K5 is a very small microkernel based on the L4 family of kernels, but designed with microcontrollers in mind and written in Rust.
Why in the world does the world need another RTOS / microkernel. There are like half a dozen Rust RTOSes alone (Hubris, Tock, RTIC, MnemOS). There are a couple of good reasons, and a few "bad".
Let's start with the good. The largest reason is that this is a capability-based microkernel. K5 most closely mimics seL4 in its design. seL4 uses a concept called capabilities as access control for IPC. The K5 scheduler is also based on seL4's MCS, though with some differences. Like seL4, but unlike many existing Rust RTOSes, K5 is meant to be formally verified. This is not done yet but is a primary goal of the kernel. The last major feature that differentiates K5 is that it is designed with support for enclaves, as first-class citizens. I am not aware of any existing RTOS that supports scheduling TrustZone-M and RISC-V PMP enclaves natively.
Now let's discuss the "bad" reasons. I've always been fascinated by kernel development, and have long wanted to try it out for myself. Of course, I could have just contributed to an existing kernel, but that wouldn't be as fun.
Right now there is some basic functionality mainly for ARMv8M. There is an example project for the STM32L5, though it should work on any v8m CPU with a little modification. You can send and receive messages between threads, the scheduler will pre-empt your tasks at set intervals, and there is a built-in logging framework using defmt.
./kernel
- Contains the kernel as a library that will be used by a host application./userspace
- The userspace library, contains functions for syscalls, startup, and logging./abi
- The ABI (application binary interface) includes shared data structures between the kernel and userspace./codegen
- Simple code-generation utility to take a list of tasks, and produce Rust./examples
- Example "apps" for various boards, right now just stm32l5../cli
- Contains thek5
build tool, it supports flashing, building, and printing logs from a k5 app
Here's a quick demo of the logging system, and two tasks sending between each other. You can find this code in ./examples
k5.demo.mov
I've got plans! So so many plans... Basically, I think the order of tasks will be as follows
- Finish the initial set of syscalls (recv, call, send, logs, caps)
- Write MPU region solver and add MPU support for Cortex-M
- Make the UX better for the kernel's APIs. Basically, make starting the kernel and threads easier.
- Support peripherals and interrupts
- Recover from panics and hard faults in tasks
- Document both the code and the design choices of the project
- Attempt to use Kani or Creusot to verify the base scheduler
- Port to RISC-V?
- Add support for building and using TZ / PMP enclave tasks
- Investigate porting to MMU-based systems
The order is subject to change based on what I feel is the shiniest object, but that's the basic roadmap. If any of those things sound interesting to you and you want to help out, please reach out.
Ok, so to be honest the name was a happy-ish accident. I needed a random name, and I knew I was building an L4 kernel. So a number and letter sounded nice together. But I now have two nice stories to back up the name. The first is that much like k8s or i18n, the 5 is a stand-in for the rest of the word "kernel". Also, mountains in the Karakoram region of a few countries (I'm not here to create geopolitical conflict), are surveyed starting with K. K5 is Gasherbrum 1, which serves as the inspiration for the logo.
This project owes a lot to seL4 and Hubris OS, both fantastic kernels and resources. The context-switching, syscall code, and build system are heavily inspired by Hubris. The scheduler and IPC system are heavily based on seL4. The logging system is just straight-up defmt, but the RTT block is accessed through the kernel. Huge thanks to the Ferrous systems and the rest of the Rust embedded community, who have helped make embedded development far friendly to work with.