Re: [PATCH v4 18/23] iommu/arm-smmu-v3-iommufd: Support implementation-defined hw_info

[Nicolin Chen <nicolinc@nvidia.com>]

On Thu, May 15, 2025 at 04:23:29PM -0300, Jason Gunthorpe wrote:
> On Thu, May 15, 2025 at 12:21:17PM -0700, Nicolin Chen wrote:
> > On Thu, May 15, 2025 at 03:56:29PM -0300, Jason Gunthorpe wrote:
> > > On Thu, May 15, 2025 at 11:52:05AM -0700, Nicolin Chen wrote:
> > > > On Thu, May 15, 2025 at 02:17:06PM -0300, Jason Gunthorpe wrote:
> > > > > On Thu, May 08, 2025 at 08:02:39PM -0700, Nicolin Chen wrote:
> > > > Should the first call return out_data_type=CMDQV while returning
> > > > the arm_smmu_v3 hw_info data? Otherwise, VMM wouldn't know what
> > > > to set in the input sub_data_type of the 2nd ioctl?
> > > 
> > > No, either set a flag in the smmu_v3 hw_info, as you were doing here,
> > > or just have the vmm probe it. Given the VMM is likely to be told to
> > > run in vCMDQ mode on the command line try-and-fail doesn't sound so
> > > bad.
> > > 
> > > And I guess we don't need a "sub type" just a "requested type" where 0
> > > means return the best one and non-zero means return a specific one or
> > > fail with EOPNOTSUPP.
> > 
> > OK. I think this would work:
> > hw_info (req_type=0)     => out_data_type=SMMU_V3, flags=HAS_CMDQV
> > hw_info (req_type=CMDQV) => out_data_type=CMDQV,   flags=0
> 
> Yeah
> 
> > Or, would it be simpler by having a sub_data_uptr:
> > hw_info => out_data_type=SMMU_V3, sub_data_type=CMDQV,
> >            data_uptr=iommu_hw_info_arm_smmuv3,
> > 	   sub_data_uptr=iommu_hw_info_tegra241_cmdqv
> > ?
> 
> I think the former is simpler to code, you can just add the req_type
> to the signatures and if the driver comes back with a type != req_type
> the core code will return EOPNOTSUPP

OK.

Maybe just turn the out_data_type to be bidirectional?

Then we would only need to update the docs:
 /**
  * enum iommu_hw_info_type - IOMMU Hardware Info Types
- * @IOMMU_HW_INFO_TYPE_NONE: Used by the drivers that do not report hardware
- *                           info
+ * @IOMMU_HW_INFO_TYPE_NONE: (for output) used by the drivers that do not report
+ *                           hardware info
+ * @IOMMU_HW_INFO_TYPE_DEFAULT: (for input) Used to request the default type
  * @IOMMU_HW_INFO_TYPE_INTEL_VTD: Intel VT-d iommu info type
  * @IOMMU_HW_INFO_TYPE_ARM_SMMUV3: ARM SMMUv3 iommu info type
+ * @IOMMU_HW_INFO_TYPE_TEGRA241_CMDQV: Subtype of ARM SMMUv3 for Tegra241 CMDQV
  */
 enum iommu_hw_info_type {
 	IOMMU_HW_INFO_TYPE_NONE = 0,
+ 	IOMMU_HW_INFO_TYPE_DEFAULT = 0,
 	IOMMU_HW_INFO_TYPE_INTEL_VTD = 1,
 	IOMMU_HW_INFO_TYPE_ARM_SMMUV3 = 2,
+	IOMMU_HW_INFO_TYPE_TEGRA241_CMDQV =3,
 };

- * @out_data_type: Output the iommu hardware info type as defined in the enum
- *                 iommu_hw_info_type.
+ * @data_type: Bidirectional property.
+ *             Input the requested iommu hardware info type as defined in the
+ *             enum iommu_hw_info_type. Requesting IOMMU_HW_INFO_TYPE_DEFAULT
+ *             lets kernel pick the default type to output, otherwise kernel
+ *             will validate the input type and may reject with -EOPNOTSUPP.
+ *             Output the supported iommu hardware info type as defined in the
+ *             same enum iommu_hw_info_type

And similarly in the iommu API kdoc too.

> Finally we end up with only one ioctl enum number space for the
> types, which seems appealing.

Yea. Avoiding a sub enum is nicer.

Thanks
Nicolin