Re: [PATCH v3 03/11] leds: led-class: Add __of_led_get() helper

[Andy Shevchenko <andy@kernel.org>]

On Fri, Dec 16, 2022 at 04:52:33PM +0100, Hans de Goede wrote:
> On 12/16/22 14:50, Andy Shevchenko wrote:
> > On Fri, Dec 16, 2022 at 12:30:05PM +0100, Hans de Goede wrote:

...

> >> +	/*
> >> +	 * For named LEDs, first look up the name in the "leds-names" property.
> >> +	 * If it cannot be found, then of_parse_phandle() will propagate the error.
> >> +	 */
> >> +	if (name)
> >> +		index = of_property_match_string(np, "leds-names", name);
> > 
> > I can't find this property anywhere in the kernel. Is it new?
> 
> Yes and no, adding a foo-names property for foo[] arrays to be
> able to get members by name is a standard template for devicetree
> bindings. See e.g. the clock bindings:
> https://github.com/devicetree-org/dt-schema/blob/main/dtschema/schemas/clock/clock.yaml
> 
> > If so, where is the bindings?
> 
> As for why not document this, there are currently no devicetree users
> and the devicetree maintainers have repeatedly let me know not to
> submit new bindings for fwnode x86 bindings ...

How above is related to fwnode as you put that? (I do see OF specific code
which is required to have a binding, right?)

> > And why entire code can't be converted
> > to use fwnode for this case?
> 
> This is a trivial change to allow the new functions to work
> with devicetree. Note this series does not introduce any devicetree
> users, hence no bindings. But it is good to have compatibility backed in
> from day 1.

AFAIU the OF properties must be documented from day 1.

> Converting to fwnode APIs would be more involved and I cannot test
> those changes.

-- 
With Best Regards,
Andy Shevchenko