Le mercredi 13 avril 2011 Ã 15:22 -0500, Christoph Lameter a Ãcrit :
> On Thu, 14 Apr 2011, Tejun Heo wrote:
>
> > On Wed, Apr 13, 2011 at 11:49:51AM -0500, Christoph Lameter wrote:
> > > Duh the retry setup if the number overflows is not correct.
> > >
> > > Signed-off-by: Christoph Lameter <cl@xxxxxxxxx>
> >
> > Can you please repost folded patch with proper [PATCH] subject line
> > and cc shaohua.li@xxxxxxxxx so that he can resolve conflicts?
> >
> > Thanks.
>
> Ok here it is:
>
>
>
>
> From: Christoph Lameter <cl@xxxxxxxxx>
> Subject: [PATCH] percpu: preemptless __per_cpu_counter_add
>
> Use this_cpu_cmpxchg to avoid preempt_disable/enable in __percpu_add.
>
> Signed-off-by: Christoph Lameter <cl@xxxxxxxxx>
>
> ---
> lib/percpu_counter.c | 27 +++++++++++++++------------
> 1 file changed, 15 insertions(+), 12 deletions(-)
>
> Index: linux-2.6/lib/percpu_counter.c
> ===================================================================
> --- linux-2.6.orig/lib/percpu_counter.c 2011-04-13 09:26:19.000000000 -0500
> +++ linux-2.6/lib/percpu_counter.c 2011-04-13 09:36:37.000000000 -0500
> @@ -71,19 +71,22 @@ EXPORT_SYMBOL(percpu_counter_set);
>
> void __percpu_counter_add(struct percpu_counter *fbc, s64 amount, s32 batch)
> {
> - s64 count;
> + s64 count, new;
>
> - preempt_disable();
> - count = __this_cpu_read(*fbc->counters) + amount;
> - if (count >= batch || count <= -batch) {
> - spin_lock(&fbc->lock);
> - fbc->count += count;
> - __this_cpu_write(*fbc->counters, 0);
> - spin_unlock(&fbc->lock);
> - } else {
> - __this_cpu_write(*fbc->counters, count);
> - }
> - preempt_enable();
> + do {
> + count = this_cpu_read(*fbc->counters);
> +
> + new = count + amount;
> + /* In case of overflow fold it into the global counter instead */
> + if (new >= batch || new <= -batch) {
> + spin_lock(&fbc->lock);
> + fbc->count += __this_cpu_read(*fbc->counters) + amount;
> + spin_unlock(&fbc->lock);
> + amount = 0;
> + new = 0;
> + }
> +
> + } while (this_cpu_cmpxchg(*fbc->counters, count, new) != count);
> }
> EXPORT_SYMBOL(__percpu_counter_add);
>
Not sure its a win for my servers, where CONFIG_PREEMPT_NONE=y
Maybe use here latest cmpxchg16b stuff instead and get rid of spinlock ?
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@xxxxxxxxxx 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@xxxxxxxxx";> email@xxxxxxxxx </a>