Hello Dmitry,

On 11/20/2015 04:46 PM, Javier Martinez Canillas wrote:

On 11/20/2015 04:32 PM, Javier Martinez Canillas wrote:

[snip]

quoted

But is not complete because the .driver_data in i2c_device_id is an
kernel_ulong_t while the .data in of_device_id is a const void * so
some casting will be needed to add an OF table in some drivers that

Some casting in the tables *and* some logic to get the .data from the
OF table entries, so something like the following will be needed:

static int foo_i2c_probe(struct i2c_client *i2c,
                         const struct i2c_device_id *id)
{
        struct of_device_id *match;
        kernel_ulong_t data;

        if (i2c->dev.of_node) {
               match = of_match_node(of_match, i2c->dev.of_node);
	       if (!match)
                        return -EINVAL;

               data = (kernel_ulong_t)match->data;
        } else {
               data = id->driver_data;
	}
	...
}

while currently I2C drivers just rely on the model part of the compatible
string to match with the entry in the I2C device ID table and the core
always passing the correct .driver_data to the probe function.

So as you can see, converting all the drivers to not rely on the I2C table
requires some refactoring before proper OF modalias reporting can be used.

Any comments about this? I'm planning to address all this at some point but
for now I think we would need $SUBJECT to have module autoloading working
on this driver when the device is registered via OF.
 

Best regards,

Best regards,
-- 
Javier Martinez Canillas
Open Source Group
Samsung Research America