Re: [PATCH v3 5/8] iommufd: Associate fault object with iommufd_hw_pgtable

[Jason Gunthorpe <jgg@ziepe.ca>]

On Fri, Mar 15, 2024 at 09:16:43AM +0800, Baolu Lu wrote:
> On 3/9/24 3:05 AM, Jason Gunthorpe wrote:
> > On Mon, Jan 22, 2024 at 03:39:00PM +0800, Lu Baolu wrote:
> > 
> > > @@ -411,6 +414,8 @@ enum iommu_hwpt_data_type {
> > >    * @__reserved: Must be 0
> > >    * @data_type: One of enum iommu_hwpt_data_type
> > >    * @data_len: Length of the type specific data
> > > + * @fault_id: The ID of IOMMUFD_FAULT object. Valid only if flags field of
> > > + *            IOMMU_HWPT_FAULT_ID_VALID is set.
> > >    * @data_uptr: User pointer to the type specific data
> > >    *
> > >    * Explicitly allocate a hardware page table object. This is the same object
> > > @@ -441,6 +446,7 @@ struct iommu_hwpt_alloc {
> > >   	__u32 __reserved;
> > >   	__u32 data_type;
> > >   	__u32 data_len;
> > > +	__u32 fault_id;
> > >   	__aligned_u64 data_uptr;
> > >   };
> > 
> > ?? We can't add fault_id in the middle of the struct??
> 
> Yes. I should add the new field at the end.
> 
> By the way, with a __u32 added, this data structure is not 64-byte-
> aligned anymore. Do we need to add another unused u32 entry, or just let
> the compiler handle it?

Yes, add a reserved u32 to ensure the structs is always without
implicit padding.

> > 
> > > +	if (cmd->flags & IOMMU_HWPT_FAULT_ID_VALID) {
> > > +		struct iommufd_fault *fault;
> > > +
> > > +		fault = iommufd_get_fault(ucmd, cmd->fault_id);
> > > +		if (IS_ERR(fault)) {
> > > +			rc = PTR_ERR(fault);
> > > +			goto out_hwpt;
> > > +		}
> > > +		hwpt->fault = fault;
> > > +		hwpt->domain->iopf_handler = iommufd_fault_iopf_handler;
> > > +		hwpt->domain->fault_data = hwpt;
> > > +		hwpt->fault_capable = true;
> > 
> > I wonder if there should be an iommu API to make a domain fault
> > capable?
> 
> The iommu core identifies a fault-capable domain by checking its
> domain->iopf_handler. Anyway, what's the difference between a fault or
> non-fault capable domain from iommu core's point of view?

From the core? Nothing. I'm just wondering from an API perspective if
we should have a little inline to indicate it.

Jason