Re: [PATCH v8 04/25] iommu/arm-smmu-v3: Move IDR parsing to common functions

On Tue, Sep 22, 2026 at 06:48:18PM -0300, Jason Gunthorpe wrote:

On Tue, Sep 22, 2026 at 12:45:20PM -0700, Nicolin Chen wrote:

quoted

Also, Jason's suggestion in v7 was hyp_arm_smmu_v3_device, which
looks nicer than ARM_SMMU_OBJ...

I guess the point is to emphasis the substition technique, I don't
mind.

quoted

quoted

+	if (coherent && !disable_msipolling &&
+	    smmu->features & ARM_SMMU_FEAT_MSI)
+		smmu->options |= ARM_SMMU_OPT_MSIPOLL;

Will pKVM ever use MSIPOLL?

Not sure I understood the approach in this patch, why was anything
left behind? I imagined everything would be moved?

Maybe a few more remarks in the commit message?

As replied to Nicolin, because it is a module param, I am ok with
adding this in the commit message, or passing disable_msipolling
as a function arg and moving the logic.

quoted

quoted

+	if (smmu->features & ARM_SMMU_FEAT_HYP &&
+	    cpus_have_cap(ARM64_HAS_VIRT_HOST_EXTN))
+		smmu->features |= ARM_SMMU_FEAT_E2H;

Why is ARM64_HAS_VIRT_HOST_EXTN left behind?

+1

And the TT_LE/BE and S1/S2 checks, and ...

If kvm doesn't want to dev_err log that's fine, we should able to
#define that away too shouldn't we? And maybe it does want to log?

The hypervisor can not print to the kernel console, so dev_err()
will not compile, I'd rather keep that outside and let the caller
decide what is a probe failure and print it's own message that keeping
the code and stubbing dev_err().

Thanks,
Mostafa

This seems pretty close otherwise

Jason