Re: [PATCH v5 2/3] arm64: dts: qcom: qcs615: Add clock nodes for multimedia clock

[Konrad Dybcio]

On 7/30/25 6:10 PM, Akhil P Oommen wrote:
> On 7/30/2025 7:07 PM, Konrad Dybcio wrote:
>> On 7/2/25 11:13 AM, Taniya Das wrote:
>>> Add support for video, camera, display and gpu clock controller nodes
>>> for QCS615 platform.
>>>
>>> Reviewed-by: Konrad Dybcio <konrad.dybcio@xxxxxxxxxxxxxxxx>
>>> Signed-off-by: Taniya Das <quic_tdas@xxxxxxxxxxx>
>>> ---
>>
>> Bjorn mentioned offline that these controllers should
>> probably have power-domains attached to them (perhaps bar
>> GPU_CC, that's under discussion..)
> 
> QCS615 has an rgmu which doesn't manage gpucc. So this is a different
> case from the other discussion. Are we talking about scaling mx and cx
> rail while setting clk rate? Downstream clk driver does that on behalf
> of the clients. I suppose you are not talking about that here.

This is also relevant, as pmdomain states are propagated up the
tree, e.g. if we have:

usb@foobar {
	...
	power-domains = <&gcc USB30_GDSC>;
};

when someone calls dev_pm_opp_set_level() (or something equivalent like
dev_pm_opp_set_rate() with required-opps defined in the table), it
will set the performance state of the GDSC (which is a NOP for the GDSC
itself), but if we have this hunk:

gcc@feedbeef {
	...
	power-domains = <&rpmhpd RPMHPD_CX>;
};

RPMHPD_CX will be declared as a parent of all GCC GDSCs and its state
will be altered too. See:

drivers/pmdomain/core.c : _genpd_set_performance_state()


TLDR: clients are responsible for ensuring vdd_levels are set

Konrad