> > I think the fallback mechanism of the core should work without any
> > module loading infrastructure. It should be there whenever possible.
> >
> 
> It's not really a fallback, is it? This is the path we'll always take
> if the driver requests a reset control on a firmware node which has a
> reset-gpios property. If the driver goes with the gpiod API, it will
> get a regular descriptor. It's deterministic enough to not warrant the
> term "fallback".

I dunno for how many drivers this is really applicable, but I really
liked the cleanup of the pca954x driver. Don't handle GPIOs internally,
just get a reset, and it might be a GPIO. I think it is very useful and
I would like to see it wherever possible.

We could now make these drivers depend on RESET_GPIO. This would make
sense in a way but is uncomfortable for the user who has not RESET_GPIO
enabled before. The driver would just disappear because of unmet
dependencies. Yes, this can happen all the time because we always find
new dependencies and describe them. I just hoped it could be avoided in
this case.

> Then I believe the platform's config should make sure the driver is
> built-in. I don't think it makes sense to just cram it into the kernel
> image for the few users it currently has.

For Morimoto-san, the PCA954x update resulted in a regression. It is
worth thinking how to avoid that. The driver is so small, I wouldn't
mind the extra space if it saves users from disappearing devices. But
mileages vary...