X86 Legacy boot loader error trapping

Question

The boot loader that I'm designing is just simply going to setup (A) Stack, (B) Segment registers (C) Load remainder of track for specified device. Conventionally, this could have been up to 4 floppy and/or fixed disk devices, but now, it may be an FDC, but most likely USB or fixed disk. As my boot sector does have an 85 byte BPB, that leaves me with 422 bytes to work with. This code is fairly weighty (105 bytes), but in return of its 8160 possibilities, it can display all the detail I need.

INT 0x13 can return a possible 16 errors, so I came up with this color/character scheme that will provide me with all the details and not comprise to much space in boot sector. As an example, we want an error code to look like this.

Bochs session

Error codes are passed in EAX as a combination of two characters with attributes

 mov eax, 0x07ce41cd
 call Err
 jz GoBackTo

The high order word is the center character and low order the outside two. Therefore 07 = Grey on Black, 0xCE is double line cross, 0x41 = Blue on Red and 0xCD = Double horizontal bars.

Prompting string can be constructed in any fashion required, but the only thing that must be is 0xFF where the error code will be inserted.

Prompt: db ' Error ['
 db -1 ; Position of error indicator
 db '] (A)bort (R)etry (I)gnore'
 db 0

Parameters

; ENTER: EAX = Two combinations of attribute & character.
; LEAVE: AL = 'N' or 'R'
; EBX, ECX, EDX modified
; FLAGS: ZF = Retry otherwise NZ to Ignore. 

Procedure

 Err: push ds
 push di
 push si
 push es
 push cs
 pop ds ; DS might be right
 push eax
 mov si, Prompt ; DS:SI now points to prompt
 mov ax, 0xb800
 mov es, ax ; Point to base of page 0
 ; Calcuate offset in page 0 where error prompting should begin.
 xor bh, bh ; BH = Page
 mov ah, GetCurPos
 int VIDEO
 xor ax, ax
 mov cx, ax
 mov cl, 160 ; Chars & Attribue / line
 mov al, dh
 and dx, 255
 shl dx, 1
 mov di, dx
 xor dl, dl ; DX needs to be null for IMUL to work
 imul cx ; Calculate offset
 add di, ax ; DI = offset in page 0
 ; Initial prompting including closing bracket will be in this attribute.
 mov ah, 14 ; BIOS color for bright yellow
 mov si, Prompt
 ; Write everything up and including opening bracket
 lodsb
 or al, al
 js $ + 5 
 stosw
 jmp $ - 6
 ; Write 3 character error indicate
 mov cx, ax ; Preserve attribute 
 pop eax
 stosd
 stosw
 ; Write closing bracket now
 mov ax, cx ; Retrieve previous attribute
 lodsb
 stosw
 ; Change attribute to bright white and display remainder of string
 ; terminated by NULL.
 mov ah, 15 ; Attribute 
 .L0: lodsb ; Read next character
 test al, 255
 jz .Done
 stosw ; Write char & attr to video memory.
 jmp .L0
; Wait for response from operator and convert to uppercase
.Done: xor ax, ax
 int KEYBOARD
 and al, 0x5f ; Converts to uppercase.
 cmp al, 'A'
 jnz $ + 4
 int ReBoot
 cmp al, 'R'
 jz .Exit ; Return with ZF set
 cmp al, 'I'
 jnz .Done
 inc al ; Clears ZF
.Exit: pop es
 pop si
 pop di
 pop ds
 ret

I haven't completely formulated how I want codes to look for disk errors, but essentially center character will reflect colors corresponding to code and outside will be blue for removable devices and bright blue for fixed assuming DL was either 00 or 0x80.

NOTE I do code a lot of hard conditionals and branches jmp $+?? to avoid a lot of unnecessary symbols in object files. Would be nice if NASM had something similar to MASM & FASM like @@:.

Answer 1

This code is fairly weighty (105 bytes)

Given all the things you want to do, and all the devices you want to interface from within this 512 bytes bootloader, you should take more steps to reduce the footprint of this error reporting routine.

---

xor dl, dl ; DX needs to be null for IMUL to work

This is a misconception. The following imul cx only depends on AX and CX for its operation. It will leave a result in DX:AX, but DX does not have to be initialized beforehand.

xor ax, ax
mov cx, ax
mov cl, 160 ; Chars & Attribue / line
mov al, dh
and dx, 255
shl dx, 1
mov di, dx
xor dl, dl ; DX needs to be null for IMUL to work
imul cx ; Calculate offset
add di, ax ; DI = offset in page

This calculation is overly complex and wastes a lot of bytes.

• Why zero the CX register and then assign CL? Use mov cx, 160
• Why zero the high byte of DX with an ugly and dx, 255 when there's a perfectly good mov dh, 0 available?
• Why use a word multiplication imul cx when both numbers are mere bytes?

This is a better version:

movzx di, dl ;DL is column
shl di, 1
mov al, 160
mul dh ;DH is row -> product in AX
add di, ax ;DI = offset in page

---

Prompting string can be constructed in any fashion required, but the only thing that must be is 0xFF where the error code will be inserted.

Because you used or al, al to test the exit condition, you denied yourself the chance to use ASCII's in the range [128,254] when constructing this prompting string. Just use the cmp al, 0xFF instruction, it's equally small.

; Write everything up and including opening bracket
lodsb
cmp al, 0xFF
je $ + 5 
stosw
jmp $ - 6

---

mov cx, ax ; Preserve attribute 
...
mov ax, cx ; Retrieve previous attribute

In stead of preserving, which uses 4 bytes, you could simply write mov ah, 14 which uses only 2 bytes.

---

and al, 0x5f ; Converts to uppercase.

This way of UCasing is dangerous! You neglect bit 7 and thus p.e. the user could select abort by typing ALT-193 or ALT-225. Similarly for ignore and retry. Best write and al, 0xDF or even and al, -33. A third possibility is and al, not 32

---

push eax
...
pop eax

In stead of preserving the dword EAX, which uses 4 bytes, you could content yourself with just push ax and pop ax because the high word is never in danger of modification in your code. This saves another 2 bytes.

---

; LEAVE: AL = 'N' or 'R'
; EBX, ECX, EDX modified
; FLAGS: ZF = Retry otherwise NZ to Ignore.

A small error here. On exit with Ignore the AL register will hold an uppercase J.
My main concern here is why you want both the AL register and the ZeroFlag express essentially the same thing? If you would stick to only return result in the ZeroFlag then you could save extra bytes and at the same time get rid of the BX, CX, and DX registers being modified which is a good thing.

 push ds
 push es
 pusha
 ...
.Exit: popa
 pop es
 pop ds
 ret

---

I haven't completely formulated how I want codes to look for disk errors, but essentially center character will reflect colors corresponding to code and outside will be blue for removable devices and bright blue for fixed assuming DL was either 00 or 0x80.

What you're saying here is that it is all about the colors that make it to the screen. Then why don't you put the ASCII's 0xCE for double line cross and 0xCD for double horizontal bars directly in the errortext? This will avoid using EAX as an argument (AX remains of course) and will save a lot of bytes on many occasions throughout the rest of the program.

mov ax, 0x0741 ;Only colors conveyed
call Err
jz GoBackTo

• Fifoernik: "Best write and al, 0xDFh or even and al, -32" I tried to change these small errors through an edit but it got rejected! I suggested: "Best write and al, 0xDF or even and al, -33. A third possibility is and al, not 32"
• Sep Roland: @Fifoernik You're right! -32 was a mistake. I copied everything you tried to edit. Thanks.