Re: [PATCH v2 1/2] Change DEFINE_SEMAPHORE() to take a number argument

[Matthew Wilcox]

On Wed, Apr 05, 2023 at 01:35:04PM -0700, Luis Chamberlain wrote:
> diff --git a/include/linux/semaphore.h b/include/linux/semaphore.h
> index 6694d0019a68..2d6aa3fd7861 100644
> --- a/include/linux/semaphore.h
> +++ b/include/linux/semaphore.h
> @@ -25,8 +25,15 @@ struct semaphore {
>  	.wait_list	= LIST_HEAD_INIT((name).wait_list),		\
>  }
>  
> -#define DEFINE_SEMAPHORE(name)	\
> -	struct semaphore name = __SEMAPHORE_INITIALIZER(name, 1)
> +/*
> + * There is a big difference between a binary semaphore and a mutex.
> + * You cannot call mutex_unlock() from IRQ context because it takes an
> + * internal mutex spin_lock in a non-IRQ-safe manner. Both try_lock()
> + * and unlock() can be called from IRQ context. A mutex must also be
> + * released in the same context that locked it.
> + */

I think this confuses cause and effect.  How about this:

/*
 * Binary semaphores and mutexes differ in that mutexes have an owner
 * so they cannot be used from interrupt context and cannot be passed
 * from one thread to another.  down_trylock() and up() can be called
 * from interrupt context.
 */

Or this:

/*
 * Unlike mutexes, binary semaphores do not have an owner, so up() can
 * be called in a different thread from the one which called down().
 * It is also safe to call down_trylock() and up() from interrupt
 * context.
 */

I'd like to mention completions as an alternative to semaphores, but
can't figure out a nice way to fit that in.