Re: [PATCH v2 1/7] riscv: dts: sophgo: cv18xx: Move RiscV-specific part into SoCs' .dtsi files

[Alexander Sverdlin <alexander.sverdlin@gmail.com>]

Hi 

On Wed, 2025-02-12 at 17:46 +0100, Krzysztof Kozlowski wrote:
> > > > Make the peripheral device tree re-usable on ARM64 platform by moving CPU
> > > > core and interrupt controllers' parts into the respective per-SoC .dtsi
> > > > files.
> > > > 
> > > > Add SOC_PERIPHERAL_IRQ() macro which explicitly maps peripheral nubering
> > > > into "plic" interrupt-controller numbering.
> > > > 
> > > > Have a nice refactoring side-effect that "plic" and "clint" "compatible"
> > > > property is not specified outside of the corresponding device itself.
> > > > 
> > > > Signed-off-by: Alexander Sverdlin <alexander.sverdlin@gmail.com>
> > > > ---
> > > > Changelog:
> > > > v2:
> > > > - instead of carving out peripherals' part, carve out ARCH-specifics (CPU
> > > > core, interrupt controllers) and spread them among 3 SoC .dtsi files which
> > > > included cv18xx.dtsi;
> > > > - define a label for the "soc" node and use it in the newly introduced DTs;
> > > > 
> > > >  arch/riscv/boot/dts/sophgo/cv1800b.dtsi    | 64 ++++++++++++---
> > > >  arch/riscv/boot/dts/sophgo/cv1812h.dtsi    | 64 ++++++++++++---
> > > >  arch/riscv/boot/dts/sophgo/cv181x.dtsi     |  2 +-
> > > >  arch/riscv/boot/dts/sophgo/cv18xx-cpu.dtsi | 57 ++++++++++++++
> > > >  arch/riscv/boot/dts/sophgo/cv18xx.dtsi     | 91 ++++++----------------
> > > >  arch/riscv/boot/dts/sophgo/sg2002.dtsi     | 64 ++++++++++++---
> > > >  6 files changed, 240 insertions(+), 102 deletions(-)
> > > >  create mode 100644 arch/riscv/boot/dts/sophgo/cv18xx-cpu.dtsi
> > > > 
> > > > diff --git a/arch/riscv/boot/dts/sophgo/cv1800b.dtsi b/arch/riscv/boot/dts/sophgo/cv1800b.dtsi
> > > > index aa1f5df100f0..eef2884b36f9 100644
> > > > --- a/arch/riscv/boot/dts/sophgo/cv1800b.dtsi
> > > > +++ b/arch/riscv/boot/dts/sophgo/cv1800b.dtsi
> > > > @@ -3,6 +3,8 @@
> > > >   * Copyright (C) 2023 Jisheng Zhang <jszhang@kernel.org>
> > > >   */
> > > >  
> > > > +#define SOC_PERIPHERAL_IRQ(nr)	((nr) + 16)
> > > > +
> > > >  #include <dt-bindings/pinctrl/pinctrl-cv1800b.h>
> > > >  #include "cv18xx.dtsi"
> > > >  
> > > > @@ -14,22 +16,62 @@ memory@80000000 {
> > > >  		reg = <0x80000000 0x4000000>;
> > > >  	};
> > > >  
> > > 
> > > > -	soc {
> > > > -		pinctrl: pinctrl@3001000 {
> > > > -			compatible = "sophgo,cv1800b-pinctrl";
> > > > -			reg = <0x03001000 0x1000>,
> > > > -			      <0x05027000 0x1000>;
> > > > -			reg-names = "sys", "rtc";
> > > 
> > > 
> > > > +	cpus: cpus {
> > > > +		#address-cells = <1>;
> > > > +		#size-cells = <0>;
> > > > +		timebase-frequency = <25000000>;
> > > > +
> > > > +		cpu0: cpu@0 {
> > > > +			compatible = "thead,c906", "riscv";
> > > > +			device_type = "cpu";
> > > > +			reg = <0>;
> > > > +			d-cache-block-size = <64>;
> > > > +			d-cache-sets = <512>;
> > > > +			d-cache-size = <65536>;
> > > > +			i-cache-block-size = <64>;
> > > > +			i-cache-sets = <128>;
> > > > +			i-cache-size = <32768>;
> > > > +			mmu-type = "riscv,sv39";
> > > > +			riscv,isa = "rv64imafdc";
> > > > +			riscv,isa-base = "rv64i";
> > > > +			riscv,isa-extensions = "i", "m", "a", "f", "d", "c", "zicntr", "zicsr",
> > > > +					       "zifencei", "zihpm";
> > > > +
> > > > +			cpu0_intc: interrupt-controller {
> > > > +				compatible = "riscv,cpu-intc";
> > > > +				interrupt-controller;
> > > > +				#interrupt-cells = <1>;
> > > > +			};
> > > >  		};
> > > >  	};
> > > >  };
> > > 
> > > Make all soc definition include the common cpu file. 
> > > Not just copy it.
> > 
> > I was acting according to Krzysztof's suggestion:
> > https://lore.kernel.org/soc/d3ba0ea5-0491-42d5-a18e-64cf21df696c@kernel.org/
> > 
> > Krzysztof, I can name the file cv18xx-cpu-intc.dtsi and pack CPU core + interrupt
> > controllers into it. Would it make sense?
> 
> 
> I don't understand the original suggestion.

This is the snippet in question:

---[ cut ]---
#define SOC_PERIPHERAL_IRQ(nr)	((nr) + 16)

/ {
	cpus: cpus {
		#address-cells = <1>;
		#size-cells = <0>;
		timebase-frequency = <25000000>;

		cpu0: cpu@0 {
			compatible = "thead,c906", "riscv";
			device_type = "cpu";
			reg = <0>;
			d-cache-block-size = <64>;
			d-cache-sets = <512>;
			d-cache-size = <65536>;
			i-cache-block-size = <64>;
			i-cache-sets = <128>;
			i-cache-size = <32768>;
			mmu-type = "riscv,sv39";
			riscv,isa = "rv64imafdc";
			riscv,isa-base = "rv64i";
			riscv,isa-extensions = "i", "m", "a", "f", "d", "c", "zicntr", "zicsr",
					       "zifencei", "zihpm";

			cpu0_intc: interrupt-controller {
				compatible = "riscv,cpu-intc";
				interrupt-controller;
				#interrupt-cells = <1>;
			};
		};
	};
};

&soc {
	interrupt-parent = <&plic>;
	dma-noncoherent;

	plic: interrupt-controller@70000000 {
		reg = <0x70000000 0x4000000>;
		interrupts-extended = <&cpu0_intc 11>, <&cpu0_intc 9>;
		interrupt-controller;
		#address-cells = <0>;
		#interrupt-cells = <2>;
		riscv,ndev = <101>;
	};

	clint: timer@74000000 {
		reg = <0x74000000 0x10000>;
		interrupts-extended = <&cpu0_intc 3>, <&cpu0_intc 7>;
	};
};
---[ cut ]---

Inochi's proposal is to put it into separate cv18xx-cpu-intc.dtsi and
include the latter in 3 other SoC-specific .dtsis. In v2 I've just
duplicated the above snippet 3 times (refer to diffstat above).

What are your thoughts? In Renesas everything is duplicated, I believe.
Sophgo outsources much smaller snippets into .dtsi (refer to cv181x.dtsi). 

> Inochi, please trim unnecessary context from replies.

-- 
Alexander Sverdlin.


_______________________________________________
linux-riscv mailing list
linux-riscv@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-riscv