Re: [PATCH v3 1/3] dt-bindings: interconnect: Remove required reg field

[Melody Olvera]

On 11/7/2022 10:28 AM, Krzysztof Kozlowski wrote:
> On 07/11/2022 15:36, Georgi Djakov wrote:
>> Hi,
>>
>> On 2.11.22 23:11, Krzysztof Kozlowski wrote:
>>> On 31/10/2022 19:29, Melody Olvera wrote:
>>>>
>>>> On 10/27/2022 8:29 AM, Krzysztof Kozlowski wrote:
>>>>> On 26/10/2022 15:05, Melody Olvera wrote:
>>>>>> Many of the *-virt compatible devices do not have a reg field
>>>>>> so remove it as required from the bindings.
>>>>> and some virt have it... This should be probably separate binding or if
>>>>> the list is small - allOf:if:then.
>>>> I attempted this; however I'm still seeing failures in dtb_check. I've added this
>>>> to the binding; does this look correct?
>>>>   allOf:
>>>>     - $ref: qcom,rpmh-common.yaml#
>>>> +  - if:
>>>> +      properties:
>>>> +        compatible:
>>>> +          contains:
>>>> +            enum:
>>>> +              - qcom,qdu1000-clk-virt
>>>> +              - qcom,qdu1000-mc-virt
>>>> +
>>>> +    then:
>>>> +      required:
>>>> +        - compatible
>>> No, because we talk about reg, not compatible. You should not require
>>> reg instead for some compatibles... but then the schema is getting
>>> complicated.
>>>
>>> It's difficult to give you recommendation because I do not know what are
>>> all these "virt" interconnects. Why some have unit address, why some do not?
>> My understanding is that the "reg" property is required for the NoCs that have
>> registers for controlling the QoS settings for the ports from Linux side.
>> Other NoCs might be controlled by some remote processor and direct access from
>> Linux may not be possible, so they do not have unit address and are outside of
>> the soc DT node.
>> Do we need to strictly define when exactly the "reg" property is required,
>> can't we just mark it as optional?
> It's preferred to make it strictly required or not allowed, so the
> bindings are specific. This also allows to validate for mistakes. It
> would be a bit different case if such test for req would make the
> bindings complicated. I think it's not the case because we could just
> split the bindings into two files:
> 1. One for controlled by AP, with reg.
> 2. One for controller by remote processors, without reg.
>
Sounds good. Will drop this change and add a new binding document.

Thanks,
Melody