From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1752954AbZGTIMT (ORCPT ); Mon, 20 Jul 2009 04:12:19 -0400 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S1752563AbZGTIMR (ORCPT ); Mon, 20 Jul 2009 04:12:17 -0400 Received: from mail-qy0-f198.google.com ([209.85.221.198]:60390 "EHLO mail-qy0-f198.google.com" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1752486AbZGTIMQ (ORCPT ); Mon, 20 Jul 2009 04:12:16 -0400 DomainKey-Signature: a=rsa-sha1; c=nofws; d=gmail.com; s=gamma; h=date:from:to:cc:subject:message-id:references:mime-version :content-type:content-disposition:in-reply-to:user-agent; b=gJI0ZKj1CQoTqmypkklyEV/mI63wzorT5rILwTmmxvMkOFD1e159YP2gFcbDo0NPWM p7ul9C1+bPOYBnIbTfAkO4R74cOpT2JOkcOFw8xHVmut3YEhNTQCDuBTD8viZ5IBF5Eq vXdNFRomPNe0JYx02CuCjn7mv0vct8/w3nMLc= Date: Mon, 20 Jul 2009 04:12:12 -0400 From: Frederic Weisbecker To: Li Zefan Cc: hpa@zytor.com, linux-kernel@vger.kernel.org, a.p.zijlstra@chello.nl, tglx@linutronix.de, mingo@elte.hu, linux-tip-commits@vger.kernel.org Subject: Re: [tip:sched/core] sched: Pull up the might_sleep() check into cond_resched() Message-ID: <20090720081210.GA5309@nowhere> References: <1247725694-6082-6-git-send-email-fweisbec@gmail.com> <4A6413AB.2050807@cn.fujitsu.com> MIME-Version: 1.0 Content-Type: text/plain; charset=us-ascii Content-Disposition: inline In-Reply-To: <4A6413AB.2050807@cn.fujitsu.com> User-Agent: Mutt/1.5.18 (2008-05-17) Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On Mon, Jul 20, 2009 at 02:50:19PM +0800, Li Zefan wrote: > > Commit-ID: 613afbf83298efaead05ebcac23d2285609d7160 > > Gitweb: http://git.kernel.org/tip/613afbf83298efaead05ebcac23d2285609d7160 > > Author: Frederic Weisbecker > > AuthorDate: Thu, 16 Jul 2009 15:44:29 +0200 > > Committer: Ingo Molnar > > CommitDate: Sat, 18 Jul 2009 15:51:44 +0200 > > > > sched: Pull up the might_sleep() check into cond_resched() > > > > might_sleep() is called late-ish in cond_resched(), after the > > need_resched()/preempt enabled/system running tests are > > checked. > > > > It's better to check the sleeps while atomic earlier and not > > depend on some environment datas that reduce the chances to > > detect a problem. > > > > Also define cond_resched_*() helpers as macros, so that the > > FILE/LINE reported in the sleeping while atomic warning > > displays the real origin and not sched.h > > > > I guess it's this patch that causes lots of "BUG" > > BUG: sleeping function called from invalid context at fs/jbd/commit.c:902 > in_atomic(): 0, irqs_disabled(): 0, pid: 64, name: kjournald > INFO: lockdep is turned off. > Pid: 64, comm: kjournald Tainted: GF 2.6.31-rc3-tip #15 > Call Trace: > [] __might_sleep+0xda/0xdf > [] journal_commit_transaction+0xb03/0xc5f > [] ? try_to_del_timer_sync+0x48/0x4f > [] kjournald+0xcf/0x1fe > [] ? autoremove_wake_function+0x0/0x34 > [] ? kjournald+0x0/0x1fe > [] kthread+0x6b/0x70 > [] ? kthread+0x0/0x70 > [] kernel_thread_helper+0x7/0x10 > BUG: sleeping function called from invalid context at fs/dcache.c:512 > in_atomic(): 0, irqs_disabled(): 0, pid: 2005, name: bash > INFO: lockdep is turned off. > Pid: 2005, comm: bash Tainted: GF 2.6.31-rc3-tip #15 > Call Trace: > [] __might_sleep+0xda/0xdf > [] __shrink_dcache_sb+0x208/0x27a > [] shrink_dcache_parent+0x2c/0xcf > [] proc_flush_task+0xa7/0x194 > [] release_task+0x29/0x3b4 > [] wait_consider_task+0x702/0xa91 > [] do_wait+0xde/0x276 > [] ? default_wake_function+0x0/0x12 > [] sys_wait4+0x8d/0xa6 > [] ? might_fault+0x85/0x87 > [] sys_waitpid+0x18/0x1a > [] sysenter_do_call+0x12/0x36 Hm, I can read that in fs/dcache.c:512 /* dentry->d_lock was dropped in prune_one_dentry() */ cond_resched_lock(&dcache_lock); Isn't it a mususe of cond_resched_lock() ? In this case, dcache.c should be fixed. Anyway a generic fix could be the following. Can you tell me if this works for you? Thanks! --- From: Frederic Weisbecker Subject: [PATCH] sched: Check if the spinlock is locked in cond_resched_lock() Some uses of cond_resched_lock() might involve an unlocked spinlock, resulting in spurious sleep in atomic warnings. Check whether the spinlock is actually locked and take that into account in the might_sleep() check. Reported-by: Li Zefan Signed-off-by: Frederic Weisbecker --- diff --git a/include/linux/sched.h b/include/linux/sched.h index cb070dc..2789658 100644 --- a/include/linux/sched.h +++ b/include/linux/sched.h @@ -2294,9 +2294,10 @@ extern int _cond_resched(void); extern int __cond_resched_lock(spinlock_t *lock); -#define cond_resched_lock(lock) ({ \ - __might_sleep(__FILE__, __LINE__, PREEMPT_OFFSET); \ - __cond_resched_lock(lock); \ +#define cond_resched_lock(lock) ({ \ + __might_sleep(__FILE__, __LINE__, spin_is_locked(lock) ? \ + PREEMPT_OFFSET : 0); \ + __cond_resched_lock(lock); \ }) extern int __cond_resched_softirq(void);