The recent change made to the irq handling in hcd_endpoint_disable()
caused a problem.  The statement

	local_irq_save (flags);

needs to be outside the rescan loop.  Otherwise, on loop iterations after
the first, flags is always set to indicate that interrupts are disabled.

In fact, since the routine ends with might_sleep() anyway, I don't see any
reason to save the interrupt state at all.  My patch just disables
interrupts at the start and enables them at the end.  I'm not sure that's
how you intended it to work, so you may want to change it a little.