Re: [PATCH v5 04/13] iommu: of_iommu: Add support for "iommu-ranges" on a device node

[Bryan O'Donoghue <bod@kernel.org>]

On 26/10/2026 12:08, Vikash Garodia wrote:
>> https://github.com/devicetree-org/dt-schema/pull/207
>>
>> Perhaps splitting this into two patches as an iommu-series would be best.
> Intention was to keep the patch to add iommu-ranges property as part of
> iris driver patches to show the usage of this new property.
> 
> I am fine either ways, whatever it takes for the maintainers of this
> file to review.
> 
> Regards,
> Vikash
> 
>>>            end = maps + size / sizeof(__be32);
>>> @@ -258,6 +367,9 @@ void of_iommu_get_resv_regions(struct device *dev,
>>> struct list_head *list)
>>>                    }
>>>                    type = iommu_resv_region_get_type(dev, &phys, iova,
>>> length);
>>> +                if (type != IOMMU_RESV_DIRECT)
>>> +                    continue;
>>> +
>> If you look in iommu_resv_region_get_type() you can emit
>>
>>           dev_warn(dev, "treating non-direct mapping [%pr] -> [%pap-%pap]
>> as reservation\n", phys,
>>                    &start, &end);
>>           return IOMMU_RESV_RESERVED;
>>
>> on a mismatch - which is not what the code will do now. Mismatch will
>> warn and then not map.
>>
>> So maybe dev_err() saying you won't map - in the new patch for reg.
>> Either way this isn't a change specific to iommu-ranges.

Sure the above needs to be addressed though because its not a straight 
change.

Before if there was a mismatch it would sill be mapped - now we print a 
warning and _don't_ map.

I'm not sure that's really what you meant to do here but if it is it 
should be stated in the git log and IMO it should be a follow on change 
_after_ the initial addition (or before it) stating why mismatched will 
no longer be mapped.

Because add iommu-ranges != don't map mismatched, basically.

---
bod