Re: [PATCH v5 06/13] iommufd: Allow pt_id to carry viommu_id for IOMMU_HWPT_ALLOC

[Nicolin Chen <nicolinc@nvidia.com>]

On Tue, Oct 29, 2024 at 12:27:46PM -0300, Jason Gunthorpe wrote:
> On Mon, Oct 28, 2024 at 07:52:10AM -0700, Nicolin Chen wrote:
> > On Mon, Oct 28, 2024 at 10:03:09AM -0300, Jason Gunthorpe wrote:
> > In iommufd_hwpt_paging_alloc(), we reject IOMMU_HWPT_FAULT_ID_VALID:
> > 	const u32 valid_flags = IOMMU_HWPT_ALLOC_NEST_PARENT |
> > 				IOMMU_HWPT_ALLOC_DIRTY_TRACKING;
> > 	...
> > 	if (flags & ~valid_flags)
> > 		return ERR_PTR(-EOPNOTSUPP);
> > 
> > In iommufd_hwpt_nested_alloc(), we mask the flag away:
> > 	if ((flags & ~IOMMU_HWPT_FAULT_ID_VALID) ||
> > 	    !user_data->len || !ops->domain_alloc_user)
> > 		return ERR_PTR(-EOPNOTSUPP);
> > 	...
> > 	hwpt->domain = ops->domain_alloc_user(idev->dev,
> > 					      flags & ~IOMMU_HWPT_FAULT_ID_VALID,
> > 					      parent->common.domain, user_data);
> > 
> > Then, in the common function it has a section of
> > 	if (cmd->flags & IOMMU_HWPT_FAULT_ID_VALID) {
> > 	...
> > 
> > It seems that this IOMMU_HWPT_FAULT_ID_VALID is for nested domains?
> 
> OK, but ARM should be blocking it since it doesn't work there.
> 
> I think we made some error here, it should have been passed in flags
> to the drivers and only intel should have accepted it.

Trying to limit changes here since two parts are already quite
large, I think a separate series fixing that would be nicer? 

> This suggests we should send flags down the viommu alloc domain path too.

Ack. Will pass it in.

Thanks
Nicolin