Re: [PATCH] [RFC] gpiolib: introduce gpio_name() helper

["Arnd Bergmann" <arnd@arndb.de>]

On Mon, Jun 29, 2026, at 17:23, Andy Shevchenko wrote:
> On Mon, Jun 29, 2026 at 03:56:29PM +0200, Arnd Bergmann wrote:
>> +const char *gpiod_name(const struct gpio_desc *desc)
>> +{
>> +	return desc ? desc->name : "(no gpio)";
>
> Can we get into here with wrong (error pointer descriptor)? Shouldn't you call
> one of validate_desc() / VALIDATE_DESC()?

Since all the callers previously call desc_to_gpio and that does
not even check desc at all, it would be a preexisting bug if
any caller passed an error pointer.

I added the NULL pointer check since many callers had that
part originally, like

      tsdata->wake_gpio ? desc_to_gpio(tsdata->wake_gpio) : -1,

> Also not sure if "(no gpio)" is a good choice. "not requested"? "not provided"?

Any of those seem fine to me, not sure.

>> +static inline const char *gpiod_name(const struct gpio_desc *desc)
>> +{
>> +	WARN_ON(desc);
>> +	return "(no gpio)";
>
> Hmm... This will be a second copy with a slight potential of going apart from
> the other case. Perhaps a #define? (Yes, yes, I understand that there are pros
> and cons, in particular readability with define is questionable.)

I was mostly trying to optimize for consistency with the other
stub functions here. Since it should not actually be used at all
without gpiolib, returning NULL or an empty string here would
also work.

     Arnd