Skip to content

HANDOFF: ND-100 <-> ND-5000 Octobus Emulation — State & Next Steps

Date: 2026-07-16. Written for the next LLM/engineer taking over this work. Read first: OCTOBUS-ND100-ND5000-REFERENCE.md (same folder) — the consolidated, evidence-marked protocol reference. Everything below assumes you have read it.

1. Mission

Emulate the complete ND-100 <-> ND-5000 (SAMSON) communication path in RetroCore so SINTRAN III VSX/500 can boot and drive a ND-5000 CPU. The octobus frame/card/ACCP layer is DONE and verified; the shared-memory mailbox layer is NOT.

2. Current state (all verified working 2026-07-16)

Code — E:\Dev\Repos\Ronny\RetroCore (branch ethernet-ii-controller-fixes, UNCOMMITTED)

File Status
Emulated.HW\ND\CPU\NDBUS\NDBusOctobus.cs ND-100 3109 card. IOX 100400-100437, idents 40B/41B..46B/47B level 13 (LIVE-verified by TPE B00 + CONFIGURATION D05 diagnostics), frame routing, station-1 talk-back adapter, loopback for TPE tests.
Emulated.HW\ND\CPU\NDBUS\OctobusFabric.cs The bus: 64-slot registry, dest->source rewrite, broadcast, timeout-as-null.
Emulated.HW\ND\CPU\NDBUS\OctobusND5000Station.cs The ACCP/Access Module: AIB/AOB flags, emergencies 241B/242B/244B, kicks 1-6, multibyte OMD routing, ACCP commands (micro 1-3, multibyte CMCPURES/CMSYSPAR), AttachCpu(IND500Cpu) via TagWritten event.
Emulated.Tests.ND100\ControllerOctobus\ 54 tests, 52 pass + 2 pre-existing skips. Includes replays of carved CH5CPUPRESENT / XRS5CPU / CON5IDENT sequences. Run: dotnet test Emulated.Tests.ND100\Emulated.Tests.ND100.csproj --filter "FullyQualifiedName~Octobus"

Documentation (this folder + corrected legacy docs)

  • OCTOBUS-ND100-ND5000-REFERENCE.md — master reference, [V]/[I]/[C] marked.
  • ND5800-MICROCODE-ACCP-OCTOBUS-CATALOG.md — every ACCP-touching microcode routine.
  • Corrected in place: ..\Devices\Octobus\OCTOBUS-PROTOCOL-REFERENCE.md, ..\ND500\ND500-BUS-OCTOBUS-HW-INTERFACE.md, the 324B carve README.

3. What is PROVEN (do not re-litigate)

  • 16-bit frame: C=15, B=14, station=13-8, E=7, K=6, M=5, S=4 — NPL + manual + carved machine code (SOCTO dispatch, SKICK builder in 026-S3IMPIT.asm).
  • Kick numbers 1-6 (N100KICK=1, CLRKICK=3, IDLEKICK=6) — NPL symbols + manual p.336 + microcode OCB_DEC_K.
  • CM* codes (CMMAC=41B, CMACO=42B, CMCPU=71B multibyte-only); emergencies = EBIT|code (241B/242B/244B).
  • Card registers: +0 FIFO read, +1 frame write, +2 status (bit 3), +3 control (4=tx-en, 1=int-en, 20B=clear); output base 100404.
  • Idents 40B/41B level 13 — settled by LIVE diagnostics (see §5 lesson 3).
  • Microcode handshake: ACCP_READ/WRITE spin on AFLAG bits 9/10; kicks/idents arrive via AOB with ATRAP+OMESS; bit 15 terminates received multi-word messages.

4. THE REMAINING WORK (agreed plan, phases 1-4)

Phase 1 — Shared-memory mailbox layer

Implement the X500DF per-CPU extension. Spec: ..\ND500\ND500-BUS-INTERFACE-REFERENCE.md §6.5: X5SEMA (offset 0, test-and-set semaphore, TSET=140123/TSETP=140516 semantics, owner word X5RES=47, -1 = ND-100), X5HEN=3 / X5FYL=4 / X5MXF=5 (FIFO head/fill/max), X5FIF=6 (element array). Backed by the same MPM RAM the ND-500 reaches via bit-31 addresses (IND500Cpu.SetMpmMemory). Model on the existing NDSharedMemory / NDBusND500IF patterns — load the nd-500-bus-interface skill for the 3022-generation equivalents.

UPDATE 2026-07-17 (from the microcode session's lossless B30 verification — E:\Dev\Ronny\ND5000UC\microcode\MAILBOX-MICROCODE-PSEUDOCODE.md §3.10 and ..\ND500\MICROCODE-ANSWER-INIT-SAMSON-AND-13B-2026-07-17.md): - [V] The 5800 microcode's message base is START_MESS = 0o20000 (+ per-CPU offset, patch constants SAMSON_CPU/OFFSET); during INIT_SAMSON, SYS_DATAF (025630) READS words at 0o20000+0x0A/+0x0C expecting the ND-100 to have written them already — so the ND-100 must populate the message area in MPM BEFORE the microcode finishes init. - [V] SYSPAR is read FROM the ACCP during INIT_SAMSON (SYS_READ 017111: three ACCP_READ halfwords → SRF 0o2006-0o2010) — the OctobusND5000Station micro-command 1 / CMSYSPAR reply path is boot-critical, not optional. - HARD REQUIREMENT (from the 13B zero-readback analysis): the mailbox backing store and the MPM-window/SetMpmMemory array MUST be the same bytes at the same offsets — a separate array or base mismatch reproduces the "reads back zeros" failure. - [V] After servicing any message the microcode writes N5STA := 3 + calls GIVEINT — and GIVEINT is NOT just a frame send: it inserts the answered message's address into the X5FIF ring under X5SEM before strobing the interrupt (full 5-step sequence in Phase 2 below). It never touches interface status registers.

Phase 2 — Wire mailbox to kicks

RE-CORRECTED 2026-07-17 per the architect session (GIVEINT decoded). Where this brief disagrees with ..\ND500\ND500-MICROCODE-INTEGRATION-ARCHITECTURE-2026-07-16.md §7.2 or E:\Dev\Ronny\ND5000UC\microcode\MAILBOX-MICROCODE-PSEUDOCODE.md §3.1b, THOSE WIN.

Activation (ND-100 -> ND-5000): SINTRAN links the message into the ex-queue at X5BEX (per-CPU extension block word 0), clears X5ACT (0 = work); kick frame is the PREEMPT path only (ACT52) — the servicer's real wakeup is a write-hook/poll on X5ACT, NOT KickReceived alone. Servicer re-arms X5ACT := 1 (IDLE_2 semantics) BEFORE consuming the queue; queue walk starts at X5BEX (-1 = empty), under X5SEM.

Answer (ND-5000 -> ND-100) — the GIVEINT sequence, ALL FIVE steps required (pseudo-C §3.1b [V]; the earlier "N5STA + frame only" description was wrong — without the ring insert SINTRAN's XN500 drain never sees the answer): 1. take X5SEM (test-and-set on global header word 0); 2. write N5STA := 3 (or 4 for trap-shaped stops) into the message; 3. insert the answered message's address into the X5FIF ring at slot[X5FYL] (32-bit slots), advance X5FYL := (X5FYL+1) mod X5MXF, full-check vs X5HEN; 4. release X5SEM; 5. send the interrupt frame to station 1 — the word is DYNAMIC, not 100401B: ((5OMDNO<<8 & 037400) >> 3) | 100001, where 5OMDNO arrives as word 1 of the CMSYSPAR multibyte the station already receives (runtime-allocated by CONOMD; 10B is merely the observed live value — NEVER hardcode 100401B). Capture it in the station's CMSYSPAR handler.

SHARED SERVICER — do NOT reimplement (drift risk): "servicer runs" means the single Nd500MicrocodeServicer extracted from NDBusND500IF.ProcessMailboxMessage by the bus-interface session (phase S0/S1 of the architecture doc), consumed via IServicerHost. Phase 2 GATES on that extraction landing. To keep moving before it lands, code Phase 2 against the IServicerHost interface with a stub — never a second MICFU engine, or every future fix lands twice.

X5PRO: SINTRAN reads it cache-bypassed to choose flag-vs-kick (ACT51/ACT52). Write it in Phase 1 (current process, -1 on idle — PRNOWR semantics); if deferred, leave a named TODO plus a skipped test, because leaving it -1 forever means the preempt/kick path is never exercised.

Drain on the ND-100 side: like the XN500 driver (ND500-BUS-INTERFACE-REFERENCE.md §7.5).

Phase 3 — MON 60B validation

Replay the bring-up order from ..\ND500\nd-500-mon\ND500-BRINGUP-BUS-INTERFACE-FEEDBACK.md: detect -> MEMDEF 040B -> CSLOAD 037B -> MICPSTART 025B -> swapper -> domain start; and DECOMESS -> MCHANDLE monitor-call returns. MON 60 subfunction table: ..\ND500\ND500-BUS-OCTOBUS-HW-INTERFACE.md §2. NOTE: MON 60 never touches octobus registers directly — it rides the mailbox; that is why Phase 1 must come first.

Phase 4 (independent) — ND-5000-originated messages

Generate TRAP_OCBM-format reports from the station: CPU-available 202B (makes SINTRAN's 5OMBREAD set 5ALIVE), HW-fault 200B with the register-dump record (format: ND5800-MICROCODE-ACCP-OCTOBUS-CATALOG.md §7 + MP-P2-N500.NPL:3372-3551). Plus Messack/Messnak replies for ACCP library commands (codes: ND-05.020.01 ch. 5.3.11).

5. LESSONS / RULES (each cost real time — follow them)

  1. Octal truncation trap: the debunked "C=bit9/B=bit8/dest=bits7-2" format came from octal values missing two trailing zeros. When a bit layout looks shifted by 6, check for dropped octal digits first.
  2. Wrong-image symbol trap: 03xxxx SINTRAN symbols resolve against the PIT-mapped overlay (026-S3IMPIT), NOT commoncode. Validate any carve by multi-symbol coherence + actual IOX usage, never by "it disassembles".
  3. Live diagnostics beat static inference: the ITB13+37B/+40B byte reading produced a wrong ident conclusion (37B/40B); TPE/CONFIGURATION diagnostics running against the emulator settled it as 40B/41B. When a real ND test program disagrees with your static analysis, the test program wins.
  4. Mark evidence: every claim is [V] verified / [I] interpretation / [C] contradiction, with file:line. Fabricated-looking constants (power-of-two enums, "formula" idents) have burned this codebase twice — never invent values to make code compile.
  5. User's coding rules (absolute): no LINQ, no foreach; keep/extend comments (never strip citations); tests must be run and 100% green before claiming success; dotnet format for whitespace; never mention Claude/AI in commits; never kill processes; full paths when naming files.

6. Open questions (evidence does not exist in the repos yet)

  • Broadcast type (BT) codes — "OCTObus Protocol Specification" document never found; MBSEND sets B when descriptor word X+2 != 0 (mechanism carved, meaning unknown).
  • ACCP EPROM dump (68000 firmware) — would unlock full ACCP library emulation.
  • ITB13 indexing scheme (why slot 37B holds the ident-40B datafield — plausibly table indexed from ident-1, UNVERIFIED).
  • Exact microcode-sent kick word encodings (100102|cpu patterns in SENKICK) vs receiving-side kick numbers.

7. Key file map

What Where
Master protocol reference E:\Dev\Ronny\NDInsight\SINTRAN\ND5000\OCTOBUS-ND100-ND5000-REFERENCE.md
Microcode catalog E:\Dev\Ronny\NDInsight\SINTRAN\ND5000\ND5800-MICROCODE-ACCP-OCTOBUS-CATALOG.md
Mailbox/FIFO spec E:\Dev\Ronny\NDInsight\SINTRAN\ND500\ND500-BUS-INTERFACE-REFERENCE.md §6.5, §7.5
MON 60 table + driver E:\Dev\Ronny\NDInsight\SINTRAN\ND500\ND500-BUS-OCTOBUS-HW-INTERFACE.md
Carved driver bodies E:\Dev\Ronny\NDInsight\tools\sintran-segment-carver\versions\L-VSX-500\re\segments-ref\026-S3IMPIT\026-S3IMPIT.asm
NPL sources E:\Dev\Ronny\NDInsight\SINTRAN\NPL-SOURCE\NPL\ (PH-P2-OPPSTART, MP-P2-N500, 5P-P2-MON60)
Hardware manual E:\Dev\Ronny\ND5000UC\manual\ND-05.020.01 EN ND-5000 Hardware Description.md (ch. 5 + App. 2)
Microcode disassemblies E:\Dev\Ronny\ND5000UC\microcode\MICRO-5800-B30.md / -A30.md (lossless since ND5000UC commit a91dff4)
Mailbox servicer pseudo-C (microcode session) E:\Dev\Ronny\ND5000UC\microcode\MAILBOX-MICROCODE-PSEUDOCODE.md — N5STA protocol, MICFU dispatch, 13B/14B field layout, §3.10 verified offsets
INIT_SAMSON / 13B answer memo E:\Dev\Ronny\NDInsight\SINTRAN\ND500\MICROCODE-ANSWER-INIT-SAMSON-AND-13B-2026-07-17.md
Emulator + tests E:\Dev\Repos\Ronny\RetroCore\Emulated.HW\ND\CPU\NDBUS\ + Emulated.Tests.ND100\ControllerOctobus\
Related skill nd-500-bus-interface (3022/5015 generation patterns, MON 60B test shapes)