From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1753524Ab2ECGnr (ORCPT ); Thu, 3 May 2012 02:43:47 -0400 Received: from terminus.zytor.com ([198.137.202.10]:47384 "EHLO mail.zytor.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751114Ab2ECGnp (ORCPT ); Thu, 3 May 2012 02:43:45 -0400 Message-ID: <4FA22915.6050400@zytor.com> Date: Wed, 02 May 2012 23:43:33 -0700 From: "H. Peter Anvin" User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:11.0) Gecko/20120329 Thunderbird/11.0.1 MIME-Version: 1.0 To: NeilBrown CC: Jim Kukunas , linux-raid@vger.kernel.org, linux-kernel@vger.kernel.org Subject: Re: [PATCH] arch/x86: use kernel_fpu_[begin|end] for RAID5 checksumming References: <1335829875-18481-1-git-send-email-james.t.kukunas@linux.intel.com> <20120503145744.4e425dd5@notabene.brown> In-Reply-To: <20120503145744.4e425dd5@notabene.brown> X-Enigmail-Version: 1.4.1 Content-Type: text/plain; charset=ISO-8859-1 Content-Transfer-Encoding: 7bit Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org -----BEGIN PGP SIGNED MESSAGE----- Hash: SHA1 On 05/02/2012 09:57 PM, NeilBrown wrote: > On Mon, 30 Apr 2012 16:51:15 -0700 Jim Kukunas > wrote: > >> Currently, the SSE and AVX xor functions manually save and >> restore the [X|Y]MM registers. Instead, we should use >> kernel_fpu_[begin|end]. >> >> This patch sacrifices some throughput,~5-10% for AVX and ~2% for >> SSE, in exchange for safety against future FPU corruption bugs. >> >> Patch applies to md/for-next. >> >> Signed-off-by: Jim Kukunas > > Hi Jim, I'm really not able to assess the validity of this patch. > I don't doubt it exactly, but I'd value a second opinion just to > be confident. Peter: can you review and comment? > > Thanks, NeilBrown > Are these functions ever called with unaligned arguments? Other than that, the patch is correct, and is the preferred way to do this, but I'm unhappy about the cost. I have suggested to Jim a few ways by which we might be able to mitigate the cost or even improve performance, but the implementation depends very much on the above question (unlike the RAID-6 code, which assumes aligned buffers at all times.) -hpa -----BEGIN PGP SIGNATURE----- Version: GnuPG v1.4.12 (GNU/Linux) Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org/ iQIcBAEBAgAGBQJPoikHAAoJEL2gYIVJO6zk9Q8P/AqcdBC2fal6Wy9RSWnky4vf BLycwd7JAK6CU6Tl/dK747gpEdzXwFOPPrGV3pbyycft8rKPvUoliBhE9esM8GN6 5wZfn9dBBXa3Gmacxdu5mj87W54LpQOCc8rGj3CSx8kfoORjBCxC0zueX3Wmno/I KdL6JtWiSEgJ93WLRr4mYfo1V8rulJ5kAxTthntyAmuIiJi4nc4ERyMLRXEH+3p2 iam8f4v4JDhm2qVxnkSX2toPjzEVLPXSa5DDhHJq8OC9o8jl6gL8nIByG7/0hYMG Z3mI4pG4GB7jdX8YxSI8sGo0rB1bXtnLEI61FAZo4vOJRs1xJr98m3nwiE4X81NV OU+xucUPpd+Y5x6oC7tZyDkoKqnkNH1nNHhhyKawe4pMWWom9m8V5skkYrXufr8G K1unlM0zfUUAH2vFbEuRqbGIMfhH2qFCWp/oywRK0M+R/HjLISVjjIIYCB87Y30K dZ+y8z6bEYRm0WyFSw/SkEulSyAu/VfHcOAmYorvu721MG2HfPkdWiviwgbf6CD2 33bZ3IUkaian619eQC+wBliGIjvR6PexxTzilIbfV4D20t9QhvBVSkkbT9cIaFeF 38gC8LxDBvrSYyutn4v6e6NUAUQlbtQnPZJQfb8uDbtOe74Er8qsZJGsCyt1bgv1 EjeLFnHOn4F9pE/g7GNh =wGxV -----END PGP SIGNATURE-----