boot/boot1.s

%include "defines.s"

[org BOOT1_LOADPOINT]
[bits 16]

%macro copy_stack_var_to_globals 2
  mov %1, [bp - %2]
  mov [GLOBALS + %2], %1
%endmacro

; boot0 loads only our first sector into memory. We must load the rest.
self_load:
  ; Now that we're not doing instruction byte golf like we were in boot0, we can afford to move
  ; the various boot0 stack variables to the globals section.
  copy_stack_var_to_globals ax, BOOT_DRIVE
  copy_stack_var_to_globals ax, SECTORS_PER_TRACK
  copy_stack_var_to_globals ax, N_HEADS
  copy_stack_var_to_globals ax, GPT_ENTRIES_START_LBA
  copy_stack_var_to_globals ax, GPT_N_ENTRIES_16
  copy_stack_var_to_globals ax, GPT_SECTOR_STRIDE
  copy_stack_var_to_globals ax, GPT_BYTE_STRIDE
  copy_stack_var_to_globals ax, GPT_ENTRIES_PER_SECTOR
  copy_stack_var_to_globals ax, GPT_CURRENT_ENTRY_IDX
  copy_stack_var_to_globals ax, GPT_SECTOR_ENTRY_IDX
  copy_stack_var_to_globals ax, GPT_SECTORS_LOADED
  copy_stack_var_to_globals ax, GPT_CURRENT_LBA
  copy_stack_var_to_globals ax, BOOT1_GPT_ENTRY_ADDR

  ; Reset the stack, now we've got everything we need from it.
  mov sp, bp

  mov si, [GLOBALS + BOOT1_GPT_ENTRY_ADDR]
  mov eax, [si + 0x20] ; Partition / boot1 start LBA lower
  mov ebx, [si + 0x24] ; Partition / boot1 start LBA upper
  mov ecx, [si + 0x28] ; Partition end LBA lower
  mov edx, [si + 0x32] ; Partition LBA upper

  ; Panic if the partition / boot1 starting LBA overflows 16 bits.
  or ebx, ebx
  jnz panic
  ror eax, 16
  or ax, ax
  jnz panic
  ror eax, 16

  ; Calculate the boot1 end LBA and panic if it overflows 16 bits.
  ; n.b. ebx is zero before this so both bx and ebx can be used as the boot1 end LBA.
  mov bx, ax
  add bx, BOOT1_TOTAL_SECTORS
  jc panic

  ; Panic if the boot1 end LBA is after the partition end LBA.
  ; If the upper 32 bits of the partition end LBA are nonzero, then it must be greater than our
  ; 16-bit boot1 end LBA.
  or edx, edx
  jnz .end_lba_ok
  ; Compare the boot1 end LBA to the lower 32 bits of the partition end LBA.
  cmp ebx, ecx
  ja panic

.end_lba_ok:

  ; The first sector has already been loaded (we're running it right now!) so increment the
  ; current LBA.
  inc ax
  push ax                        ; Current LBA
  push bx                        ; boot1 end LBA
  mov ebx, BOOT1_LOADPOINT + 512 ; Current sector load address

.self_load_loop:
  mov ax, [bp - 0x02]      ; Load current LBA
  cmp word [bp - 0x04], ax ; Compare to boot1 end LBA
  jb .self_load_done

  mov ecx, ebx
  call read_sector
  jc panic
  
  add ebx, 512
  inc word [bp - 0x02]
  jmp .self_load_loop

.self_load_done:

  mov ebx, cafebabe
  call addr32_to_addr16
  mov eax, es:[ebx]
  hlt

  call test_a20
  add al, 0x30
  mov byte fs:[0x0000], al
  hlt

pr_womble:
  ; Reveal the true nature of jen
  xor bx, bx
.loop:
  cmp bx, WOMBLE_LEN
  jae .done
  lea si, [womble + bx]
  mov dh, [si]
  mov di, bx
  shl di, 1
  mov byte fs:[di], dh
  mov byte fs:[di + 0x01], 0x1f
  inc bx
  jmp .loop
.done:
  hlt


; Converts a 32-bit address to a 16-bit sector and offset.
; Arguments:
; - ebx: 32-bit address
; Return:
; - es: 16-bit address segment (unchanged on failure)
; - ebx: 16-bit address offset
; - cf: unset on success, set on failure
; Clobber: none
addr32_to_addr16:
  push bp
  mov bp, sp
  push es
  push eax

  mov eax, ebx
  ; Divide addr by 16 and saturate to 16 bits to get the segment.
  shr eax, 4
  ror eax, 16
  or ax, ax
  jz .segment_ok
  mov eax, 0xffff0000
.segment_ok:
  ror eax, 16
  mov es, ax

  ; Calculate offset = addr - (16 * segment), failing if the offset doesn't fit in 16 bits.
  shl eax, 4
  sub ebx, eax
  ror ebx, 16
  or bx, bx
  jnz .fail
  ror ebx, 16
  
  pop eax
  add sp, 2 ; Discard the original es from the stack
  pop bp
  clc
  ret

.fail:
  pop eax
  pop es
  pop bp
  stc
  ret
  

; Reads a single sector at the given LBA into memory.
; Arguments:
; - ax: start LBA
; - ecx: address to read sector to
; Return:
; - cf: unset on success, set on failure
; Clobber: eax, ecx, edx
read_sector:
  ; sector - 1 = LBA  % sectors_per_track
  ; temp       = LBA  / sectors_per_track
  ; head       = temp % n_heads
  ; cylinder   = temp / n_heads

  push bp
  mov bp, sp
  push es
  push ebx
  
  mov ebx, ecx
  call addr32_to_addr16
  jc .return
  
  ; Calculate sector and temp
  xor dx, dx
  ; Divide by sectors per track. dx = mod (sector - 1), ax = div (temp)
  div word [GLOBALS + SECTORS_PER_TRACK]
  ; Put the sector into cx (the bios call will use cl)
  mov cx, dx
  inc cx
  
  ; Calculate head and cylinder
  xor dx, dx
  ; Divide by number of heads. dx = mod (head), ax = div (cylinder)
  div word [GLOBALS + N_HEADS]
  mov dh, dl
  mov ch, al

  mov dl, byte [GLOBALS + BOOT_DRIVE]
  mov ah, 0x02
  mov al, 1
  ; Read sector
  int 0x13

.return:
  pop ebx
  pop es
  pop bp
  ret


%if ($ - $$) > 512
%error "boot1 self-loader exceeded sector size"
%endif


test_a20:
  push gs
  
  ; Restore the boot sector identifier in case it was overwritten by anything.
  mov word [0x7dfe], 0xaa55

  mov ax, 0xffff
  mov gs, ax
  xor ax, ax

  ; If the word at 0x107dfe (1 MiB after the boot sector identifier) is different to the boot
  ; sector identifier, than A20 must be enabled.
  cmp word gs:[0x7e0e], 0xaa55
  setne al
  jne .return

  ; Even if A20 was enabled, the two words may have been equal by chance, so we temporarily swap
  ; the boot sector identifier bytes and test again.
  ror word [0x7dfe], 8
  cmp word gs:[0x7e0e], 0x55aa
  setne al
  ror word [0x7dfe], 8
  jmp .return

.return:
  pop gs
  ret


panic:
  mov ax, 0x0003
  int 0x10
  mov word fs:[0x0000], 0x4f21
  hlt


womble     db "jen is a womble!"
WOMBLE_LEN equ $ - womble

  times 409600 db 0
cafebabe:
  dd 0xfeedface

BOOT1_TOTAL_LEN     equ $ - $$
BOOT1_TOTAL_SECTORS equ (BOOT1_TOTAL_LEN + 511) / 512

%if (BOOT1_LOADPOINT + BOOT1_TOTAL_LEN) > EBDA_START
%error "boot1 too large to be loaded"
%endif