On Tue, 2012-05-29 at 12:06 +0200, Christian Dietrich wrote:

Since nw_gpio_lock is a raw_spinlock_t it should be used with the raw_spinlock_* functions and not the spinlock_* variants. Functionally this is equivalent at the moment, because the raw_spinlock_t is the first field of spinlock_t, and therefore &nw_gpio_lock == &(nw_gpio_lock->rlock). But when other spinlock_t functions use other field they read and write random memory.

Hm, why are we exposing a raw spinlock to drivers?

Should we export a helper function (or macro, I suppose) which does the appropriate locking *and* the GPIO operation?