Re: [PATCH v4 07/16] iommu/amd: Introduce per-device domain ID to workaround potential TLB aliasing issue

[Jason Gunthorpe <jgg@ziepe.ca>]

On Tue, Jan 16, 2024 at 10:38:16PM +0530, Vasant Hegde wrote:
> Joson,
> 
> 
> On 1/16/2024 7:30 PM, Jason Gunthorpe wrote:
> > On Tue, Jan 16, 2024 at 04:22:16PM +0530, Vasant Hegde wrote:
> > 
> >>> The *device* has a list of PASID's that point to iommu_domains. This
> >>> is stored in an xarray inside the iommu_group.
> >>>
> >>> The *iommu_domain* has a list of *devices & PASIDs* that can use this
> >>> domain for translation (ie that it was attached to)
> >>
> >> IIUC domain will be having device/PASID combination something like below:
> >>
> >> UNMANAGED_DOMAIN_A with IO Page table :
> >>   - DevA + PASID zero (say non PASID capable device)
> >>   - devB + PASID <M>
> >>   - devC + PASID <N>
> >>
> >> We will *not* have same device with different PASID in same domain like
> >>   - DevB + PASID M
> >>   - DevB + PASID N
> >>
> >> Is that correct?
> > 
> > No, all combinations are allowed in the API.
> 
> What is the use case for same device with two or more different PASID to point
> to same page table?

Possibly something with SIOV like techniques where you have two
virtual functions, each being DMA isolated with PASID, installed into
the same VM. The guest physical map would be the same iommu_domain but
the PASID would be different.

Regardless, the API doesn't require this and I don't want to see
drivers inventing their own restrictions without a reason..

Jason