, Jan 05, 2024 at 08:39:56PM +0700, Suthikulpanit, Suravee wrote:
> > Hi Kevin
> >
> > On 12/15/2023 2:35 PM, Tian, Kevin wrote:
> > > > From: Suravee Suthikulpanit <suravee.suthikulpanit@xxxxxxx>
> > > > Sent: Wednesday, December 13, 2023 12:02 AM
> > > >
> > > > AMD HW-vIOMMU feature requires IOMMU driver to specify a unique 16-bit
> > > > Guest ID (GID) for each VM. This ID is used to index into various
> > > > data structures for configuring the hardware.
> > > >
> > > > Introduce amd_iommu_vminfo_hash hashtable to store per-vm
> > > > configuration,
> > > > which uses 16-bit GID as a hash key along with helper functions.
> > >
> > > somehow it's unclear to me whether this series is only for hw
> > > supporting vf or broader hw supporting nested capability. for
> > > the latter case is GID still necessary?
> >
> > I am restructuring the series and might be moving GID stuff until later when
> > introduce broader hw support for AMD vIOMMU.
>
> I'm hoping you can just skip enabling the viommu features and still
> have nesting? That should be OK right? The SW will manage the
> invalidations.

Correct, the part 1 only add support for nested translation (w/o 
HW-vIOMMU feature). So, SW would need to manage the invalidation.

Suravee