From mboxrd@z Thu Jan 1 00:00:00 1970 Return-Path: Received: (majordomo@vger.kernel.org) by vger.kernel.org via listexpand id S933470AbXCJPoS (ORCPT ); Sat, 10 Mar 2007 10:44:18 -0500 Received: (majordomo@vger.kernel.org) by vger.kernel.org id S933472AbXCJPoS (ORCPT ); Sat, 10 Mar 2007 10:44:18 -0500 Received: from mail.gmx.net ([213.165.64.20]:59128 "HELO mail.gmx.net" rhost-flags-OK-OK-OK-OK) by vger.kernel.org with SMTP id S933470AbXCJPoR (ORCPT ); Sat, 10 Mar 2007 10:44:17 -0500 X-Authenticated: #14349625 X-Provags-ID: V01U2FsdGVkX1+31htJut23nrRKJNPrimYsMWv4PfcK96EaPcWY9G pElXNdKy1bIXLP Subject: Re: kref refcounting breakage in mainline From: Mike Galbraith To: Greg KH Cc: Andrew Morton , Tejun Heo , Kay Sievers , linux-kernel@vger.kernel.org, Adrian Bunk In-Reply-To: <1173245937.6613.26.camel@Homer.simpson.net> References: <20070302005833.949be737.akpm@linux-foundation.org> <20070306002521.GA12164@kroah.com> <1173159802.6955.6.camel@Homer.simpson.net> <20070306210445.GB29164@kroah.com> <1173245937.6613.26.camel@Homer.simpson.net> Content-Type: text/plain Date: Sat, 10 Mar 2007 16:44:06 +0100 Message-Id: <1173541446.6561.32.camel@Homer.simpson.net> Mime-Version: 1.0 X-Mailer: Evolution 2.8.2 Content-Transfer-Encoding: 7bit X-Y-GMX-Trusted: 0 Sender: linux-kernel-owner@vger.kernel.org X-Mailing-List: linux-kernel@vger.kernel.org On Wed, 2007-03-07 at 06:39 +0100, Mike Galbraith wrote: > On Tue, 2007-03-06 at 13:04 -0800, Greg KH wrote: > > On Tue, Mar 06, 2007 at 06:43:22AM +0100, Mike Galbraith wrote: > > > On Mon, 2007-03-05 at 16:25 -0800, Greg KH wrote: > > > > > > > Mike, I've reverted this patch, and I don't see any references leaking. > > > > And, as your patch released the reference on the driver, and the > > > > module_add_driver() call would not grab a reference to the driver, only > > > > the module kobject, I don't see what you were trying to fix with this > > > > patch. > > > > > > > > Do you have a test case that this fixes? > > > > > > What it fixed for me was the hard hang reported below. > > > > > > http://lkml.org/lkml/2007/2/16/96 > > > > What specific module are you trying to unload that causes the hang? I > > think it might just be a problem with that module, and not with all > > others. > > It's ipmi_si that's hanging, waits for completion that never comes. > > > So, I'm going to revert your patch and work to try to find the real > > cause of this problem. > > Yeah, my stab at it seems busted. I'll take another poke at it to see > if I can find out why (post 725522b5453dd680412f2b6463a988e4fd148757) > I'm left with a reference. Ok, stab #2. My reference count woes stem from module_remove_driver() not removing the link created in module_add_driver(). With the below, my box boots fine. Since I obviously know spit about driver layer glue, I'll just call this one a diagnostic, and head for the hills :) --- linux-2.6.20-rc3/kernel/module.c.org 2007-03-10 15:16:47.000000000 +0100 +++ linux-2.6.20-rc3/kernel/module.c 2007-03-10 15:43:09.000000000 +0100 @@ -2411,14 +2411,28 @@ void module_remove_driver(struct device_ return; sysfs_remove_link(&drv->kobj, "module"); - if (drv->owner && drv->owner->mkobj.drivers_dir) { - driver_name = make_driver_name(drv); - if (driver_name) { - sysfs_remove_link(drv->owner->mkobj.drivers_dir, + driver_name = make_driver_name(drv); + if (!driver_name) + return; + if (drv->owner && drv->owner->mkobj.drivers_dir) + sysfs_remove_link(drv->owner->mkobj.drivers_dir, driver_name); - kfree(driver_name); - } + else if (drv->mod_name) { + struct module_kobject *mk; + struct kobject *mkobj; + + /* Lookup built-in module entry in /sys/modules */ + mkobj = kset_find_obj(&module_subsys.kset, drv->mod_name); + if (!mkobj) + goto out_free; + mk = container_of(mkobj, struct module_kobject, kobj); + module_create_drivers_dir(mk); + sysfs_remove_link(mk->drivers_dir, driver_name); + /* Release reference taken via lookup */ + kobject_put(mkobj); } +out_free: + kfree(driver_name); } EXPORT_SYMBOL(module_remove_driver); #endif