PRB: MASM Generates Unexpected Prefix Bytes 66h, 67h

• Microsoft Macro Assembler (MASM) for MS-DOS, versions 5.0, 5.1, 5.1a, 6.0, 6.0a, 6.0b
• Microsoft Macro Assembler for OS/2, versions 5.1, 5.1a, 6.0, 6.0a, 6.0b

SYMPTOMS

When the .386, .386p, .486, or .486p directive is used in MASM, the source code listing or the mixed source and assembly code listing in the CodeView debugger may indicate that code for 32-bit instructions has been generated incorrectly. For example, when either of the following instructions is assembled, the prefix bytes 66h and 67h may or may not appear before the instruction opcode:

      mov eax, [ebx]
      mov ax, [bx]

CAUSE

The 386 and 486 processors can operate with either 32-bit operations or 16- bit operations as the default mode. The descriptor for each memory segment controls the default mode for operations in the segment. In either mode, a prefix byte (66h or 67h) at the beginning of an instruction overrides the default mode for the current memory segment for a given instruction. In other words, when the processor encounters a prefix byte on an instruction in a 16-bit segment, the instruction executes as a 32-bit operation.

The condition described above occurs when the source code indicates that a code segment is a 16-bit segment or a 32-bit segment, and the segment loaded into memory is of the opposite type.

RESOLUTION

When MASM assembles a file that uses full segment directives, it treats all segments as 32-bit segments and all operations as 32-bit operations. To change this treatment, specify the USE16 modifier for each SEGMENT directive that describes a 16-bit segment. In MASM version 6.0 or later, the OPTION SEGMENT:USE16 directive specifies that all segments are 16-bit segments.

When MASM assembles a file that uses simplified segment declarations, it treats all segments as 32-bit segments when a .386, .386p, .486, or .486p processor directive precedes the .MODEL statement. If the processor directive follows the .MODEL statement, MASM treats all segments as 16-bit segments and all operations as 16-bit operations.

MORE INFORMATION

Note that to run code in 32-bit segments, the operating system with which the application runs support 32-bit code segments. Most PC-based operating systems, including MS-DOS version 5.0, Microsoft Windows version 3.1, and OS/2 versions 1.x do not support 32-bit code segments. Later versions of operating systems and MS-DOS extender programs may provide this support.

For more information on the differences between 16-bit and 32-bit segments, refer to chapter 16 of the "Intel 80386 Programmer's Reference Manual" (1986 edition).

The following code samples demonstrate this problem when CodeView for MS-DOS is used to debug these programs.

Sample Code

; Assembler options needed: /Zi

.386

DOSSEG                           ; Force DOS segment order
DGROUP GROUP _DATA, STACK        ; Stack and data in DGROUP

    DB 100h DUP (?)              ; 256-byte stack

_DATA SEGMENT WORD PUBLIC 'DATA' ; Data segment with string data msg DB "Hello, world.", 13, 10, "$" _DATA ENDS

_TEXT SEGMENT WORD PUBLIC 'CODE' ; Code segment ASSUME cs:_TEXT, ds:DGROUP, ss:DGROUP

start:

    mov ax, DGROUP               ; Initialize data segment
    mov ds, ax

    cli                          ; Turn off interrupts
    mov ss, ax                   ; Make SS and
    mov sp, OFFSET STACK         ;   SP relative to DGROUP
    sti                          ; SS now equals DS

    mov ah, 9h                   ; Request DOS Function 9
    mov dx, OFFSET msg           ; Load DX with offset of string
                                 ;   (segment already in DS)
    int 21h                      ; Display String to Standard Output

    mov ax, 4C00h                ; Exit functions with 0 in AL
    int 21h                      ; Terminate Process with Return Code

END start                        ; End with reference to first
                                 ;   executable statement (CS:IP)

Sample Code

; Assembler options needed: /Zi

.386

.MODEL small, c                  ; Small model (could be any model)

.STACK 100h                      ; 256-byte stack

.DATA                            ; Data segment

.CODE                            ; Code segment

    mov ax, DGROUP               ; Initialize data segment
    mov ds, ax

    cli                          ; Turn off interrupts
    mov ss, ax                   ; Make SS and
    mov sp, OFFSET STACK         ;   SP relative to DGROUP
    sti                          ; SS now equals DS

    mov ah, 9h                   ; Request DOS Function 9
    mov dx, OFFSET msg           ; Load DX with offset of string
                                 ;   (segment already in DS)
    int 21h                      ; Display String to Standard Output

    mov ax, 4C00h                ; Exit functions with 0 in AL
    int 21h                      ; Terminate Process with Return Code

END start                        ; End with reference to first
                                 ;   executable statement (CS:IP)