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 S932406AbbELIps (ORCPT <rfc822;w@1wt.eu>);
	Tue, 12 May 2015 04:45:48 -0400
Received: from bombadil.infradead.org ([198.137.202.9]:47668 "EHLO
	bombadil.infradead.org" rhost-flags-OK-OK-OK-OK) by vger.kernel.org
	with ESMTP id S932296AbbELIpo (ORCPT
	<rfc822;linux-kernel@vger.kernel.org>);
	Tue, 12 May 2015 04:45:44 -0400
Date: Tue, 12 May 2015 10:45:29 +0200
From: Peter Zijlstra <peterz@infradead.org>
To: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Douglas Hatch <doug.hatch@hp.com>, Thomas Gleixner <tglx@linutronix.de>,
        Ingo Molnar <mingo@kernel.org>, Waiman Long <Waiman.Long@hp.com>,
        Linux Kernel Mailing List <linux-kernel@vger.kernel.org>,
        "Norton, Scott J" <scott.norton@hp.com>, Peter Anvin <hpa@zytor.com>,
        "linux-tip-commits@vger.kernel.org" 
	<linux-tip-commits@vger.kernel.org>
Subject: Re: [tip:locking/core] locking/pvqspinlock: Replace xchg() by the
 more descriptive set_mb()
Message-ID: <20150512084529.GC21418@twins.programming.kicks-ass.net>
References: <tip-52c9d2badd1ae4d11c29de57d4e964e48afd3cb4@git.kernel.org>
 <20150511145408.GU27504@twins.programming.kicks-ass.net>
 <CA+55aFxre4JPFoPUC1UqsCvG8=nSka=AJMvFgqKzG8XiNWoD=A@mail.gmail.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <CA+55aFxre4JPFoPUC1UqsCvG8=nSka=AJMvFgqKzG8XiNWoD=A@mail.gmail.com>
User-Agent: Mutt/1.5.21 (2012-12-30)
Sender: linux-kernel-owner@vger.kernel.org
List-ID: <linux-kernel.vger.kernel.org>
X-Mailing-List: linux-kernel@vger.kernel.org

On Mon, May 11, 2015 at 10:50:42AM -0700, Linus Torvalds wrote:
> On Mon, May 11, 2015 at 7:54 AM, Peter Zijlstra <peterz@infradead.org> wrote:
> >
> > Hmm, so I looked at the set_mb() definitions and I figure we want to do
> > something like the below, right?
> 
> I don't think you need to do this for the non-smp cases. 

Well, its the store tearing thing again, we use WRITE_ONCE() in
smp_store_release() for the same reason. We want it to be a single
store.

> The whole
> thing is about smp memory ordering, so on UP you don't even need the
> WRITE_ONCE(), much less a barrier.

No, we actually need both still on UP.

Imagine the following sequence:

	for (;;) {
		set_current_state(TASK_KILLABLE);
		if (cond)
			break;

		schedule();
	}
	__set_current_state(TASK_RUNNING);

vs

	<IRQ>
		  wake_up_process(p);

As we know, set_current_state() is set_mb(), and thus will look like:

	current->state = TASK_KILLABLE;
	smp_mb();
	if (cond)
		break;

So without the WRITE_ONCE() we can get store tearing, and suppose our
compiler is insane and translates the store into 4 byte stores.

	current->state[0] = TASK_UNINTERRUPTIBLE;
	current->state[1] = TASK_WAKEKILL >> 8;
	current->state[2] = 0;
	current->state[3] = 0;

The obvious fail here is to get the wakeup interrupt between [0] and
[1].

	current->state[0] = TASK_UNINTERRUPTIBLE;

	<IRQ>
		wake_up_process(p);
		p->state = TASK_RUNNING;

	current->state[1] = TASK_WAKEKILL >> 8;
	current->state[2] = 0;
	current->state[3] = 0;

With the end result that ->state == TASK_WAKEKILL, from which we'll not
wake up unless killed.

Similarly, without the barrier(), our friendly compiler is allowed to
do:

	if (cond)
		break
	current->state = TASK_KILLABLE;
	schedule();

Which we all know to be broken.

So no, set_mb() (or smp_store_mb()) very much does need the WRITE_ONCE()
and a barrier() on UP.