From mboxrd@z Thu Jan  1 00:00:00 1970
Return-Path: <linux-kernel-owner+willy=40w.ods.org-S267302AbUHPANN@vger.kernel.org>
Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand
	id S267302AbUHPANN (ORCPT <rfc822;willy@w.ods.org>);
	Sun, 15 Aug 2004 20:13:13 -0400
Received: (majordomo@vger.kernel.org) by vger.kernel.org id S267301AbUHPAND
	(ORCPT <rfc822;linux-kernel-outgoing>);
	Sun, 15 Aug 2004 20:13:03 -0400
Received: from colin2.muc.de ([193.149.48.15]:27653 "HELO colin2.muc.de")
	by vger.kernel.org with SMTP id S267302AbUHPAMl (ORCPT
	<rfc822;linux-kernel@vger.kernel.org>);
	Sun, 15 Aug 2004 20:12:41 -0400
Date: 16 Aug 2004 02:12:38 +0200
Date: Mon, 16 Aug 2004 02:12:38 +0200
From: Andi Kleen <ak@muc.de>
To: Christoph Lameter <christoph@lameter.com>
Cc: Christoph Lameter <clameter@sgi.com>, linux-ia64@vger.kernel.org,
       linux-kernel@vger.kernel.org
Subject: Re: page fault fastpath: Increasing SMP scalability by introducing pte locks?
Message-ID: <20040816001238.GA6978@muc.de>
References: <2ttIr-2e4-17@gated-at.bofh.it> <2tzE4-6sw-25@gated-at.bofh.it> <2tCiw-8pK-1@gated-at.bofh.it> <m31xi8dkm2.fsf@averell.firstfloor.org> <Pine.LNX.4.58.0408151654240.4346@server.home>
Mime-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <Pine.LNX.4.58.0408151654240.4346@server.home>
User-Agent: Mutt/1.4.1i
Sender: linux-kernel-owner@vger.kernel.org
X-Mailing-List: linux-kernel@vger.kernel.org

On Sun, Aug 15, 2004 at 04:55:57PM -0700, Christoph Lameter wrote:
> On Mon, 16 Aug 2004, Andi Kleen wrote:
> 
> > Christoph Lameter <clameter@sgi.com> writes:
> >
> > > On Sun, 15 Aug 2004, David S. Miller wrote:
> > >
> > >>
> > >> Is the read lock in the VMA semaphore enough to let you do
> > >> the pgd/pmd walking without the page_table_lock?
> > >> I think it is, but just checking.
> > >
> > > That would be great.... May I change the page_table lock to
> > > be a read write spinlock instead?
> >
> > That's probably not a good idea. r/w locks are extremly slow on
> > some architectures. Including ia64.
> 
> I was thinking about a read write spinlock not an readwrite
> semaphore. Look at include/asm-ia64/spinlock.h.

I was also talking about rw spinlocks.

> The implementations are almost the same. Are you sure
> about this?

Yes. Try the cat /proc/net/tcp test. It will take >100k read locks
for the TCP listen hash table, and on bigger ppc64 and ia64 machines this
can take nearly a second of system time.

-Andi