On Thu, Aug 15, 2024 at 08:24:05PM -0300, Jason Gunthorpe wrote:
On Wed, Aug 07, 2024 at 01:10:49PM -0700, Nicolin Chen wrote:
@@ -946,4 +947,40 @@ struct iommu_viommu_unset_vdev_id { __aligned_u64 vdev_id; }; #define IOMMU_VIOMMU_UNSET_VDEV_ID _IO(IOMMUFD_TYPE, IOMMUFD_CMD_VIOMMU_UNSET_VDEV_ID)
+/**
- enum iommu_viommu_invalidate_data_type - VIOMMU Cache Invalidate Data Type
- @IOMMU_VIOMMU_INVALIDATE_DATA_ARM_SMMUV3: Invalidation data for ARM SMMUv3
- */
+enum iommu_viommu_invalidate_data_type {
- IOMMU_VIOMMU_INVALIDATE_DATA_ARM_SMMUV3,
+};
=1 here I think. Lets try to avoid 0 for the types..
And this shouldn't be in this patch
But also we can probably just use reuse enum iommu_hwpt_invalidate_data_type here?
Would that force IOMMU drivers to implement both hwpt and viommu invalidations? SMMUv3 driver would implement both anyway though..
+struct iommu_viommu_invalidate {
- __u32 size;
- __u32 viommu_id;
- __aligned_u64 data_uptr;
- __u32 data_type;
- __u32 entry_len;
- __u32 entry_num;
- __u32 __reserved;
+}; +#define IOMMU_VIOMMU_INVALIDATE _IO(IOMMUFD_TYPE, IOMMUFD_CMD_VIOMMU_INVALIDATE)
I wonder if we should use IOMMU_HWPT_INVALIDATE? the hwpt_id can tell which mode it is in. The ioctl becomes badly named but these have identical arguments.
Mostly they would be identical. So, I think that's doable.
If someday we need something hwpt-specific or viommu-specific, we can always duplicate a different structure.
Thanks Nicolin