From mboxrd@z Thu Jan  1 00:00:00 1970
Return-Path: <linux-kernel-owner@vger.kernel.org>
Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand
	id S1756646Ab0EYJvs (ORCPT <rfc822;w@1wt.eu>);
	Tue, 25 May 2010 05:51:48 -0400
Received: from bombadil.infradead.org ([18.85.46.34]:40171 "EHLO
	bombadil.infradead.org" rhost-flags-OK-OK-OK-OK) by vger.kernel.org
	with ESMTP id S1756627Ab0EYJvp convert rfc822-to-8bit (ORCPT
	<rfc822;linux-kernel@vger.kernel.org>);
	Tue, 25 May 2010 05:51:45 -0400
Subject: Re: BUG: using smp_processor_id() in preemptible [00000000] code:
 icedove-bin/5449
From: Peter Zijlstra <peterz@infradead.org>
To: Ingo Molnar <mingo@elte.hu>
Cc: piotr@hosowicz.com, linux-kernel@vger.kernel.org,
       Jens Axboe <jens.axboe@oracle.com>, Divyesh Shah <dpshah@google.com>,
       Andrew Morton <akpm@linux-foundation.org>
In-Reply-To: <1274780834.5882.730.camel@twins>
References: <4BF9EC69.5030709@example.com> <1274777422.5882.591.camel@twins>
	 <20100525094347.GA7881@elte.hu>  <1274780834.5882.730.camel@twins>
Content-Type: text/plain; charset="UTF-8"
Content-Transfer-Encoding: 8BIT
Date: Tue, 25 May 2010 11:51:35 +0200
Message-ID: <1274781095.5882.743.camel@twins>
Mime-Version: 1.0
X-Mailer: Evolution 2.28.3 
Sender: linux-kernel-owner@vger.kernel.org
List-ID: <linux-kernel.vger.kernel.org>
X-Mailing-List: linux-kernel@vger.kernel.org

On Tue, 2010-05-25 at 11:47 +0200, Peter Zijlstra wrote:
> On Tue, 2010-05-25 at 11:43 +0200, Ingo Molnar wrote:
> > > Subject: sched_clock: Add local_clock()
> > > From: Peter Zijlstra <a.p.zijlstra@chello.nl>
> > > Date: Tue May 25 10:48:51 CEST 2010
> > > 
> > > For people who otherwise get to write: cpu_clock(smp_processor_id()),
> > > there is now: local_clock().
> > 
> > This doesnt fix the whole issue. cpu_clock() is local, while the measurements 
> > done in the blk code are global ...
> > 
> > While the warning is fixed this way, the far more serious issue is still 
> > there: time can go backwards if two points of time measurement are on 
> > different CPUs and can mess up the statistics with negative values, etc...
> 
> cpu_clock() is synced on each tick, so the inter-cpu-drift should not
> exceed 2 jiffies.
> 
> But yeah, if they want anything better, they'll have to start caring on
> what cpu which timestamp got taken and use cpu_clock().

For completeness, the previously used sched_clock() isn't synced between
cores at all, and can exhibit unbounded drift.