[PATCH RFCv1 12/14] iommufd: Add IOMMUFD_OBJ_VQUEUE and IOMMUFD_CMD_VQUEUE_ALLOC

[Nicolin Chen]

On Fri, May 24, 2024 at 05:26:50AM +0000, Tian, Kevin wrote:
> External email: Use caution opening links or attachments
> 
> 
> > From: Nicolin Chen <nicolinc at nvidia.com>
> > Sent: Friday, May 24, 2024 12:42 PM
> >
> > On Thu, May 23, 2024 at 06:57:15AM +0000, Tian, Kevin wrote:
> > > > From: Jason Gunthorpe <jgg at nvidia.com>
> > > > Sent: Sunday, May 12, 2024 11:02 PM
> > > >
> > > > On Fri, Apr 12, 2024 at 08:47:09PM -0700, Nicolin Chen wrote:
> > > >
> > > > > +/**
> > > > > + * struct iommu_vqueue_alloc - ioctl(IOMMU_VQUEUE_ALLOC)
> > > > > + * @size: sizeof(struct iommu_vqueue_alloc)
> > > > > + * @flags: Must be 0
> > > > > + * @viommu_id: viommu ID to associate the virtual queue with
> > > > > + * @out_vqueue_id: The ID of the new virtual queue
> > > > > + * @data_type: One of enum iommu_vqueue_data_type
> > > > > + * @data_len: Length of the type specific data
> > > > > + * @data_uptr: User pointer to the type specific data
> > > > > + *
> > > > > + * Allocate an virtual queue object for driver-specific HW-accelerated
> > > > queue
> > > > > + */
> > > > > +
> > > > > +struct iommu_vqueue_alloc {
> > > > > +   __u32 size;
> > > > > +   __u32 flags;
> > > > > +   __u32 viommu_id;
> > > > > +   __u32 out_vqueue_id;
> > > > > +   __u32 data_type;
> > > > > +   __u32 data_len;
> > > > > +   __aligned_u64 data_uptr;
> > > >
> > > > Some of the iommus will want an IPA here not a user pointer. I think
> > > > it is fine API wise, we'd just add a flag to indicate data_uptr is an
> > > > IPA.
> > > >
> > >
> > > Presumably each driver will create its own type data which can
> > > include any IPA field as vcmdq_base in this patch?
> >
> > You mean putting a base address field in this common vqueue
> > structure? Hmm, I think we could, assuming every queue must
> > have a set of base and size info in the guest level.
> >
> 
> We could, but my original point was just that it's confusing to
> change the meaning of data_uptr. Let's stick it to be driver defined
> data and any driver-specific IPA field is defined in that data.

Oh, that's copied from IOMMU_HWPT_ALLOC:
* @data_uptr: User pointer to the type specific data

And it actually makes sense to me to be "type specific" since a
driver (AMD VIOMMU for example) might have two different vqueue
types (invalidation queue v.s. fault queue)?

Thanks
Nicolin