From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S1756772Ab0EaJkh (ORCPT ); Mon, 31 May 2010 05:40:37 -0400 Received: from ozlabs.org ([203.10.76.45]:45285 "EHLO ozlabs.org" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with ESMTP id S1756529Ab0EaJkg (ORCPT ); Mon, 31 May 2010 05:40:36 -0400 From: Rusty Russell To: Brandon Philips Subject: Re: [Regression] Crash in load_module() while freeing args Date: Mon, 31 May 2010 19:10:31 +0930 User-Agent: KMail/1.13.2 (Linux/2.6.32-21-generic; KDE/4.4.2; i686; ; ) Cc: "Rafael J. Wysocki" , Linus Torvalds , LKML , Andrew Morton , Jon Masters , Tejun Heo , Masami Hiramatsu References: <201005252300.07739.rjw@sisk.pl> <201005271456.20003.rusty@rustcorp.com.au> <20100527184602.GF7351@jenkins.ifup.org> In-Reply-To: <20100527184602.GF7351@jenkins.ifup.org> MIME-Version: 1.0 Content-Type: Text/Plain; charset="iso-8859-1" Content-Transfer-Encoding: 7bit Message-Id: <201005311910.32407.rusty@rustcorp.com.au> Sender: linux-kernel-owner@vger.kernel.org List-ID: X-Mailing-List: linux-kernel@vger.kernel.org On Fri, 28 May 2010 04:16:02 am Brandon Philips wrote: > On 14:56 Thu 27 May 2010, Rusty Russell wrote: > > On Thu, 27 May 2010 08:26:25 am Rafael J. Wysocki wrote: > > > On Wednesday 26 May 2010, Rusty Russell wrote: > > > > I suspect that the increased parallelism enabled by this patch uncovered this > > > > bug. Does this fix it? > > > > > > Since the commit has been reverted, do you still want me to test this patch? > > > Quite frankly I'd prefer to test a complete replacement for that commit on top > > > of current -git. > > > > OK, combo meal deal below, against Linus' latest. I'd really appreciate > > a report, since AFAIK you're the only one hitting it, and only when that > > other (now reverted) patch was applied. > > I tested this patch on my machine on top of Linus's latest and it > fixes the issue. Without the patch and using Linus's latest I > reproduce the original issue: > > [ 60.836022] bnx2: gave up waiting for init of module libcrc32c. > [ 60.847997] bnx2: Unknown symbol crc32c > > Note: Again, since I don't have bnx2x hardware I forced bnx2.ko to > depend on libcrc32c as bnx2x does: > http://ifup.org/~philips/review/bnx2-hack-to-use-libcrc32c.patch > > > As an side to Brandon: I can see how my patch fixed an explicit > > request_module inside module_init (that's how I tested it). I can't > > see how we have a problem with an implicit dependency such as > > bne2x->crc32. Modules go into the live state without retaking the > > lock. > > libcrc32c is doing an explicit request_module inside of its > module_init. Follow the call chain in libcrc32c_mod_init() Thanks for confirmation, I figured that must be the case as I pondered it on the weekend after sending my query. Linus didn't like dropping the lock, so I'll create a more ambitious patch which reduces the lock coverage to those places which really need it. Cheers, Rusty.