oxide2

A Linux-class kernel + minimal userspace, written in Rust.

Targets x86_64-unknown-oxide-kernel and aarch64-unknown-oxide-kernel. Userspace built for upstream *-unknown-linux-musl (per docs/29a). Linux-binary-compatible at the modern syscall ABI: uname.sysname reports Linux, release reports 5.15.0-oxide.

Status

Phases 0–6 (read-driver layer) closed per 00§3 master-plan ladder. Both arches boot Limine → kernel_main. SMP brings up multi-CPU at -smp 4 (cross-CPU IPI + load balancer verified). A real-musl static-PIE shell runs as PID 1 — you can type at it and it executes builtins against the kernel's filesystem.

Run an interactive shell

cargo run -p xtask -- qemu --arch x86_64 --features debug-all

Boot scrolls through ACPI / PMM / VMM / sched / smoke output, then drops to:

oxide-sh: builtins exit / echo / help / ls / cat / pwd
oxide$

Type past the canned smoke (ls /proc, cat /proc/version, exit) — once the smoke completes the prompt is yours. Builtins:

ls [path]	openat(O_DIRECTORY) + getdents64 — works on /, /proc, /dev, /etc, etc.
cat <path>	read + write to stdout — works on /proc/version, /proc/cpuinfo, /etc/passwd, …
echo <args>	write args back
pwd	prints / (no real cwd yet)
help, exit	what they look like

To leave QEMU: Ctrl-A x.

What works

• Boot: ACPI parse → PMM (buddy) → demand-paged VMM → LAPIC/GIC → real timer IRQs → preemptive scheduler → user-as activation → ELF static-PIE loader (DT_RELA self-relocs, real auxv).
• SMP: Limine MP request → AP startup → per-CPU runqueue + IDTR + LAPIC + sti+hlt idle → cross-CPU resched IPI (vec 0x41) → load balancer migrates CFS tasks. Verified at -smp 4: cpus=4 aps_started=3 resched_ipis_received=3 migrated_total=2.
• Userspace: Real-musl static-PIE binaries run as PID 1. Hand-rolled orchestrator init exec's yo/hi/echo/cat then the real-musl oxide-sh takes over.
• Syscalls: ~280 slots wired (read/write/openat/close/getdents64/exit/fork/execve/wait4/clock_*/etc).
• Signals: rt_sigaction, sa_handler dispatch + signal frame + rt_sigreturn, SIGSTOP/SIGCONT scheduler hooks, kill(-pgid) POSIX semantics.
• Procfs: 80+ paths (status / cmdline / stat / maps / cpuinfo / version / uptime / meminfo / loadavg, full per-pid surface, /proc/sys sysctl tree).
• Devfs / etc: /dev/{null,zero,console,tty*,ptmx,pts/*}, /etc/{passwd,group,issue,os-release,…}.
• PTY: /dev/ptmx + /dev/pts/, full termios round-trip, cooked-mode line discipline (ICANON / ECHO / ISIG / Erase / Kill / EOF).
• Tmpfs: /tmp/<name> create-on-write.
• ext4 RO (crate-level): superblock + GDT + inode + extent + dir + Mount, 45 hosted tests against a real mke2fs-built image.

Not yet

• busybox / login / getty: the in-kernel shell is the only userspace binary today. Real busybox ships once the userspace build pipeline lands.
• Real disk I/O: ext4 RO crate works against a memory-backed image; no boot-disk integration yet (Limine module / virtio-blk). Phase 6 final mile.
• ext4 RW + JBD2: Phase 7b, not started.
• Networking: Phase 8, ENOSYS.
• Hardening / observability / modules: Phase 9, ongoing.

Status numbers

44 of 46 spec docs FROZEN; 45 ext4 + 60+ sched + 100+ kernel hosted tests pass; CI runs lint + tests + both arches default + --features debug-all on every PR.

For the live snapshot see state.md. For per-session history see CHANGELOG.md.

Quick start

make ci                                                       # full PR gate
cargo run -p xtask -- qemu --arch x86_64 --features debug-all # interactive shell
cargo run -p xtask -- qemu --arch x86_64 --smp 4 --features debug-all  # multi-CPU
make help                                                     # all make targets

Where to start

• docs/00-master-plan.md — top-level plan, phases, exit criteria.
• docs/MANIFEST.md — index of every spec.
• docs/02-spec-discipline.md — how specs evolve.
• docs/03-modernity.md — what's in v1 / what's deferred.
• state.md — current snapshot (read first when picking up work).
• CHANGELOG.md — per-session history of what landed on main.
• CLAUDE.md — project rules for Claude Code sessions.

License

MIT