From mboxrd@z Thu Jan  1 00:00:00 1970
From: Benjamin Herrenschmidt <benh@kernel.crashing.org>
Subject: Re: MMIO and gcc re-ordering issue
Date: Fri, 13 Jun 2008 10:07:04 +1000
Message-ID: <1213315624.14478.56.camel@pasglop>
References: <1211852026.3286.36.camel@pasglop> <4843C3D7.7000609@sgi.com>
	 <200806031433.12460.nickpiggin@yahoo.com.au>
	 <200806030952.10360.jbarnes@virtuousgeek.org> <4847A690.302@sgi.com>
	 <1212655433.9496.109.camel@pasglop>
	 <20080612150716.GX30405@parisc-linux.org>
Reply-To: benh@kernel.crashing.org
Mime-Version: 1.0
Content-Type: text/plain
Content-Transfer-Encoding: 7bit
Return-path: <linux-arch-owner@vger.kernel.org>
Received: from gate.crashing.org ([63.228.1.57]:44381 "EHLO gate.crashing.org"
	rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP
	id S1757454AbYFMAIc (ORCPT <rfc822;linux-arch@vger.kernel.org>);
	Thu, 12 Jun 2008 20:08:32 -0400
In-Reply-To: <20080612150716.GX30405@parisc-linux.org>
Sender: linux-arch-owner@vger.kernel.org
List-ID: <linux-arch.vger.kernel.org>
To: Matthew Wilcox <matthew@wil.cx>
Cc: Jes Sorensen <jes@sgi.com>, Jesse Barnes <jbarnes@virtuousgeek.org>, Nick Piggin <nickpiggin@yahoo.com.au>, Jeremy Higdon <jeremy@sgi.com>, Roland Dreier <rdreier@cisco.com>, Arjan van de Ven <arjan@infradead.org>, linux-arch@vger.kernel.org, linux-kernel@vger.kernel.org, tpiepho@freescale.com, linuxppc-dev@ozlabs.org, scottwood@freescale.com, torvalds@linux-foundation.org, David Miller <davem@davemloft.net>, alan@lxorguk.ukuu.org.uk

On Thu, 2008-06-12 at 09:07 -0600, Matthew Wilcox wrote:
> On Thu, Jun 05, 2008 at 06:43:53PM +1000, Benjamin Herrenschmidt wrote:
> > Note that the powerpc implementation currently clears the flag
> > on spin_lock and tests it on unlock. We are considering changing
> > that to not touch the flag on spin_lock and just clear it whenever
> > we do a sync (ie, on unlock, on explicit mmiowb, and possibly even
> > on readl's where we happen to do sync's).
> 
> Your current scheme sounds like it's broken for
> 
> spin_lock(a)
> writel();
> spin_lock(b);
> spin_unlock(b);
> spin_unlock(a);

Which is why we are considering changing it :-)

But as Paulus said before, he did some measurement and we came to the
conclusion that (pending more measurements on a wider range of HW) we
may as well drop the whole scheme and make writel fully synchronous
instead.

Then, we can get some nice weakly ordered accessors and start adding
them with appropriate explicit barriers to the hot path of perf.
critical drivers we care about.

Cheers,
Ben.