[PATCH 0/9] Remove SMMUv3 struct arm_smmu_cmdq_ent

[Pranjal Shrivastava]

On Fri, May 01, 2026 at 11:29:09AM -0300, Jason Gunthorpe wrote:
> [ This is part of the patch pile to move SMMUv3 over to the generic page
> table:
> 1) Introduction of new gather items and RISCV usage
>   https://patch.msgid.link/r/0-v1-54e7264d71b4+17cc3-iommu_riscv_inv_jgg@nvidia.com
> 2) Remove SMMUv3 struct arm_smmu_cmdq_ent
> 3) Organize the SMMUv3 invalidation flow so iommupt can use it
> 4) Use the generic iommu page table for SMMUv3
> 
> The whole branch is here:
>    https://github.com/jgunthorpe/linux/commits/iommu_pt_arm64/
> ]
> 
> The invalidation logic has this multi-step process where it first
> writes the command into a 32 byte struct arm_smmu_cmdq_ent, then it
> calls a function which converts it into a 16 byte HW struct, and
> sometimes it then edits the HW struct a little bit before passing it
> off to the batch or submission functions.
> 
> Instead just generate the HW struct directly by moving the FIELD_PREP
> blocks out of the big case statement and into helper functions. Call the
> right function in all the places that were building arm_smmu_cmdq_ent.
> 
> Add a type for the CMDQ entry similar to the STE/CD types that wraps the
> two u64s for clarity and use it everywhere.
> 
> This is intended to have no functional change. It makes the following
> patches work better and removes a bunch of LOC. I've run several AI tools
> with instruction to look for functional changes, which did find one subtle
> mistake in PRI response.
> 
> The removal of arm_smmu_cmdq_build_cmd() also achieves what Mostafa is
> doing in the pkvm series by making the command formation entirely header
> based with the arm_smmu_make_cmd_*() mini inlines.
> 
> This series has no dependencies. Several people have already tested this
> on various ARM systems along with the full iommupt conversion.
> 
> Jason Gunthorpe (9):
>   iommu/arm-smmu-v3: Add struct arm_smmu_cmd to represent the HW format
>     command
>   iommu/arm-smmu-v3: Use the HW arm_smmu_cmd in cmdq selection functions
>   iommu/arm-smmu-v3: Use the HW arm_smmu_cmd in cmdq submission
>     functions
>   iommu/arm-smmu-v3: Convert arm_smmu_cmdq_batch cmds to struct
>     arm_smmu_cmd
>   iommu/arm-smmu-v3: Remove CMDQ_OP_CFGI_CD_ALL from
>     arm_smmu_cmdq_build_cmd()
>   iommu/arm-smmu-v3: Directly encode simple commands
>   iommu/arm-smmu-v3: Directly encode CMDQ_OP_ATC_INV
>   iommu/arm-smmu-v3: Directly encode CMDQ_OP_SYNC
>   iommu/arm-smmu-v3: Directly encode TLBI commands
> 
>  .../arm/arm-smmu-v3/arm-smmu-v3-iommufd.c     |  24 +-
>  drivers/iommu/arm/arm-smmu-v3/arm-smmu-v3.c   | 475 +++++++-----------
>  drivers/iommu/arm/arm-smmu-v3/arm-smmu-v3.h   | 248 +++++----
>  .../iommu/arm/arm-smmu-v3/tegra241-cmdqv.c    |  16 +-
>  4 files changed, 350 insertions(+), 413 deletions(-)
> 

I was able to test this with 7.1-rc3 on an Arm server with the 
iova_stress test [1]. For the series:

Tested-by: Pranjal Shrivastava <praan at google.com>

Thanks,
Praan

[1] https://github.com/soleen/iova_stress