Marc Balmer <mbal...@netbsd.org> wrote: > > > >> + mutex_enter(&sc->sc_mtx); > >> + if (sc->sc_opened) > >> + return EBUSY; > > > > This leaks the lock. > > Ooops... But it is not needed here anymore, see below...
Are underlying drivers MP-safe? > >> int sc_opened; > >> + kmutex_t sc_mtx; > > > > Preferred suffix is _lock, rather than _mtx, so please use sc_lock. > > I have seen both forms in the tree. The ones which use _mtx should rather be fixed, instead of sprinkling them even more. Also, for condvars we use _cv suffix. > I reworked the patch again. This version changes the behaviour as well: > It is now possible that multiple thready/processes have a gpio device > open, but only one thread can be in ioctl(). Two new functions are > added, gpio_lock() and gpio_nolock() which can be used by child drivers > as well. Locking everything with a single fat lock is inefficient. As initial step, while learning, that is suitable. Otherwise locking should be fine grained. -- Mindaugas
