Re: [PATCH] ppp: add 64 bit stats

[Eric Dumazet <eric.dumazet@gmail.com>]

On Wed, 2012-07-25 at 10:43 -0400, Kevin Groeneveld wrote:
> On Wed, Jul 25, 2012 at 12:04 AM, Eric Dumazet <eric.dumazet@gmail.com> wrote:
> > It there are spinlocks already, why even adding u64_stats_sync ?
> 
> That is a good question I have already wondered about myself.
> 
> >include/linux/u64_stats_sync.h
> >
> >* 3) Write side must ensure mutual exclusion or one seqcount update could
> > *    be lost, thus blocking readers forever.
> > *    If this synchronization point is not a mutex, but a spinlock or
> > *    spinlock_bh() or disable_bh() :
> 
> It seems the u64_stats_sync requires some form of mutual exclusion.
> So why bother ever using it at all?  Maybe there are cases where the
> required mutual exclusion can be cheaper than a spinlock?  Maybe it is
> just to avoid the spinlocks on the read side of things?
> 
> I hope you don't mind all my questions...

u64_stats_sync is good for write sides without any locking,
for example using percpu data.

So if you use shared counters, u64_stats_sync has no value.