On Fri, May 11, 2012 at 11:40:11AM -0600, Grant Likely wrote:
> > +static int __init tps65910_gpio_init(void)
> > +{
> > +	return platform_driver_register(&tps65910_gpio_driver);
> > +}
> > +subsys_initcall(tps65910_gpio_init);

> module_platform_driver()

This seems a little premature when we've not got many (any?) drivers
implementing -EPROBE_DEFER.  If we got the change I posted to make
gpio_request() return -EPROBE_DEFER if there was no GPIO registered
that'd get most cases covered though, the drivers would hopefully pass
the error through.

I've not done this for regulators since the cpufreq drivers are still
deviceless so couldn't cope.