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 port map, the three banks and the bank-switch mechanism are confirmed twice over — read from the schematic and the ROM source, then measured on the running machine (tools/check_o1_map.sh, #64). What this document does not carry is the decode granularity below a 256-byte page: the chip selects check A11 plus one of A8/A9/A10, which is a schematic-level fact recorded in docs/o1-mainboard-schematic.md §7 and §7a.

The machine’s 64 KB address space is overlaid by three banks (§2), selected by two latched signals. Every onboard peripheral is memory-mapped inside the bank-2 window 2000–2FFF (§3). The bank switch is four writes to I/O ports 0x00–0x03, whose data byte is ignored (§4). Three IRQs and the front-panel RESET reach the Z80 (§5). §6 is the pinout reference for every socket and connector a tap or a plug has to mate with; §7 is the free-window map and the recommended virtual-peripheral base; §8 is what the map decides.

Sources: the OCC mainboard schematic 1A2011-00 Rev E (research/osborne1/OCC1_1A2011-00_Schem_RevE.pdf), the Osborne 1 Technical Manual 2F00153-01 (1982), the ROM 1.44 / CBIOS source, and the MAME osborne1.cpp driver as an independent cross-check.

Serves #40, #62, #64 and #72. Unblocks: #11 (hardware), #16–#24 (firmware), the memory-resident vs pure-I/O decision (gap #15), the 6850 ACIA modem (#21, gap #10), the WD179x floppy interposer (#20, gap #11), and the ScreenPac personality (#24).


1. CPU and clocking

Item Value
CPU NEC μPD780C (≡ Zilog Z80A), DIP-40, location UC11
Crystal 15.9744 MHz (X1)
CPU clock 4 MHz (÷4 via UD3 LS161)
Dot clock 8 MHz
Memory / char clock 2 MHz / 1 MHz
Bus width 8-bit data (DATA0–7), 16-bit address (ADR0–15)

Wait-state generation: RAM accesses insert wait states (≈375 ns for non-M1 cycles, 188 ns for the first M1); ROM executes without added delay.

1a. One crystal, no free lunch — and the 2× “business turbo”

Per Lee Felsenstein’s rule of thumb — “the number of crystals in a machine design is equivalent to the number of designers” — the O1 has one oscillator, and the whole timing economy (video, ACIA baud, floppy data separator, DRAM refresh, software delay loops) is divided down from it. Consequences:

  • Serial quirk already visible: the J1 600/2400 jumper is ~8.3% fast (650/2600 actual) because the video-tap ÷13 ÷(16|64) chain can’t land exactly on 600/2400 from 15.9744 MHz. The factory’s own alternate baud rate shows the seams.
  • No clean general turbo: 3× has no integer relationship to the video chain (breaks video), and any CPU speedup drags the ACIA baud rate, the FDC data separator, and software timing loops along with it.
  • The clean-ish 2×: doubling the CPU clock to 8 MHz keeps the ÷2 dot-clock tap coherent (15.9744 ÷ 2 = 8 MHz), so video timing survives. It is viable for business apps if the three side effects are engineered around:
    1. Floppy — the WD1793 data separator is timed for 4 MHz; in turbo mode physical floppies are out. Use the interposer’s storage instead (#20/#22: image-library / IEEE-488 storage answers on the bus at whatever speed the host runs), or gate turbo off when the FDC is selected.
    2. Timing loops — accept that games/delays run double-speed; scope the feature to calculation-bound work (SuperCalc recalc, dBASE II, compilers, assemblers). A classic PC-era “TURBO” switch selects 4 MHz ↔ 8 MHz.
    3. Serial — no clean fix at the ACIA. Mitigation: build a baud-rate buffering corrector into the WiFi modem (#21). The modem’s RP2354B talks to the far end at a rock-solid standard rate (2400/9600) and tolerates whatever the O1 ACIA is actually doing: on send it buffers until a batch accumulates before transmitting; on receive it buffers a frame, then re-clocks it out to the O1 at the rate the O1 expects. The O1-side rate drift becomes the modem’s problem, not the link’s.
  • Net: a 2× business turbo is a documented, buildable option (mostly on the replica #49, where SRAM removes the DRAM-refresh constraint); it is not a general speedup and doesn’t pretend to be.

2. Memory banks

The 64 KB address space is overlaid by three banks, selected by the ROM MODE* signal and the BIT 9 latch (see §4).

Bank 1 — main RAM (“normal mode”)

64 KB of 4116 dynamic RAM. The top 4 KB is dual-purpose as video RAM.

Range Contents
0000–3FFF RAM (boot ROM overlays the low 16 KB when bank 2 is active)
4000–EFFF RAM — CP/M TPA, BDOS/CCP, BIOS live here
F000–FFFF Video RAM low 8 bits (128 cols × 32 rows char data)

Page zero. CP/M owns 0000–00FF while a program runs, but the Osborne 1 User’s Guide “Memory Allocation” chart (p. R-243) marks 0050H–005BH — 12 bytes — as “not used”, and at least one published program relied on that: the date facility in The Portable Companion #7 keeps the system date there in both binary and ASCII form (docs/rt60a-analysis.md). Treat it as documented free scratch space — worth remembering for the interposer (#12).

Bank 2 — BIOS ROM + memory-mapped I/O (“shadow mode”)

Selected when ROM MODE* is active (low). Only 0000–3FFF is affected; 4000–FFFF still reads bank 1 RAM.

Range Contents
0000–0FFF Monitor ROM (4 KB, 2732 at UD11; earlier 2× 2716)
1000–1FFF Unused — reads as mirror of the ROM
2000–2FFF Memory-mapped I/O window (see §3)
3000–3FFF Unused — reads as mirror of 2000–2FFF
4000–FFFF Mirrors bank 1 RAM

Bank 3 — video attribute bit (“dim bit”)

Selected when BIT 9 latch is set. A 9th bit per video cell, stored in a dedicated 4116. Only F000–FFFF is affected.

Range Contents
F000–FFFF Video attribute RAM, 4K×1 (dim attribute = bit 8 of each cell)

2a. Video memory geometry — 128 columns wide, 52 displayed (#97)

The 4 KB at F000–FFFF is 32 rows of 128 characters, of which 24 rows of 52 characters are displayed at a time:

Memory from 0F000 hex to 0FFFF hex is used for the video memory map. That memory is considered to be 32 rows of 128 characters, even though only 24 rows of 52 characters are actually displayed on the monitor screen at a time. The character matrix is 7 × 9 in an 8 × 10 box. — Osborne 1 Technical Manual, §5.2 “Video Memory”, p. 23

This is the answer to the “64 vs 52” question in #97. The screen is a window onto a buffer 2.5× wider and 1.3× taller than what is shown, and the displayed width is a property of the window, not of the memory. That is why the same 4 KB shows 52 columns on stock hardware, 80 on Nuevo and 104 with the ScreenPac (mame-emulation.md §2) — the text layout is the CBIOS’s, the window is the machine’s.

It is also why the scrolling feels different from a terminal’s. Holding Ctrl and using the arrow keys slides the window over the memory rather than moving the memory, which is a register change and not a rewrite of the buffer — the same mechanism mame-emulation.md §6 calls a ring buffer. A terminal scroll does move the lines, which is why that stays chunky while the viewport slide is smooth. (The Ctrl+arrow slide is a hardware observation, reported on the real machine; the geometry above is the manual’s.)

Bit fields (§5.2, figure): bits 0–6 are the ASCII character, bit 7 is the underline attribute, bit 8 is dim. Bits 0–7 are the byte in the video RAM proper; bit 8 is the separate 4116 in bank 3. This is why the MAME probe masks with % 128 to read characters (tools/o1_mame_probe.lua).

The window width is software-selectable. The SCREEN-PAC user guide (p. 13) gives the mechanism, and it is in the stock BIOS rather than on the board:

At BIOS location E1A5H, POKE or store 0 for 52 columns / 1 for 104 columns / 3 for 80 columns. CALL BIOS location E4E5H. … In these same formats, SCREEN-PAC provides horizontal scrolling by 32 column blocks.

So 52, 80 and 104 are three widths of one window onto the same 128×32 buffer, and the 32-column blocks are why the slide lands on round numbers. Without the SCREEN-PAC only 52 is available, because the wider windows need the board’s dual-port video RAM — the ROM contains the routine either way.

Logical width is not physical width — this is what “64 vs 52” turns on. The SCREEN-PAC guide (p. 13) separates the two, and it is the piece the technical manual’s geometry leaves implicit:

Logical screen size refers to the number of characters you can enter on a line without automatically inserting a carriage return. Logical screen size can range from 1 up to a maximum of 128 characters.

Physical screen size refers to the number of columns actually displayed on your screen. With SCREEN-PAC, this will be 52, 80, or 104. — Osborne 1 SCREEN-PAC User Guide, p. 13

So a program’s line width is a software choice up to 128 — the width of the video memory row — while the window is 52, 80 or 104. A drawing 64 columns wide is a logical line of 64 characters: it fits the row, and the window slides over it. The guide also warns against the obvious mistake, which is worth knowing before writing anything that sets a width: “logical screen size should NOT be less than physical screen size”, or the display scrambles.

The width is set from the configuration screen (Item C, “Screen Size”, in the Osborne 1 Configuration Program) or from software by poking E1A5H and calling E4E5H as above. Both do the same thing.

The cursor is hardware, and it can be reversed (#24). The stock cursor is an underline, and OCC knew it was poor — Thom Hogan argued against it while at the company and was refused twice (The Portable Companion #13, June/July 1984, “Reverse Video and a Faster Serial Port”, p. 24): he could not get the change made permanent, and could not get permission to publish it, “as the company was worried about the legal implications of publishing modifications”. The modification makes the cursor a solid white block, and the article locates the circuit precisely: a 74LS86 piggybacked on the 74LS00 at E22, with pin 1 of the 74LS74 at UE21 lifted, plus changes around the VIDEO and DIM signals.

So the cursor is generated in hardware — not by the character generator, and not by the attribute bit — which matters to a video personality: it has to reproduce a cursor the machine makes outside video RAM, and the block cursor is the variant OCC’s own engineer preferred. Extract: research/technical-articles/TPC-Issue13-1984-pp26-31-reverse-video-and-9600-baud.pdf.

And it works with the SCREEN-PAC fitted, leaving the board in place. Hogan’s version is for a stock machine. A reader’s fix, printed in The Portable Companion #15 (Jan/Feb 1985, letters, PDF p. 32), makes the block cursor work on a machine with the SCREEN-PAC 80 installed — “and it can be carried out without removing the piggy-back board containing the Screen-Pac 80 electronics”. What changes is where the logic is: the modification is done on the upgrade board’s own chips — U2 (74HC00) and U23 (74HC74) — not the main board’s, and the letter warns in capitals that “THESE PINS ARE NOT THE SAME ONES LISTED IN THOM HOGAN’S ARTICLE!”.

That the SCREEN-PAC carries a 74HC00 and a 74HC74 of its own is useful in itself: the cursor logic is duplicated on the piggy-back board, which fits the service manual’s description of the upgrade as an IC-harness board soldered over the main logic (2F00040 §6.12, the source used by firmware/screenpac-video/). Extract: research/technical-articles/TPC-Issue15-1985-pp32-block-cursor-with-screenpac.pdf.

2b. The monitor RAM and the system clock (#7, #81)

The top of bank 1 RAM (EF00–EFFF, 256 bytes) is the monitor RAM: the ROM’s own storage, above the CP/M TPA. The CBIOS RAM map (occram25.asm) and the ROM’s absolute equates (rom144.asm) agree on its layout, and it contains the machine’s system clock:

Address Cell Meaning
EF00 TEM boot-routine scratch
EF08 ROMRAM ROM/RAM flag
EF60 ESCH ESC-holding flag
EF61 / EF62 PIAAD / PIABD last PIA-A / PIA-B data
EF63 / EF64 / EF65 IDAY / IMONTH / IYR calendar date, binary
EF66 HOURS hours, 2-digit BCD
EF67 MINS minutes, 2-digit BCD
EF68 SECS seconds, 2-digit BCD
EF69 SEC6 sixtieths, 2-digit BCD
EF6A DACTVE disk-active countdown
EF6B BELCNT bell countdown
EF6C LLIMIT max columns in a logical line
EF6E LDTRK last track used on the non-selected drive
EF6F–EF70 IESTK saved stack pointer
EFD0 SAVTYP disk type (density + sector size)
EFD1 RWCMD sectors to read/write
EFEF VRTOFF last vertical offset
EFF0–EFFF INTBL 8-word IM 2 interrupt vector table

HOURS–SEC6 is a real clock. The stock ROM’s 60 Hz interrupt handler (GKEY) calls UPTIM, which increments SEC6, SECS, MINS and HOURS in BCD with a 24-hour wrap. It starts at midnight on every boot, nothing sets it and nothing displays it in a stock machine — but OZROM 1E documents the same four cells and adds a setter, a display and an alarm. The EFF0 vector table is where the two clock products hook: INTBL+(4*2) = EFF8 is the keyboard-interrupt vector the RT-60A’s CLOCK.COM replaces, while OZROM calls its own hook at EFF2. See docs/rt60a-analysis.md (“The third shape”) and docs/ozrom-1e.md.

3. I/O decode (bank 2, “memory-mapped I/O”)

All onboard peripherals are memory-mapped, reached through the bank-2 window. Chip-selects are generated from address lines within 2000–2FFF.

Address Device Chip Registers
2100–2103 Floppy disk controller Fujitsu MB8877 ≡ WD1793 (UB7) 2100 stat/cmd, 2101 track, 2102 sector, 2103 data
2201–2280 Keyboard (read) 81LS95 + 74LS05 (UE12/13/14) One address bit per row: 2201=row0 … 2280=row7; columns on DATA0–7
2900–2903 IEEE-488 PIA MC6821 (UC7) 2900 PA data/dir, 2901 PA ctrl, 2902 PB data/dir, 2903 PB ctrl
2A00–2A01 Serial / modem ACIA MC6850 (UC4) 2A00 status(R)/control(W), 2A01 RX(R)/TX(W)
2C00–2C03 Video PIA MC6821 (UC15) scroll/offset latches, bell, brightness, dim control
2400 (read) SCREEN-PAC presence — bit 2 LOW = SCREEN-PAC fitted, HIGH = stock machine (user guide p. 13). Read-only status bit, not a register

6850 ACIA detail (#21 modem, gap #10)

  • Base 2A00: read = status, write = control. 2A01: read = RX, write = TX.
  • Baud clock from divider UC3 (÷13) fed by a 2/4 µs tap off the video counter (UA13). ACIA (UC4) divides by 16 or 64 → 1200 or 300 baud.
  • Jumper J1 selects 300/1200; Rev E+ can be re-jumpered to 600/2400.
  • Faster rates, from inside OCC (#21): The Portable Companion #13 (June/July 1984), “Reverse Video and a Faster Serial Port” by Thom Hogan — the same Hogan whose AUTOCPM.ASM is the source of the site wordmark (#97) — written as former OCC manager of software and publications: “developed internally by the company, [but] not promoted … the primary … was the problem of warranty repair on modified machines.”
    • Removing J1 doubles 300/1200 to 600/2400 — exactly as above — and he adds a detail the schematic cannot show: the SETUP program cannot sense J1, so it still displays 300/1200 after the modification.
    • For 2400/9600 the mod is to the baud divider: the LS161/LS163 at C3, with pin 2 moved to the middle pin of J2 (“for 2400/9600 baud rates rather than 1200/4800”) and pins 3, 7 and 10 lifted. Figure 3 in the article has the detail. Extract: research/technical-articles/TPC-Issue13-1984-pp26-31-reverse-video-and-9600-baud.pdf.
  • Pin level, from the schematic (sheet 7/9 of 9, read at 600–1200 dpi): the ACIA’s RxCLK (pin 3) and TxCLK (pin 4) are strapped together on the board and driven by UC3’s QD (pin 11). UC3 is a 74LS161 preset-load counter — RCO (pin 15) is inverted by UB3 (74LS04) into /LD (pin 9) — so the count length comes from the preset pads rather than being a power of two; that is where the ÷13 of the stock 19.2 kHz clock lives. Pin numbers as drawn: RXDATA 2, RxCLK 3, TxCLK 4, /RTS 5, TXDATA 6, /IRQ 7, CS2 8, /CS1 9, CS0 10, RS 11, R/W 13, E 14, D7…D0 15–22, /DCD 23, /CTS 24 (Vcc/Vss, pins 12/1, are not drawn on the symbol).
  • The factory’s own 9600/2400 modification is documented, which the archive had only from inside OCC: Osborne 1 Technical Manual Figure 4.4 “Baud Rate Hardware Modification”, p. 22 (PDF page 36 — the manual’s printed-page offset of +14 holds here too, see docs/o1-chargen-graphics.md). It draws the LS161/LS163 at C3 with pin 2 moved between a “FOR 9600/2400 BAUD” pad and a “FOR 4800/1200 BAUD” pad, jumpers J2 and J3, J1 marked “DON’T CARE”, and the caption box 300/1200 NO MOD - J1 IN / 600/2400 NO MOD - J1 OUT / 1200/4800 MOD W/O PIN 2 (J1 = X) / 2400/9600 MOD W/PIN 2 (J1 = X), bracketed “SPEED SELECTED W/IN 6850“. That is the same modification Hogan describes from inside OCC in #13 — the 6850 choosing within each pair (÷64/÷16). It is also the “method shown on page 22 in the Technical Manual” that The Portable Companion #8 points a would-be 80-column terminal user at (see docs/o1-modem-serial.md).
  • A fifth route, and a third-party one: Advent Products’ Osbaud (1983, $59.95, 16 DIP-switch rates) is soldered to the ACIA’s own clock and supply pins and replaces the clock net outright — see docs/o1-modem-serial.md.
  • The ACIA’s control-register init byte is in the BIOS at 1FE9H. The vendor’s own “Programming the serial port” patch (The Portable Companion #7, June 1983, PDF pp. 56–57) rewrites 1FE9H/1FEAH to move off the 8-N-1 default, with a table of 6850 control codes (55H = 8-N-1 @1200). A second handle on the ACIA besides the port pair — see docs/o1-modem-serial.md §”Framing”.
  • Modem status (DTE): MSB (pin 4), MCB (pin 8), RI (pin 9); BIOS LISTST at 0E12D returns 0FFH/00H.
  • IRQ0 = ACIA interrupt (one of three IRQ sources; see §5).

WD179x floppy detail (#20 floppy interposer, gap #11)

  • Controller UB7, registers 2100–2103 as above.
  • Drive-select / density lines are driven from the video PIA (UC15) outputs (DRIVE1/DRIVE2/DDEN), not a dedicated latch — see schematic sheet 5. DDEN selects double density when the add-on DD board is present.
  • Stock format: single-sided single-density, 40 tracks × 10×256-byte sectors (102 KB/drive), FM. DD option board adds 5×1024 Osborne-DD, 8×512 IBM, 18×128 Xerox-820, 9×512 DEC-182 formats.

Keyboard detail

  • 8×8 keyswitch matrix; rows driven on address lines A0–A7 (one-hot), columns returned on DATA0–7.
  • Read strobes at 2201, 2202, 2204, 2208, 2210, 2220, 2240, 2280 (i.e. 2200 + (1<<row)).
  • 14 programmable keys (Ctrl-0…9, four arrows) via BIOS + SETUP.

4. Bank switching (the mechanism)

Banking is controlled by writes to I/O space, decoded by UD10 (LS138) on the two low address bits; the data bus value is ignored.

I/O write Effect
0x00 Clear UB4 → ROM MODE* low → select bank 2 (ROM+I/O). Also forced by CPU reset.
0x01 Preset UB4 → ROM MODE* high → select bank 1 (main RAM), if MREQ low.
0x02 Set BIT 9 latch (UE10A) high → map bank 3 (dim bit) into F000–FFFF.
0x03 Clear BIT 9 latch → map bank 1/2 back into F000–FFFF.

ROM MODE* is also set by hardware: power-on reset, and the NMI flip-flop chain (UB6/UB4). Bank selection between 1 and 2 is additionally qualified by M1 and IRQ-acknowledge conditions via three flip-flops — so interrupt acknowledge and the front-panel RESET reselect the ROM bank automatically.

Interposer note: because only the two low address bits are decoded and data is ignored, any OUT (0),A / OUT (1),A toggles banks. The interposer must never drive I/O addresses 0x00–0x03.

Measured on the running machine (tools/check_o1_map.sh, #64). All four writes behave exactly as tabulated — the port number selects, the data byte is ignored, and ports 0x04–0x07 alias 0x00–0x03 (MAME’s I/O space reports mask=0x0003, i.e. only A0/A1 reach the decode). The M1/IRQACK qualification above is not a footnote: write 0x01 and the RAM bank is selected immediately, but 10 ms of emulated time later the machine is back in the ROM bank, because the bank is re-derived on every instruction fetch and the video PIA interrupts at 60 Hz.

Consequences for an interposer:

  • The bank state cannot be latched and assumed — it has to be sampled per bus cycle, and a bank-2 I/O access is only meaningful while the CPU is actually running with bank 2 selected.
  • At the idle A> prompt the machine is parked in the ROM polling loop with bank 2 live, so the live bank depends on what is running. That does not make the ROM readable from CP/M: a program in the TPA runs in bank 1 with the ROM banked out, which is why OUT (0) is needed to select it and why DUMP/DDT return nothing useful (docs/o1-rom-read-from-machine.md §2).

5. Interrupts

Three IRQ sources into the Z80:

  • IRQ0 — serial ACIA (UC4)
  • IRQ1 — video PIA IRQA (UC15)
  • IRQ2 — IEEE-488 PIA IRQA (UC7)

NMI is generated by the front-panel RESET switch S1 (debounced by UE20, sequenced through UB6/UB4 on M1): NMI restarts the ROM program at 0000 and forces bank 2. There is no periodic/60 Hz NMI — the “60 SEC” signal is a video-tick line into the video PIA CA1 (used by the console routine’s real-time clock), not an NMI.

6. Connector / socket pinouts (mechanical tap points)

Z80 CPU — DIP-40 (UC11)

Standard Z80A pinout (Zilog Z80 Microprocessor Family Databook; the O1 fits an NEC µPD780C, pin-compatible), and the O1’s own sheet says the same thing pin for pin — the socket is wired as a plain Z80 DIP-40, so a board that mates with it can use the datasheet table (docs/o1-mainboard-schematic.md §4, read at 600 dpi). tools/check_schematics.py asserts this table against every connector that mates with the O1’s Z80 socket.

Pin(s) Signal
1–5 A11, A12, A13, A14, A15
6 CLK (4 MHz)
7–10 D4, D3, D5, D6
11 +5 V
12–15 D2, D7, D0, D1
16 INT
17 NMI
18 HALT
19 MREQ
20 IORQ
21 RD
22 WR
23 BUSAK
24 WAIT
25 BUSRQ
26 RESET
27 M1
28 RFSH
29 GND
30–34 A0, A1, A2, A3, A4
35–40 A5, A6, A7, A8, A9, A10

Character generator ROM — 24-pin (UA15)

2 KB (2716) character ROM at UA15 — on sheet 4 of 9 (PDF page 7, the RAM sheet), not the video sheet; see o1-mainboard-schematic.md §4a for the extraction. 128 chars × 8×10 in an 8×10 box (7×9 visible), addressed by (char code + scan line), output to the video shift register (UA14, 74166). This is the second tap point the ScreenPac personality (#24) uses. (It is the third tap on the interposer, which also leads to the main ROM socket UD11 — see docs/virtual-peripherals-block-diagram.md §1 and docs/o1-rom-2.0.md §4.1a.)

Pin Signal
1–8 A7–A0 (A0–A6 = char code from UA18; A7 = SCAN0)
9, 10, 11, 13–17 O0–O7
12 GND — bussed with 18 and 20 to ground
18 /CE — grounded
19 A10 (SCAN3)
20 /OE — grounded
21 strapped +5 V (A11 on a 2732)
22 A9 (SCAN2)
23 A8 (SCAN1)
24 VCC (+5)

The socket is the plain Intel 2716 pinout, read at 1200 dpi (docs/o1-mainboard-schematic.md §4a carries the extraction). Because /CE (18) and /OE (20) are hard-grounded, the ROM always drives O0–O7: a tap may buffer and read those pins, but it cannot tristate the ROM to inject its own font without intercepting pins 18 and 20 between socket and ROM.

Other onboard connectors (for reference)

Every connector on an Osborne 1 is on the FRONT of the machine — there is nothing to plug into at the back. That is not a detail: it is why fitting a period peripheral meant unclipping the keyboard, which docks to seal that same face (the O1A design note in hardware/next-integrated/), and it is why every standalone device in this project attaches at the front (#47, #56, #84).

  • P3 — IEEE-488, 26-pin edge (pins 25/26 unused)
  • P2 — RS-232, DB-25 (DTE)
  • P1 — Modem
  • P4 — Keyboard, 20-pin IDC (1=GND, 2–9 rows 4,0,3,6,2,5,1,7, 10–17 cols 0–7, 18 NC, 19=+12V, 20 GND). On the later O1A (“blue/grey”) machines the coiled-cable plug has 24 holes but only the inner 20 mate with pins. Pin 19 carries +12V only if jumper J6 is bridged (a 2-pin link, open from the factory) through R21 (22Ω) — the intended way to power an accessory off the keyboard cable. Rows are driven low (open-collector inverters, one-hot); columns are pulled up to +5V via a 3.3K resistor pack and buffered onto the data bus. Reference: O1 Technical Manual; pinout per pinouts.ru (verified against a PCB), archived in-repo at research/osborne1/keyboard/osborne_keyboard_pinout.html (provenance alongside) — see #47 for the full matrix map.
  • P8 — Floppy drive, 34-pin
  • P5 — Video, 10-pin inline

The bank-2 I/O window is sparsely decoded; many sub-ranges alias the real devices. Conservatively free (no onboard device) within 2000–2FFF:

Range Status
2000–20FF Free — and clean: no device mask matches it (see below)
2104–21FF Free (above floppy)
2300–28FF Free (above keyboard, below IEEE) — not all clean: 2300–23FF and 2500–27FF alias the floppy and the keyboard, and 2400–27FF is the SCREEN-PAC’s own select (its presence bit is 2400h)
2904–29FF Free (above IEEE PIA)
2A02–2BFF Free (above serial, below video PIA)

The O1 decode is “sloppy” (MAME): the chip selects do not check all address bits, so large regions alias real devices. Treat only page-aligned blocks between the listed devices as safe, and validate on hardware before committing.

Measured — the aliasing is real, and it has a 256-byte answer. The five selects check A11 plus one of A8/A9/A10 each, so of the region above only three page-aligned blocks are selected by nothing: 2000–20FF, 2400–24FF and 2800–28FF (offsets 000, 400, 800 — A11 set with A8/A9/A10 clear, or all four clear). 0x2301 is the floppy’s track register and the keyboard’s row 0 (docs/o1-mainboard-schematic.md §7a), which is what moved the ROM↔Pico door from 2300h to 2800h (docs/rom-pico-protocol.md §2). The SCREEN-PAC’s window test in the driver is (off & 0xc00) == 0x400 — 2400–27FF — so 2400–24FF is clean of the stock machine but not of a SCREEN-PAC one, 2400h being its presence bit. The two blocks free on every machine are 2000–20FF and 2800–28FF.

And the one device that is free by mutual exclusivity claims no bank-2 memory at all. Drive C — the standalone reproduction (#84) and the interposer’s own drive-c-gpib personality — cannot be fitted at the same time as the interposer (docs/mailbox-protocol.md §1), which invites the question of whether its space is reusable. Checked against the characterisation (docs/drive-c-protocol.md §1–§2, docs/drive-c-design.md §2): it is not a GPIB device and has no window. The Drive C connector (P3) is used as a parallel port, and the host drives it by bit-banging the mainboard’s own MC6821 PIA at 2900–2903 — machine hardware, always fitted, not borrowable. Nothing in the Drive C material names a 2xxx address other than that PIA. What its exclusivity does free is the drive-2 identity in the mailbox (which the interposer already owns) and the P3 connector itself — a bus, not an address. So there was no Drive C window to put the ROM↔Pico door in; 2800h is inside this document’s own recommended region instead, chosen from the blocks above.

Recommendation

  • Primary virtual-peripheral block: 2800–28FF (256 bytes). It is inside the clean gap between keyboard (2201–2280) and IEEE-488 (2900–2903) and it is one of the three blocks no select touches (above). The ROM↔Pico door takes 2800h/2801h (docs/rom-pico-protocol.md §2); suggested sub-allocation for the rest: CoPower-88 emulation, network/modem FIFO, RTC, status regs.
  • 2000–20FF is equally clean and equally page-aligned if a second disjoint window is needed; 2400–24FF is clean only of a stock machine, because 2400h is the SCREEN-PAC presence bit.
  • Secondary: 2B00–2BFF (between serial 2A01 and video PIA 2C00) if a third window is needed — but note that 2B00–2BFF does alias the serial ACIA’s select, so it is free of the device, not of the decode.
  • Avoid 0x00–0x03 (bank switch) entirely.
  • The interposer should decode bank-2 I/O space (memory-mapped) rather than Z80 IN/OUT port space, since onboard devices are memory-mapped and the plain I/O port space 0x00–0xFF is almost entirely consumed by the bank-switch decoder aliases (only low 2 bits decoded → 0x00–0x03 repeats every 4 bytes across the whole low port page).

8. Decisions this unblocks

  • Gap #15 (memory-resident vs pure I/O): → pure I/O, memory-mapped in bank 2 in the §7 window (2800–28FF). No need to steal RAM or fight the bank flip-flops.
  • Gap #10 (modem ACIA ports): → 2A00/2A01 confirmed; the WiFi modem can either emulate the 6850 there or live at a FIFO in the same §7 window with a small BIOS hook.
  • Gap #11 (floppy variant): → MB8877 ≡ WD1793 confirmed; registers at 2100–2103; drive/density selects are PIA-driven (emulate via PIA shadow).
  • #24 (ScreenPac): → two tap points confirmed: Z80 DIP-40 (UC11) and char-gen 24-pin (UA15). The interposer adds a third — the main ROM socket UD11 — for ROM 2.00 (#83); ScreenPac itself never needed it.

Sources

  • OCC schematic 1A2011-00 Rev E (research/osborne1/OCC1_1A2011-00_Schem_RevE.pdf), sheets 1–5 (CPU, RAM, video, IEEE-488, serial/disk).
  • Osborne 1 Technical Manual 2F00153-01 (1982), ch. 3 (memory), ch. 4 (interfaces), ch. 5 (video), ch. 7 (disk), ch. 8 (keyboard), ch. 11 (theory of operations: ROM MODE/bank select, NMI).
  • MAME src/mame/osborne/osborne1.cpp (bank/I/O map cross-check).