Well we are discussing models in general.

If you do not want preemption disable it. For example in posix threads use sched_fifo and pin to one core.

Loom might provide similar features if needed.