On Mon, Aug 19, 2024 at 05:02:16PM +0200, Marc Gonzalez wrote: > On 19/08/2024 14:57, Robin Murphy wrote: > > > Luckily, in this case it seems straightforward enough to be able to see > > that if we have a "qcom,msm8996-smmu-v2" with 13 context banks then we > > should just treat it as if it has 12 - it's also notable that it only > > reports NUMSMRG=12, so we couldn't use more than that many S1 context > > banks at once anyway. > > This is what the hypervisor reports: > > [ 2.550974] arm-smmu 5100000.iommu: probing hardware configuration... > [ 2.557309] arm-smmu 5100000.iommu: SMMUv2 with: > [ 2.563815] arm-smmu 5100000.iommu: stage 1 translation > [ 2.568494] arm-smmu 5100000.iommu: address translation ops > [ 2.573791] arm-smmu 5100000.iommu: non-coherent table walk > [ 2.579434] arm-smmu 5100000.iommu: (IDR0.CTTW overridden by FW configuration) > [ 2.585088] arm-smmu 5100000.iommu: stream matching with 12 register groups > [ 2.592132] arm-smmu 5100000.iommu: 13 context banks (0 stage-2 only) > [ 2.619316] arm-smmu 5100000.iommu: Supported page sizes: 0x63315000 > [ 2.626225] arm-smmu 5100000.iommu: Stage-1: 36-bit VA -> 36-bit IPA > [ 2.632645] arm-smmu 5100000.iommu: preserved 0 boot mappings > > > smmu->num_mapping_groups = 12 Ignore num_mapping_groups, they are used to define which streams should be mapped to which context bank. But there's no relationship between these numbers. > smmu->num_context_banks = 13 > > > Are you saying that > > smmu->num_context_banks > smmu->num_mapping_groups > > does not make sense? > > > Would a well-placed > > if (smmu->num_context_banks > smmu->num_mapping_groups) > smmu->num_context_banks = smmu->num_mapping_groups; > > be a proper work-around? No, something like this would apply your quirk to other targets (and specifically it would be wrong, per above). > > (Probably in qcom_smmu_cfg_probe() so as to not interfere with other platforms.) > > > Maybe to limit the side effects even more: > > if (of_device_is_compatible(smmu->dev->of_node, "qcom,msm8998-smmu-v2") && > smmu->num_context_banks > smmu->num_mapping_groups)) > smmu->num_context_banks = smmu->num_mapping_groups; If we don't want to introduce a more specific compatible for this SMMU instance, then let's add this to qcom_smmu_cfg_probe(): /* MSM8998 LPASS SMMU reports 13 context banks, but only 12 are accessible */ if (of_device_is_compatible(smmu->dev->of_node, "qcom,msm8998-smmu-v2") && smmu->num_context_banks == 13) smmu->num_context_banks = 12; Regards, Bjorn > > > Neither work-around would require changing the binding. > > Is either work-around acceptable, Robin? > > Regards > >
