Reading the monitor ROM off a real machine
The machine · reference
Status: Partially solved. The identification is
settled and needs no tool: the upper line of the sign-on is framed by the
underline escape and survives a faulty dim bit, so a machine’s ROM revision is
readable off the screen (§1), and ^D at the boot prompt separates the early
ROMs from the 1.40-and-later ones (§1a). The copy is written down to the byte
(§2, §4) but has not been run end to end, and the machine this was written for
is out of play (§7a), so there is at present no hardware route to any NO_DUMP
slot. Issues: #205, #70.
MAME carries BIOS A, 1.2, 1.2.1 and 1.3 as NO_DUMP, and the one thing that
would settle 1.3 is a dump from a machine that has one. This is the procedure
for reading it off a real Osborne 1 — what can be done from inside the machine,
what each route needs, and what the bytes should look like so the answer is
unambiguous.
Nothing here writes to a diskette. Every route below is read-only on the media.
Vocabulary:
rev Bin this document is the Osborne 1’s monitor ROM (9/12/81, theOCCROM0Blineage) — it is not a board revision. Board revisions in this project belong to the boards; the register and the naming rule are in [hardware/README.md](https://github.com/Myrcurial/o1next/blob/main/hardware/README.md).
1. First, the free answer: read the upper line of the sign-on
The sign-on has two lines and they are framed by different video escapes —
the upper one by ESC 'l' … ESC 'm', the lower one by ESC ')' … ESC '('.
They are the underline pair and the half-intensity pair respectively, and
that is why a machine with a faulty dim bit loses exactly the revision line and
nothing else. Confirmed on a real machine: the upper line renders
underlined, and the revision line is the one that goes.
The two attributes live in different places, which is the whole reason the
fault is so selective. Underline is bit 7 of the main video RAM; dim is
bit 8 — the dedicated 4116 in bank 3 (docs/o1-memory-io-map.md §2). So a
fault in the bank-3 attribute path (that 4116, or the BIT 9 latch that maps
it) costs the dim line and leaves the underlined one perfectly legible. One
chip, not the video RAM.
So: the upper line is readable even when the version line is not, and the upper line differs by revision.
| Upper line reads | ROM |
|---|---|
OSBORNE SYSTEM ONE. (all caps) |
rev B, 9/12/81 |
Osborne System ONE. (mixed case) |
1.3, 12/16/81 |
OSBORNE 1 |
1.4 / 1.43 / 1.44 and later |
The lower line, for completeness: ` MONITOR 1.3 ` on 1.3,
` Rev 1.44 (c) 1983 OCC ` on 1.44, REV B 227 on rev B.
This is derived from the sources, not from a guess: occrom1b.asm line 338 is
DB 'OSBORNE SYSTEM ONE.'; the 1.3 source (occrom13.asm, IBMSG) is
DB 'Osborne System ONE.' and then DB ' MONITOR 1.3 '; and
research/roms/os1-144.rom at 0x1C3 is OSBORNE 1.
If the upper line reads Osborne System ONE., stop and dump it — that is
the ROM nobody has.
1a. The second free test, and the one that matters most: press ^D
The early ROMs carry a resident diagnostic section, entered from the boot
prompt with ^D (the 1.3 source says so in occrom73.asm’s header). The later
ones do not have it at all. Checked against every ROM we hold:
| ROM | ^D diagnostic menu? |
|---|---|
| rev B, 9/12/81 | yes — ` ROM DIAGNOSTICs` |
| 1.3, 12/16/81 | yes — ` ROM DIAGNOSTICs` |
| 1.40 | no |
| 1.43 | no |
| 1.44 | no |
So, at the boot prompt, before pressing RETURN:
- a
ROM DIAGNOSTICsmenu appears → the ROM is an early one: rev B, 1.2, 1.2.1 or 1.3. Combined with the title test above, that excludes rev B and 1.3 — so it is 1.2 or 1.2.1, both of which areNO_DUMP. Dump it. - nothing happens → it is 1.40, 1.43 or 1.44, all of which we already hold. Nothing is gained and the trip is not worth making.
The menu is bright text, so the faulty dim bit does not hide it, and the test
needs no disk, no tool and no typing beyond ^D. The 1.3-era menu reads
A,B Boot Sys / D isk / K ey-vdt / M emory / R ead in test; a different set of
entries is itself evidence of an even earlier ROM than 1.3.
2. Reading the bytes: the ROM is not visible from CP/M
This is the part that catches people out. DUMP.COM and DDT’s D command
will not show the ROM: the monitor ROM is banked into 0000–0FFF only when
ROM MODE* is selected, and a CP/M program runs with the RAM bank in place.
The mechanism is in docs/o1-memory-io-map.md §2 and §4, measured on the
running machine by tools/check_o1_map.sh (#64):
OUT (0),Aselects bank 2 — the ROM at0000–0FFF, the memory-mapped I/O window at2000–2FFF, and4000–FFFFstill bank 1 RAM;OUT (1),Aselects bank 1 — main RAM;- the port number selects and the data byte is ignored;
- the selection is re-derived, not latched (
tools/o1_mame_probe.luacheck C10: write the RAM bank, let emulated time pass, and the ROM bank is back) — so the bank is selected and used inside the copy, never assumed to hold; - an interrupt acknowledge reselects the ROM bank, so
DIfor the duration of the copy, or a 60 Hz video interrupt will put RAM back under the read; - the program itself must live above
4000, because with bank 2 selected2000–3FFFis the I/O window and code there would vanish.
So reading the ROM is a three-instruction preamble plus a copy: select bank 2,
copy 0000–0FFF to somewhere in 4000+, select bank 1, return.
3. What is on the disks, so nothing is planned around a missing tool
Checked against the in-repo images, both single-density:
| Disk | Tools |
|---|---|
OS1SYSS.IMD — CP/M 2.2 System Disk, SSSD 92K |
autost copy help movcpm pip setup stat sysgen xdir — no ddt, no asm, no dump |
OS1UTLS.IMD — CP/M 2.2 Utilities Disk, SSSD 92K |
asm autost ddt dump ed install load movcpm stat submit sysgen xdir xsub |
OS1XUTLS.IMD — Extended utilities (80-col card) |
autost copy movcpm pip setup sysgen xdir |
So: DDT is on the Utilities disk, not the System disk. Boot the System
disk, then swap the Utilities disk into drive A and run DDT from it. (CP/M is
already in RAM; nothing is written to either disk.)
4. The DDT procedure (needs only the two SD disks above)
DDT’s A command assembles straight into memory, so the program is typed in
mnemonics — no assembler, no source file, no disk write. Type it at 0x4000
and it copies the ROM to 0x5000, where DDT’s D can show it.
A>DDT
-DDT
-A4000
4000 DI
4001 XRA A
4002 OUT 0
4004 LXI H,0000
4007 LXI D,5000
400A LXI B,1000
400D MOV A,M
400E STAX D
400F INX H
4010 INX D
4011 DCX B
4012 MOV A,B
4013 ORA C
4014 JNZ 400D
4017 MVI A,1
4019 OUT 1
401B EI
401C RET
401D <- a blank line ends assembly
-G4000
-D51C0,5200 <- the banner, copied
Bytes, so the assembled listing can be checked by eye:
4000 F3 DI
4001 AF XRA A
4002 D3 00 OUT 0 ; bank 2: ROM at 0000-0FFF
4004 21 00 00 LXI H,0000
4007 11 00 50 LXI D,5000
400A 01 00 10 LXI B,1000
400D 7E MOV A,M <- loop
400E 12 STAX D
400F 23 INX H
4010 13 INX D
4011 0B DCX B
4012 78 MOV A,B
4013 B1 ORA C
4014 C2 0D 40 JNZ 400D
4017 3E 01 MVI A,1
4019 D3 01 OUT 1 ; bank 1: back to RAM
401B FB EI
401C C9 RET
D51C0 then shows 0000+1C0’s bytes — the banner. MONITOR 1.3 in hex is
4D 4F 4E 49 54 4F 52 20 31 2E 33; OSBORNE 1 is
4F 53 42 4E 4F 52 4E 45 20 31 — either is unmistakable in a dump.
How far this is checked. The mechanism is measured — docs/o1-memory-io-map.md
§4 was measured on the running machine with tools/check_o1_map.sh — and the byte
offsets in §5 come from the shipped ROMs. The DDT route itself is reasoned from
those and has not been run end to end: the attempt under MAME failed on the
harness, tools/research/o1keys.lua segfaulting partway through typing the
program in, not on the program.
5. Fingerprints, so the answer is not a judgement call
Offsets are into the 4096-byte image.
| ROM | Offset | Bytes | Text |
|---|---|---|---|
| 1.44 | 0x1C3 |
4F 53 42 4F 52 4E 45 20 31 |
OSBORNE 1 |
| 1.44 | 0x1F7 |
52 65 76 20 31 2E 34 34 … |
Rev 1.44 (c) 1983 OCC |
| 1.43 | 0x1F7 |
52 65 76 20 31 2E 34 33 … |
Rev 1.43 (c) 1982 OCC |
| 1.3 | 0x19E |
4F 73 62 6F 72 6E 65 20 53 79 73 74 65 6D 20 4F 4E 45 2E |
Osborne System ONE. |
| 1.3 | 0x1D2 |
20 4D 4F 4E 49 54 4F 52 20 31 2E 33 20 20 |
` MONITOR 1.3 ` |
| 1.3 | 0x0F4 |
63 2E 20 31 39 38 31 20 4F 43 43 |
c. 1981 OCC |
| rev B | — | from occrom1b.asm |
OSBORNE SYSTEM ONE. (caps) |
And check the encoding of the messages, not just the text. The banner is
plain ASCII in every revision. What differs is how the message strings are
written: rev B and 1.3 use the DBE/DCE pseudo-ops, so each byte is stored
doubled (value * 2, which is why ` ROM DIAGNOSTICs appears as
40 40 40 A4 9E 9A …), while 1.44 uses DB/DC throughout and stores plain
ASCII. Shift a doubled string right one bit to read it. See
[docs/rom-13-reconstruction.md`](https://github.com/Myrcurial/o1next/blob/main/docs/rom-13-reconstruction.md) §4.
The 1.3 offsets come from the reconstructed image, and not from the damaged
regions (0x1F0, 0x8B0, 0xA80, 0xAB0, 0xFA0, 0xFD0 — see
docs/rom-13-reconstruction.md §7), so they are trustworthy. c. 1981 OCC at
0x0F4 is a COPYR: label the 1.3 ROM never references — dead data, but a
unique and very convenient fingerprint.
A rev-B image from OCCROM.IMD (OCCROM0B.COM, CRC32 f5f697c7) is also on
record, and it boots — docs/o1-rom-variants.md.
6. A dead end, recorded so it is not repeated
SETUP.COM does not identify the ROM. It offers “press M to get source
from memory”, which reads the running system’s CBIOS configuration block,
not the monitor ROM — its six items (printer, baud, screen size, auto-scroll,
function keys, arrow keys) are the CBIOS’s, and its own banner (Rev 2.2) is
SETUP’s version, not the ROM’s. Checked by running it under MAME. The ROM and
the system disk version independently; that is the same trap as reading
CBIOSV out of the CBIOS.
7. For a real dump — what MAME actually wants
Everything above is a software copy through the machine: fine for identification, and usable as evidence, but it is not a chip read.
UD11 is a socketed 2732. Pulling it and reading it in an EPROM programmer
is the only true dump, and it needs no software, no banking subtleties and no
serial transfer. That is the artifact to hand upstream; the DDT copy is the
thing to do first because it takes two minutes and tells you whether the trip
is worth making.
Getting a 4 KB file off the machine otherwise means either transcribing a hex
dump from the screen, or a serial capture — the ROM’s own SIO init is
SI.S16, “16x clock for 1200 baud” (occrom13.asm), and CP/M’s PIP can send
a file to PUN:/LST:, but binary down a 1200-baud line wants a cable and
care.
7a. The machine this was written for is out of play
The single-density machine that prompted this doc is out of play — it blew
a RIFA capacitor in the PSU. The ^D test in §1a was never run, so its ROM
was never identified, and the question of whether it held a NO_DUMP
revision is still open. The other two machines on hand are 1.44, which is
already held (docs/o1-rom-variants.md), so there is at present no hardware
route to any NO_DUMP slot — §1a and §4 are procedures waiting on a machine,
not instructions for one.
One repository fact belongs with it. The tree holds the supply’s repair manual —
research/osborne1/o1astecpsac8151.pdf, the Repair Manual for Astec Power Supply
AC8151, with the test set-up, the 115/230 V input-select wire and the +5/+12/−12
rail table (#52; a row in research/README.md) — but no supply schematic and no
board-level view, and the Technical Manual in-repo
(research/osborne1/2F00153-01_Osborne1TechnicalManual_1982.pdf) carries only
§1.2 “Power On”, two pages of procedure. What is missing is exactly what a
photograph of the board would supply. The one design note is
docs/design/o1next-lunchbox-concept.svg, which records that the original supply
“and its RIFA caps” are retired in that concept.
8. What the result would settle
- A 1.3 dump fills
osb13.binand closes theNO_DUMPslot outright. - It would also grade the reconstruction in
docs/rom-13-reconstruction.md. That doc’s control build reproduces a shipped Osborne ROM to 3759 of its 4096 bytes, and the same toolchain’s 1.3 image is uncertified for the same reason; a real dump would say whether the recoveredDBE/DCErule, theocctxt6.astassumption and the 28-byte question were right — a positive answer would validate the method for rev B, BIOS A, 1.2 and 1.2.1 as well, and a negative one would say exactly where the method is wrong.
Related: #205, #70, #72, #222, docs/o1-rom-variants.md,
docs/o1-memory-io-map.md, docs/rom-13-reconstruction.md.