I've made the mistake of presenting the counter value directly instead of
converting it to nanoseconds. We need to do that, by grabbing the period from
the device and using it.

Since hpet is already our slow fallback anyway, I am using simple
multiplication and not bothering with complex mult / shift stuff.