Re: [RFC] embedded struct device Re: [patch] IDE driver model update

[Alan Cox]

> > That's because normally pci_unregister_driver() or whatever is called in
> > cleanup_module(), but obviously to be able to call it the refcount has to 
> > be zero already...
> 
> That's true for drivers, but not for devices. Nonetheless, it's a big 
> problem that I've hoped would magically go away. Of course it won't, but I 
> don't have a solution for it off hand...

Is it actually cleanly solvable without sticking the refcounts in the
layer they belong not in driverfs ?