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

[David Schwartz]

Pual Mackerras is said to have opined:

> Patrick Mochel writes:

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

> BZZZT.  If another CPU is also doing atomic_inc_and_read you could end
> up with both calls returning the same value.
>
> You can't do atomic_inc_and_read on 386.  You can on cpus that have
> cmpxchg (e.g. later x86).  You can also on machines with load-locked
> and store-conditional instructions (alpha, ppc, probably most other
> RISCs).

	You can also do it with a conditional move instruction, but it's kind of
ugly. No help on a '386 though.

	There are ways to do it that work on a 386, but they are all basically
equivalent to (or worse than) acquiring a spinlock, doing the deed, and then
releasing it.

	You could also do (in pseudo-code):

top:
 ret <- v->counter
 inc ret
 LOCK incl v->counter
 cmp v->counter, ret
 jz end
 LOCK decl v->counter
 jmp top:
end:
 return ret

	This does not strictly guarantee in order return values, but that's
meaningless without a lock anyway.

	DS