Re: [PATCH rc v2] iommu/arm-smmu-v3: Fix inconsistent ATS state tracking

[Pranjal Shrivastava <praan@google.com>]

On Tue, May 05, 2026 at 02:44:27PM -0700, Nicolin Chen wrote:
> On Tue, May 05, 2026 at 09:23:15PM +0000, Pranjal Shrivastava wrote:
> > On Tue, May 05, 2026 at 01:21:37PM -0700, Nicolin Chen wrote:
> > > /**
> > >  * pci_prepare_ats - Setup the PS for ATS
> > >  * @dev: the PCI device
> > >  * @ps: the IOMMU page shift
> > >  *
> > >  * This must be done by the IOMMU driver on the PF before any VFs are created to
> > >  * ensure that the VF can have ATS enabled.
> > > 
> > > But it returns 0 directly on dev->is_virtfn.
> > > 
> > > I wonder if this function itself should just fail if dev->is_virtfn
> > > and !pci_physfn(dev)->ats_stu?
> > > 
> > 
> > Just wondering if there's a use-case that might break?
> 
> Well, even a break would be what the kdoc statement demands :-/
> 

Yea

> > Otherwise, in
> > current context I agree that we should fail if ats_stu == 0 / mis-match
> 
> Yea, should check mis-match. "== 0" doesn't guarantee the enable
> success.
> 
> 
> > > Then, arm_smmu_probe_device() must fail if pci_prepare_ats() fails.
> > 
> > I don't think probe device should fail, we should just mark that ATS
> > isn't supported for the function? right?
> 
> PF must be probed() first with ats_stu set correctly. If VF gets
> probed before that or with a different ats_stu, something must be
> wrong.
> 
> Again, following the kdoc statement, VF should not be "created"
> in this case. Or should we read it differently? :-/
> 

Yea, VFs shouldn't be created without PFs.

Praan