Skip to content

HANDOFF - HLE Ethernet II: DEF-NETWORK-CONN SOLVED, now prove send/receive

Date: 2026-08-11 Status: DEF-NETWORK-CONN returns "Ok" on the native-C# HLE card (committed). Send/receive proven at the CARD BOUNDARY. Full node-to-node conn-to NOT yet proven over real Ethernet. Live-test boot config PREPARED but NOT booted (a test was running on D100; user said hold).

Full path of this file: E:\Dev\Ronny\NDInsight\SINTRAN\XMSG\DOC\COSMOS-RE\HANDOFF-HLE-SEND-RECEIVE-TEST-2026-08-11.md

What is DONE and committed

DEF-NETWORK-CONN <system> ENNS0 now returns "Ok" on NDBusEthernetIIHle (the native C# Ethernet II card). Committed to RetroCore as e8df4f132. Validated on a real SINTRAN boot (run-64680): D102 / D200 / D19999 all return "Ok"; List-Routing-Info shows 102/200/19999 -> *->WAN?->*; device log shows [0x0400]x12 -> [0x054A]x6 -> [0x0500]x6. 8/8 descriptor unit tests pass.

The two defects that were fixed (file E:\Dev\Repos\Ronny\RetroCore\Emulated.HW\ND\CPU\NDBUS\EthernetII\NDBusEthernetIIHle.cs):

  1. Idle re-poll. An XROUT-routed letter arrives with NO doorbell, so the Recv listen gate that only fired on the doorbell never consumed it. Added a fallback poll (CA_IDLE_REPOLL_TICKS = 300_000, _caIdlePollLeft): consume immediately if _caReceiveArmed, else count down and re-poll.
  2. Route-define answer. The card must NOT build the XSDSY itself - it is unprivileged, so the kernel answers XRPRV (0x0A privilege error). Instead the card matches the route-define letter (rhdr == 0x0441) BEFORE the connect-letter test, sends the verbatim oracle [0x0400] directory reply (BuildRouteDefineDirectoryReply), and lets the PRIVILEGED command program build the XSDSY. Reply descriptors: XFSCM, XFWRI 40 bytes from ROUTE_REPLY_BUF = 0x1d600, XFSND|XFSEC to the requester magic.

What is PROVEN about send/receive

  • At the card boundary: Nd100EthernetIIHleHubJoinTests.HleCard_JoinsTcpHub_AndFramesCrossBothDirections passes - frames cross both directions when the HLE card joins a TCP hub. This proves the card's network attach and framing, not a full COSMOS conn-to.
  • Full two-node guest-ring conn-to is NOT proven. The harness Nd100TwoNodeEthernetIIHleHarnessTests moves 0 packets because both emulated nodes boot the SAME disk image, so both program MAC 00:00:00:00:00:00 - a same-image identity collision. The harness log says so explicitly: "both cards programmed the SAME MAC ... connect-to cannot distinguish the nodes. Give ImagePathB a distinct-sysno image." This is a harness limitation, not a defect in the fix.

The live-test plan (pivot away from the two-node harness)

Because the two-node harness collides on identity, the plan is a LIVE test: boot one local RetroCore ND machine with the HLE card on the operator's Ethernet hub (127.0.0.1:5010), NO HDLC line, and conn-to the REAL D102 that is already on that hub.

CONFIG PREPARED (uncommitted, in RetroCore working copy)

  • E:\Dev\Repos\Ronny\RetroCore\RetroCore\scripts\ND100\ND100-configs.ini new config ND100-ETHHLE-HUB: cpu=ND100CX, device add ETHHLE 0 --net=tcp:127.0.0.1:5010, FX/SMD/TERM 5/TERM 6, and NO device add HDLC (that is the "kill off hdlc1" test condition - proves ENNS0 + conn-to work over Ethernet ALONE).
  • E:\Dev\Repos\Ronny\RetroCore\RetroCore\INI\ND100Script.ini new boot label ND-BOOT-ETHHLE-HUB (attaches %BD0_L% = D:\BIGDISK0-L.IMG, boots BD0), and menu choice 9: "SINTRAN HLE Ethernet on hub (no HDLC)".

Verified (not assumed): the device add parser (DebugCommands.Devices.cs line 760) extracts --net= and passes it to the ETHHLE case (ND100Machine.cs line 444) -> FromSpec("tcp:127.0.0.1:5010") -> joins the hub.

*** CORRECTION 2026-08-11 - cpu_number/system_type do NOT set identity on ND-100 ***

An earlier draft of this config added --cpu_number=100 --system_type=100 with a comment claiming it relabels an already-installed pack to D100 (MAC ..64..) at boot. That is WRONG and has been removed. Those flags are used only when the machine is FIRST INSTALLED - they seed the identity that then lives on the SINTRAN disk. Booting an already-installed pack ignores them; the system number SINTRAN reports (and the number the card programs into its station MAC at bring-up) comes from the INSTALLED DISK IMAGE (BD0). The identity CAN be changed later (re-install / SINTRAN config), it is just not re-applied by re-passing the flags at boot. User stated this directly (2026-08-11: "cpu number and system type in ini file doesnt change what sintran sees, that's in the sintran disk after system has been installed"; 2026-08-17: "--cpu_number/--system_type only used when machine is installed first time, can change it later"). See memory [[nd100cx-identity-comes-from-disk-not-ini]].

Consequence for the test: the BD0 pack attached in ND-BOOT-ETHHLE-HUB must be installed with a system number DIFFERENT from D102's, or the local node and the real D102 collide on identity exactly like the two-node harness did. %BD0_L% (D:\BIGDISK0-L.IMG) is the current attach - its installed system number must be confirmed (or a distinct-number pack chosen) before booting. THIS IS THE OPEN ITEM.

NEXT ACTIONS (in order, once the user gives the go-ahead)

  1. Confirm which BD0 pack has a system number distinct from D102's (ask the user; the two-node collision proves this matters). Point ND-BOOT-ETHHLE-HUB at that pack.
  2. Launch RetroCore, Boot menu -> choice 9.
  3. Drive over the retroterm MCP (ONE session, ESC first, SYSTEM\r\r login):
    • start-net-server enns0
    • DEF-NETWORK-CONN D102 ENNS0 (expect "Ok")
    • conn-to d102
  4. Watch for real frames crossing to D102 over Ethernet (card TX/RX counters moving, and D102 answering the conn-to).

HARD CONSTRAINTS from the user this session

  • Do NOT kill D100 (9010) - a test was running on it.
  • Do NOT boot the emulator until the user says the running test is clear.
  • The three live processes (hub 5010, D100 9010, D102 9102) must be left untouched.
  • Leftover testhosts on this box are from E:\Dev\Repos\Ronny\RetroCore-spin\ (a DIFFERENT working copy) - NOT ours, leave them alone.