When you put together documentation for your function reference and that reference turns out to be incorrect on two counts for one function, and the combination of those two errors makes me lose two weeks worth of work tracking down a bizarre bit of behavior, I'm bound to get a little testy.
In case you're wondering, I managed to vanquish the whack-a-mole bug. It turns out that the function that I use to reset the interrupt vector uses a different value than the one that is documented; however, when the incorrect value is used, it does not fail liek it is supposed to, but reports a success.
The end result of this was that my interrupt handler was deleted, but the reference to my interrupt handler in the kernel was not, and whenever I would enable interrupts on the system, it would crash, but only the second time. After that, the previously loaded value just happened to be correct in memory.