Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1762324AbXHONqT (ORCPT ); Wed, 15 Aug 2007 09:46:19 -0400 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S1757715AbXHONqB (ORCPT ); Wed, 15 Aug 2007 09:46:01 -0400 Received: from Chamillionaire.breakpoint.cc ([85.10.199.196]:34314 "EHLO Chamillionaire.breakpoint.cc" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1754245AbXHONqA (ORCPT ); Wed, 15 Aug 2007 09:46:00 -0400 Date: Wed, 15 Aug 2007 15:45:55 +0200 From: Sebastian Siewior To: Satyam Sharma Cc: Herbert Xu , Andi Kleen , Linux Kernel Mailing List Subject: Re: [patch 1/2] i386: use asm() like the other atomic operations already do. Message-ID: <20070815134555.GA29761@Chamillionaire.breakpoint.cc> References: MIME-Version: 1.0 Content-Type: text/plain; charset=iso-8859-15 Content-Disposition: inline In-Reply-To: X-Key-Id: FE3F4706 X-Key-Fingerprint: FFDA BBBB 3563 1B27 75C9 925B 98D5 5C1C FE3F 4706 User-Agent: Mutt/1.5.16 (2007-06-09) Sender: linux-kernel-owner@vger.kernel.org X-Mailing-List: linux-kernel@vger.kernel.org Content-Length: 2803 Lines: 62 * Satyam Sharma | 2007-08-15 18:32:10 [+0530]: >On Wed, 15 Aug 2007, Herbert Xu wrote: > >> Andi Kleen wrote: >> >> My config with march=pentium-m and gcc (GCC) 4.1.2 (Gentoo 4.1.2): >> >> text data bss dec hex filename >> >> 3434150 249176 176128 3859454 3ae3fe atomic_normal/vmlinux >> >> 3435308 249176 176128 3860612 3ae884 atomic_inlineasm/vmlinux >> > >> > What is the difference between atomic_normal and atomic_inlineasm? >> >> The inline asm stops certain optimisations from occuring. > >Yup, the "__volatile__" after "__asm__" shouldn't be required. The >previous definitions allowed the compiler to optimize certain atomic >ops away, and considering we haven't seen any bugs due to the past >behaviour anyway, changing it like this sounds unnecessary. > >The second behavioral change that comes about here is the force- >constraining of v->counter to memory in the extended asm's constraint >sets. But that's required to give "volatility"-like behaviour, which >I suspect is the goal of this patch. exactly. >Sebastian, could you look at the kernel images in detail and see why, >how and what in the text expanded by 1158 bytes with these changes? Wow, easy. "normal" is Linus' git. It has no volatile or anything. I suspect the compiler to "optimize" some of the reads and sets away (which remain after my patch is applied). Anyway, I tried to find out the source of this and it is not that easy. If the text is getting bigger, all references to the data section are changed and I can't simply diff. Than I searched for two "equal" .o files which differ in size. This was not better because gcc reorganized the whole .o file and the functions that were at the beginning in the first file, were at a different position in the second one. For those who have plenty of time in their hands, I posted the four kernels [1] and my four O= folders [2]. Once I have some time, I try to diff the specific function within the .o file which used the atomic_{read|set} operation. >> I'm still unconvinced why we need this because nobody has >> brought up any examples of kernel code that legitimately >> need this. > >Me too, but I do find these more palatable than the variants using >"volatile", I admit. Yes. The inline asm is even smaller than the volatile ones, what sounds like better optimization by the gcc (and it does exactly what volatile was used for). [1] http://download.breakpoint.cc/kernel_builds_vmlinux.tbz2 8.3 MiB [2] http://download.breakpoint.cc/kernel_builds.tbz2 117 MiB > >Satyam Sebastian - To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.kernel.org More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/