Re: rmmod uhci_hcd -> BUG: atomic counter underflow

[Alan Stern]

On Mon, 12 Mar 2007, Jiri Slaby wrote:

> Jiri Slaby napsal(a):
> > Alan Stern napsal(a):
> >> On Mon, 12 Mar 2007, Jiri Slaby wrote:
> >>> After rmmoding of uhci_hcd on fresh booted 2.6.21-rc3-mm2 I got this:
> >>>
> >>> BUG: atomic counter underflow at:
> > [...]
> >>>   [<c01db754>] kobject_put+0x14/0x16
> >>>   [<c01db8a3>] kobject_unregister+0x22/0x25
> >>>   [<c024c987>] bus_remove_driver+0x75/0x82
> >>>   [<c024d3b8>] driver_unregister+0xb/0x18
> >>>   [<c01e7020>] pci_unregister_driver+0x13/0x73
> >>>   [<f88dbbd9>] uhci_hcd_cleanup+0xd/0x2d [uhci_hcd]
> > [...]
> >> Would it be possible for you to add the atomic counter underflow check 
> >> to 2.6.21-rc3 and see if the problem still occurs?  If it doesn't, 
> >> that's a good indication the USB stack isn't guilty -- the bus 
> >> registration code hasn't changed for several kernel releases.
> > 
> > Yes.
> 
> I can confirm, that this issue went upstream and is currently present there.

I found the problem, but I don't know how to fix it.  Hopefully Greg or 
Rusty will know.

There are two bugs in kernel/module.c:module_remove_driver(), right at the 
end of the routine:

	/*
	 * Undo the additional reference we added in module_add_driver()
	 * via kset_find_obj()
	 */
	if (drv->mod_name)
		kobject_put(&drv->kobj);

It's supposed to undo this code in module_add_driver():

	if (mod)
		mk = &mod->mkobj;
	else if (drv->mod_name) {
		struct kobject *mkobj;

		/* Lookup built-in module entry in /sys/modules */
		mkobj = kset_find_obj(&module_subsys.kset, drv->mod_name);
		if (mkobj)
			mk = container_of(mkobj, struct module_kobject, kobj);
	}

The first bug is in the condition of the "if (drv->mod_name)" statement.  
If "mod" isn't NULL then kset_find_obj() doesn't get called, but the "if" 
can succeed anyway.  (Maybe this isn't a real bug -- it would depend on 
all the callers; I don't know.)

The second bug is the argument to kobject_put().  drv->kobj is _not_ the 
kobject whose refcount gets incremented by the kset_find_obj() call.  
That's why the BUG occurred; the kobject's refcount was getting 
decremented without being incremented first.

Alan Stern

-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/