9
\$\begingroup\$

I just started with deep learning of x86 architecture and assembly so i decided to go from bare metal and write my own bootloader. To add some "functionality" i decided to make quine (from binary perspective) and i have some questions:

  1. Could this be considered as valid quine ?

  2. Is mechanic and design of my code acceptable?

  3. Is there some asm patterns that generally should be followed (i come from Java background where patterns are everywhere)

  4. How to properly format asm code ?

My code :

start:
mov si, 7F00h   ;; set stack pointer after our bootloader 
mov ax, 0h      
mov ds, ax      ;; set DS to 0;
mov di, 7C00h   ;; set Data pointer to memory location where is our bootloader loaded 

.printMemoryValue:
mov al, 0          ;;  Using int15 to simulate pause for real time output, ax,ah,cx,dx dictate pause length 
mov ah, 86h         
mov cx, 0006H
mov dx, 8480H
int 15h
mov ah, 0Eh     ;; Ah to 0eh setting for teletype output ( int 10h)
mov dx, [ds:di]     ;; moving content of first memory location in dx (7c00h)
push dx             ;; save dx in our stack
xor  cx, cx        ;; for (int i = 0; ;; for loop used to output 16 bit string (0 & 1)

.loopstart:
cmp  cx, 00010h     ;; i < 16;
je  .loopend    ;; break if i >= 16     
pop dx              
rol dx, 1           ;; rotating left 1 bit so we can extract MSF bit with our mask. because we write on screen from right to left
push dx             ;; save our curent dx to stack
and dx, 0000000000000001   ;; mask   
add dx, 30h         ;; adding (30h) for ASCII ( 0 or 1)
push cx             ;; saving our counter in stack because cx is volotile register      
mov al, dl          ;; moving low 8 bits from dx (dl) to our teletype output register Al for calling int 10h
int 10h         ;; int 10h ( al = character to output if ah = 0eH, teletype )
pop cx
add cx, 1       ;; i++
jmp .loopstart  

.loopend:
mov al, 000Ah   ;; new Row
int 10h
mov al, 000Dh       ;; carriage return
int 10h
add di, 2h          ;; adding 2 to our di pointer ( because we are in 16bit mode)
cmp di, 7E00h       ;; are we at the end of our bootloader (7c00h + 200h )?
je .hlt             ;; if yes halt
jmp .printMemoryValue   ;; print next memory location ( di is increased by 2 )
.hlt:
hlt             ;; ende   

times 510-($-$$) db 0 
dw 0xAA55           ; => 0x55 0xAA (little endian byte order)

U can try code with mountable image available here

Any suggestions are welcome.

\$\endgroup\$

3 Answers 3

6
\$\begingroup\$

First correct some bugs.

start:
 mov si, 7F00h   ;; set stack pointer after our bootloader 
 mov ax, 0h      
 mov ds, ax      ;; set DS to 0;
 mov di, 7C00h   ;; set Data pointer to memory location where is our bootloader loaded 

The stackpointer on x86 is held in the SP register. You've loaded SI, a general purpose register that your program doesn't use at all hereafter.

Your bootloader program sits in memory from 7C00h to 7DFFh. If you put your stack behind the programcode and start with a stackpointer at 7F00h, you will have 256 bytes of stack (from 7E00h to 7EFFh) before you run into the program itself. If the forementioned error were corrected, this is exactly what would happen because of a second error further down in the program.

Each word you read from the program memory, you push it on the stack, but you leave it there and so the stack fills up to the point that you're pushing on top of the program memory! Either pop dx to balance the stack on each iteration of the program (write it just above .loopend:), or much better don't use the stack this way since there are enough registers free to store that particular value.

Given that your program is a bootloader, there are no segment registers that you can trust to have a defined value! If you're going to setup a stack, you'll need to also initialize the SS segment register. It's important to do that in the instruction directly above initializing the SP register.

start:
    xor  ax, ax      ; This is shorter/faster than "mov ax, 0h"
    mov  ds, ax
    mov  ss, ax
    mov  sp, 7F00h   ; set stack pointer after our bootloader 
    mov  di, 7C00h   ; set Data pointer to memory location where is our bootloader loaded 

Keep things together.

You've put the setup of the BIOS.Teletype function number outside of the loop. Although of course this is not wrong, it does diminish the readability of the program. I prefer to always have the mov ah, 0Eh just above of the int 10h.

Get rid of redundant things.

The mov al, 0 instruction doesn't serve any purpose for the BIOS delay function.

The ds: segment override prefix in mov dx, [ds:di] is redundant since DS is the segment that will be used by default. Writing it increases the code size by 1 byte.

The BIOS.Teletype function does not require nor clobber the CX register. You don't have to push cx ... pop cx. Your mention of CX being a volatile register does not apply when dealing with an API like BIOS. Just look at the registers in and registers out.
If you look up the Teletype function you will notice that it also requires you to setup the BH register with the DisplayPage and if on a graphics screen the BL register with the Color.

Use the best loop construct.

Currently your program uses a While constructs that requires 2 jumps on each iteration. Jumps are expensive in terms of execution time and so we always try to have as few as possible.

If you know that the loop is going to be run at least once then a Repeat-Until construct is better suited.

    xor  cx, cx
.loopstart:
    ...             ; Counting upward [0,15]
    inc  cx
    cmp  cx, 16
    jb   .loopstart

If you know that the body of this loop does not depend on the actual value in the loop counter then a slightly better version will be:

    mov  cx, 16
.loopstart:
    ...             ; Counting downward [16,1]
    dec  cx
    jnz  .loopstart

See the opportunities to write compacter code.

pop dx
rol dx, 1
push dx
and dx, 0000000000000001
add dx, 30h
push cx
mov al, dl
int 10h
pop cx

When you know that the BIOS.Teletype function expects the character in the AL register, you should strive to do these calculation straight on the accumulator which additionaly uses shorter encodings.

pop  ax
rol  ax, 1
push ax
and  al, 1
add  al, "0"
mov  ah, 0Eh
int  10h

We can go a bit further here. Instead of placing the data on the stack (push dx), we can hold it in the BP register (mov bp, [di]). Use the registers that are at your disposal!
With some more clever re-arranging the above snippet becomes:

rol  bp, 1       ; Produces a CF
mov  ax, 0E00h   ; Function number in AH, zeroing AL
adc  al, "0"     ; 0 + "0" + CF=0 ==> "0"
int  10h         ; 0 + "0" + CF=1 ==> "1"

The conditional jumps can jump 128 bytes backwards (x86-16).

add di, 2h
cmp di, 7E00h       ;; are we at the end of our bootloader (7c00h + 200h )?
je .hlt             ;; if yes halt
jmp .printMemoryValue   ;; print next memory location ( di is increased by 2 )
.hlt:
hlt             ;; ende   

Inverse the conditioncode and you'll no longer need the direct jump nor the extra label.

add  di, 2
cmp  di, 7E00h
jne  .printMemoryValue
hlt

For robustness the jne should become jb. Sometimes things go wrong and so it could be, that the expected value of 7E00h never occurs producing an infinite loop! That's why prudent programmers prefer testing for less/below and greater/above conditions.


This is the complete code with all of the above applied:

start:
    xor   ax, ax      ; This is shorter/faster than "mov ax, 0h"
    mov   ds, ax
    mov   ss, ax
    mov   sp, 7F00h   ; set stack pointer after our bootloader 
    mov   di, 7C00h   ; set Data pointer to where is our bootloader loaded 
.printMemoryValue:
    mov   cx, 0006H   ; CX:DX = 00068480h Pause for about 0.4 sec
    mov   dx, 8480H
    mov   ah, 86h     ; BIOS.Delay
    int   15h
    mov   bp, [di]    ; Moving content of memory location in BP
    mov   bx, 0007h   ; Display page 0 in BH, Attribute WhiteOnBlack in BL 
    mov   cx, 16
.loopstart:
    rol   bp, 1       ; Produces a CF
    mov   ax, 0E00h   ; Function number in AH, zeroing AL
    adc   al, "0"     ; 0 + "0" + CF=0 ==> "0"
    int   10h         ; 0 + "0" + CF=1 ==> "1"
    dec   cx
    jnz   .loopstart
    mov   ax, 0E0Dh   ; Newline is carriage return plus linefeed
    int   10h
    mov   ax, 0E0Ah
    int   10h
    add   di, 2
    cmp   di, 7E00h
    jb    .printMemoryValue
    hlt               ; ende

    times 510-($-$$) db 0
    dw    0AA55h      ; => 55h 0AAh (little endian byte order)

4.How to properly format asm code ?

Everybody has a personal style. Most people however like to use the nice tabular format that you see in my code examples.

What you should do is being consistent when it comes to number representations.

  • Choose between the hex prefix 0x or the hex suffix h, but try not to mix both in the same program. This is especially true in a short program.

  • Don't forget to write the affix for numbers that need it.

    and dx, 0000000000000001  --> and dx, 0000000000000001b
    
  • Write as many hex digits as the register can take.

    mov ax, 0h      --> mov ax, 0000h
    cmp cx, 00010h  --> cmp cx, 0010h
    add dx, 30h     --> add dx, 0030h
    mov al, 000Ah   --> mov al, 0Ah
    mov al, 000Dh   --> mov al, 0Dh
    add di, 2h      --> add di, 0002h
    
  • Don't express numbers that don't really need it in hexadecimal.

    mov ax, 0h      --> mov ax, 0
    cmp cx, 00010h  --> cmp cx, 16
    add di, 2h      --> add di, 2
    
  • Sometimes expressing a number as a character improves readability.

    add dl, 30h     --> add dl, "0"
    
\$\endgroup\$
6
\$\begingroup\$
  • Since cx is not used other than a loop counter, you may consider making a countdown loop. That would spare 2 instructions:

        move cx, 16
    .loopstart:
        pop dx
        ....
        dec cx
        jnz .loopstart
    
  • The MSB operations can also be shortened:

        xor al,al    # Clear al
        sal dx,1     # MSB lands in CF
        adc al,30h   # Add with carry. al becomes 30h + MSB
    
  • Yet another jump could be spared by jne printMemoryValue.

\$\endgroup\$
2
\$\begingroup\$

As far as code formatting and as pointed out by @Sep Roland there lacks a definitive standard other than tabulating each instruction on a single line. I've gone a step further and you may have noticed how I nest my comments and group things that logically go together.

It may seem like a lot of work, but amply describe what your program is doing. I still struggle with coming up with meaningful statements, but you'd be surprised without detailed description how hard it is deciphering your code a year or two later. I've found this method too, solidifies concepts and makes debugging much easier.

; Read E820 map into a temporary buffer just above boot sector @ 7E0:0

            mov     ax, BOOT_SEG + 32   ; So MSB of EAX is nullified
            mov     es, ax
            mov     ds, ax              ; So segment overrides are not required

            xor     di, di              ; ES:DI = Pointer to base address of map
            mov     bx, di              ; Initial continuation value
            mov    edx, 'PAMS'          ; Function signature
            push    edx

    ; Top of loop to read first or next map entry

ReadNext:   inc     byte [MAP_ENTR]     ; Bump number of map entries = 0 first iteration.
   .skip:   mov     cl, 48              ; Let function call know how big entry can be.
            mov     ax, 0xe820          ; System Service function number.
            int     SYS_SERVICE

    ; Assert the possible error and termination conditions

            jc      .done               ; CY = 1 can happen in all cases
            cmp     bl, 1               ; Is this the first entry
            jb      .done               ; If zero, no more entries
            ja      .J0                 ; Next code only needs happen on first iteration

        ; This need only happen on first iteration

            pop     edx
            sub     eax, edx            ; Does BIOS even support this function
            jz      .J0 - 3
            dec     byte [MAP_ENTR]     ; Bump value back to -1
            jmp     .done

            mov     [MAP_SIZE], cl      ; Save actual size of entry returned by function.

    .J0:    jcxz    .skip               ; Ignore any null length entries
            cmp     cl, 20
            jbe     .J1

            test    byte [di + 20], 1   ; Ignore ACPI entries
            jz      .skip

        ; Test 64 bit value representing length for zero

    .J1:    mov     eax, [di + 8]       ; Get low order DWORD of length
            or      eax, [di + 12]      ; Determine if QWORD value is zero
            jz      .skip

        ; Bump ES:DI pointer to next entry

            add     di, cx
            jmp     ReadNext

    .done:  or      di, di              ; Was a map even created
            jnz     .movemap

    ; Screen is completely blank now, so to indicate there was a problem with E820

            mov     ax, 0xb800          ; Point to video
            mov     es, ax
            mov     di, 0x7CE           ; Offset to vertical & horizontal center of screen

    ; This will display flashing "[ ]" in yellow with "E" between in high intensity white

            mov    eax, 0xf458e5b
            stosd
            inc     al
            inc     al                  ; AL = "]"
            stosw
            push    ss                  ; Define upper for calculating total sectors
            jmp     MoveBlock           ; Dont need to move map as it doesnt exist

; Move E820 entries immediately below bottom of stack frame.

.movemap:   mov     cx, di              ; Get copy of total bytes in E820 map
            shr     cx, 2               ; CX / 4 = Total DWORDS to move

    ; Only every 4th entry is segment aligned (16 bytes), so offset in DI needs to be
    ; calculated so last entry of map terminates at bottom of stack.

            mov     ax, ss              ; Get base of stack frame
            sub     bx, di
            and     bx, 15              ; BL = 0, 4, 8, 12
            jz      $ + 3

        ; Because BL <> 0 segment has to be skewed by one

            dec     ax                  ; Bump back one more segment

    ; Now offset can be saved and moved into DI

            mov     [MAP_ADDR], bx      ; Lower half of long pointer
            xchg    di, bx              ; Move offset into index
            shr     bx, 4               ; BL = Total # of 16 byte segments
            sub     ax, bx
            mov     [MAP_ADDR+2], ax    ; Upper half of long pointer
            mov     es, ax              ; ES:DI = Destination buffer
            push    ax                  ; Define upper for calculating total sectors

    ; Establish source pointer and then move CX DWORDS

            xor     si, si              ; DS:SI = Source
            rep     movsd
\$\endgroup\$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge that you have read and understand our privacy policy and code of conduct.

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