On Sun, 29 Jun 2008 12:24:39 -0400, Sean MacLennan wrote:

On Sun, 29 Jun 2008 09:17:25 +0200
"Jean Delvare" [off-list ref] wrote:

quoted

Ah, OK. If you use i2c_new_device() then it's alright.

Correct.

I have done the same thing for the i2c-ibm_iic.c driver. Jean, I think
you will like this. It gets rid of the index and the numbered drivers.
And the walking of the device tree is very clean because the dts knows
all the devices.

For example here is the relevant portion of the dts for the Warp:

IIC0: i2c@ef600700 {
	compatible = "ibm,iic-440ep", "ibm,iic-440gp", "ibm,iic";
	reg = <ef600700 14>;
	interrupt-parent = <&UIC0>;
	interrupts = <2 4>;
	#address-cells = <1>;
	#size-cells = <0>;

	ad7414@4a {
		compatible = "adi,ad7414";
		reg = <4a>;
		interrupts = <19 8>;
		interrupt-parent = <&UIC0>;
	};
};

It clearly shows that first i2c controller (IIC0) contains one ad7414 device at address 4A.

That's fine with me. I expected the dts to be converted to platform
initialization data (i2c_board_info structures) being registered with
i2c_register_board_info() and numbered adapters. But if you prefer
unnumbered adapters and the platform code or the bus driver itself
calls i2c_new_device() based on the dts, that should work too.

-- 
Jean Delvare