Alan wrote:

>Unfortunately ten years ago someone created 'register_and_activate_irq'
>calling it 'register_irq', and it hasn't yet been fixed.
>  
>
Then it's time to fix that.

I would propose something like the attached patch - it handles all archs 
that support disable_irq on an unregistered interrupt. The remaining 
arch [which one, btw] must implement request_irq_disabled().

--
    Manfred