Operating Systems 1 — kernel project

Course project for Operating Systems 1 (Operativni sistemi 1) at the University of Belgrade. The assignment is to implement a small bare-metal kernel for RISC-V (rv64ima, lp64) that runs under QEMU on the virt machine, without a general-purpose OS underneath.

Scope

The kernel provides a minimal runtime for user-level code: trap handling (supervisor trap entry in assembly), thread scheduling with timer-based preemption (TCB, dispatcher, context switch in contextSwitch.S), and a system call layer exposed as:

• C API (syscall_c.hpp): heap (mem_alloc / mem_free and related), threads (thread_create, thread_dispatch, thread_exit, …), semaphores (sem_open, sem_wait, sem_signal, …), timed sleep (time_sleep), and console I/O (getc, putc).
• C++ API (syscall_cpp.hpp): RAII-style wrappers around the same facilities.

Supporting pieces include a memory allocator, kernel console (KConsole), scheduler and queues, RISC-V control/status helpers, and tests under test/ (threads, producer–consumer, sleep, system mode, C/C++ API variants). The build produces a single linked kernel image (Makefile, kernel.ld) for QEMU; optional GDB debugging is wired via the qemu-gdb target and .gdbinit.tmpl-riscv.

Build and run

Requires a RISC-V bare-metal toolchain (riscv64-unknown-elf-* or riscv64-linux-gnu-*, auto-detected by the Makefile) and qemu-system-riscv64.

• Build: make all — produces kernel and kernel.asm.
• Run in QEMU: make qemu

Development notes

• Before the trap / interrupt path is fully installed, avoid relying on system calls in code that runs too early during bring-up.
• Thread / Semaphore (C++ API) and dynamic allocation: do not construct these or call new in global scope; global constructors run before the kernel is ready and can conflict with teardown ordering.
• Early putc-style output may use paths that do not depend on the full console pipeline; once the console subsystem is initialized, use the documented APIs consistently.