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: Partially solved. The port pair, the handshake and the block transfer are read out of the two RAMDISK.COM builds and corroborated by the board’s own schematic; what is left is the 8088-side command set — the monitor ROM is in the repo (research/roms/swp-p88.rom, #70) but has not been disassembled for it — and the meaning of every status bit other than the two the drivers test.

The CoPower-88 is an 8088 co-processor board sitting on the host Z80’s bus, and it presents the host with a two-port mailbox. The even address of the pair is the data/command port; the odd address is status. Every transfer is a poll of the status port followed by an OUTI/INI block. The pair is configurable twice over, and both have to agree: the board takes it from an 8-position switch (docs/copower88-schematics.md, §”Port decode”), and the driver is told the same pair at install time, defaulting to the two bytes in the .COM header.

Serves #3 (the Z80 socket and pinout this board sits in), #4 (the CoPower-88’s own ROM, which is in the archive), #5 (the port decode) and #23 (the virtual CoPower-88 personality this contract is written for).

Port Address Direction What it is
Data / command even — D in the driver (Kaypro 7Eh, Zorba FEh) both OUT latches host→8088; IN reads the 8088→host latch
Status odd — E in the driver (7Fh, FFh) read bit 0 set = the host must wait before a write; bit 7 set = a byte is waiting for the host
Doorbell the same odd address write sets INTR on the 8088 (sheet 4). Neither RAMDISK driver uses it: every write they make goes to the even port, and that write is itself the doorbell

The pair is stored as one word at 0107h — low byte (E) = status, high byte (D) = data — so a port override has to keep that order. RAMDISK.COM asks for it directly: CoPower_88 port address or <CR> to use default.

Sources: research/copower88/kaypro/extracted_cpm/ (copwrcpm.td0) and research/copower88/zorba/extracted/ (swpdos.td0). Disassemblies in */disasm/ (z80dasm, org 0x100). Schematics: research/copower88/zorba/611-0003-{1..4}.pdf — the port decode is settled in docs/copower88-schematics.md (#5).

The transfer protocol, instruction by instruction

Both drivers carry the same subroutine at 0B82h (host → 8088) and 0BE9h (8088 → host) — the machine-specific parts of RAMDISK.COM are the port defaults and the install-time patching, not the transfer:

Step The code What it means
1 LD DE,(0DB6h) / LD C,E load the port pair; C := the status (odd) port
2 IN A,(C) / RRA / JP C wait while status bit 0 is set
3 LD C,D / OUT (C),A write the command byte to the data port
4 LD A,005h … LD A,001h the two-byte preamble 05h, 01h, each behind a fresh poll
5 OUTI, with LD B,080h the block, 128 bytes at a time — one poll per byte
6 IN A,(C) / JP P / LD C,D / IN A,(C) / CP 005h on a read, wait while status bit 7 is clear, then expect 05h
7 CP 001h, then INI the same 05h, 01h preamble, then the block read

Three consequences worth stating plainly:

  • 05h/01h is a preamble, not a direction code. Both directions open with it, so it frames the transfer rather than selecting it; what distinguishes a read from a write is which subroutine the driver called. The length lives in the driver’s own work cell (0CBDh), and the 080h in the block loop is CP/M’s 128-byte record.
  • The status port is read for two different bits at two different polarities — bit 0 set means wait before a write, bit 7 clear means wait before a read. The schematic accounts for one host-visible flag, bit 0, toggled by the 8088’s write (docs/copower88-schematics.md §”Mailbox registers”); what drives bit 7 is not shown on the sheet.
  • The header bytes are the configuration (identical in both images): 0103h = the assigned drive number, zero-based — SUB 41h at 037Bh, ADD A,41h at 0753h — default 0Ch = drive M; 0104h = the erase-directory answer (59h = Y); 0107h/0108h = the port pair above. RAMDISK asks for all three at install time: drive letter (A–P), erase-directory Y/N, load address (page-aligned, ??00) and the port pair, then installs a BIOS wedge — “not standard CP/M system, cannot install” if the BIOS jump table is not where it expects it.

Machine glue differences (Kaypro MSDOS.COM vs Zorba PCDOS.COM)

  • Kaypro MSDOS.COM touches Kaypro motherboard system ports (0x10-0x14 SIO/CTC/system, 0x1C/0x1D/0x1F heavily) IN ADDITION to the CoPower ports — screen/keyboard/clock glue for the “BIOS calls only” PC emulation environment.
  • Zorba PCDOS.COM error strings show the watchdog protocol: “INVALID 8088 REQUEST #XX AT XXXX”, “8088 DOES NOT RESPOND”, “8088 SYNC ERROR, PRESS R TO RETRY” — the Z80 host polls the 8088’s request queue; the 8088 ROM monitor services requests asynchronously.
  • Both drivers are the same vintage SWP codebase (identical io subroutine offsets) — the machine-specific parts are: port defaults, console/screen BIOS glue, and install-time BIOS vector patching.

Implications for the Osborne 1 port (#26)

  1. The Osborne board’s pair is not a free choice, and it is not the Kaypro default. The CoPower-88 decodes the pair on its own board — it snoops only IORQ, one address bit and the data bus (docs/copower88-schematics.md) — but the host’s decode sees the same write cycles, and the O1 decodes only A1:A0 of an OUT (docs/o1-memory-io-map.md §4, measured), so every port aliases one of 00h–03h. The member that matters is the even one, because that is the one written on every byte: 7Eh/FEh alias 02h = set BIT 9, which maps the dim-bit bank over video. A port override therefore has to land on a pair whose written address is harmless under A1:A0 — 00 (bank-2 select, which the machine re-derives) or 11 (clear BIT 9, idempotent) — never 10. docs/mailbox-protocol.md §2.1 dodges the same hazard by the same rule, and docs/o1-memory-io-map.md §7’s “avoid 0x00–0x03 entirely” is the general form of it. What the pair is set to on an Osborne board is a measurement, not a guess — see “Open questions”.
  2. What is settled is that a port override needs no change to the driver’s code: the pair is one word at 0107h, so once the board’s setting is known the existing driver talks to it with a different constant.
  3. MSDOS.COM’s Kaypro system-port access must be mapped to Osborne equivalents (bank-2 memory-mapped ACIA/PIA) — the BIOS-call translation layer is where the porting effort lives.
  4. The 8088-side firmware is on the board’s ROM (not on these disks); the DOS disks (copwrdos.td0, dos211-*.td0) are PC-side payloads. That ROM is in the repo as research/roms/swp-p88.rom (#70) — 8088 code, dumped from an SWP board.

Open questions

  • The 8088-side command set beyond the 05h/01h preamble — sector read/write, identify, sync. research/roms/swp-p88.rom (#70) is that monitor, in 8088 code and already in the repo, so the answer is a disassembly away.
  • What drives status bit 7, the read-side “data available” flag. Sheet 4 shows one host-visible flag, on bit 0; the read path waits for bit 7.
  • The Osborne variant’s port pair — and whether it could keep the Kaypro default at all, given the A1:A0 alias above. The board photographed is a KAYPRO-88 (docs/photo-survey.md §”CoPower-88 8088 board” — the photo drop for #9), and that document reads the base as a comparator with a hardwired reference rather than the 8-position switch the drawing shows; the drawing is the primary source here, and no Osborne board’s setting is recorded either way.
  • The other six status bits — the drivers never test them, and the schematic drives only one.