From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1756487AbYGNGkO (ORCPT ); Mon, 14 Jul 2008 02:40:14 -0400 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S1751621AbYGNGkC (ORCPT ); Mon, 14 Jul 2008 02:40:02 -0400 Received: from one.firstfloor.org ([213.235.205.2]:42680 "EHLO one.firstfloor.org" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1750766AbYGNGkB (ORCPT ); Mon, 14 Jul 2008 02:40:01 -0400 Message-ID: <487AF4BB.6050805@firstfloor.org> Date: Mon, 14 Jul 2008 08:39:55 +0200 From: Andi Kleen User-Agent: Thunderbird 1.5.0.12 (X11/20060911) MIME-Version: 1.0 To: "H. Peter Anvin" CC: Ingo Molnar , Yinghai Lu , Arjan van de Ven , Thomas Gleixner , Suresh Siddha , LKML Subject: Re: [PATCh] x86: overmapped fix when 4K pages on tail - 64bit References: <200807080141.05436.yhlu.kernel@gmail.com> <200807080143.27997.yhlu.kernel@gmail.com> <200807092015.03004.yhlu.kernel@gmail.com> <20080710071640.5035cd70@infradead.org> <874p6t25n5.fsf@basil.nowhere.org> <86802c440807131117g3ba9e61chea61af81b7537bb0@mail.gmail.com> <487A4DFC.5090701@firstfloor.org> <20080713203250.GA6925@elte.hu> <487A6AD0.5000506@firstfloor.org> <487A9804.8090307@zytor.com> In-Reply-To: <487A9804.8090307@zytor.com> 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 H. Peter Anvin wrote: > Andi Kleen wrote: >> >> First I was only commenting on one specific patch, nothing more. >> >> My point is full rounding to 4K on all corners is wasteful because the >> CPUs have to handle that case anyways and every split costs precious >> TLB entries in direct mapping accesses. >> > > Well, the CPU *does* handle them... by splitting the larger pages into > smaller pages. They still end up in the small-page TLB, so there is no > real difference if done in the CPU or in software. There's actually a difference in some cases, but that's a different issue. Only when the hole is a real hole. But when it's just some firmware area or similar that's not needed. > Memory state transitions are (fortunately) relatively rare and > long-lived, That's not true today with several 3d driver setups. Also in general I would expect more PAT use in the future and that is usually a split. -Andi