History / Reference Error codes seen

The boot floppy, the BitBLT engine, and a keystroke-by-keystroke install guide The install guide was actively misleading: it still told you to patch the distribution CD and make a pre-`go` checkpoint, both of which were retired on 15 September. Somebody following it today would take a dead route and find out an hour in. That is now the boot-floppy route, with a screenshot of every screen and the exact key to press at each one. New: **The boot floppy** -- what is on the disk, how to build one, the two lines you type, the machine it needs, and a section on running it in the web UI. Rewritten: * **Running text-mode Setup** -- fourteen screens, thirteen screenshots, every keystroke. It calls out the step that is skipped most: pressing `S` at the mass-storage screen to add the ADB driver, without which Setup dies several screens later with "Setup did not find a keyboard" and the run is lost. * **Booting the installed system** -- the `bootdisk.of` route, GUI-mode Setup page by page, and the three pages that are not Return: Registration's boxes do not auto-advance, the repair disk defaults to Yes when there is no floppy NT can write, and the networking page leads to an adapter search that walks between two pages for ever. * **Making an OEM CD** -- kept, banner'd as retired. Updated: the video-stack page gains the BitBLT engine and the legacy-window mirror, including the evidence (0 of 307,200 pixels differ from a CPU-drawn reference) and the byte-lane measurement that turned out to be the instrument rather than the driver. The ledger retires rows 8 and 10 -- group 2 is now empty, and the prediction that a real OEM driver disk would retire them is left on the record because it was right. Status and roadmap notes the complete install built from nothing and NT's restart working. Error codes gains four messages that each mean something other than what they say. Images are the project's own captures, in images/.

pappadf committed Sep 17, 2026

Update for the desktop milestone, the cold boot, and the floppy investigation Eight pages. The ones that were actively wrong first: Machines-Desktop-7500-8500-9500 gains the measurement that answers most of its own open question 1. Comparing the ANS NT ROM's Open Firmware packages against a plain ANS ROM: pe-loader is the ONLY difference, and little-endian? is a settable variable on both. So Microsoft's veneer cannot be started on a machine without an NT ROM -- every patch in the catalogue is structurally ANS-only -- but the CPU-mode switch is not NT-specific, and xcoff-loader is present everywhere, so a loader of your own is the route. Caveat recorded: 1.1.22 is an ANS ROM, not a desktop one. Project-Status-and-roadmap said Setup's graphical half 'copies files'. It now reaches a logged-in desktop, and a patched CD cold-boots Setup from power-on with no pokes and no breakpoints. Project-Ledger-of-workarounds: rows 1-5 are bytes in the image now, not memory pokes, and two new rows join -- the i8042prt Start and kbdclass Group edits that GUI Setup's keyboard depends on. They are flagged as suspicious as a pair: two wrong values, both keyboard, both from one Setup pass, while every mouse equivalent is correct. Reference-Veneer-patch-catalogue was missing three sites (the two claim nops and 0x53DB0) and still claimed the file is never modified. Reference-Error-codes-seen gains VrOpen returned 8 = EIO, with the ARC ordering corroborated by wall 22's ENODEV=13. Debugging-When-your-instrumentation-lies gains three sections, all from mistakes made this week: a grep that counts only its target cannot distinguish 'absent' from 'never ran' (twice, a dead daemon and a checkpoint invalidated by rebuilding); a negative result is only as strong as the condition it was taken under (the veneer floppy claim, published without its condition); and map-space exists only inside the pe-loader package, so 'bad address to DMA-MAP-IN' is usually a missing mapping rather than a hardware limit. Emulator-Checkpoints-and-deltas: rebuilding the emulator invalidates every checkpoint, restamp-ckpt.py fixes it conditionally, and the pre-go checkpoint is now worth only ~2 minutes against mkcoldboot.py.

pappadf committed Sep 15, 2026

Write the wiki: 47 pages covering theory, guides and reference Seeds the wiki with the full structure rather than an outline. Five groups of pages, each written to be read by a person rather than consulted by a tool: Theory (11) Open Firmware, ARC, the veneer, the NT boot chain, the HAL contract, the NT PowerPC ABI, little-endian PowerPC, the video stack, how Setup chooses a HAL, and why any of this is hard Emulator (10) getting Granny Smith through to capturing and booting an installed image Debugging (5) the emulator shell, bugchecks, reading NT binaries, recipes, and what to do when the instrumentation itself is lying to you Hardware (4) what is known and unknown about real machines, what would have to change, serial consoles, and the risks Machines (4) the Network Server, the TNT desktops, PReP reference systems Reference (6) ARC variables, the address and interrupt map, error codes seen, HAL exports, the veneer patch catalogue, the VrDebug bitmask Project (5) status and roadmap, the ledger of workarounds, provenance, contributing, the story Plus Home, FAQ, Glossary, _Sidebar and _Footer. Everything here is drawn from this repository's own STORY.md, docs/ write-ups, tools and PR descriptions. No page reproduces Microsoft, Apple or third-party documentation; where a fact came from someone else's work it is attributed on the provenance page. The status claims are deliberately conservative and the workarounds are stated where a reader would otherwise be misled - in particular the VideoPortVerifyAccessRanges bypass that graphics-mode Setup currently depends on.

pappadf committed Sep 14, 2026