Oxide RFD 284: Loading the Host Operating System

7 hours ago

The Pico Host Boot Loader (phbl) is the component responsible for loading and starting the host OS image on Oxide machines.
phbl is loaded from SPI flash by the PSP after DRAM training and runs from the BSC reset vector, bringing the processor to 64-bit mode.
It loads a compressed CPIO archive (ZLIB) containing the host kernel, extracts the ELF image, loads it into RAM, and calls the kernel's entry point.
Virtual memory mappings use the largest possible page sizes (1GiB, 2MiB, 4KiB) and respect segment protection bits; UART is mapped uncached.
phbl minimizes system state changes, passing only the CPIO archive's physical address and length to the kernel, and maintains a defined system state on entry.
The current implementation compresses the boot archive into the phbl image, but future separation is possible without backward compatibility concerns.
Security relies on the service processor validating host images in flash; phbl performs no runtime verification.

PFL is an experimental project aiming to translate pf's firewall language and stateful semantics to Linux's XDP (eXpress Data Path) for efficient packet processing.
Built on FreeBSD's pfSense legacy, PFL uses Rust for the control plane (pfld) and eBPF dataplane, leveraging XDP's driver-level execution for high packet rates.
Key architectural features include a tail-call graph pipeline to comply with eBPF verifier constraints, LRU hash maps for connection state, and per-CPU counters.

The author collects disposable vapes and discovered they contain a PUYA PY32 microcontroller (ARM Cortex-M0+).
The chip is a PY32F002B with 24 MHz Cortex M0+, 24 KiB flash, and 3 KiB RAM.
They used semihosting (debugger syscalls) and SLIP to emulate a modem over serial, enabling IP networking.

ZeroFS introduces a native kernel client: an out-of-tree Linux filesystem module written in Rust, with a small amount of C for netfslib helpers.
The kernel client registers with VFS, uses netfslib for I/O, and speaks the 9P2000.L.Z protocol directly over TCP or Unix sockets, eliminating the userspace round trip of FUSE-based mounts.
Compared to Linux v9fs, the ZeroFS kernel client supports custom metadata operations, reconnection after server failover, and leader/standby pairs.

Author switched from openSUSE Tumbleweed to Void Linux for finer control and native ZFS support.
Guide targets experienced administrators and uses Void Linux (glibc) with ZFSBootMenu for a FreeBSD-like boot experience.
Steps include: booting from hrmpf rescue system, partitioning disk, setting up ZFS encryption, creating pool and datasets, and installing base system.

IBM i (originally AS/400) is a unique operating system with an integrated Db2 database at the kernel level, treating database and OS as one inseparable unit.
It uses an object-based architecture instead of file-centric, with Single Level Storage (SLS) treating all memory and disk as one flat address space.
The TIMI abstraction layer ensures decades-long backward compatibility, allowing programs from 1990 to run on modern hardware without recompilation.

PFL is an experimental project aiming to translate pf's firewall language and stateful semantics to Linux's XDP (eXpress Data Path) for efficient packet processing.
Built on FreeBSD's pfSense legacy, PFL uses Rust for the control plane (pfld) and eBPF dataplane, leveraging XDP's driver-level execution for high packet rates.
Key architectural features include a tail-call graph pipeline to comply with eBPF verifier constraints, LRU hash maps for connection state, and per-CPU counters.

The author collects disposable vapes and discovered they contain a PUYA PY32 microcontroller (ARM Cortex-M0+).
The chip is a PY32F002B with 24 MHz Cortex M0+, 24 KiB flash, and 3 KiB RAM.
They used semihosting (debugger syscalls) and SLIP to emulate a modem over serial, enabling IP networking.

ZeroFS introduces a native kernel client: an out-of-tree Linux filesystem module written in Rust, with a small amount of C for netfslib helpers.
The kernel client registers with VFS, uses netfslib for I/O, and speaks the 9P2000.L.Z protocol directly over TCP or Unix sockets, eliminating the userspace round trip of FUSE-based mounts.
Compared to Linux v9fs, the ZeroFS kernel client supports custom metadata operations, reconnection after server failover, and leader/standby pairs.

Author switched from openSUSE Tumbleweed to Void Linux for finer control and native ZFS support.
Guide targets experienced administrators and uses Void Linux (glibc) with ZFSBootMenu for a FreeBSD-like boot experience.
Steps include: booting from hrmpf rescue system, partitioning disk, setting up ZFS encryption, creating pool and datasets, and installing base system.

IBM i (originally AS/400) is a unique operating system with an integrated Db2 database at the kernel level, treating database and OS as one inseparable unit.
It uses an object-based architecture instead of file-centric, with Single Level Storage (SLS) treating all memory and disk as one flat address space.
The TIMI abstraction layer ensures decades-long backward compatibility, allowing programs from 1990 to run on modern hardware without recompilation.

Hasty Briefsbeta