Re: [PATCH] udev enhancements to use kernel event queue

[Robert Love]

On Thu, 2003-06-12 at 16:34, Patrick Mochel wrote:

> > Nice thinking. It is a shame we need a lock for this, but we don't
> > have an atomic_inc_and_return().
> 
> Those were my sentiments exactly:

Heh.

> +static inline int atomic_inc_and_read(atomic_t *v)
> +{
> +	__asm__ __volatile__(
> +		LOCK "incl %0"
> +		:"=m" (v->counter)
> +		:"m" (v->counter));
> +	return v->counter;
> +}

What prevents a race between the increment and the return (i.e. you
return v->counter after another person also increments it)? Only the
increment is guaranteed atomic.

I think you need to copy the result of the increment into a local
variable _inside_ of the LOCK and return that. Whether or not that will
work sanely on all architectures I dunno.

	Robert Love