Most times whene I call EDIT on my XT-class machine with MS-DOS 5.0 installed on the hard disk, the menus of EDIT appear, but the window that should contain the file contents stays blank and the machine stops responding to any input.

The BIOS on the machine is made by PHOENIX, and its version string says 2.27.


Analysis of the problem

The cause is an incompatibility with many pieces in it. It is centered around what the bit 4 of 40:96 in the BIOS data area actually means. This bit is documented to mean that a "101/102 key keyboard" is connected to the computer. This is the case in my XT system. The keyboard I use has a switch (8088/80286) to choose the protocol, and in XT mode, it uses the XT keyboard protocol, but it still sends E0 prefixes like in AT mode when extended keys (like the separate cursor keys) are pressed.

The BIOS on this computer does not know about 101/102-key keyboards, and also does not use 40:96 at all, so it is zeroed out on boot. The KEYB command included with MS-DOS 5.0 overrides the BIOS implementation of the keyboard IRQ handler (INT 09) with a variant that does handle extended keys. If the MS-DOS 5.0 keyboard driver recognizes E0 or E1 scan codes (no matter on what machine), it automatically sets the bit indicating the presence of an "extended keyboard". This makes sense.

On the other hand, the keyboard abstraction layer in QBASIC (and EDIT on DOS 5 is implemented by running QBASIC with a special command line switch) attaches more meaning to that bit. If it sees this bit set, QBASIC implies that the keyboard BIOS is AT-like, and thus INT 16 supports the extended BIOS services 10h, 11h and 12h that IBM introduced in the AT. These services are needed to make best use of the 101/102-key keyboard, as they allow (e.g.) to differentiate between the numpad cursor keys with numlock off and the separate cursor keys.

You likely can see where this is heading: As the DOS keyboard driver replaces only INT 09, the hardware IRQ driver is now 101/102-key aware, and the awareness is indicated in the BIOS data area after the first extended key has been pressed. The BIOS API at INT 16 is still the PC-class service that only supports subfunctions 0 to 2. As QBASIC now tries to interface to the keyboard using the unsupported subfunctionss 10h and 11h, it reads garbage inputs and seems to not draw the file to the screen while some keys (possibly scrolling keys) are pressed.

A way to fix the problem

Actually, the QBASIC keyboard abstraction layer has an internal flag to disable AT keyboard BIOS support. It does not seem to be set anywhere in QBASIC though. The configuration flags for the driver are in QBASIC.EXE in the data segment, but as QBASIC is compressed by Microsoft's linker (/EXEPACK), the flag byte can not be patched. What can be patched is the instruction that checks the 101-key flag.

26 F6 06 96 04 10       test    es:keybd_flags_3, 10h ; 101 key keyboard?
74 16                   jz      NoATSupport
8B 1D                   mov     bx, [di+KbdState.Environment]
83 7F 18 00             cmp     [bx+KbdEnvironment.InhibitATSupport], 0
75 0E                   jnz     NoATSupport

The byte sequence 26 F6 06 96 04 10 has only one hit in QBASIC.EXE. Patch the last byte to 00 to make sure the zero flag is always set after this instruction.


An easier way to fix this problem is to execute a small program called "Keyfix.com" that was supplied, largely un-documented, with some OEM distributions of MSDOS 3.2 and 3.3. Syntax is:

keyfix /i

Just typing:


will check the keyboard in use and advise whether keyfix needs to be installed.

Hope this helps.

  • 4
    The only keyfix I could find in archives using google is a keyfix tool by compaq distributed with a pre-release of MS-DOS 3.1 for their Deskpro series computers. This keyfix utitlity uses 286 opcodes and seems to be intended to fix a different problem Do you have any pointers to the keyfix utility you are talking about? Not necesserily a link to a questionable abondonware site, but maybe you can remember which OEM(s) provided keyfix. – Michael Karcher Mar 23 at 14:30

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.