Rob Herring wrote at Tuesday, December 06, 2011 12:07 PM:

On 12/06/2011 12:22 PM, Stephen Warren wrote:

...

Stephen Warren wrote at Friday, December 02, 2011 3:09 PM:

...

Document the device tree binding for the WM8903 codec, and modify the driver to extract platform data from the device tree, if present.

Mark,

I just realized that when I was re-organizing all the WM8903 patches, I dropped the part that added the of_match table to the driver:

+static const struct of_device_id wm8903_of_match[] __devinitconst = {

• { .compatible = "wlf,wm8903", },
• {},

+}; +MODULE_DEVICE_TABLE(of, wm8903_of_match);

Now, everything still works without this. Looking at the Linux OF code, it works by retrieving the compatible property, taking everything after the comma if present, and then creating an i2c_board_info with that type, which in this case is "wm8903" and matches wm8903.c's i2c_device_id table. See drivers/of/of_i2c.c:of_i2c_register_devices() and the call to base.c:of_modalias_node().

So, the question is: Should I go back and add the of_match table, or is I2C intended to work without it perpetually? I notice that you added an of_match table for all the other WM codecs.

It definitely tries to match first with the OF match table, so it should probably be added back.

OK, I have no issues putting it back.

But, I certainly can't find any code in drivers/i2c or drivers/of/of_i2c.c that matches the compatible property against the of_match_table. Can you point it out please. As best I can tell, I2C bus/controller drivers call of_i2c_register_devices() which calls of_modalias_node() which only operates in the manner I described. I don't think any generic instantiation code can be operating instead, since there's nothing in of_i2c_register_devices() that skips devices that have already been instantiated via other means. Nothing in drivers/i2c seems to touch of_node or the compatible property.