[balp]

Yes, thata one of the RTOS thing, but almost always, when trying to build this down into what these timing limits are, they are kinda arbitrary, gut feeling put into a number. There are exceptions, yes, usally these exceptions as well are only for a limity subpart of the system. So for this using RTOS for everything is as stupid as not using RTOS, (or possibly even puting these parts into hardware, ASIC of FPGA) for these small subsystems. These timing used, also is mostly up to scheduler and Linux is possible to run with a scheduler that gives it similar capabilities ans most RTOS systems. It's a blury grey area at best.

[planteen]

Yes, this is so true. When someone has a true jitter-sensitive task that needs to run on say microsecond or sub-microsecond accuracy, that is not for the realm of a high-performance CPU with caches even if it is on a RTOS. My first question is if that tight of a bound is truly necessary or just an over-specified requirement. If it is necessary, I say do that in hardware or on a simple microcontroller (e.g., Cortex-R) if that is truly your requirement.