On Wed, Jul 04, 2012 at 07:41:09PM +0900, Magnus Damm wrote:

> Right, so what's wrong with having a "select REGULATOR_FIXED_VOLTAGE"
> on boards that make use of them? That seems to be the most
> straightforward way to me.

I tend to agree; as I've said several times now the only sensible
options I see are:

 - Add the selects per board.
 - Just enable the fixed voltage driver whenever the regulator core is
   enabled.

Anything else seems silly.

> Also, I know too little about the regulator subsystem, but to me it is
> somewhat special that REGULATOR=n behaves differently than REGULATOR=y
> and REGULATOR_DRIVER=n.

What's happening is that the regulator API stubs itself out when it's
not in use in order to avoid having lots of ifdefs every time anyone
tries to use the API in generic code or drivers that get used on
multiple platforms.  This is the same behaviour as gpiolib, it's a
similar sort of utility API.

If the power on defaults for the system weren't suitable for whatever is
being done you'd still have an issue even with the stubs but a lot of
the time the main effect of the API is to allow drivers to do additional
power optimisations.