On Tue, Sep 09, 2014 at 05:45:17AM -0700, Jacob Pan wrote:
> > > -static int axp20x_i2c_probe(struct i2c_client *i2c,
> > > -			 const struct i2c_device_id *id)
> > > +static int axp2xx_match_device(struct axp2xx_dev *axp2xx, struct
> > > device *dev) {
> > > -	struct axp20x_dev *axp20x;
> > > +	const struct acpi_device_id *acpi_id;
> > >  	const struct of_device_id *of_id;
> > > -	int ret;
> > >  
> > > -	axp20x = devm_kzalloc(&i2c->dev, sizeof(*axp20x),
> > > GFP_KERNEL);
> > > -	if (!axp20x)
> > > -		return -ENOMEM;
> > > +	of_id = of_match_device(axp2xx_of_match, dev);
> > > +	if (of_id) {
> > > +		axp2xx->variant = (long) of_id->data;
> > > +		goto found_match;
> > > +	}
> > >  
> > > -	of_id = of_match_device(axp20x_of_match, &i2c->dev);
> > > -	if (!of_id) {
> > > -		dev_err(&i2c->dev, "Unable to setup AXP20X
> > > data\n");
> > > +	acpi_id = acpi_match_device(dev->driver->acpi_match_table,
> > > dev);
> > > +	if (!acpi_id || !acpi_id->driver_data) {
> > > +		dev_err(dev, "Unable to setup AXP2XX ACPI data\n");
> > >  		return -ENODEV;
> > >  	}
> > > -	axp20x->variant = (long) of_id->data;
> > > +	axp2xx->variant = (long) acpi_id->driver_data;
> > 
> > Shouldn't that be in the if statement above? I guess acpi_id will be
> > null on a DT-based system.
> > 
> I am not sure what you mean. if acpi_id == NULL, then it will return
> -ENODEV since of_match_device() already found no match. If acpi_id !=
> NULL, then id must contain variant info.

Hmm, never mind. I read it backward and thought you were still in this
code path. I guess I need more coffee.

That goto isn't very intuitive though.

Maybe something like

if (of_id) {
   /* DT case */
} else if (acpi_id) {
  /* ACPI case */
} else {
  return;
}

would be better?

Maxime

-- 
Maxime Ripard, Free Electrons
Embedded Linux, Kernel and Android engineering
http://free-electrons.com