RE: [PATCH v5 6/9] iommufd: Fault-capable hwpt attach/detach/replace

[Tian, Kevin]

> From: Baolu Lu <baolu.lu@xxxxxxxxxxxxxxx>
> Sent: Monday, May 20, 2024 10:10 AM
> 
> On 5/15/24 4:43 PM, Tian, Kevin wrote:
> >> From: Lu Baolu <baolu.lu@xxxxxxxxxxxxxxx>
> >> Sent: Tuesday, April 30, 2024 10:57 PM
> >> +
> >> +int iommufd_fault_domain_replace_dev(struct iommufd_device *idev,
> >> +				     struct iommufd_hw_pagetable *hwpt,
> >> +				     struct iommufd_hw_pagetable *old)
> >> +{
> >> +	struct iommu_attach_handle *handle;
> >> +	int ret;
> >> +
> >> +	if (hwpt->fault)
> >> +		ret = iommufd_fault_iopf_enable(idev);
> >> +	else
> >> +		iommufd_fault_iopf_disable(idev);
> >> +
> >> +	ret = iommu_group_replace_domain(idev->igroup->group, hwpt-
> >>> domain);
> >> +	if (ret)
> >> +		goto out_cleanup;
> >> +
> >> +	iommufd_auto_response_faults(old, idev);
> >> +	handle = iommu_attach_handle_get(idev->igroup->group,
> >> IOMMU_NO_PASID, 0);
> >> +	handle->idev = idev;
> >
> > why is auto response required in replace? new requests can come
> > after the auto response anyway...
> >
> > The user should prepare for faults delivered to the old or new hwpt
> > in the transition window.
> 
> The current design of replace allows switching between one that is not
> IOPF-capable and one that is. This implies that if we switch from an
> IOPF-capable hwpt to a non-IOPF-capable one, the response queue needs to
> be auto responded.
> 

then do it only for that scenario?