From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1752103Ab1HMOXo (ORCPT ); Sat, 13 Aug 2011 10:23:44 -0400 Received: from merlin.infradead.org ([205.233.59.134]:50622 "EHLO merlin.infradead.org" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1751987Ab1HMOXk convert rfc822-to-8bit (ORCPT ); Sat, 13 Aug 2011 10:23:40 -0400 Subject: Re: [ANNOUNCE] 3.0.1-rt11 From: Peter Zijlstra To: Mike Galbraith Cc: linux-kernel , Thomas Gleixner , "Paul E. McKenney" , linux-rt-users Date: Sat, 13 Aug 2011 16:23:13 +0200 In-Reply-To: <1313243965.4486.36.camel@marge.simson.net> References: <1313232790.25267.7.camel@twins> <1313236135.4486.10.camel@marge.simson.net> <1313236713.25267.10.camel@twins> <1313243965.4486.36.camel@marge.simson.net> Content-Type: text/plain; charset="UTF-8" Content-Transfer-Encoding: 8BIT X-Mailer: Evolution 3.0.2- Message-ID: <1313245393.25267.15.camel@twins> Mime-Version: 1.0 Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On Sat, 2011-08-13 at 15:59 +0200, Mike Galbraith wrote: > Hohum. rt10 did change the symptom. Box no longer gripes at some > random point while just idling along, now it gripes (and dies as well) > during boot. > > First boot, it choked on sr0 a wee bit later, second boot here. > > [ 40.582256] igb 0000:01:00.1: Intel(R) Gigabit Ethernet Network Connection > [ 40.582260] igb 0000:01:00.1: eth1: (PCIe:2.5Gb/s:Width x4) 08:00:69:15:c1:d5 > [ 40.582335] igb 0000:01:00.1: eth1: PBA No: FFFFFF-0FF > [ 40.582338] igb 0000:01:00.1: Using MSI-X interrupts. 8 rx queue(s), 8 tx queue(s) > [ 100.409012] INFO: rcu_preempt_state detected stalls on CPUs/tasks: {} (detected by 21, t=60002 jiffies) > > Guess I should try x3550 M3 or Q6600. They were griping the same way UV > box did earlier this morning (with an earlier -rt though), and they make > much smaller gripes. > > Gripe attached. Looks a lot like the old gripes to me, just earlier and > deadlier. But I don't speak rcu. Nothing stands out, but then, I'm tired and it are 32 cpu traces to look through. There is definitely something still fishy with RCU, the last patch in -rt10 is as far as we understand things a complete bandaid at best and horridly wrong otherwise. I think I'll try Paul's WARN_ON(t->rcu_read_lock_nesting == 0 && t->rcu_read_unlock_special); in __rcu_read_lock() sometime later. I also haven't read through my IRC backlog in which Paul and Steven discussed things way after I fell over and got keyboard face. /me needs to go out in the rain to get some shopping done, shops in this country I live in are closed on sunday :/