Re: [RFC] iommufd: Use accurate dev_id in the PRI forwarding path

[Jason Gunthorpe <jgg@nvidia.com>]

On Thu, Mar 06, 2025 at 01:01:30PM +0800, Yi Liu wrote:
> On 2025/3/6 12:10, Tian, Kevin wrote:
> > > From: Liu, Yi L <yi.l.liu@intel.com>
> > > Sent: Thursday, March 6, 2025 11:30 AM
> > > 
> > > On 2025/3/6 11:10, Tian, Kevin wrote:
> > > > > From: Jason Gunthorpe <jgg@nvidia.com>
> > > > > Sent: Thursday, March 6, 2025 1:19 AM
> > > > > 
> > > > > I'm confused about how that can even work if the RID is erased in the
> > > > > fabric, how do we deliver the PRI response to the correct entity?
> > > > > 
> > > 
> > > If the RID is detached, kernel should response any pending PRIs. If user
> > > sends a response, it won't suit any pending PRI then dropped.
> > > 
> > 
> > Jason talked about the case where the iommu group is created e.g
> > for a legacy PCI bridge, with all devices behind sharing a same RID.
> > Then there is no way to know which device should receive the response.
> 
> hmmm. such devices shall not support PRI I suppose. But yes, this also
> pushes us to not supporting PRI for devices that have alias. We might
> need an iommu API to count the device number of a given iommu group.

iommu groups are not the same thing as aliases..

Aliased devices go into groups, but groups are not just consistent of
aliased devices..

Jason