[PATCH 3/4] ARM: dts: rockchip: add rk3288 dma controllers

Am Montag, 11. August 2014, 19:37:29 schrieb Andreas F?rber:

Am 11.08.2014 19:01, schrieb Heiko St?bner:

quoted

Also for an upcoming v2, I've also changed the structure a bit, as the
first dma-controller has both a secure and non-secure version of it.

So currently the rk3288.dtsi looks like [0]:
	amba {
	
		compatible = "arm,amba-bus";
		#address-cells = <1>;
		#size-cells = <1>;
		ranges;
		
		/* dma1 in secure state */
		dma-controller at ffb20000 {
		
			compatible = "arm,pl330", "arm,primecell";
			reg = <0xffb20000 0x4000>;
			interrupts = <GIC_SPI 0 IRQ_TYPE_LEVEL_HIGH>,
			
				     <GIC_SPI 1 IRQ_TYPE_LEVEL_HIGH>;
			
			#dma-cells = <1>;
			clocks = <&cru ACLK_DMAC1>;
			clock-names = "apb_pclk";
			status = "disabled";
		
		};
		
		/* dma1 in non-secure state */
		dma-controller at ffb60000 {
		
			compatible = "arm,pl330", "arm,primecell";
			reg = <0xffb60000 0x4000>;
			interrupts = <GIC_SPI 0 IRQ_TYPE_LEVEL_HIGH>,
			
				     <GIC_SPI 1 IRQ_TYPE_LEVEL_HIGH>;
			
			#dma-cells = <1>;
			clocks = <&cru ACLK_DMAC1>;
			clock-names = "apb_pclk";
			status = "disabled";
		
		};
		
		dmac2: dma-controller at ff250000 {
		
			compatible = "arm,pl330", "arm,primecell";
			reg = <0xff250000 0x4000>;
			interrupts = <GIC_SPI 2 IRQ_TYPE_LEVEL_HIGH>,
			
				     <GIC_SPI 3 IRQ_TYPE_LEVEL_HIGH>;
			
			#dma-cells = <1>;
			clocks = <&cru ACLK_DMAC2>;
			clock-names = "apb_pclk";
		
		};
	
	};

and the board is responsible for enabling the correct variant [1], as most

likely the bootloader decides in which mode to start the dma controller:
	amba {
	
		/* dma1 in secure state */
		dmac1: dma-controller at ffb20000 {
		
			status = "okay";
		
		};
	
	};

This is based on some mailinglist discussion, I found at some point, about
this but for the life of me am not able to find anymore. So of course
feedback would appreciated there too.

I would suggest to give labels such as dmac1_s and dmac1_ns like you did
for dmac2 in the former snippet, so that you can override the status in
the latter without replicating the amba/dma-controller hierarchy.

I was hoping to just keep it to one handle, so that dma consumers would just 
work and not have to care which dma controller variant was available :-) .

However, for the Zynq SoC we chose to just model the secure DMAC though
[*]. I was told that either the bootloader or the user should change the
DT when running in such a non-secure scenario.

ok, this sounds interesting too, as I haven't seen a board that used the dmac 
in question in non-secure mode, yet.

So I guess one way to handle it could be to declare the non-secure dma, but 
set it to a disabled state and use the secure one as default for everything:

		dmac_bus_s: dma-controller at ffb20000 {
			compatible = "arm,pl330", "arm,primecell";
			reg = <0xffb20000 0x4000>;
			interrupts = <GIC_SPI 0 IRQ_TYPE_LEVEL_HIGH>,
				     <GIC_SPI 1 IRQ_TYPE_LEVEL_HIGH>;
			#dma-cells = <1>;
			clocks = <&cru ACLK_DMAC1>;
			clock-names = "apb_pclk";
		};

		dmac_bus_ns: dma-controller at ffb60000 {
			compatible = "arm,pl330", "arm,primecell";
			reg = <0xffb60000 0x4000>;
			interrupts = <GIC_SPI 0 IRQ_TYPE_LEVEL_HIGH>,
				     <GIC_SPI 1 IRQ_TYPE_LEVEL_HIGH>;
			#dma-cells = <1>;
			clocks = <&cru ACLK_DMAC1>;
			clock-names = "apb_pclk";
			status = "disabled";
		};

and a board wanting to use the non-secure variant would then be required to 
override the status and the dma channels in question (this dmac does mainly 
i2s and spdif):

&dmac_bus_s {
	status = "disabled";
}:

&dmac_bus_ns {
	status = "okay";
};

&i2s0 {
	dmas = <&dmac_bus_ns 6>, <&dmac_bus_ns 7>;
};

[*] https://patchwork.kernel.org/patch/4620251/

quoted

[0]
https://github.com/mmind/linux-rockchip/blob/devel/workbench/arch/arm/boo
t/dts/rk3288.dtsi [1]
https://github.com/mmind/linux-rockchip/blob/devel/workbench/arch/arm/boo
t/dts/rk3288-evb.dtsi

[snip]