Re: [PATCH v2 1/3] dt-bindings: clock: qcom: Add SM8750 GPU clocks

[Konrad Dybcio]

On 7/28/25 7:05 AM, Krzysztof Kozlowski wrote:
> On 25/07/2025 11:23, Konrad Dybcio wrote:
>> On 7/24/25 4:42 PM, Krzysztof Kozlowski wrote:
>>> On 24/07/2025 12:53, Konrad Dybcio wrote:
>>>> On 7/24/25 10:18 AM, Krzysztof Kozlowski wrote:
>>>>> On Wed, Jul 23, 2025 at 10:38:48PM +0200, Konrad Dybcio wrote:
>>>>>> From: Konrad Dybcio <konrad.dybcio@xxxxxxxxxxxxxxxx>
>>>>>>
>>>>>> The SM8750 features a "traditional" GPU_CC block, much of which is
>>>>>> controlled through the GMU microcontroller. Additionally, there's
>>>>>> an separate GX_CC block, where the GX GDSC is moved.
>>>>>>
>>>>>> Add bindings to accommodate for that.
>>>>>>
>>>>>> Signed-off-by: Konrad Dybcio <konrad.dybcio@xxxxxxxxxxxxxxxx>
>>>>>> ---
>>
>> [...]
>>
>>> Yes, qcom,gcc. If that was missing intentionally, it is fine assuming
>>> you implement the rest of comments.
>>
>> With the description addition that you suggested above, should I keep
>> this file in clocks/ after all?
> 
> Good point, I don't know, this is unusual case. The question is whether
> there could be user of this binding/DTS, which would need/use
> clock-cells? If none of possible users could use it as a clock
> controller, I think it is not a clock controller from how SW sees it.
> IOW, it does not matter what it is fully (in bigger picture) if it
> cannot be used in that way.
> 
> If all users of the binding can use it only as power domain provided, I
> would move it to power with rest of power domains. Also rename the node
> name to power-controller or power-domain.

The hardware block can be accessed from the CPU directly, skipping
the microcontroller (although that is undesirable and the only "real" use
for it I can think about is someone trying to get rid of a blob).

I can add clock/reset-cells to describe the hardware accurately, but
the Linux driver(s - this is a block that exists on many >=2024 SoCs as
you may imagine) will continue to only provide a single power domain.
With that, I think clock/ makes sense, as this is essentially the same
hardware template as other instances of QCOM_*CC

Konrad