Re: [PATCH 1/1] iommu/arm-smmu: Implement qcom,skip-init

[Will Deacon]

[Adding Bjorn, Jordan and John because I really don't want a bunch of
different ways to tell the driver that the firmware is screwing things up]

On Sat, Jul 04, 2020 at 02:28:09PM +0200, Konrad Dybcio wrote:
> This adds the downstream property required to support
> SMMUs on SDM630 and other platforms (the need for it
> most likely depends on firmware configuration).
> 
> Signed-off-by: Konrad Dybcio <konradybcio@xxxxxxxxx>
> ---
>  .../devicetree/bindings/iommu/arm,smmu.yaml       | 10 ++++++++++
>  drivers/iommu/arm-smmu.c                          | 15 +++++++++------
>  2 files changed, 19 insertions(+), 6 deletions(-)
> 
> diff --git a/Documentation/devicetree/bindings/iommu/arm,smmu.yaml b/Documentation/devicetree/bindings/iommu/arm,smmu.yaml
> index d7ceb4c34423..9abd6d41a32c 100644
> --- a/Documentation/devicetree/bindings/iommu/arm,smmu.yaml
> +++ b/Documentation/devicetree/bindings/iommu/arm,smmu.yaml
> @@ -102,6 +102,16 @@ properties:
>        access to SMMU configuration registers. In this case non-secure aliases of
>        secure registers have to be used during SMMU configuration.
>  
> +  qcom,skip-init:
> +    description: |
> +      Disable resetting configuration for all context banks
> +      during device reset.  This is useful for targets where
> +      some context banks are dedicated to other execution
> +      environments outside of Linux and those other EEs are
> +      programming their own stream match tables, SCTLR, etc.
> +      Without setting this option we will trample on their
> +      configuration.

It would probably be better to know _which_ context banks we shouldn't
touch, no? Otherwise what happens to the others?

> +
>    stream-match-mask:
>      $ref: /schemas/types.yaml#/definitions/uint32
>      description: |
> diff --git a/drivers/iommu/arm-smmu.c b/drivers/iommu/arm-smmu.c
> index 243bc4cb2705..a5c623d4caf9 100644
> --- a/drivers/iommu/arm-smmu.c
> +++ b/drivers/iommu/arm-smmu.c
> @@ -1655,13 +1655,16 @@ static void arm_smmu_device_reset(struct arm_smmu_device *smmu)
>  	 * Reset stream mapping groups: Initial values mark all SMRn as
>  	 * invalid and all S2CRn as bypass unless overridden.
>  	 */
> -	for (i = 0; i < smmu->num_mapping_groups; ++i)
> -		arm_smmu_write_sme(smmu, i);
>  
> -	/* Make sure all context banks are disabled and clear CB_FSR  */
> -	for (i = 0; i < smmu->num_context_banks; ++i) {
> -		arm_smmu_write_context_bank(smmu, i);
> -		arm_smmu_cb_write(smmu, i, ARM_SMMU_CB_FSR, ARM_SMMU_FSR_FAULT);
> +	if (!of_find_property(smmu->dev->of_node, "qcom,skip-init", NULL)) {
> +		for (i = 0; i < smmu->num_mapping_groups; ++i)
> +			arm_smmu_write_sme(smmu, i);
> +
> +		/* Make sure all context banks are disabled and clear CB_FSR  */
> +		for (i = 0; i < smmu->num_context_banks; ++i) {
> +			arm_smmu_write_context_bank(smmu, i);
> +			arm_smmu_cb_write(smmu, i, ARM_SMMU_CB_FSR, ARM_SMMU_FSR_FAULT);
> +		}
>  	}

Do we not need to worry about the SMRs as well?

Will