From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1751562AbaJRUga (ORCPT ); Sat, 18 Oct 2014 16:36:30 -0400 Received: from mga02.intel.com ([134.134.136.20]:2270 "EHLO mga02.intel.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751459AbaJRUg2 (ORCPT ); Sat, 18 Oct 2014 16:36:28 -0400 X-ExtLoop1: 1 X-IronPort-AV: E=Sophos;i="5.04,745,1406617200"; d="scan'208";a="621126754" Date: Sat, 18 Oct 2014 13:22:07 -0400 From: Matthew Wilcox To: Mathieu Desnoyers Cc: Matthew Wilcox , Matthew Wilcox , linux-fsdevel@vger.kernel.org, linux-mm@kvack.org, linux-kernel@vger.kernel.org Subject: Re: [PATCH v11 04/21] mm: Allow page fault handlers to perform the COW Message-ID: <20141018172207.GN11522@wil.cx> References: <1411677218-29146-1-git-send-email-matthew.r.wilcox@intel.com> <1411677218-29146-5-git-send-email-matthew.r.wilcox@intel.com> <20141016091136.GC19075@thinkos.etherlink> <20141016194815.GD11522@wil.cx> <289646725.10903.1413560101974.JavaMail.zimbra@efficios.com> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <289646725.10903.1413560101974.JavaMail.zimbra@efficios.com> User-Agent: Mutt/1.5.23 (2014-03-12) Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On Fri, Oct 17, 2014 at 03:35:01PM +0000, Mathieu Desnoyers wrote: > > > The page fault handler being very much performance sensitive, I'm > > > wondering if it would not be better to move cow_page near the end of > > > struct vm_fault, so that the "page" field can stay on the first > > > cache line. > > Although it's pretty much always true that recent architectures L2 cache > lines are 64 bytes, I was more thinking about L1 cache lines, which are, > at least on moderately old Intel Pentium HW, 32 bytes in size (AFAIK > Pentium II and III). > > It remains to be seen whether we care about performance that much on this > kind of HW though. Oh, I just remembered ... this data structure is on the stack, so if it's not cache-hot, something has gone horribly wrong.