> > fault: divide by zero, debugging traps, NMI, into, bound and illegal
> > instruction, all others starting at interrupt 7 which is the quite critical
> > (FPU context switches) should not experience any performance degradation.
>
> Yep
Okay.. I feel better..
>
> > OTOH it makes virtually impossible to use NMI for any interprocessor
> > interrupts on SMP machines (at least if I understand correctly APIC's doc,
> > what a mess).
>
> Linux doesnt use NMI between CPU's - we use IRQ 13 and 16 in 2.0.x
>
Great, so wheres the patch?