Skip to content

roadmap.md

Latest commit

 

History

History
190 lines (174 loc) · 9.02 KB

roadmap.md

File metadata and controls

190 lines (174 loc) · 9.02 KB
Raw

HXNU Roadmap

Release target:

  • 2608 as the active first version marker for August 2026
  • As of 2026-06-11, the first public bootstrap release is expected to wait for closure of Phase 3.5 and a stable x86_64 SMP baseline from Phase 4

Phase 0

  • Separate Rust kernel repository
  • x86_64 target definition
  • Minimal ELF kernel entry
  • Early serial logging

Phase 1

  • Limine handoff wrappers
  • Physical memory map parsing
  • Early logging and panic reporting
  • Frame allocator bootstrap
  • Kernel heap bootstrap

Phase 2

  • GDT/IDT
  • Exception handlers
  • APIC or timer bring-up
  • Interrupt dispatch
  • Basic scheduler skeleton
  • Structured kernel diagnostics and panic reports

Current status:

  • GDT/IDT activation is online on x86_64
  • CPUID inventory is online on x86_64
  • CPUID topology leaf inventory from 0x0B/0x1F is online on x86_64
  • UEFI framebuffer or GOP handoff is online on x86_64
  • Output-only TTY console bootstrap is online on x86_64
  • Local APIC timer one-shot bring-up is online on x86_64
  • Local APIC periodic tick and scheduler bootstrap are online on x86_64
  • Minimal ACPI discovery with RSDP, XSDT, MADT, and FADT parsing is online on x86_64
  • MADT processor, IO APIC, and interrupt-override topology summaries are online on x86_64
  • FADT power and reset-register summaries are online on x86_64
  • SMP topology inventory and AP bring-up target discovery are online on x86_64
  • Read-only procfs snapshot bootstrap is online on x86_64
  • Read-only devfs namespace bootstrap is online on x86_64
  • Minimal VFS mount and read facade is online on x86_64
  • VFS normalized path resolution and node lookup facade are online on x86_64
  • cpio newc initrd discovery and /initrd read path are online on x86_64
  • Initrd-backed bootstrap block device, GPT partition discovery, and read-only /fat VFS mount smoke path are online on x86_64
  • /initrd/init executable candidate discovery and format probe are online on x86_64
  • /initrd/init ELF64 header and program-header inspection skeleton is online on x86_64
  • /initrd/init ELF PT_LOAD vm-map planning with RWX and BSS accounting is online on x86_64
  • Early Unix-like shebang interpreter fallback from /bin/* to /initrd/bin/* is online on x86_64
  • Partial Linux + Ghost + HXNU-native syscall compatibility dispatcher bootstrap is online on x86_64
  • x86_64 int 0x80 syscall gate, register-frame dispatch, and entry self-test are online
  • Bootstrap uaccess copyin/copyout validation facade is online on x86_64
  • Bootstrap openat/read/close (Linux) and open/read/close (Ghost, HXNU) VFS-backed syscall paths are online
  • exit_group syscall path is connected to scheduler thread-exit request handling
  • Scheduler-backed getpid/getppid/gettid identity path is online for bootstrap syscall personalities
  • Open-file table ownership is now process-scoped, and exit_group purges owned descriptors
  • exit_group now tears down the current thread-group and advances to the next runnable scheduler entry
  • Multiple virtual TTY screen foundation is online on x86_64
  • Scheduler thread table and runqueue skeleton are online on x86_64
  • Bootstrap to idle-thread context switching is online on x86_64
  • Initial user page-table creation and bootstrap ELF segment mapping are online on x86_64
  • /initrd/init first userspace handoff is online on x86_64
  • HXNU-native ring3 bootstrap payload can issue int 0x80 syscalls on x86_64
  • Styled framebuffer console output is online on x86_64
  • Breakpoint, page fault, and general protection fault self-tests are working
  • Power-reset self-test reaches the FADT reset-register path on x86_64
  • Trap-frame-aware ring3 scheduler handoff is online on x86_64
  • First user fatal-exception kill-and-resume policy is online on x86_64
  • Bootstrap init exit-triggered restart policy is online on x86_64

Cross-repo status (as of 2026-03-29):

  • External compiler repository hxnu-rustc-compiler-x86_64 is online and versioned separately
  • Rust-first SDK v0.1.0 is tagged and includes hxnu-rustc, hxnu-cargo, hxnu-sdk, and x86_64-unknown-hxnu target spec
  • SDK bundle flow (build, pack, install) and ELF verification flow are automated in the compiler repository
  • Kernel integration model is consumer-style (PATH + hxnu-cargo), with no monorepo coupling

Phase 3

  • Virtual memory manager
  • Kernel virtual address-space management
  • User virtual address-space management
  • Page-fault resolution path
  • Process and thread core
  • Syscall entry path
  • User-kernel memory copy and validation path
  • IPC fast path
  • ELF loader
  • VFS core
  • devfs
  • procfs
  • TTY core and console plumbing
  • Multiple virtual TTY screens or virtual consoles
  • Early keyboard or console input path
  • UEFI framebuffer or GOP handoff and framebuffer console
  • Block device layer
  • Partition discovery
  • cpio-compatible initrd support
  • FAT16/32 support
  • Minimal ACPI discovery on x86_64
  • MADT and FADT parsing
  • Reboot and poweroff plumbing
  • Userspace ABI planning

Phase 3.5

  • Trap-frame-aware user thread context save and restore
  • Timer-driven ring3 preemption and real sched_yield handoff
  • Scheduler bookkeeping for user threads beyond bootstrap, idle, and single-init bring-up
  • Page-fault classification plus first resolve-or-kill policy for user faults
  • Stable init lifetime, exit, and restart semantics
  • Early keyboard or console input path
  • Block device, partition, and FAT bootstrap needed to move beyond initrd-only boot

Current focus (as of 2026-06-11):

  • Harden the ring3 scheduler and first user-fault handling path before broad compatibility work
  • Treat Phase 4 entry as a narrow SMP/per-CPU milestone first
  • Keep POSIX, PTY, and driver-loading work behind stable process and SMP foundations

Phase 4

  • Stable SMP bring-up on x86_64
  • Stable BSP to AP startup flow
  • Per-CPU data areas
  • IPI support
  • TLB shootdown path
  • POSIX personality
  • Legacy Ghost compatibility layer
  • Core virtualization or LVE hooks
  • Full Linux or Unix-like init startup contract
  • PTY and POSIX terminal semantics
  • Active TTY switching and console session routing
  • Driver object model
  • Device enumeration and bus framework
  • Driver loading infrastructure for external driver directories
  • Driver discovery and load policy for filesystem-backed modules
  • Driver trust and load policy
  • ext4 driver
  • exFAT driver

Phase 5

  • aarch64 bring-up
  • PL011 early UART
  • DTB parsing
  • Exception vectors and GIC
  • aarch64 SMP topology bring-up
  • Heterogeneous CPU topology support
  • big.LITTLE or hybrid-core scheduling awareness
  • Minimal busybox-rust bootstrap trials after stable exec/spawn, writable VFS, and PTY foundations are in place
  • Basic Ethernet bring-up
  • Early network driver model
  • Loopback and packet path groundwork
  • Minimal userspace networking boundary

Phase 6

  • Rust cross compiler support with x86_64 and aarch64 as first-class targets (x86_64 bootstrap release is online in external compiler repo)
  • C and C++ cross compiler support with x86_64 and aarch64 as first-class targets
  • Additional architectures after the main two are stable
  • PPC 32-bit bring-up
  • Audio stack entry point
  • Additional driver families loaded from external driver directories
  • AHCI, NVMe, or virtio-blk expansion
  • Richer Ethernet and audio driver families
  • Debug monitor, symbol lookup, and crash dump direction

Architecture Direction

  • HXNU is a hybrid kernel
  • Native HXNU primitives come first
  • POSIX and legacy Ghost support are compatibility personalities, not the native kernel model
  • Boot-critical and virtualization-critical pieces stay in kernel
  • Replaceable services and policy should move to user space
  • FAT16/32 can live in kernel if that keeps early boot and recovery simpler
  • ext4 and exFAT are expected to work well as separate drivers or service modules
  • devfs and procfs should arrive early with the VFS core
  • TTY and framebuffer console support should be available before broader userspace compatibility work
  • Multiple virtual TTY screens should sit between the early console path and full PTY/session semantics
  • UEFI framebuffer support should be treated as a boot-critical display path
  • Minimal ACPI discovery and power-state plumbing belong in kernel
  • Full power-policy logic should stay outside the kernel when practical
  • SMP comes before broad userspace compatibility work
  • Heterogeneous CPU scheduling belongs after base SMP and timer stability
  • The syscall and user-kernel boundary should be treated as a first-class kernel milestone
  • Storage needs a block layer before filesystem work can scale
  • Driver loading from dedicated filesystem directories should be supported after the base VFS and init path are stable

Toolchain Priorities

  • Rust cross compilation: x86_64, then aarch64
  • C and C++ cross compilation: x86_64, then aarch64
  • Other architectures only after the main two toolchains are reliable
  • Compiler development continues in a dedicated repository: https://github.com/neonix-bmx/hxnu-rustc-compiler-x86_64
  • Kernel repository tracks integration contract and acceptance checks, not compiler internals