On 2021-01-25 21:52, Isaac J. Manjarres wrote:
When extracting the mask for a SMR that was programmed by the bootloader, the SMR's valid bit is also extracted and is treated as part of the mask, which is not correct. Consider the scenario where an SMMU master whose context is determined by a bootloader programmed SMR is removed (omitting parts of device/driver core):
->iommu_release_device() -> arm_smmu_release_device() -> arm_smmu_master_free_smes() -> arm_smmu_free_sme() /* Assume that the SME is now free */ -> arm_smmu_write_sme() -> arm_smmu_write_smr() /* Construct SMR value using mask and SID */
Since the valid bit was considered as part of the mask, the SMR will be programmed as valid.
Ah, right, because ARM_SMMU_SMR_{ID,MASK} are 16-bit fields to accommodate EXIDS, which doesn't matter normally when the IDs are strictly validated in arm_smmu_probe_device()...
Fix the SMR mask extraction step for bootloader programmed SMRs by masking out the valid bit when we know that we're already working with a valid SMR.
This seems like the neatest approach to me.
Reviewed-by: Robin Murphy robin.murphy@arm.com
Fixes: 07a7f2caaa5a ("iommu/arm-smmu-qcom: Read back stream mappings") Signed-off-by: Isaac J. Manjarres isaacm@codeaurora.org Cc: stable@vger.kernel.org
drivers/iommu/arm/arm-smmu/arm-smmu-qcom.c | 2 ++ 1 file changed, 2 insertions(+)
diff --git a/drivers/iommu/arm/arm-smmu/arm-smmu-qcom.c b/drivers/iommu/arm/arm-smmu/arm-smmu-qcom.c index bcda170..abb1d2f 100644 --- a/drivers/iommu/arm/arm-smmu/arm-smmu-qcom.c +++ b/drivers/iommu/arm/arm-smmu/arm-smmu-qcom.c @@ -206,6 +206,8 @@ static int qcom_smmu_cfg_probe(struct arm_smmu_device *smmu) smr = arm_smmu_gr0_read(smmu, ARM_SMMU_GR0_SMR(i)); if (FIELD_GET(ARM_SMMU_SMR_VALID, smr)) {
/* Ignore valid bit for SMR mask extraction. */smr &= ~ARM_SMMU_SMR_VALID; smmu->smrs[i].id = FIELD_GET(ARM_SMMU_SMR_ID, smr); smmu->smrs[i].mask = FIELD_GET(ARM_SMMU_SMR_MASK, smr); smmu->smrs[i].valid = true;