PC-DOS 1.1 Hacking - Playing with PC-DOS 1.10 bios/dos and command.com

Sources

The experiment started from the vintage MS-DOS code opened by Microsoft:

https://github.com/microsoft/MS-DOS/blob/master/v1.25/source/COMMAND.ASM

ibmbio.com sources are retrieved from https://www.os2museum.com/wp/pc-dos-1-1-from-scratch

MIT License (In line with Source License):

• https://github.com/tsupplis/pcdos11-hacking/blob/master/LICENSE.md
• https://github.com/microsoft/MS-DOS/blob/master/LICENSE.md

First Contact

There is a first failure on EQU symbol redefined. This prevents compilation.

156c156
< ZERO    EQU     $
---
> ZERO    =       $

To get to the official PC-DOS binary production:

A triple NOP replaces the MS instruction MOV [COMFCB],AL. This change fixes reload from command.com on drive A: by not overriding the first FCB byte with 0 (Default drive).

469a474,479
>         IF IBMVER
>         NOP
>         NOP
>         NOP
>         ENDIF
>         IF MSVER
470a481
>         ENDIF
2166d2176
< 

The 2 previous changes allow recreation of the PC-DOS exact command.com, when MSVER is set to FALSE and IBMVER is set to TRUE

A third small change has been added in MSVER mode, before opening the command.com file. To be checked if this is really necessary ...

>         IF MSVER
>         MOV     AL, 0
>         MOV     [COMFCB], AL
>         ENDIF

Those variations beetween IBM and Microsoft seem to be linked to the command.com file being searched on the default drive (Microsoft) or on drive A: (IBM).

Configs

• msdosenh: slightly enhanced PC-DOS 1.1 command.com marked as 1.25A/1.17A
  • VER command
  • Addition of command.com search on A: after default drive search failed
• msdosorg: MS-DOS 1.25 command 1.17 vanilla command.com rebuilt
  • Addition of FCB Drive 0'ing from Zenith ZDOS before opening command.com
• pcdosenh: slightly enhanced PC-DOS 1.1 command.com marked as 1.10A
  • VER command
  • CLS command (BIOS)
  • Check for command in A: if not found
  • External MEM command
• pcdosorg: PC-DOS 1.1 command.com rebuilt
  • Checked against PC-DOS 1.1 distribution
  • https://github.com/microsoft/MS-DOS/blob/master/v1.25/bin/COMMAND.COM
  • SHA256(pcdosorg.com)= 84034261608f7b9d38a6b81b6896bb5cc45dc5ebfae0b0927081f86f354aa571

TODO: Other little changes

• pcdosenh:
  • Search path on A: if default drive seach fails
  • Investigate further other MS-DOS 1.25 variations
• bios and dos
• format.com has been patch to format disk that can work with all emulators including qemu. the signature 55AA is added at the end of the 512 byte boot sector.

Build/Test Dependencies

Linux/macOS

• pce emulator (test)
• emu2 (for build from macosx/linux)
• pc-dos 1.1 floppy image (for test if available)
• exe2bin.exe from https://github.com/microsoft/MS-DOS/blob/master/v1.25/bin/EXE2BIN.EXE
• ibm link.exe 1.10 from https://github.com/microsoft/MS-DOS/blob/master/v1.25/bin/LINK.EXE
• microsoft masm.exe 1.10 (patched, cf below) https://github.com/microsoft/MS-DOS/blob/master/v2.0/bin/MASM.EXE

DOS

• exe2bin.exe from https://github.com/microsoft/MS-DOS/blob/master/v1.25/bin/EXE2BIN.EXE
• ibm link.exe from https://github.com/microsoft/MS-DOS/blob/master/v1.25/bin/LINK.EXE
• microsoft masm.exe 1.10 (patched, cf below) https://github.com/microsoft/MS-DOS/blob/master/v2.0/bin/MASM.EXE

Build

• Use Makefile on Linux, macOS
• Use Makefile.dos on Win32, DOS (builds only PC-DOS configs)

MASM Fix

The 1.10 version of MASM part of the MS-DOS 2.00 distribution hangs on emulators, including dosbox and emu2.

There is a detailed explanation at: https://slions.net/threads/debugging-ibm-macro-assembler-version-1-00.33/

The execution of the Faulty code looks like this:

11AD:0000 B83B0F           MOV     AX,0F3B
11AD:0003 8ED8             MOV     DS,AX
11AD:0005 8C062400         MOV     [0024],ES
11AD:0009 FA               CLI
11AD:000A 8ED0             MOV     SS,AX
11AD:000C 268B1E0200       MOV     BX,ES:[0002]
11AD:0011 2BD8             SUB     BX,AX
11AD:0013 81FB0010         CMP     BX,1000
11AD:0017 7E03             JLE     001C

The fix is just about changing the JLE into JBE (7E03 -> 7603)

11AD:0000 B83B0F           MOV     AX,0F3B
11AD:0003 8ED8             MOV     DS,AX
11AD:0005 8C062400         MOV     [0024],ES
11AD:0009 FA               CLI
11AD:000A 8ED0             MOV     SS,AX
11AD:000C 268B1E0200       MOV     BX,ES:[0002]
11AD:0011 2BD8             SUB     BX,AX
11AD:0013 81FB0010         CMP     BX,1000
11AD:0017 7603             JBE     001C

The masm binary checked in git is fixed.