From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1754270Ab1HRVlH (ORCPT ); Thu, 18 Aug 2011 17:41:07 -0400 Received: from smtp1.linux-foundation.org ([140.211.169.13]:42338 "EHLO smtp1.linux-foundation.org" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1754112Ab1HRVlE (ORCPT ); Thu, 18 Aug 2011 17:41:04 -0400 Date: Thu, 18 Aug 2011 14:40:25 -0700 From: Andrew Morton To: Greg Thelen Cc: linux-kernel@vger.kernel.org, linux-mm@kvack.org, KAMEZAWA Hiroyuki , Balbir Singh , Daisuke Nishimura , linux-arch@vger.kernel.org Subject: Re: [PATCH] memcg: remove unneeded preempt_disable Message-Id: <20110818144025.8e122a67.akpm@linux-foundation.org> In-Reply-To: <1313650253-21794-1-git-send-email-gthelen@google.com> References: <1313650253-21794-1-git-send-email-gthelen@google.com> X-Mailer: Sylpheed 3.0.2 (GTK+ 2.20.1; x86_64-pc-linux-gnu) Mime-Version: 1.0 Content-Type: text/plain; charset=US-ASCII Content-Transfer-Encoding: 7bit Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org (cc linux-arch) On Wed, 17 Aug 2011 23:50:53 -0700 Greg Thelen wrote: > Both mem_cgroup_charge_statistics() and mem_cgroup_move_account() were > unnecessarily disabling preemption when adjusting per-cpu counters: > preempt_disable() > __this_cpu_xxx() > __this_cpu_yyy() > preempt_enable() > > This change does not disable preemption and thus CPU switch is possible > within these routines. This does not cause a problem because the total > of all cpu counters is summed when reporting stats. Now both > mem_cgroup_charge_statistics() and mem_cgroup_move_account() look like: > this_cpu_xxx() > this_cpu_yyy() > > ... > > --- a/mm/memcontrol.c > +++ b/mm/memcontrol.c > @@ -664,24 +664,20 @@ static unsigned long mem_cgroup_read_events(struct mem_cgroup *mem, > static void mem_cgroup_charge_statistics(struct mem_cgroup *mem, > bool file, int nr_pages) > { > - preempt_disable(); > - > if (file) > - __this_cpu_add(mem->stat->count[MEM_CGROUP_STAT_CACHE], nr_pages); > + this_cpu_add(mem->stat->count[MEM_CGROUP_STAT_CACHE], nr_pages); > else > - __this_cpu_add(mem->stat->count[MEM_CGROUP_STAT_RSS], nr_pages); > + this_cpu_add(mem->stat->count[MEM_CGROUP_STAT_RSS], nr_pages); > > /* pagein of a big page is an event. So, ignore page size */ > if (nr_pages > 0) > - __this_cpu_inc(mem->stat->events[MEM_CGROUP_EVENTS_PGPGIN]); > + this_cpu_inc(mem->stat->events[MEM_CGROUP_EVENTS_PGPGIN]); > else { > - __this_cpu_inc(mem->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]); > + this_cpu_inc(mem->stat->events[MEM_CGROUP_EVENTS_PGPGOUT]); > nr_pages = -nr_pages; /* for event */ > } > > - __this_cpu_add(mem->stat->events[MEM_CGROUP_EVENTS_COUNT], nr_pages); > - > - preempt_enable(); > + this_cpu_add(mem->stat->events[MEM_CGROUP_EVENTS_COUNT], nr_pages); > } On non-x86 architectures this_cpu_add() internally does preempt_disable() and preempt_enable(). So the patch is a small optimisation for x86 and a larger deoptimisation for non-x86. I think I'll apply it, as the call frequency is low (correct?) and the problem will correct itself as other architectures implement their atomic this_cpu_foo() operations.