Nanoseconds if you can respond inside the interrupt handler ('top half' in Linux.) Microseconds to milliseconds with significant variability if you need to be scheduled ('bottom half'.)

And that only if your codepath is deterministic. Linux can do pretty funny stuff at times.

I've run a plasma cutter under Linux on a run-of-the-mill VIA SBC, it was actually pretty easy: just make a program setuid root, use that to disable interrupts after you've read all your data in and then just go wild on polled and precision timed IO until you're done. Re-enable interrupts and return as if nothing ever happened.