RTOS Kernel

Project Requirements

Develop a real-time kernel on ARM based chips from scratch capable of admission control, task scheduling, isolation, and synchronization.
In simpler terms, this means ensuring tasks are schedulable, actually scheduling tasks on the fly, memory protection (per thread), mutexes and avoiding deadlocks.

Mar 2020 – Apr 2020

Developed a real-time kernel capable of admission control, task scheduling, isolation, and synchronization.
In simpler terms, the kernel ensured that the tasks are schedulable, actually scheduling tasks on the fly, memory protection (per thread), mutexes and avoiding deadlocks.
Kernel features:
- Context switching and task management
- Fixed priority rate-monotonic scheduling using a Priority queue
- Isolation (using MPU) and real-time synchronization, using Original Priority Ceiling Protocol to avoid unbounded priority inversion and deadlocks

Timers, Interrupts and System Calls

Feb 2020 – Mar 2020

Built the first part of our RTOS kernel with:

User space isolation: a kernel capable of running arbitrary programs in userspace with an exception driven jump from Handler mode to Thread mode.
Interrupts: a kernel with the ability to handle requests from user programs via system calls.
Servo motor control: servo motor control program, using a few custom system calls.

STM32: Device Drivers and MMIO

Jan 2020 – Feb 2020

Used Memory Mapped IO (MMIO) to interface with peripheral embedded devices.
Implemented the drivers for UART and I2C serial communication as well as the onboard ADC.
Interfaced with our custom PCB to measure sound and light levels and display them to the seven-segment LED display.

ARM Bootloader Bootstrapping and Assembly optimization

Jan 2020

Wrote a simple bootloader for the STM32. Assembly code to load and initialise global data (.data and .bss sections; to be copied onto SRAM) for an ARM Cortex-M4 architecture with code residing on Flash memory. JTAG and OpenOCD used for debugging and transferring binaries.
Optimized an ARM assembly challenge with 50x speedup (compared to the original code)