Re: [PATCH 03/11] iommu: Compute iommu_groups properly for PCIe switches

[Alex Williamson <alex.williamson@redhat.com>]

On Mon, 22 Sep 2025 20:51:31 -0400
Donald Dutile <ddutile@redhat.com> wrote:

> On 9/22/25 7:15 PM, Jason Gunthorpe wrote:
> > On Mon, Sep 22, 2025 at 04:32:00PM -0600, Alex Williamson wrote:  
> >> The ACS capability was only introduced in PCIe 2.0 and vendors have
> >> only become more diligent about implementing it as it's become
> >> important for device isolation and assignment.  
> PCIe-2.0 spec-wise, was released in 2007, 18 years ago.
> If hw is on a 3-yr lifecycle, that's 6 generations (7th including this year releases, assuming
> gen-1 was 2007); assuming a 5yr hw cycle, that's 4 generations of hardware.
> 
> Maybe a more interesting date is when DC servers implemented device-assignment/SRIOV
> in full-scale, and then, determine number of hw generations from that point on as
> 'learning -> devel-changing' years.
> I recall we had in in 'enterprise' customers in 2010, which only shaves one generation
> from above counts.

I don't see the relevance of these timelines.  A vendor with their head
in the sand still has their head in the sand regardless of time
passing.  Device assignment has a heavy non-enterprise user base.

> > IDK about this, I have very new systems and they still not have ACS
> > flags according to this interpretation.
> >   
> >> IMO, we can't assume anything at all about a multifunction device
> >> that does not implement ACS.  
> > 
> > Yeah this is all true.
> > 
> > But we are already assuming. Today we assume MFDs without caps must
> > have internal loopback in some cases, and then in other cases we
> > assume they don't.
> > 
> > I've sent and people have tested various different rules - please tell
> > me what you can live with.
> > 
> > Assuming the MFD does not have internal loopback, while not entirely
> > satisfactory, is the one that gives the least practical breakage.
> > 
> > I think it most accurately reflects the majority of real hardware out
> > there.
> > 
> > We can quirk to fix the remainder.
> > 
> > This is the best plan I've got..
> > 
> > Jason
> >   
> 
> +1 to Jason's conclusions.
> We should design the quirk hook to add ACS hooks for MFDs that do
> not adhere to the spec., which should be the minority, and that's what
> quirks are suppose to handle -- the odd cases.

Sorry, I can't agree.  I think we're conflating lack of a specific ACS
p2p capability to imply lack of internal p2p with lack of an ACS
capability at all.  I don't believe we can infer anything from the
latter.  Thanks,

Alex