On 19/07/2022 10:17, Tinghan Shen wrote:
>>> + syscon@10006000 {
>>> + compatible = "mediatek,scpsys", "syscon", "simple-mfd";
>>
>> This should be a SoC-specific compatible (and filename).
>
> Ok. I think that you mean "mediatek,mt8195-scpsys".
> I'll update it in next version.
Yes.
>
>>
>>> + reg = <0x10006000 0x100>;
>>> +
>>> + spm: power-controller {
>>
>> I think you created before less-portable, quite constrained bindings for
>> power controller. You now require that mt8195-power-controller is always
>> a child of some parent device which will share its regmap/MMIO with it.
>>
>> And what if in your next block there is no scpsys block and power
>> controller is the scpsys alone? It's not possible with your bindings.
>
> Do you mean a power controller node that looks like this?
>
> scpsys: power-controller@10006000 {
> compatible = "mediatek,mt6797-scpsys";
> #power-domain-cells = <1>;
>
> // ...
> };
Yes, I mean, with an unit address.
>
>>
>> Wouldn't it be better to assign some address space to the
>> power-controller (now as an offset from scpsys)?
>
> Is this mean adding an offset after the node name?
>
> spm: power-controller@0 {
This or above. I think it does not matter for the bindings - it's an
implementation detail, whether you give to the child absolute SoC
address or you give an bus-specific (scpsys) sub-address/offset.
The point is that you have an unit address, thus in the future this
could be a device node separate from scpsys.
Best regards,
Krzysztof
