From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1752005Ab3HHEFz (ORCPT ); Thu, 8 Aug 2013 00:05:55 -0400 Received: from moutng.kundenserver.de ([212.227.126.187]:58413 "EHLO moutng.kundenserver.de" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1750832Ab3HHEFx (ORCPT ); Thu, 8 Aug 2013 00:05:53 -0400 Message-ID: <1375934704.5403.60.camel@marge.simpson.net> Subject: Re: [PATCH V3]hrtimer: Fix a performance regression by disable reprogramming in remove_hrtimer From: Mike Galbraith To: Peter Zijlstra Cc: Ethan Zhao , Thomas Gleixner , Ingo Molnar , LKML , johlstei@codeaurora.org, Yinghai Lu , Jin Feng , Youquan Song , LenBrown Date: Thu, 08 Aug 2013 06:05:04 +0200 In-Reply-To: <1375863939.12157.27.camel@marge.simpson.net> References: <1374955447-5051-1-git-send-email-ethan.kernel@gmail.com> <20130730093519.GP3008@twins.programming.kicks-ass.net> <1375774140.5412.9.camel@marge.simpson.net> <1375863939.12157.27.camel@marge.simpson.net> Content-Type: text/plain; charset="UTF-8" X-Mailer: Evolution 3.2.3 Content-Transfer-Encoding: 7bit Mime-Version: 1.0 X-Provags-ID: V02:K0:QjmhOedxadKGL+XWNbFtJRisR3gD8DHLZ378+AIso4m kqp2nILLWE4xhjcImcRIwTxN/5a1itOV36IMIffrSFv0X4OtcS HGil7Orlf/2xA33bNESoDTVbqHmFnKMHw0EaNiHN60QjJYd6V9 Sune8Fly20SejkCbeQwcHvPpe1q/1ckRFf3PT4zxGkCnXoKn6D JupAx1+lBmHa5FZXuSrSuXDi7Nei9DU+M8C9oX9m5mPLkfaIDj gVff1z8o6e/TWvQ45t170GKfRE8CJ8hXywcCmNustbcHl27jqO WBeQVo/4Xb53KwhkhalCsBxRfAfsE91dZ05ocaoxjp/BvcOawR P7rOv7+RTrsDcOMBIsVrEfaXoR7iLkDPteYZaQ7cHU+iLquhVC 8iQ3Jb39Jaj+g== Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On Wed, 2013-08-07 at 10:25 +0200, Mike Galbraith wrote: > E5620 (2.4 GHz Westmere) throttle > > v3.11-rc4-27-ge4ef108 v3.11-rc4-27-ge4ef108+peterz > 6.76% [k] __schedule 7.68% [k] reschedule_interrupt (find this little bastard) ^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^ > 5.90% [k] reschedule_interrupt 7.18% [k] __schedule > 4.12% [k] __switch_to 4.52% [k] __switch_to > 3.58% [k] resched_tas 4.24% [k] cpuidle_enter_state > 3.24% [k] cpuidle_enter_state 2.96% [k] cpu_idle_loop > 3.01% [k] _raw_spin_lock_irqsave 2.96% [k] resched_task > 2.67% [k] cpu_idle_loop 2.79% [k] _raw_spin_lock_irqsave > 2.49% [k] task_waking_fair 2.45% [k] ktime_get > 2.38% [k] copy_user_generic_string 2.34% [k] copy_user_generic_string > 2.27% [k] mutex_lock 2.32% [k] get_typical_interval 7d1a941731fabf27e5fb6edbebb79fe856edb4e5 is the first bad commit commit 7d1a941731fabf27e5fb6edbebb79fe856edb4e5 Author: Thomas Gleixner Date: Thu Mar 21 22:50:03 2013 +0100 x86: Use generic idle loop So, in summary, seems we need to forget all about scheduling anything remotely fast cross core these days, even on core2 with its lovely shared L2, as otherwise we'll beat it all up. Boxen with no nohz throttle maybe don't notice how bad all this progress hurt because they're used to being in agony. Your patch (modulo any bustage, haven't run any testsuite, only ran for regression hunt/test purposes) should make smiles for those who notice large chunk of that pain going away. My beloved ole Q6600 is currently not particularly happy, and that's best case, shared L2, where nearly any dinky overlap used to be reclaimable. Even with your patch, we'll still be losers at network fast movers across the board on any box methinks. -Mike