phys with subnodes and #phy-cells requirement

[mmind]

the phy-provider pattern matches against ...phy@... but for example
on some Rockchip socs the actual phy node might be in a subnode of it,
like when the phy is composed of multiple phys.
Which then triggers an error about #phy-cells being missing.

For example in arch/arm64/boot/dts/rockchip/rk3399.dtsi:

  u2phy0: usb2-phy@e450 {
  	compatible = "rockchip,rk3399-usb2phy";
  	reg = <0xe450 0x10>;
  	clocks = <&cru SCLK_USB2PHY0_REF>;
  	clock-names = "phyclk";
  	#clock-cells = <0>;
  	clock-output-names = "clk_usbphy0_480m";
  	status = "disabled";

  	u2phy0_host: host-port {
  		#phy-cells = <0>;
  		interrupts = <GIC_SPI 27 IRQ_TYPE_LEVEL_HIGH 0>;
  		interrupt-names = "linestate";
  		status = "disabled";
  	};

  	u2phy0_otg: otg-port {
  		#phy-cells = <0>;
  		interrupts = <GIC_SPI 103 IRQ_TYPE_LEVEL_HIGH 0>,
  			     <GIC_SPI 104 IRQ_TYPE_LEVEL_HIGH 0>,
  			     <GIC_SPI 106 IRQ_TYPE_LEVEL_HIGH 0>;
  		interrupt-names = "otg-bvalid", "otg-id",
  				  "linestate";
  		status = "disabled";
  	};
  };

Not sure how to solve this though

[robherring]

Ideally only phy providers are named 'phy' and the parent something else (not sure what to suggest). I'd probably rename the child nodes to phy@e454 and phy@e458 as those happen to be the unique register for each phy. Of course, that wouldn't work if all the registers were interleaved. The other option would have been get rid of the child nodes and just combine the interrupts to a single property in the parent node (with #phy-cells == 1).