[patch] Re: 2.6.20.git regression: 'PCI: add the sysfs driver name to all modules' causes hard hang on boot

[Mike Galbraith <efault@gmx.de>]

On Sun, 2007-02-18 at 10:27 +0100, Mike Galbraith wrote:
> On Sun, 2007-02-18 at 09:02 +0100, Mike Galbraith wrote:
> 
> > The reason it's hanging is that nobody releases the driver, so we wait
> > forever in driver_unregister().  With the below, box boots fine...
> > 
> > --- drivers/base/bus.c.org	2007-02-18 08:38:57.000000000 +0100
> > +++ drivers/base/bus.c	2007-02-18 08:39:09.000000000 +0100
> > @@ -593,6 +593,7 @@ void bus_remove_driver(struct device_dri
> >  	driver_detach(drv);
> >  	module_remove_driver(drv);
> >  	kobject_unregister(&drv->kobj);
> > +	driver_release(&drv->kobj);
> >  	put_bus(drv->bus);
> >  }
> >  
> > 
> > ...but that can't be right given that the darn thing booted just fine
> > prior to the naming patch with an equally unhappy init_ipmi_si().  Hmm.
> 
> Ok.  The path it's supposed to take to driver_release() goes like so....
> 
> [   17.495312] bus platform: add driver ipmi
> [   17.506560] ipmi message handler version 39.1
> [   17.518099] ipmi device interface
> [   17.528491] device class 'ipmi': registering
> [   17.539854] bus platform: add driver ipmi_si
> [   17.551210] IPMI System Interface driver.
> [   17.562242] bus pci: add driver ipmi_si
> [   17.583686] bus pci: remove driver ipmi_si
> [   17.594721] BUG: at drivers/base/bus.c:65 driver_release()
> [   17.607224]  [<c0105136>] show_trace_log_lvl+0x1a/0x30
> [   17.619434]  [<c0105862>] show_trace+0x12/0x14
> [   17.630822]  [<c0105906>] dump_stack+0x16/0x18
> [   17.642098]  [<c034b632>] driver_release+0x37/0x39
> [   17.653703]  [<c02c73b9>] kobject_cleanup+0x43/0x64
> [   17.665359]  [<c02c73e5>] kobject_release+0xb/0xd
> [   17.676748]  [<c02c8017>] kref_put+0x28/0x8c
> [   17.687626]  [<c02c7374>] kobject_put+0x14/0x16
> [   17.698712]  [<c02c74c4>] kobject_unregister+0x22/0x25
> [   17.710359]  [<c034b7e0>] bus_remove_driver+0x95/0xa5
> [   17.721911]  [<c034c87b>] driver_unregister+0xe/0x47
> [   17.733317]  [<c02d59ac>] pci_unregister_driver+0x13/0x73
> [   17.745149]  [<c033e141>] init_ipmi_si+0x798/0x7ba
> [   17.756339]  [<c065b58c>] init+0x114/0x23c
> [   17.766748]  [<c0104dab>] kernel_thread_helper+0x7/0x1c
> 
> ...so I guess it's a ref counting problem somewhere.

The below fixes a reference counting bug exposed by commit
725522b5453dd680412f2b6463a988e4fd148757.  If driver.mod_name exists, we
take a reference in module_add_driver(), and never release it.  Undo
that reference in module_remove_driver().

My box now boots fine, and modprobe/rmmod didn't explode, so I'll add a
blame line.

Signed-off-by: Mike Galbraith <efault@gmx.de>

--- a/kernel/module.c.org	2007-02-19 06:41:02.000000000 +0100
+++ b/kernel/module.c	2007-02-19 06:49:08.000000000 +0100
@@ -2417,6 +2417,12 @@ void module_remove_driver(struct device_
 			kfree(driver_name);
 		}
 	}
+	/*
+	 * Undo the additional reference we added in module_add_driver()
+	 * via kset_find_obj()
+	 */
+	if (drv->mod_name)
+		kobject_put(&drv->kobj);
 }
 EXPORT_SYMBOL(module_remove_driver);