Re: [PATCH v5 7/9] iommufd: Associate fault object with iommufd_hw_pgtable

[Jason Gunthorpe]

On Mon, May 20, 2024 at 03:39:54AM +0000, Tian, Kevin wrote:
> > From: Baolu Lu <baolu.lu@xxxxxxxxxxxxxxx>
> > Sent: Monday, May 20, 2024 10:19 AM
> > 
> > On 5/15/24 4:50 PM, Tian, Kevin wrote:
> > >> From: Lu Baolu <baolu.lu@xxxxxxxxxxxxxxx>
> > >> Sent: Tuesday, April 30, 2024 10:57 PM
> > >>
> > >> @@ -308,6 +314,19 @@ int iommufd_hwpt_alloc(struct iommufd_ucmd
> > >> *ucmd)
> > >>   		goto out_put_pt;
> > >>   	}
> > >>
> > >> +	if (cmd->flags & IOMMU_HWPT_FAULT_ID_VALID) {
> > >> +		struct iommufd_fault *fault;
> > >> +
> > >> +		fault = iommufd_get_fault(ucmd, cmd->fault_id);
> > >> +		if (IS_ERR(fault)) {
> > >> +			rc = PTR_ERR(fault);
> > >> +			goto out_hwpt;
> > >> +		}
> > >> +		hwpt->fault = fault;
> > >> +		hwpt->domain->iopf_handler = iommufd_fault_iopf_handler;
> > >> +		hwpt->domain->fault_data = hwpt;
> > >> +	}
> > >
> > > this is nesting specific. why not moving it to the nested_alloc()?
> > 
> > Nesting is currently a use case for userspace I/O page faults, but this
> > design should be general enough to support other scenarios as well.
> 
> Do we allow user page table w/o nesting?
> 
> What would be a scenario in which the user doesn't manage the
> page table but still want to handle the I/O page fault? The fault
> should always be delivered to the owner managing the page table...

userspace always manages the page table, either it updates the IOPTE
directly in a nest or it calls iommufd map operations.

Ideally the driver will allow PRI on normal cases, although it will
probably never be used.

Jason