multiple interface test case

[Duncan Sands]

Hi Roman,

> Weird that it hasn't bitten many other people yet.

only a very few drivers do this kind of grab-a-bunch-of-other-devices
stuff.

> BTW do you have SMP / preemption enabled?  I may be missing something,
> but it looks like that list_empty(&dev->driver_list) test is not
> protected by a barrier from list_del_init(&dev->driver_list) in
> device_release_driver().  But my understanding in these matters is less
> than poor...

I turned off all SMP / preempt stuff, and it didn't make any difference.
All the locking has been changed in -mm by the way.  Anyway, what
happens is that after the disconnect and everything, the driver has a
reference count of 1, not 0.  So I just have to work out where the refs
are taken and where they are released...

Ciao,

D.