OpenBIC

Summary This is an experimental proof-of-concept, not a request to merge a finished board. It demonstrates that OpenBIC's core stack - IPMI dispatch, IPMB transport, MCTP, PLDM, the sensor framework, FRU - is portable to mainline Zephyr ( v4.4.0), running on an SoC family (NXP MCX-N9xx) with no Aspeed/Nuvoton relationship at all, instead of the vendor-forked Zephyr trees every existing OpenBIC board depends on. I'm opening this mainly to get a read from maintainers on whether this direction is of any interest to the project, not because I think it's ready to merge as-is. Full writeup, including exactly what's real vs. stubbed and why, is in on this branch. What's genuinely verified on real hardware Real // code (not stand-ins) for watchdog, timer, I2C, sensor framework, FRU, MCTP, PLDM. A real IPMI/IPMB transport: and needed almost no changes - the only Aspeed-specific piece was the bottom-edge I2C-slave integration, replaced with mainline's own API. A shell self-test round-trips a real Get Device ID request through the actual dispatch pipeline (queues, threads, netfn routing, handler) and every response byte matches expected values exactly. Boots clean, IPMB TX/RX threads start on real target-mode I2C hardware, dual-core mailbox IPC between the SoC's two Cortex-M33 cores, NVS persistent storage, GPIO monitoring, watchdog-triggered reset - all demonstrated on physical hardware, not just compiled. What's explicitly NOT done / still open No real sensors wired up - the EVK has none, so the sensor/SDR tables are genuinely empty. Wire-level IPMB transaction: untested, not confirmed working. A jumper-wire test between this board's own two I2C instances showed an unresolved signal-integrity anomaly (documented in the README) - properly resolving it needs an oscilloscope or a second real device, neither available here. IPMI transport for a real host (KCS/SSIF) isn't attempted - no mainline equivalent exists, and there's no host on this EVK to talk to. Meta's proprietary OEM-1S command set (fan/GPIO/JTAG/PECI/APML) is stubbed, not ported, for the same "no applicable hardware" reason. I3C, and a real GPIO driver for this SoC, are out of scope - see the README for the specific mainline driver issues hit. This currently depends on an unmerged upstream fix** (zephyrproject-rtos/hal_nxp#799) for a real divide-by-zero in NXP's vendored HAL, pulled in via a manifest override pointing at my fork in the meantime. Test plan [x] builds clean from a fresh (no manual patching). [x] Flashed and verified boot banner, shell, and all subsystem init log lines on a real MCX-N9XX-EVK over its MCU-Link VCOM console. [x] shell command verified correct, byte-exact IPMI Get Device ID response through the real dispatch pipeline. [ ] Wire-level IPMB transaction between two real devices - not possible with the hardware available for this port; see README. Happy to answer questions or adjust scope/direction based on maintainer feedback.