ROM and firmware
The machine
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.VIOis written as pairs —STO A,H.VIO+1thenSTO A,H.VIO(“send data”), andSTO A,H.VIO+3thenSTO A,H.VIO+2(lines 382–410).- Reading
H.VIOclears 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/DISDIMtoggle the dim bit throughOUT 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:
research/shipped-software/OS1SYSD.IMD→ raw (imdinfo.py --raw),AUTOST.COMremoved,ACT80.COMadded → drive A. It is a 250 kbps MFM 40 × 5 × 1024 disk, andcpmtools’osborne1diskdef 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/AL1marking the two directory blocks, offset 3 tracks).- The same disk emptied of files, with
rom144.asmnormalised to CRLF and written asROM144.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. o1raw2imd.py --ddwraps both raw images (MAME will not open a bare sector image),o1keys.shboots A with B in drive B and types the command, andacthex.pyturns the hex back into an image and compares it against the shipped ROM.
Four things the emulated build teaches about the source
rom144.asmas 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\tPAGEis 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–0x0065,0x00E1–0x00FFand0x0FFFare gaps — and the ROM holds00there, notFF. Filling the gaps withFFinstead 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, notACT.COM. The vendor’s memo says to typeACT, 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
PIPcopy 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
FFat0FC0produce, as the neighbouring all-FFrecords confirm (41at0FB0,21at0FD0,11at0FE0). 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 at0EFF0H; - the GPIB control codes and PIA handshake constants for the IEEE port.
Open questions / next steps
- 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. - Read the IEEE driver section (lines 3113–3400) against
docs/drive-c-protocol.mdto confirm the exact handshake the Drive C firmware expects (#6/#22). - 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_1ROM/Source_Code_v144/(all 18 file sizes match). Seeresearch/README.md. - The 1.44 binary used for the CRC comparison:
research/roms/os1-144.rom(Don Maslin’s archive) — identical to MAME’s3a10082-00rev-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.txtretains Osborne’s 1985 FOG letter verbatim; the terms are recorded inresearch/README.md.