Re: [PATCH v2 05/10] arm64: dts: rockchip: rk3568: add XPCS and fixed-clock nodes

[Coia Prant <coiaprant@gmail.com>]

On August 2, 2026 11:19:37 PM GMT+08:00, Andrew Lunn <andrew@lunn.ch> wrote:
>On Sun, Aug 02, 2026 at 11:28:12AM +0800, Coia Prant wrote:
>> Hi Andrew,
>> 
>> Thanks for the detailed questions. I've looked into the clock tree
>> more carefully.
>> 
>> > Do you have documentation for these Combo PHYs?
>> 
>> Yes, the information is available in the Rockchip RK3568 TRM
>> (Technical Reference Manual).
>> The relevant clock tree details are also mentioned in upstream
>> discussions and patches.
>> 
>> > Can these clocks be enabled/disabled?
>> 
>> The 100 MHz input clock to the Combo PHY is a prerequisite for its operation.
>> It is generated by the SoC's internal PPLL and fed to the PHY through
>> a fixed divider.
>> It is not a gate that can be individually enabled or disabled by
>> software; it is part of the
>> PHY's fundamental operating conditions.
>> 
>> > And how is the Combo PHY fed with a clock? Does it have an external oscillator/crystal?
>> 
>> The Combo PHY is fed by an internal 100 MHz clock derived from the PPLL.
>> Specifically, the PPLL runs at 200 MHz, and this is divided by 2 to provide the
>>  100 MHz clock input to the Combo PHY.
>> So, it does not use an external oscillator or crystal for its main
>> reference clock;
>> it's generated internally by the SoC.
>> 
>> To summarize:
>> - The 125 MHz clocks (xpcs_gmac0_clk / xpcs_gmac1_clk) are outputs
>> from the Combo PHY.
>> - The Combo PHY itself is clocked by an internal 100 MHz clock from the PPLL.
>> - This is all internal to the SoC and does not involve a board-level oscillator.
>> 
>> I hope this clarifies the clock architecture.
>
>So what is the purpose of the fixed-clock DT nodes?
>
>If you want to describe the hardware, is the PPLL in DT? Should you
>then add a fixed divider to give 100MHz? And then a fixed multiplier
>to give 125MHz?
>
>   Andrew

Hi Andrew,

Thanks for the follow-up.

The fixed-clock nodes (`xpcs_gmac0_clk` and `xpcs_gmac1_clk`) serve as
clock inputs to the GMAC controller. From the GMAC driver's perspective,
the 125 MHz clock is a fixed, external input. The fact that it's
generated internally by the Combo PHY from the PPLL is an implementation
detail of the SoC, not something the GMAC driver needs to be aware of.

In the device tree, the principle is to describe the hardware
interfaces, not the entire internal clock generation pipeline. The
PPLL configuration (200 MHz) is already set up in the CRU node via
assigned-clocks. The Combo PHY's internal PLL that generates 125 MHz is
part of the PHY's own operation, not something that the system clock
framework needs to manage or represent as a separate clock node.

If we were to fully model PPLL -> divider (100 MHz) -> Combo PHY PLL
(125 MHz) in DT, we would be overcomplicating the DT for no real gain.
It would also make board variations harder to handle (e.g., if a board
uses an external oscillator instead of the internal path, we'd have to
rewrite the entire clock tree). This approach is consistent with how
other similar SoCs (e.g., pcs-xpcs-plat) handle fixed reference clocks.

I think keeping these as fixed-clock nodes is the right balance between
accuracy and simplicity. If you still have concerns, I can add a comment
in the DT to clarify that these clocks are outputs of the Combo PHY.

Thanks,
Coia