From: Andi Kleen <ak@suse.de>

John Stultz noticed that kirqd didn't start because of another logic error,
which broke irq balancing.  This was still a fallout from the generic
subarchitecture changes.

Actually it still refuses to balance anything on my test box, but perhaps
I'm just not able to generate enough interrupts.

Anyways, with this patch the thread is running again at least.