Re: [PATCH v4 7/7] arm64: dts: qcom: mahua: Switch pcie5_phy ref clock to RPMH_CXO_CLK

[Konrad Dybcio]

On 6/9/26 3:55 PM, Krzysztof Kozlowski wrote:
> On 09/06/2026 15:06, Konrad Dybcio wrote:
>> On 5/28/26 4:29 AM, Qiang Yu wrote:
>>> PCIe5 PHY on Mahua gets refclk from CXO0 pad directly, so no QREF
>>> clkref_en voting is required. Override the clock list to use RPMH_CXO_CLK
>>> directly instead.
>>
>> This is the last piece of the puzzle that this series is missing.
>> There's no QREF clkref_en, but there is a refgen that needs to be
>> powered. For PCIe5 on Mahua this would be L2F_E0 (0p9) and L4H_E0
>> (1p2).
>>
>> I think the easiest (laziest?) solution would be to add dummy clocks
>> in the clkref driver and only toggle the required regulators. Another
>> option is to defer back to individual drivers (such as PCIe QMPPHY).
>>
>> I kinda like the "one central node to drive power" approach, but I'm
>> not sure others agree, since it stretches truth just a tiny bit
>> (although not as much as one would think since there are *some*
>> controls for the transparent-to-the-OS hw pieces in these paths still
>> in TCSR).. Dmitry, Krzysztof, would you object to that?
> 
> Not sure what you ask here... the tcsr will get the refgen supplies and
> that's all what is needed from DT.

What I'm saying is that there are always-on clock branches (no register
to control them) which go through controllable repeaters (with a reg in
tcsr, although we never control them since the reset state is OK)

Both the repeaters and the associated refgens need power supplies. My
suggestion was to take care of that by adding "fake" clocks. Currently,
we have e.g.:

&usb_0_qmpphy {
        vdda-phy-supply = <&vreg_l4h_e0_1p2>;
        vdda-pll-supply = <&vreg_l3f_e0_0p72>;
        refgen-supply = <&vreg_l2f_e0_0p82>; // the other 1.2v refgen is missing

        status = "okay";
};

The refgen supply is deferred to the QMPPHY node itself, and the
repeaters on the way are ignored. The "ref" clock points to RPMH_XO.

What I'm suggesting is that we add "fake" (i.e. uncontrollable) TCSR
clocks, such as the inexistent TCSR_USB4_0_CLKREF_EN, to which we can
link the required regulators, both for the repeaters and the refgen
source

Konrad