Re: [PATCH 5/6] iommu: init pasid array while doing domain_replace and iopf is active

[Joel Granados]

On Thu, Sep 12, 2024 at 12:17:35PM +0800, Baolu Lu wrote:
> On 9/11/24 5:55 PM, Joel Granados wrote:
> > On Thu, Sep 05, 2024 at 11:30:05AM +0800, Baolu Lu wrote:
> >> On 9/4/24 9:17 PM, Joel Granados via B4 Relay wrote:
> >>> From: Joel Granados<j.granados@xxxxxxxxxxx>
> >>>
> >>> iommu_report_device_fault expects a pasid array to have an
> >>> iommu_attach_handle when a fault is detected.
> >> The iommu_attach_handle is expected only when an iopf-capable domain is
> >> attached to the device or PASID. The iommu_report_device_fault() treats
> >> it as a fault when a fault occurs, but no iopf-capable domain is
> >> attached.
> > I don't follow. The way that I read it: if the pasid_array x-array does
> > not have an iommu_attach_handle indexed by either fault->prm.pasid or
> > IOMMU_NO_PASID, it will follow the err_bad_iopf and return -EINVAL
> > (please correct me if I'm wrong). So the iommu_attach_handle is*always*
> > expected.
> > 
> > Would it be more clear for it to be:
> > """
> > The iommu_report_device_fault function expects the pasid_array x-array
> > to have an iommu_attach_handle indexed by a PASID. Add one indexed with
> > IOMMU_NO_PASID when the replacing HWPT has a valid iommufd fault object.
> > Remove it when we release ownership of the group.
> 
> Can you please explain why iommu core needs to remove the attach handle
> when the ownership of the group is changed?

It does not. Probably left from a previous version of the patch. Sorry
for the noise. I have reamoved the xa_erase from
__iommu_release_dma_ownership. Will send a V2 once we finish discussing
the rest of your comments.

Best and thankyou for your feedback

-- 

Joel Granados