On 01/03/2025 14:46, Krzysztof Kozlowski wrote:
>>>> Signed-off-by: Wilson Ding <dingwei@xxxxxxxxxxx>
>>>> ---
>>>> arch/arm64/boot/dts/marvell/armada-cp11x.dtsi | 8 ++++++++
>>>> 1 file changed, 8 insertions(+)
>>>>
>>>> diff --git a/arch/arm64/boot/dts/marvell/armada-cp11x.dtsi
>>> b/arch/arm64/boot/dts/marvell/armada-cp11x.dtsi
>>>> index 161beec0b6b0..c27058d1534e 100644
>>>> --- a/arch/arm64/boot/dts/marvell/armada-cp11x.dtsi
>>>> +++ b/arch/arm64/boot/dts/marvell/armada-cp11x.dtsi
>>>> @@ -226,6 +226,8 @@ CP11X_LABEL(rtc): rtc@284000 {
>>>> CP11X_LABEL(syscon0): system-controller@440000 {
>>>> compatible = "syscon", "simple-mfd";
>>>> reg = <0x440000 0x2000>;
>>>> + #address-cells = <1>;
>>>> + #size-cells = <1>;
>>>>
>>>> CP11X_LABEL(clk): clock {
>>>
>>> Wait, no unit address here.
>>
>> This subnode came from the existing code. I didn't touch this subnode
>> in my patch. As you can see, the system-controller has a wide address
>> range, which includes clock, GPIO registers as well as the unit-softreset
>> register.
>>
>>>
>>>> compatible = "marvell,cp110-clock";
>>>> @@ -273,6 +275,12 @@ CP11X_LABEL(gpio2): gpio@140 {
>>>> <&CP11X_LABEL(clk) 1 17>;
>>>> status = "disabled";
>>>> };
>>>> +
>>>> + CP11X_LABEL(swrst): reset-controller@268 {
>>>
>>>
>>> So why here it appeared? This is wrong and not even necessary. Entire
>>> child should be folded into parent, so finally you will fix the
>>> incomplete parent compatible.
>>
>> We do need the reset-controller as a subnode under system-controller node
>> for the following reasons:
>>
>> - We need to have 'reg' property in this subnode so that we can get the offset
>> to system-controller register base defined in parent node. This is suggested
>> by Rob in V2 comments.
>> And we need to know the register size to calculate the number of reset lines.
>> This is suggested by Philipp in V1 comments.
>
> You do not need and you received that comment as well. It is implied by
> compatible.
>
>>
>> - We also need to define the 'reset-cells' in this subnode. And the consumer of
>> the reset controller uses the label of this subnode for the phandle and reset
>> specifier pair.
>
> reset-cells will be in the parent once you fold it.
>
>>
>> As I mentioned in my reply to the first comment, the reset-controller is not the
>> only device within the system-controller register spaces. Do you still think I
>
> You provided very little hardware description of the device. So based on
> hardware description you provided: yes.
and to clarify - by device I mean the parent node, the system controller.
Your commit even mentions driver, not the hardware:
"Add device-tree binding documentation for the Armada8K reset driver..."
not hardware, so you will get the review as good as you describe things.
Bindings are about hardware so if you disagree with the review here,
please provide arguments in terms of hardware.
>
>> should fold it into the parent node. And what I proposed is exactly same as
>> that the armada_thermal driver did (See below). I wonder why what was accepted
>> in the past become not accepted now.
>
> We did not discuss here drivers, but if you insist talking about
> "marvell,armada-cp110-thermal" then point me to review or ack from DT
> people. You claim it was accepted so how did we accept it?
>
> It was 2013 so that's another answer: many things done 12 years ago were
> done not according to best practices. Also best practices evolved.
Best regards,
Krzysztof
