Re: SMP syncronization on AMD processors (broken?)

[Linus Torvalds]

On Thu, 6 Oct 2005, Kirill Korotaev wrote:
> 
> The question raised because the situation we observe on AMD processors is
> really strange and makes us believe that something is wrong in kerne/in
> processor or our minds. Below goes an explanation:

Your code is buggy.

> The whole story started when we wrote the following code:
> 
> void XXX(void)
> {
> 	/* ints disabled */
> restart:
> 	spin_lock(&lock);
> 	do_something();
> 	if (!flag)
> 		need_restart = 1;
> 	spin_unlock(&lock);
> 	if (need_restart)
> 		goto restart;	<<<< LOOPS 4EVER ON AMD!!!
> }
> 
> void YYY(void)
> {
> 	spin_lock(&lock);	<<<< SPINS 4EVER ON AMD!!!
> 	flag = 1;
> 	spin_unlock(&lock);
> }

If you want to notify another CPU that you want the spinlock, then you 
need to set the "flag" variable _outside_ of the spinlock.

Spinlocks are not fair, not by a long shot. They never have been, and they 
never will. Fairness would be extremely expensive indeed.

> Other observations:
> - This does not happen on Intel processors, more over on Intel 2 CPUs take
> locks in a fair manner, exactly one by one!

It depends entirely on the cache coherency protocol. I bet you'd find 
differences even within Intel CPU's.

		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/