rounding up CPUs on debug exception

[Benjamin Herrenschmidt]

> >From what I understand, the smp_call_function() will block 
> while the other cpu has interrupts disabled.  Ideally we 
> want to halt that cpu before it does anything else.  Also, if 
> the other cpu is waiting on a spin lock the current cpu 
> holds, there will be a deadlock.
> 
> BTW, the current implementation uses smp_call_function() and 
> the system hangs occasionally on the call to smp_call_function().

There is no NMI generically available. On some machines, you can trigger
a system reset (0x100) exception but that isn't always recoverable.

Ben.