Can't see where you save with microkernel for embedded systems. You get the same as pre-configured monokernel plus the IPC overhead.

QNX, vxWorks, INTEGRITY,...

While QNX is mentioned, does anyone know of any materials detailing the implementation of MX tables in QNX? I've stumbled upon a brief synopsis[1] of what they do, but it lacked implementation details (such as: do the entries need to be aligned on page boundaries? how many copies are made at the end? etc. etc.).

[1]: <https://cseweb.ucsd.edu/~voelker/cse221/papers/qnx-paper92.p...>

That doesn't address my point. Do you really appreciably save on binary footprint simply by using QNX versus trimmed down Linux kernel build?

You mean the binary footprint of a type 1 hypervisor that Linux requires when deployed in the same high integrity computing scenarios that QNX is usually used for?

The use case suggested by the GP was IoT/Home Assistant scenarios. So no, I absolutely do not mean that.

Now that you have safe AND performant languages like Rust or Oberon, what is the reason to not have exokernels for these applications?

I guess a mix of industry not wanting it until security is a legal liability, type 1 hypervisors, microservices and containers being used to tame monolithic kernels into pseudo-microkernels, hiring practices,....