I've realized this (and am certainly not alone in that, it's rather obvious, when one think about the nature of timing) -- but does anyone have some links on implementing this? Are there some (sequence of) x86_64 instructions that can be used to bound a procedure (in the pascal sense of the word) to a quantum, regardless of things like when the procedure is called (assuming the procedure is short enough, and depending on branching behaviour, I suppose instruction fetch/decode, data fetch/write and accompanying cache hit/miss can make it hard to a) select a worst-run time in terms of clocks cycles to target (and if so, for which concrete cpus) and b) be hard to make sure the cpu is actually busy for exactly that many cycles...)? Is this even possible to approach in this portable C99?

I suppose if one ignore the information leak due to possible change in cpu load, one might device a kind of evented "call back" model, where one wait to return the result of a procedure until an interrupt is triggered?

I don't expect a full answer, but if anyone has a link to some source code that isn't too complicated, I'd be very happy (either "real-world" or some good "example" code).