[PATCH v2 7/9] iommu/arm-smmu-v3: Directly encode CMDQ_OP_ATC_INV

[Nicolin Chen]

On Wed, May 13, 2026 at 08:57:46PM -0300, Jason Gunthorpe wrote:
> Add a new command make function and convert all the places using
> ATC_INV.
> 
> Split out full invalidation to directly make the cmd instead of
> overloading size=0 to mean full invalidation.
> 
> In section "3.9.1 ATS Interface" of F.b the specification says:
> 
>   When the SMMU returns an ATS Translation Completion for a request that
>   had a PASID, the Global bit of the Translation Completion Data Entry
>   must be zero.
> 
> Even though it faithfully forwards the G bit through to the ATS
> invalidation command there is no way to create G mappings so there is
> never any need to send a G invalidation. Thus don't expose global in the
> new helpers and leave CMDQ_ATC_0_GLOBAL unused.
> 
> Reviewed-by: Mostafa Saleh <smostafa at google.com>
> Reviewed-by: Pranjal Shrivastava <praan at google.com>
> Tested-by: Pranjal Shrivastava <praan at google.com>
> Tested-by: Mostafa Saleh <smostafa at google.com>
> Signed-off-by: Jason Gunthorpe <jgg at nvidia.com>

Reviewed-by: Nicolin Chen <nicolinc at nvidia.com>

With a nit ..

> -static void
> -arm_smmu_atc_inv_to_cmd(int ssid, unsigned long iova, size_t size,
> -			struct arm_smmu_cmdq_ent *cmd)
> +static struct arm_smmu_cmd
> +arm_smmu_atc_inv_to_cmd(u32 sid, int ssid, unsigned long iova, size_t size)
>  {
>  	size_t log2_span;
>  	size_t span_mask;
> @@ -2395,17 +2386,6 @@ arm_smmu_atc_inv_to_cmd(int ssid, unsigned long iova, size_t size,
>  	 * This has the unpleasant side-effect of invalidating all PASID-tagged
>  	 * ATC entries within the address range.
>  	 */
> -	*cmd = (struct arm_smmu_cmdq_ent) {
> -		.opcode			= CMDQ_OP_ATC_INV,
> -		.substream_valid	= (ssid != IOMMU_NO_PASID),
> -		.atc.ssid		= ssid,
> -	};
> -
> -	if (!size) {
> -		cmd->atc.size = ATC_INV_SIZE_ALL;
> -		return;
> -	}

.. should !size jump to arm_smmu_make_cmd_atc_inv_all()?

Or maybe add a note to disallow this function being used with !size
and WARN_ON? The function name itself doesn't sound very restricted
though.

Nicolin