From owner-freebsd-current  Wed Jan 17 12:26:18 2001
Delivered-To: freebsd-current@freebsd.org
Received: from netau1.alcanet.com.au (ntp.alcanet.com.au [203.62.196.27])
	by hub.freebsd.org (Postfix) with ESMTP
	id 2678137B698; Wed, 17 Jan 2001 12:25:57 -0800 (PST)
Received: from mfg1.cim.alcatel.com.au (mfg1.cim.alcatel.com.au [139.188.23.1])
	by netau1.alcanet.com.au (8.9.3 (PHNE_22672)/8.9.3) with ESMTP id HAA27841;
	Thu, 18 Jan 2001 07:23:07 +1100 (EDT)
Received: from gsmx07.alcatel.com.au by cim.alcatel.com.au
 (PMDF V5.2-32 #37641) with ESMTP id <01JZ1LYNYTNKEMY2IF@cim.alcatel.com.au>;
 Thu, 18 Jan 2001 07:23:05 +1100
Received: (from jeremyp@localhost)	by gsmx07.alcatel.com.au (8.11.0/8.11.0)
 id f0HKN2212246; Thu, 18 Jan 2001 07:23:02 +1100 (EST envelope-from jeremyp)
Content-return: prohibited
Date: Thu, 18 Jan 2001 07:23:02 +1100
From: Peter Jeremy <peter.jeremy@alcatel.com.au>
Subject: Re: Atomic breakage?
In-reply-to: <XFMail.010117102803.jhb@FreeBSD.org>; from jhb@FreeBSD.ORG on
 Wed, Jan 17, 2001 at 10:28:03AM -0800
To: John Baldwin <jhb@FreeBSD.ORG>
Cc: Poul-Henning Kamp <phk@critter.freebsd.dk>, current@FreeBSD.ORG,
	Garrett Wollman <wollman@khavrinen.lcs.mit.edu>,
	Robert Drehmel <robert@gizmo.quizbot.org>
Mail-followup-to: John Baldwin <jhb@FreeBSD.ORG>,
 Poul-Henning Kamp <phk@critter.freebsd.dk>, current@FreeBSD.ORG,
 Garrett Wollman <wollman@khavrinen.lcs.mit.edu>,
 Robert Drehmel <robert@gizmo.quizbot.org>
Message-id: <20010118072302.E98607@gsmx07.alcatel.com.au>
MIME-version: 1.0
Content-type: text/plain; charset=us-ascii
Content-disposition: inline
User-Agent: Mutt/1.2.5i
References: <53512.979753444@critter> <XFMail.010117102803.jhb@FreeBSD.org>
Sender: owner-freebsd-current@FreeBSD.ORG
Precedence: bulk
X-Loop: FreeBSD.ORG

On 2001-Jan-17 10:28:03 -0800, John Baldwin <jhb@FreeBSD.ORG> wrote:
>  So as long as we keep all
>atomically-accessed 64-bit integers within a single page we should be fine for
>CX8 on all pentiums should we even want 64-bit atomic ops.

This comes free of charge (along with better performance) if all
64-bit objects are aligned on 8-byte boundaries.  By default our gcc
only enforces long (4-byte) alignment (apparently, this is part of the
IA32 ABI).  We can get 8-byte alignment by adding
"__attribute__((aligned(8)))" to the object definition.

This is one area where I think we would be better off changing our
gcc to default to aligning "long long" (and maybe "double") at
8-byte boundaries.

We should at least have an alignment expansion in <sys/cdefs.h>

Peter



To Unsubscribe: send mail to majordomo@FreeBSD.org
with "unsubscribe freebsd-current" in the body of the message