From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: from mail191.messagelabs.com (mail191.messagelabs.com [216.82.242.19]) by kanga.kvack.org (Postfix) with SMTP id 308136B01B3 for ; Tue, 23 Mar 2010 13:57:41 -0400 (EDT) Received: by fxm10 with SMTP id 10so1004732fxm.30 for ; Tue, 23 Mar 2010 10:57:38 -0700 (PDT) Subject: Re: [Bugme-new] [Bug 15618] New: 2.6.18->2.6.32->2.6.33 huge regression in performance Mime-Version: 1.0 (Apple Message framework v1077) Content-Type: text/plain; charset=us-ascii From: Anton Starikov In-Reply-To: Date: Tue, 23 Mar 2010 18:57:34 +0100 Content-Transfer-Encoding: quoted-printable Message-Id: <54F3A3FB-E99F-4278-AAAB-5B6A09247C4B@gmail.com> References: <20100323102208.512c16cc.akpm@linux-foundation.org> <20100323173409.GA24845@elte.hu> Sender: owner-linux-mm@kvack.org To: Linus Torvalds Cc: Ingo Molnar , Andrew Morton , linux-mm@kvack.org, linux-kernel@vger.kernel.org, bugzilla-daemon@bugzilla.kernel.org, bugme-daemon@bugzilla.kernel.org, Peter Zijlstra List-ID: On Mar 23, 2010, at 6:45 PM, Linus Torvalds wrote: >=20 >=20 > On Tue, 23 Mar 2010, Ingo Molnar wrote: >>=20 >> It shows a very brutal amount of page fault invoked mmap_sem spinning=20= >> overhead. >=20 > Isn't this already fixed? It's the same old "x86-64 rwsemaphores are = using=20 > the shit-for-brains generic version" thing, and it's fixed by >=20 > 1838ef1 x86-64, rwsem: 64-bit xadd rwsem implementation > 5d0b723 x86: clean up rwsem type system > 59c33fa x86-32: clean up rwsem inline asm statements >=20 > NOTE! None of those are in 2.6.33 - they were merged afterwards. But = they=20 > are in 2.6.34-rc1 (and obviously current -git). So Anton would have to=20= > compile his own kernel to test his load. Thanks for info, I will try it now. > We could mark them as stable material if the load in question is a = real=20 > load rather than just a test-case. On one of the random page-fault=20 > benchmarks the rwsem fix was something like a 400% performance=20 > improvement, and it was apparently visible in real life on some crazy = SGI=20 > "initialize huge heap concurrently on lots of threads" load. It is not just a test-case, it is real-life code. With real-life = problems on 2.6.32 and later :) Anton.= -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to majordomo@kvack.org. For more info on Linux MM, see: http://www.linux-mm.org/ . Don't email: email@kvack.org