From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1032518AbXEHWgq (ORCPT ); Tue, 8 May 2007 18:36:46 -0400 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S968440AbXEHWgm (ORCPT ); Tue, 8 May 2007 18:36:42 -0400 Received: from smtp-out.google.com ([216.239.45.13]:14573 "EHLO smtp-out.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S968474AbXEHWgk (ORCPT ); Tue, 8 May 2007 18:36:40 -0400 DomainKey-Signature: a=rsa-sha1; s=beta; d=google.com; c=nofws; q=dns; h=received:date:from:x-x-sender:to:cc:subject:in-reply-to: message-id:references:mime-version:content-type; b=F18/xHjWC31B4KWMNJYD7LHVGmGDNNtOf7OEbqBEK/fn9Pmpasw0d/3wTi8PrOJ+K JnsTbHJ7e2QV+LWUx3TUg== Date: Tue, 8 May 2007 15:35:31 -0700 (PDT) From: David Rientjes X-X-Sender: rientjes@chino.kir.corp.google.com To: Jeremy Fitzhardinge cc: Randy Dunlap , Andrew Morton , Paul Sokolovsky , linux-kernel@vger.kernel.org, Andi Kleen Subject: Re: [RFC/PATCH] doc: volatile considered evil In-Reply-To: <4640F9C4.8050804@goop.org> Message-ID: References: <516386418.20070501080839@gmail.com> <20070430235642.e576e917.akpm@linux-foundation.org> <20070508121404.17bd97a6.randy.dunlap@oracle.com> <4640E98B.4030107@goop.org> <4640ED7D.708@goop.org> <4640F3FD.80001@goop.org> <4640F9C4.8050804@goop.org> MIME-Version: 1.0 Content-Type: TEXT/PLAIN; charset=us-ascii Sender: linux-kernel-owner@vger.kernel.org X-Mailing-List: linux-kernel@vger.kernel.org On Tue, 8 May 2007, Jeremy Fitzhardinge wrote: > Sounds like it's referring to micro-architectural reordering, which is > distinct from compiler reordering. In other words, even if you > specified "-mvolatile-asm-stop" I would assume that the compiler could > still reorder the asm statements. Am I right, or should I read more > into the manual description than it actually says? > The ia64 architecture does not include any interlocks for asm constructs and allow them to be executed in parallel without stop bits (although it should emit a diagnostic message from gcc, although that is not a constraint that was introduced in C99). Adding -mvolatile-asm-stops will add these stop bits for any asm volatile construct that does not already have them so that gcc is aware that it is unsafe to execute such constructs in parallel. [ This is a tangent because we currently don't have -mvolatile-asm-stop support and it's ia64-specific. If they ever request such support, then this will simply alter your warning about reordering asm volatile constructs to exclude the ia64 case. ] David