Re: SMP syncronization on AMD processors (broken?)

[Linus Torvalds]

On Thu, 6 Oct 2005, Hugh Dickins wrote:
> 
> That reminds me: ought cond_resched_lock to be doing something more?

I suspect the "need_resched()" case is sufficient in practice: it might 
not be perfect, but let's face it, there's not a lot we can do.

It's true that the cpu_relax() basically does nothing except perhaps by 
luck (and depending on CPU and memory setup). There's no good alternative, 
though. You could make it more complex and change the cpu_relax() to 
something like

	i = random_backoff();
	for (;;) {
		cpu_relax();
		if (spin_is_locked(lock))
			break;
		if (!need_lockbreak(lock))
			break;
		if (!--i)
			break;
	}

but I'd argue vehemently that you shouldn't do this unless you can see 
real problems.

I doubt it's really a problem in practice. You have to have a really hot 
spinlock that stays hot on one CPU (and most of them you can't even _make_ 
hot from user space), and I suspect we've fixed the ones that matter.

		Linus
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/