[net-next PATCH v5 04/15] leds: Provide stubs for when CLASS_LED is disabled

[Christian Marangi]

On Mon, Mar 20, 2023 at 08:31:36PM +0100, Andrew Lunn wrote:
> On Mon, Mar 20, 2023 at 06:52:47PM +0100, Christian Marangi wrote:
> > On Sun, Mar 19, 2023 at 11:49:02PM +0100, Andrew Lunn wrote:
> > > > +#if IS_ENABLED(CONFIG_LEDS_CLASS)
> > > >  enum led_default_state led_init_default_state_get(struct fwnode_handle *fwnode);
> > > > +#else
> > > > +static inline enum led_default_state
> > > > +led_init_default_state_get(struct fwnode_handle *fwnode)
> > > > +{
> > > > +	return LEDS_DEFSTATE_OFF;
> > > > +}
> > > > +#endif
> > > 
> > > 0-day is telling me i have this wrong. The function is in led-core.c,
> > > so this should be CONFIG_NEW_LEDS, not CONFIG_LEDS_CLASS.
> > > 
> > 
> > Any idea why? NEW_LEDS just enable LEDS_CLASS selection so why we need
> > to use that? Should not make a difference (in theory)
> 
> 0-day came up with a configuration which resulted in NEW_LEDS enabled
> but LEDS_CLASS disabled. That then resulted in multiple definitions of 
> led_init_default_state_get() when linking.
> 
> I _guess_ this is because select is used, which is not mandatory. So
> randconfig can turn off something which is enabled by select.
> 
> I updated my tree, and so far 0-day has not complained, but it can
> take a few days when it is busy.
> 

BTW yes I repro the problem.

Checked the makefile and led-core.c is compiled with NEW_LEDS and
led-class is compiled with LEDS_CLASS.

led_init_default_state_get is in led-core.c and this is the problem with
using LEDS_CLASS instead of NEW_LEDS...

But actually why we are putting led_init_default_state_get behind a
config? IMHO we should compile it anyway.

So my suggestion is to keep the LEDS_CLASS and just remove the part for 
led_init_default_state_get.

Also why IS_ENABLED instead of a simple ifdef? (in leds.h there is a mix
of both so I wonder if we should use one or the other)

-- 
	Ansuel