Re: [PATCH v1 2/4] dt-bindings: PCI: qcom,pcie-sa8775p: document link_down reset

[Krzysztof Kozlowski <krzk@kernel.org>]

On 04/06/2025 12:05, Qiang Yu wrote:
>>>>> Signed-off-by: Ziyue Zhang <quic_ziyuzhan@quicinc.com>
>>>>> ---
>>>>>   .../devicetree/bindings/pci/qcom,pcie-sa8775p.yaml  | 13 +++++++++----
>>>>>   1 file changed, 9 insertions(+), 4 deletions(-)
>>>>>
>>>>> diff --git a/Documentation/devicetree/bindings/pci/qcom,pcie-sa8775p.yaml b/Documentation/devicetree/bindings/pci/qcom,pcie-sa8775p.yaml
>>>>> index e3fa232da2ca..805258cbcf2f 100644
>>>>> --- a/Documentation/devicetree/bindings/pci/qcom,pcie-sa8775p.yaml
>>>>> +++ b/Documentation/devicetree/bindings/pci/qcom,pcie-sa8775p.yaml
>>>>> @@ -61,11 +61,14 @@ properties:
>>>>>         - const: global
>>>>>     resets:
>>>>> -    maxItems: 1
>>>>> +    minItems: 1
>>>>> +    maxItems: 2
>>>> Shouldn't we just update this to maxItems:2 / minItems:2 and drop
>>>> minItems:1 from the next clause?
>>> Hi Dmitry,
>>>
>>> link_down reset is optional. In many other platforms, like sm8550
>>> and x1e80100, link_down reset is documented as a optional reset.
>>> PCIe will works fine without link_down reset. So I think setting it
>>> as optional is better.
>> You are describing a hardware. How can a reset be optional in the
>> _hardware_? It's either routed or not.
> 
> I feel a bit confused. According to the theory above, everything seems to
> be non-optional when describing hardware, such as registers, clocks,
> resets, regulators, and interrupts—all of them either exist or do not.

Can you construct a DTS being fully complete and correct picture of
hardware without these? If not, they are not optional, because correct
hardware representation would need them.

> 
> Seems like I misunderstand the concept of 'optional'? Is 'optional' only
> used for compatibility across different platforms?
> 
> Additionally, we have documented the PCIe global interrupt as optional. I
> was taught that, in the PCIe driver, this interrupt is retrieved using the
> platform_get_irq_byname_optional API, so it can be documented as optional.
> However, this still seems to contradict the theory mentioned earlier.

ABI is just one side of the required properties.



Best regards,
Krzysztof