From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1423269AbWJYLJn (ORCPT ); Wed, 25 Oct 2006 07:09:43 -0400 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S1423270AbWJYLJn (ORCPT ); Wed, 25 Oct 2006 07:09:43 -0400 Received: from mail.dsa-ac.de ([62.112.80.99]:18448 "EHLO mail.dsa-ac.de") by vger.kernel.org with ESMTP id S1423269AbWJYLJm (ORCPT ); Wed, 25 Oct 2006 07:09:42 -0400 Date: Wed, 25 Oct 2006 13:09:37 +0200 (CEST) From: Guennadi Liakhovetski To: Ingo Molnar , linux-kernel@vger.kernel.org Subject: Re: [2.6.18-rt6] BUG / typo In-Reply-To: Message-ID: References: MIME-Version: 1.0 Content-Type: TEXT/PLAIN; charset=US-ASCII; format=flowed Sender: linux-kernel-owner@vger.kernel.org X-Mailing-List: linux-kernel@vger.kernel.org Hi again Just a bit more detail to the way the bug occurs: you just need CONFIG_LATENCY_TIMING=y and in my case I had 2 consoles: tty1 and ttyS0. tty1 is a framebuffer. I don't know where irqs get disabled (the original BUG was BUG: scheduling with irqs disabled: posix_cpu_timer/0x00000001/2 caller is rt_spin_lock_slowlock+0xd8/0x1c8 but is it at all ok to schedule in kmalloc(GFP_ATOMIC)? Which is exactly what happens here as kmalloc tries to acquire the per-cpu "spinlock" / mutex slab_irq_locks and is forced into the slow path. The bug often triggers when I reset /proc/sys/kernel/preempt_max_latency with 0 and if I have high enough console logging level in /proc/sys/kernel/printk. Thanks Guennadi --------------------------------- Guennadi Liakhovetski, Ph.D. DSA Daten- und Systemtechnik GmbH Pascalstr. 28 D-52076 Aachen Germany