Re: [PATCH v8 07/10] iommufd: Fault-capable hwpt attach/detach/replace

[Jason Gunthorpe <jgg@nvidia.com>]

On Tue, Oct 22, 2024 at 10:30:10PM +0800, Zhangfei Gao wrote:
> On Fri, 18 Oct 2024 at 21:53, Jason Gunthorpe <jgg@nvidia.com> wrote:
> >
> > On Wed, Oct 16, 2024 at 12:25:03PM -0300, Jason Gunthorpe wrote:
> > > > > smmu-v3 needs some more fixing to move that
> > > > > arm_smmu_master_enable_sva() logic into domain attachment.
> > > >
> > > > Will think about this, Thanks Jason
> > >
> > > Can you test it if a patch is made?
> >
> > Here it is:
> >
> > https://github.com/jgunthorpe/linux/commits/smmuv3_nesting/
> >
> > fa1528253d2210 iommu: Remove IOMMU_DEV_FEAT_SVA
> > 5675560a272cf5 iommu/vt-d: Check if SVA is supported when attaching the SVA domain
> > 94bc2b9525b508 iommu/arm-smmu-v3: Put iopf enablement in the domain attach path
> >
> > Let me know..
> 
> With these patches, do we still need ops->user_pasid_table?

It makes no change - you need user_pasid_table to make
IOMMU_DOMAIN_NESTED work.

If you aren't using IOMMU_DOMAIN_NESTED you shouldn't need it.

> if (fault->prm.flags & IOMMU_FAULT_PAGE_REQUEST_PASID_VALID) {
>                 attach_handle = iommu_attach_handle_get(dev->iommu_group,
>                                 fault->prm.pasid, 0);
> 
> // is attach_handle expected effect value here?
>                 if (IS_ERR(attach_handle)) {
>                         const struct iommu_ops *ops = dev_iommu_ops(dev);
>
>                         if (!ops->user_pasid_table)
>                                 return NULL;
>                         /*
>                          * The iommu driver for this device supports user-
>                          * managed PASID table. Therefore page faults for
>                          * any PASID should go through the NESTING domain
>                          * attached to the device RID.
>                          */
>                         attach_handle = iommu_attach_handle_get(
>                                         dev->iommu_group, IOMMU_NO_PASID,
>                                         IOMMU_DOMAIN_NESTED);
> 
> Now I still need set ops->user_pasid_table, since attach_handle can not
> return from the first iommu_attach_handle_get with fault->prm.pasid = 1,
> but the second iommu_attach_handle_get with  IOMMU_NO_PASID,
> suppose it is not expected?

The second handle_get will always fail unless you are using
IOMMU_DOMAIN_NESTED in userspace with iommufd.

What testing are you doing exactly?

Jason