Re: [PATCH] percpu: preemptless __per_cpu_counter_add

[Shaohua Li <shaohua.li@intel.com>]

On Fri, 2011-04-29 at 00:35 +0800, Eric Dumazet wrote:
> Le jeudi 28 avril 2011 A  11:17 -0500, Christoph Lameter a A(C)crit :
> > On Thu, 28 Apr 2011, Eric Dumazet wrote:
> > 
> > > > If _sum() notices seqcount was changed too much, restart the loop.
> > 
> > This does not address the issue of cpus adding batch -1 while the
> > loop is going on.
> 
> 
> Yes, it does, I left the needed changes to write side as an exercice ;)
> 
> 
> For example, based on current linux-2.6 code
> 
> 
> void __percpu_counter_add(struct percpu_counter *fbc, s64 amount, s32 batch)
> {
>         s64 count;
> 
>         preempt_disable();
>         count = __this_cpu_read(*fbc->counters) + amount;
>         if (count >= batch || count <= -batch) {
>                 spin_lock(&fbc->lock);
> 		fbc->seqcount++;
>                 fbc->count += count;
>                 __this_cpu_write(*fbc->counters, 0);
>                 spin_unlock(&fbc->lock);
>         } else {
>                 __this_cpu_write(*fbc->counters, count);
>         }
>         preempt_enable();
> }
yep, this can resolve Tejun's concern. The problem is how do you
determine maxfuzzy. This can mislead user too. Because in the worst
case, the deviation is num_cpu*batch. If a user uses a small maxfuzzy
and expect the max deviation is maxfuzzy, he actually will get a bigger
deviation in worst case.

Thanks,
Shaohua

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Fight unfair telecom internet charges in Canada: sign http://stopthemeter.ca/
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>