Draft. Draft — assembled from the working notes by `tools/build_book.py`. The source document is the authority; this page is a copy for reading, and it is regenerated from that source.

Status: Solved. The source is indexed and every port and vector table it holds agrees with docs/o1-memory-io-map.md §3, which had been derived from the schematic and from the running machine; and rom144.asm reassembles with the vendor’s own assembler to the shipped ROM, byte for byte (The 1.44 reassembly, below). What is not done: the 1.41 CBIOS source has never been diffed against the CBIOS on a shipped system disk, and the video register semantics are still read from the schematic rather than from the source (Open questions / next steps). Issues: #72, #40, #203.

Osborne’s own source for the machine’s firmware — the thing the rest of the project had been inferring — and the primary source for the interfaces the rest of docs/ builds against:

Confirmed Section
The five memory-mapped I/O bases — floppy 2100, keyboard 2200, IEEE-488 PIA 2900, 6850 ACIA 2A00, video 2C00 — agreeing with docs/o1-memory-io-map.md §3 entry for entry The memory-mapped I/O map, as the source states it
The monitor RAM map (0EA80–0EFFF) and the interrupt vector table at 0EFF0H Monitor RAM and vectors
The WD179x floppy command set the BIOS issues (#20) Floppy command set
The IEEE-488 register programming, the GPIB command bytes, the inverted bus, and the “IEEE always appears to be ready” stub (#6/#22, #19) IEEE-488 — and the parallel printer is the same port
The video write pairs (+1→+0, +3→+2) and ENADIM/DISDIM through OUT 2/OUT 3 Video, as the ROM drives it
A negative result: the firmware never selects its own bank — there is no OUT 0 or OUT 1 anywhere in the ROM Bank switching: what the ROM actually does

It is a safety copy of material already public in several places; see Provenance at the end.

The two components

The machine’s firmware is two separate programs, and the source tree has one of each:

Component Lives in Source here Rev
Monitor ROM — boot, video, keyboard, disk, IEEE-488, interrupt handlers the 2732 at UD11 rom144.asm 1.44
CBIOS — CP/M 2.2 character and disk BIOS loaded into RAM from the system disk occbio*.asm 1.41

They version independently, so do not assume the 1.41 CBIOS source is what ships inside the 1.44 ROM. The monitor ROM is the 4 KB in the socket; the CBIOS is a RAM-resident overlay assembled to a specific system size (below).

File index

bios141.gen is the build recipe, and it lists exactly the files present here, so the set is complete:

File Lines What it is
rom144.asm 4655 The monitor ROM, Rev 1.44, ORG 0. Z80. Roger W. Chapman, 2/4/1983. Title: “DOUBLE DENSITY ROM”. Contains the memory-mapped I/O equates, RAM storage locations, IEEE/disk equates, monitor main loop, video, keyboard, disk drivers, the IEEE-488 interface and drivers, the parallel (Centronics) printer path, and the interrupt handlers.
occtxt6.ast 374 The shared equates file (S=OCCTXT6.AST in the build): port bases, ROM vectors, monitor RAM bases, character codes, disk-type constants. occbio05.asm says “FOR USE WITH OCCTXT6.AST ONLY”.
occbio05.asm … occbio95.asm 47, 145, 80, 70, 341, 141, 246, 60, 505, 66 The CBIOS listing, split into ten sections. 05 is the title block and layout; 45 holds the disk control blocks (DPBASE, DPHGEN); 85 the I/O-byte dispatch and character devices.
occram15.asm, occram25.asm 31, 175 “Monitor RAM Storage” — the RAM-resident data area, ORG MRAM, assembled into both the ROM and the CBIOS.
bios141.gen 56 The generation procedure: needs ACT 3.5e, DDT 2.2, SYSGEN 2.1 and MOVCPM, and the twelve files above.
tran.txt 61 The 1.44 ROM release memo (Sai Kit, 2/4/1983): how the HEX and print files were generated, on a DD machine with a Corvus hard disk.
release.txt 53 The 1985 Osborne letter granting FOG permission to reprint the schematics and BIOS listing. Retained verbatim — see research/README.md.
source.md 2 Provenance (BrettHallen/Osborne_1 ROM/Source_Code_v144/).

Revision lineage

Monitor ROM 1.44 — the header carries the delta from 1.43, which is the useful part (it is a change log for the DD machine):

CBOOT  - added seek 10 tracks and home drive : tsk
GKEY   - fixed bell bug                      : tsk
SENDEN - changed the number of retrys to NRETRY
PSEKC  - added disk head settle delay
SELDRV - changed start up delay to 500 ms    : tsk

NRETRY = 10 (disk retry count), SLD_RCT = 3 (slide-key repeat), and BRTBIT/DIMBIT (80h/00h) are defined at the top of the file. The 1.44 image is research/roms/os1-144.rom (CRC32 c0596b14) — see docs/o1-rom-variants.md.

CBIOS 1.41 — from occbio05.asm/occbio15.asm:

MSIZE = 59              ; 59K system
CCP   = 0CB00H          ; location of the CCP
BIOS  = ccp+1600h
BDOS  = ccp+806h
DDT R (relocation value) = 1F80H-BIOS

So the CBIOS is relocated for a 59K system, and occbio15.asm carries the CP/M disk-type tables — HSTSIZ = 1024 (blocking/deblocking buffer), FPYSIB = 2048/128 (SD block = 2K), FPYDIB = 1024/128 (DD block = 1K), DSKS1 = 5 (single density, 256-byte sectors, single sided).

The memory-mapped I/O map, as the source states it

Straight out of rom144.asm (;MEMORY MAPPED I/O, around line 83) and occtxt6.ast:

Address Name in source What it is
2100H D.CMDR / D.STSR Floppy: command (write) / status (read)
2101H D.TRKR Floppy: track
2102H D.SECR Floppy: sector
2103H D.DATR Floppy: data (read/write)
2200H H.KEY Keyboard
2900H CPDRA / PA.DTA / PA.DIR IEEE-488 PIA: port A data / direction
2901H CCRA / PA.CTL IEEE-488 PIA: control A
2902H CPDRB / PB.DTA / PB.DIR IEEE-488 PIA: port B data / direction
2903H CCRB / PB.CTL IEEE-488 PIA: control B
2A00H H.SCTRL (W) / H.SSTS (R) 6850 ACIA: control / status
2A01H H.SXMT (W) / H.SREC (R) 6850 ACIA: transmit / receive
2C00H H.VIO Video memory controls (the ROM also uses +1, +2, +3)

occtxt6.ast names the same bases — H.FDC = 2100h, H.KEY = 2200h, H.IEEE = 2900h, H.SIO = 2A00h, H.VIO = 2C00h — which is worth noting because the CBIOS refers to the IEEE port as H.IEEE, not as a PIA.

Cross-check against docs/o1-memory-io-map.md §3

Map doc §3 Source Verdict
2100–2103 floppy, 2100 stat/cmd, 2101 track, 2102 sector, 2103 data D.CMDR/STSR/TRKR/SECR/DATR agrees
2201–2280 keyboard, one address bit per row H.KEY = 2200H, read with A0–A7 one-hot agrees
2900–2903 IEEE-488 PIA, MC6821 (UC7) CPDRA/CCRA/CPDRB/CCRB — a PIA with A/B data+direction+control, i.e. 6821-style agrees
2A00–2A01 serial ACIA, MC6850 (UC4) H.SCTRL/H.SSTS at 2A00, H.SXMT/H.SREC at 2A01 agrees
2C00–2C03 video PIA, MC6821 (UC15) H.VIO = 2C00H, ROM touches +0…+3 agrees

No contradictions. The I/O map in docs/o1-memory-io-map.md §3 is now sourced from the firmware as well as the schematic and the running machine.

Free space within the window. The firmware touches exactly five bases (2100, 2200, 2900, 2A00, 2C00), so the clear gaps between the device blocks are 2300–28FF, 2B00–2BFF and 2D00–2FFF. That is where a virtual peripheral belongs — but confirm the decode granularity against the schematic before relying on an address inside one of the used blocks, because several decode more than the registers the firmware names (the keyboard decodes A0–A7 one-hot, for instance). This narrows the recommendation in the map doc §7 and closes most of gap 1 in docs/research-gaps.md.

Monitor RAM and vectors

The ROM’s scratch area sits immediately below video RAM, and the emulator, the video personality and any bus-level assertion need to know about it:

Address Name Note
0EA80H MRAM Base of the monitor RAM area (ORG MRAM in occram*.asm)
0EF00H TEM Used in the boot routines
0EF05H RTRY Retry counter
0EF08H ROMRAM ROM/RAM flag
0EF09H DSTSB Six bytes of disk info
0EF0FH DMADR Disk DMA address
0EF13H SEKDEL Disk step delay
0EF14H SAVSEC / SAVTRK / SDISK Saved sector / track / disk
0EF50H HSTACT, UNASEC, LOGSEC Host active, unallocated/logical sector
0EF59H KEYLCK, CURS, LKEY, ESCH Keyboard lock, cursor, last key, escape
0EF61H PIAAD, PIABD Saved PIA A/B data — the IEEE port’s shadow
0EF6AH DACTVE, BELCNT, LLIMIT Disk active, bell counter, line limit
0EFF0H INTBL Interrupt vector table
0F000H FWAVM First address of video memory

Also from occtxt6.ast: ROMVEC = 100H (start of the ROM vector table), NMIA = 66H (NMI vector), BKPI = 0CFH (breakpoint interrupt = 1), SEKTMO = 0FEH (seek time-out status).

Bank switching: what the ROM actually does

The ROM defines exactly two bank-related macros:

ENADIM: MACRO          DISDIM: MACRO
        OUT 2                  OUT 3
        ENDM                   ENDM

OUT 2 / OUT 3 set and clear the BIT 9 latch (the “dim bit”, bank 3) — matching docs/o1-memory-io-map.md §4. There is no OUT 0 or OUT 1 anywhere in the ROM: the firmware never selects the RAM bank itself, so bank 1 ↔ bank 2 must be driven by hardware (reset and the NMI flip-flop chain), exactly as §4 describes. This is a useful negative result for the interposer: the firmware gives us no example of a software bank switch to model, and the “never drive 0x00–0x03” rule in §4 stands.

Floppy command set

The WD179x commands the BIOS issues (rom144.asm, ;DISK EQUATES):

Constant Value Command
D.SEK 010H Seek
D.STP 020H Step
D.STPI 040H Step in
D.STPO 060H Step out
D.RDS 080H Read sector
D.WRTS 0A0H Write sector
D.RDA 0C0H Read address
D.FINT 0D0H Force interrupt
D.RDT 0E0H Read track
D.WRTT 0F0H Write track

These are the type I/II/III/IV encodings of a WD179x-class part, which is what the floppy interposer (#20) has to speak. The MB8877-vs-WD1793 part question in docs/o1-memory-io-map.md §3 remains schematic-level and is unaffected.

IEEE-488 — and the parallel printer is the same port

The ROM carries a complete IEEE-488 implementation (;IEEE-488 INTERFACE., line 2606, drivers from 3113), and it is the register-level contract for Drive C (#6/#22) and for the virtual printer (#19).

Three things worth extracting immediately:

The bus is inverted. Line 2668: “ALL SIGNALS ARE LOW ON THE IEEE BUS WHEN PIA REGISTER CONTAINS ‘1’.” An emulated device that assumes positive logic will be wrong.

GPIB command bytes (;IEEE control codes):

Constant Value Meaning
IE_TALK 40H make talker
IE_UTLK 5FH make untalk
IE_LSTN 20H make listener
IE_ULST 3FH make unlisten

The PIA handshake programming (;IEEE EQUATES), including PA.CDT = 00101110b — “to address port a data and set port a in input program handshake mode” — plus the direction values (PA.DRO = 0FFH, PA.DRI = 00H, PB.DR = 0BFH, PB.DTO = 00000010b, PB.DTI = 00001011b) and the ready/strobe bits (PP.ORDY = 01000000b in port B, PP.IRDY = 10000000b in the PIA control register, STRB = 00100000b in port B). The current PIA state is shadowed in RAM at PIAAD/PIABD (0EF61H/0EF62H).

And one behaviour that matters more than the register values: line 3133 says “IEEE always appears to be ready” — the BIOS’s device-status routine is effectively a stub. The O1 will not poll a GPIB device for readiness, so an emulated Drive C must be ready when addressed.

Finally, line 3249: “The Parallel port is actually the IEEE port driven with the centronix [protocol]”. The MX-80 personality (#19) is therefore an IEEE-488 device speaking Centronics on the same PIA, not a separate port.

Video, as the ROM drives it

  • H.VIO is written as pairs — STO A,H.VIO+1 then STO A,H.VIO (“send data”), and STO A,H.VIO+3 then STO A,H.VIO+2 (lines 382–410).
  • Reading H.VIO clears the video interrupt (line 2025, LD A,H.VIO ;clear interrupt) — the video PIA is one of the three IRQ sources at 60 Hz (docs/o1-memory-io-map.md §5).
  • ENADIM / DISDIM toggle the dim bit through OUT 2 / OUT 3 (above).

Individual register semantics stay in docs/o1-memory-io-map.md §3; this pass records what the ROM does with them, not a re-derivation.

The 1.44 reassembly: the source is the shipped ROM (#203)

Measured 2026-10-13. rom144.asm assembles with the vendor’s own assembler — SORCIM’s ACT80 — to an image identical, byte for byte, to research/roms/os1-144.rom (CRC32 c0596b14). That settles #83’s deliverable 2: the source in this repository is the source of the shipped 1.44 monitor ROM, and the toolchain that produced it runs under MAME.

   
Assembler ACT80.COM — SORCIM “Assembly Code Translator”, version 3.0, safety copy in research/act-assembler/
Command ACT80 B:ROM144 H=B: — the vendor’s own, from tran.txt
Reported no ERRORs, 473 Labels, Program LWA = 0FFFh
Output 11 520 bytes of Intel HEX, 255 records, ROM144.HEX on drive B
Image 4096 bytes, CRC32 c0596b14 — the shipped ROM
Reproduce tools/research/rom144-reassemble.sh — the happy path, end to end

This is the happy path, and it is self-contained in the repository. The assembler is the vendor’s own, the command is the vendor’s own, and both live here: research/act-assembler/ACT80.COM and tran.txt. No substitution, no porting, no search for a missing version — the toolchain that built the part is the toolchain that rebuilds it.

ACT 3.0 and 3.2 — and no 3.5e

The archive holds nine copies of ACT80.COM, and they are three distinct binaries over two versions:

Version Copies under …/rlee/ md5
3.0 S/SORCIM/ACT Cross Assemblers/, M/MORROW/OUTSIDE SOFTWARE/ACT/, L/LOOSECPM/{079,109,262}/ 4f76a911a66ddc8e1e02a1676c232d45
3.2 (serial 279) L/LOOSECPM/{073,087,119}/ cf73590466736fd1960e829b5f2221aa
3.2 (no serial line) D/DIGITAL RESEARCH/ECA/ 88256731a9399dee68a29054865d072c

ACT 3.5e, the version bios141.gen names, is not in the archive. The search was bounded rather than open-ended: every ACT80.COM under the archive tree found by name, hashed and string-searched for its banner, and the tree searched for an ACT .DOC or manual — the only hit was an unrelated accounts-receivable file. READ.ME in the kit lists the 3.0 files exactly as the archive holds them.

All three builds are now safety copies in this repository — research/act-assembler/, with the 3.0 kit’s own READ.ME, DOCOP80.ASM (every ACT mnemonic, which is the file to diff against if a future source is rejected) and its LINK include files; the provenance rows are in research/README.md, and the licence notice in the kit’s READ.ME is recorded there rather than quietly dropped. The 3.2 builds sit beside it, so the byte-identical claim above can be re-checked against a local file rather than the mirror:

O1_ACT80=research/act-assembler/3.2/ACT80.COM tools/research/rom144-reassemble.sh

tools/research/rom144-reassemble.sh defaults to the 3.0 copy, so the build no longer needs the local archive to be present. The third build — the one that circulated inside a third-party bundle, 3.2/ACT80-eca-bundle.COM — is not a drop-in: it returns to the prompt printing nothing, where both SORCIM builds assemble cleanly. It is held as a curiosity, not as a spare.

The version gap turns out not to matter for this source. 3.2 assembles rom144.asm to a byte-identical hex file — same no ERRORs, same 473 labels, same LWA = 0FFFh, same 11 520 bytes. The only difference visible at all is one line of the summary, where 3.2 says 8230h bytes not used and 3.0 says 8232h; that is ACT’s own bookkeeping, not the object code. So 3.0 is not merely “close enough” — for 1.44 it is exact, and the question of searching further for 3.5e can be closed as not-needed.

So the reassembly needs neither z80asm — the wrong tool, because this is an ACT job and not an 8080-syntax one — nor ACT 3.5e: the vendor’s own assembler was in the local archive all along, and 3.0 and 3.2 both give the object code exactly (REQUIREMENTS.md).

ROM 1.3 is the same exercise one revision earlier, and the assembler is the one thing that did not carry over. The source disk reads — it is an 8-inch 77-track FM/MFM disk with CP/M skew 3, which tools/research/o1imd8.py handles — the source is extracted to research/source-code/ROM-13-source/, and the gap fill, the terminator and the CRLF handling all carried over from this build. But the 1.3 source uses the DBE/DCE pseudo-ops, which only ACT 3.5E has, and the archive holds 3.0/3.2. The bytes those pseudo-ops emit are recovered from the shipped rev-B ROM on OCCROM.IMD, and a 1.3 image can be built with a mechanical source rewrite, but it is not certified — see docs/rom-13-reconstruction.md and #205. The project still has no use for a 1.3 ROM — we build from the 1.44 source — so this remains upstream’s gain and nothing here depends on it.

The caveats in Four things the emulated build teaches about the source, and the damaged-record handling below, are the ones to carry into any other Osborne source assembled this way.

How the build is driven

The vendor’s procedure (tran.txt) assumes two floppies: ACT80 in drive A, the source in drive B, ACT B:ROM144 H=B:. Reproduced under MAME with two DD images, and tools/research/rom144-reassemble.sh is the whole of it:

  1. research/shipped-software/OS1SYSD.IMD → raw (imdinfo.py --raw), AUTOST.COM removed, ACT80.COM added → drive A. It is a 250 kbps MFM 40 × 5 × 1024 disk, and cpmtools’ osborne1 diskdef is the right description of it: the CBIOS’s own DPB agrees field for field (occbio95.asm, DPBGEN DSKD1,8*5,3,7,0,D1DSM,64,1100000000000000B,3 — 1 K blocks, AL0/AL1 marking the two directory blocks, offset 3 tracks).
  2. The same disk emptied of files, with rom144.asm normalised to CRLF and written as ROM144.ASM → drive B. The source is 83 KB and the hex file is 12 KB; a full system disk has room for neither, which is the one thing that has to be arranged deliberately.
  3. o1raw2imd.py --dd wraps both raw images (MAME will not open a bare sector image), o1keys.sh boots A with B in drive B and types the command, and acthex.py turns the hex back into an image and compares it against the shipped ROM.

Four things the emulated build teaches about the source

  • rom144.asm as captured is LF-only, with three stray CRs (offsets 56 819, 59 483, 60 178 — lines 3198, 3334, 3383) where a line break lost its LF: \tRET\r\tPAGE is two lines, and so are the other two. CP/M text wants CRLF, so the disk copy is normalised CRLF → LF → CRLF, treating each stray CR as the break it was — 4658 lines rather than 4655. The repository file is left exactly as captured.
  • The hex records do not cover the whole address space. ACT emits a record only where the source put a byte, so 0x005C–0x0065, 0x00E1–0x00FF and 0x0FFF are gaps — and the ROM holds 00 there, not FF. Filling the gaps with FF instead leaves 42 bytes wrong (acthex.py --fill). Worth knowing before the same exercise is run against ROM 1.3.
  • ACT’s terminator is not Intel’s. The file ends with :00010010 — a four-byte record carrying no data — rather than :00000001FF. Both 3.0 and 3.2 emit it identically, so it is ACT’s.
  • The program is ACT80.COM, not ACT.COM. The vendor’s memo says to type ACT, which on this kit is CP/M’s <name>? — command not found — and reads exactly like an ACT usage error. It is not one.

The one record that arrives damaged, and why it does not matter

On the floppy image, one record — :100FC000…, ROM addresses 0FC0–0FCF — reaches the host with six non-hex bytes in place of six characters (c0 de c0 de c0 0d, at file offset 11 264: the first bytes of the file’s last 1024-byte sector). The line is still 43 characters long, so nothing looks wrong at a glance. Three facts pin it down:

  • It is not the assembler’s doing. ACT 3.0 and ACT 3.2 are different binaries and produce byte-identical output, including those six bytes. Two assemblers do not invent the same garbage.
  • It is not a general fault in the floppy path. A PIP copy of the 83 KB source through the same disk — guest write, MAME image save — came back byte-identical, so writing a file whose last sector is partial is fine.
  • The intended content is fixed by ACT’s own bytes. The record’s checksum is intact, and it is exactly the value 16 bytes of FF at 0FC0 produce, as the neighbouring all-FF records confirm (41 at 0FB0, 21 at 0FD0, 11 at 0FE0). The record therefore says what it should have held, and the range is restored from that (acthex.py --recover-ff).

With the range restored, the assembled image is identical to research/roms/os1-144.rom in all 4096 bytes — and before the restore it already agreed in the other 4080. The cause of the six bytes is not isolated; it is deterministic, it lands in the file’s last sector, and it is recorded here as an emulation observation rather than explained away.

What this closes, and what it does not

Gap (docs/research-gaps.md) Status after this pass
1 — exact memory & I/O map, free decode windows Substantially closed. Every address the firmware touches is now sourced, and the gaps between the five device blocks (2300–28FF, 2B00–2BFF, 2D00–2FFF) are clear. Decode granularity is still schematic-level.
3 — char-gen socket pinout Not addressed — that is a hardware pinout, not firmware. Still open, still needed by #24.
10 — 6850 ACIA location for the WiFi modem Closed — 2A00/2A01 confirmed by an independent source.
11 — floppy controller variant, port map, DD changes Port map and command set closed; the DD changes are now documented (1.44 is the DD ROM, DD block = 1K, disk-type tables in the CBIOS). MB8877-vs-WD1793 remains open.

Two things this pass adds that no existing doc had:

  • the monitor RAM map (0EA80–0EFFF) and the interrupt vector table at 0EFF0H;
  • the GPIB control codes and PIA handshake constants for the IEEE port.

Open questions / next steps

  1. Does the 1.44 ROM’s CBIOS match the 1.41 source? The CBIOS is disk-loaded, so extract it from a shipped system disk (OS1SYSD.IMD / OS1SYSS.IMD) and diff it against the 1.41 build.
  2. Read the IEEE driver section (lines 3113–3400) against docs/drive-c-protocol.md to confirm the exact handshake the Drive C firmware expects (#6/#22).
  3. Pin down the video register semantics from the ROM’s write pairs (+1→+0, +3→+2) against the schematic (#24).

Provenance

  • The source tree: github.com/BrettHallen/Osborne_1 ROM/Source_Code_v144/ (all 18 file sizes match). See research/README.md.
  • The 1.44 binary used for the CRC comparison: research/roms/os1-144.rom (Don Maslin’s archive) — identical to MAME’s 3a10082-00rev-e.ud11.
  • The OCC engineering source disks (research/source-code/*.IMD, bitsavers) are the disk copies of the same material and have not yet been extracted — that is #2’s pipeline work.
  • Licensing: release.txt retains Osborne’s 1985 FOG letter verbatim; the terms are recorded in research/README.md.