Re: [PATCH v6 07/10] iommufd: Add a nested HW pagetable object

[Yi Liu]

On 2023/10/25 02:19, Nicolin Chen wrote:
> On Tue, Oct 24, 2023 at 03:00:49PM -0300, Jason Gunthorpe wrote:
> > On Tue, Oct 24, 2023 at 10:50:58AM -0700, Nicolin Chen wrote:
> >
> > > > The point is for the user_data to always be available, the driver
> > > > needs to check it if it is passed.
> > > >
> > > > This should all be plumbed to allow drivers to also customize their
> > > > paging domains too.
> > >
> > > We don't have a use case of customizing the paging domains.
> > > And our selftest isn't covering this path. Nor the case is
> > > supported by the uAPI:
> >
> > But this is the design, it is why everything is setup like this - we
> > didn't create a new op to allocate nesting domains, we made a flexible
> > user allocator.
> >
> > > 458- * A kernel-managed HWPT will be created with the mappings from the given
> > > 459- * IOAS via the @pt_id. The @data_type for this allocation must be set to
> > > 460: * IOMMU_HWPT_DATA_NONE. The HWPT can be allocated as a parent HWPT for a
> > > 461- * nesting configuration by passing IOMMU_HWPT_ALLOC_NEST_PARENT via @flags.
> > > 462- *
> >
> > Yes, that is the reality today. If someone comes to use the more
> > complete interface they need to fix that comment..
>
> Ack.
>
> > > Also, if we do passing in the data, we'd need to...
> >   
> > > 280-static struct iommu_domain *
> > > 281-mock_domain_alloc_user(struct device *dev, u32 flags,
> > > 282-		       struct iommu_domain *parent,
> > > 283:		       const struct iommu_user_data *user_data)
> > > 284-{
> > > 285-	struct mock_iommu_domain *mock_parent;
> > > 286-	struct iommu_hwpt_selftest user_cfg;
> > > 287-	int rc;
> > > 288-
> > > 289:	if (!user_data) {	/* must be mock_domain */
> > >
> > > ...change this to if (!parent)...
> >
> > Yes, this logic is not ideal. The parent is the request for nesting,
> > not user_data. user_data is the generic creation parameters, which are
> > not supported outside nesting
> >   
> > Like this:
> >
> > --- a/drivers/iommu/iommufd/selftest.c
> > +++ b/drivers/iommu/iommufd/selftest.c
> > @@ -286,14 +286,12 @@ mock_domain_alloc_user(struct device *dev, u32 flags,
> >          int rc;
> >   
> >          /* must be mock_domain */
> > -       if (!user_data) {
> > +       if (!parent) {
> >                  struct mock_dev *mdev = container_of(dev, struct mock_dev, dev);
> >                  bool has_dirty_flag = flags & IOMMU_HWPT_ALLOC_DIRTY_TRACKING;
> >                  bool no_dirty_ops = mdev->flags & MOCK_FLAGS_DEVICE_NO_DIRTY;
> >   
> > -               if (parent)
> > -                       return ERR_PTR(-EINVAL);
> > -               if (has_dirty_flag && no_dirty_ops)
> > +               if (user_data || (has_dirty_flag && no_dirty_ops))
> >                          return ERR_PTR(-EOPNOTSUPP);
> >                  return __mock_domain_alloc_paging(IOMMU_DOMAIN_UNMANAGED,
> >                                                    has_dirty_flag);
>
> Yea.. Then the vt-d driver needs a similar change too (@Yi) as I
> found it almost doing the same:
> https://lore.kernel.org/linux-iommu/20231024151412.50046-8-yi.l.liu@xxxxxxxxx/
yes. mock driver is kind of sample code, so the intel iommu driver is doing
almost the same thing. will follow up with the branch Jason shared.

--
Regards,
Yi Liu