From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S263640AbUHJJJE (ORCPT ); Tue, 10 Aug 2004 05:09:04 -0400 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S263429AbUHJJIk (ORCPT ); Tue, 10 Aug 2004 05:08:40 -0400 Received: from holomorphy.com ([207.189.100.168]:23015 "EHLO holomorphy.com") by vger.kernel.org with ESMTP id S263640AbUHJJGv (ORCPT ); Tue, 10 Aug 2004 05:06:51 -0400 Date: Tue, 10 Aug 2004 02:06:33 -0700 From: William Lee Irwin III To: Ingo Molnar Cc: Jesse Barnes , Andrew Morton , linux-kernel@vger.kernel.org, Nick Piggin Subject: Re: 2.6.8-rc3-mm2 Message-ID: <20040810090633.GL11200@holomorphy.com> Mail-Followup-To: William Lee Irwin III , Ingo Molnar , Jesse Barnes , Andrew Morton , linux-kernel@vger.kernel.org, Nick Piggin References: <200408091217.50786.jbarnes@engr.sgi.com> <20040809195323.GU11200@holomorphy.com> <20040809204357.GX11200@holomorphy.com> <20040809211042.GY11200@holomorphy.com> <20040809224546.GZ11200@holomorphy.com> <20040810063445.GE11200@holomorphy.com> <20040810080206.GF11200@holomorphy.com> <20040810083018.GA27270@elte.hu> <20040810085639.GJ11200@holomorphy.com> <20040810090051.GA28403@elte.hu> Mime-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <20040810090051.GA28403@elte.hu> User-Agent: Mutt/1.5.6+20040722i Sender: linux-kernel-owner@vger.kernel.org X-Mailing-List: linux-kernel@vger.kernel.org * William Lee Irwin III wrote: >> Actually, what I just narrowed it down to was *only* the printk change >> fixes it. On Tue, Aug 10, 2004 at 11:00:51AM +0200, Ingo Molnar wrote: > when i've seen such things on x86 it was usually some race with > interrupts on the other CPU. Where do all the ia64 interrupts go to > during bootup? > the other possibility is messed up completion logic - some stuff is > still on this CPU's kernel stack and the printk delays its > corruption/destruction. printk() seems to only have a few possible effects. I just tried mdelay() for the delay effect and not much appears to have happened. I'll probably try fiddling with schedule(), yield(), and local_irq_enable() and so on next. Your advice is very much like what I have in mind for possibilities, except I consider the messed up completion logic ruled out since backing out the completion removal part of the printk() "fix" didn't break it. -- wli