os_debug_dump_rax clobbers upper 32 bits of RAX

[macpacheco]

	mov   rax,0x1234567890ABCDEF
	call  os_debug_dump_rax
	call  os_debug_dump_rax

Dumps correctly first time, then dumps 0000000090ABCDEF
Here's the culprit:
mov rax,0x1234567890ABCDEF
rol eax, 8
rol eax, 8
rol eax, 16
; now rax contains: 0000000090ABCDEF
Need to add a push RAX, pop RAX on os_debug_dump_rax to fix. However the clever way RET was reused for RAX, RAX, AX, AL dump, the code needs to be refactored.

[IanSeyler]

Thank you for reporting this! I’ll look into the fix. Anytime you use an opcode against the low 32-bit of a 64-bit register it will clear the upper 32-bits.

At least it’s handy for clearing a 64-bit register:

xor rax, rax
xor eax, eax

Both do the exact same but the xor against eax is a slightly shorter instruction.

[macpacheco]

Never rolling EAX (use RAX instead) fixes it:

; -----------------------------------------------------------------------------
; os_debug_dump_(rax|eax|ax|al) -- Dump content of RAX, EAX, AX, or AL
;  IN:	RAX = content to dump
; OUT:	Nothing, all registers preserved
os_debug_dump_rax:
	rol rax, 8
	call os_debug_dump_al
	rol rax, 8
	call os_debug_dump_al
	rol rax, 8
	call os_debug_dump_al
	rol rax, 8
	call os_debug_dump_al
	rol rax, 32
os_debug_dump_eax:
	rol rax, 40			; use RAX instead to avoid clobbering RAX upper half
	call os_debug_dump_al
	rol rax, 8
	call os_debug_dump_al
	rol rax, 16
os_debug_dump_ax:
	rol ax, 8
	call os_debug_dump_al
	rol ax, 8
os_debug_dump_al:
	push rax
	push ax				; Save AX for the low nibble
	shr al, 4			; Shift the high 4 bits into the low 4, high bits cleared
	or al, '0'			; Add "0"
	cmp al, '9'+1			; Digit?
	jl os_debug_dump_al_h		; Yes, store it
	add al, 7			; Add offset for character "A"
os_debug_dump_al_h:
	mov [tchar+0], al		; Store first character
	pop ax				; Restore AX
	and al, 0x0F			; Keep only the low 4 bits
	or al, '0'			; Add "0"
	cmp al, '9'+1			; Digit?
	jl os_debug_dump_al_l		; Yes, store it
	add al, 7			; Add offset for character "A"
os_debug_dump_al_l:
	mov [tchar+1], al		; Store second character
	pop rax
	push rsi
	push rcx
	mov rsi, tchar
	mov rcx, 2
	call b_output
	pop rcx
	pop rsi
	ret
; -----------------------------------------------------------------------------

	mov   rax,0x1234567890ABCDEF
	call  os_debug_dump_rax
	call  os_debug_newline
	call  os_debug_dump_rax

Output:
1234567890ABCDEF
1234567890ABCDEF

[IanSeyler]

That’ll do it. I’ll update the code with a fix. I just want to verify that dump_eax still behaves the same.

[macpacheco]

    mov rax, 0x30B7BA143
call  os_debug_newline
call  os_debug_dump_rax
call  os_debug_space
call  os_debug_dump_eax
call  os_debug_space
call  os_debug_dump_rax

000000030B7BA143 0B7BA143 000000030B7BA143

As you can see, os_debug_dump_rax and os_debug_dump_eax keeps the registers untouched and dump_eax properly dumps only the low 32 bits.

; -----------------------------------------------------------------------------
; os_debug_space -- Output a space
;  IN:	Nothing
; OUT:	Nothing, all registers preserved
os_debug_space:
	push rsi
	push rcx
	mov rsi, space
	mov rcx, 2
	call b_output
	pop rcx
	pop rsi
	ret
; -----------------------------------------------------------------------------

src/BareMetal/src/sysvar.asm:

; Strings
newline:                db 13, 10, 0
space:                  db ' ', 0

[IanSeyler]

Excellent! I've merged that change and added the os_debug_space call for the kernel.

Thank you for reporting this