Re: SMP syncronization on AMD processors (broken?)

[Eric Dumazet <dada1@cosmosbay.com>]

Hugh Dickins a écrit :
> On Thu, 6 Oct 2005, Linus Torvalds wrote:
> 
>>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.
> 
> 
> That reminds me: ought cond_resched_lock to be doing something more?
> 
> int cond_resched_lock(spinlock_t *lock)
> {
> 	int ret = 0;
> 
> 	if (need_lockbreak(lock)) {
> 		spin_unlock(lock);
> 		cpu_relax();
> 		ret = 1;
> 		spin_lock(lock);
> 	}
> -

Isnt it funny that some bugs can spot other bugs ? :)

break_lock should be declared atomic_t and used like that :

void __lockfunc _##op##_lock(locktype##_t *lock)
{
     preempt_disable();
     for (;;) {
         if (likely(_raw_##op##_trylock(lock)))
             break;
         preempt_enable();
	atomic_inc(&(lock)->break_lock);
         while (!op##_can_lock(lock))
             cpu_relax();
         preempt_disable();
	atomic_dec(&(lock)->break_lock);
     }
}


Eric