Relative Addressing errors - Mac 10.10

Question

I'm trying to learn how to write assembly code and I'm doing it with the help of http://gnu.mirrors.pair.com/savannah/savannah//pgubook/ProgrammingGroundUp-0-8.pdf. It's an excellent resource and I'm trying to write the code as 64bit for my Mac in Macho64 format.

I've run into some trouble with absolute and relative addressing.

This is my code:

    DEFAULT REL
;PURPOSE:   This program finds the maximum number of a set of data items
;

;VARIABLES: The registers have the following uses
;
;   rbx - Holds the index of the data item being examined
;   rdi - Largest data item found
;   rax - Current data item
;
;   The following memory locations are used:
;
;   data_items - contains the item data. A 0 is used to terminate the data
;

global _main

section .data
    data_items: dw  3,67,34,222,45,75,54,34,44,33,22,11,66,0
    ;These are the data items

section .text

_main:              
    mov rdi, 0          ;move 0 into index register
    mov rax, [data_items+rbx*4] ;load the first data byte
    mov rdi, rax        ;since this is the first item, eax is biggest

start_loop:         ;start loop
    cmp 0, rax          ;check to see if we've hit the end
    je loop_exit
    inc rdi
    mov rax, [data_items+rbx*4]
    cmp rdi, rax
    jle start_loop

    mov rdi,rax
    jmp start_loop    

loop_exit:
    mov rax, 0x2000001          ;1 is the exit() syscall
    syscall

and these are the error messages I get:

Samuels-MBP:Starting sam$ make
src/maximum.s:26: error: Mach-O 64-bit format does not support 32-bit absolute addresses
src/maximum.s:30: error: invalid combination of opcode and operands
src/maximum.s:33: error: Mach-O 64-bit format does not support 32-bit absolute addresses

So I was wondering if anyone can help me. I looked up Relative Addressing, but I can't find anything that explains in simple language what I am doing wrong.

I do know the cmp statement is wrong as well, but I think I can fix that myself.

Answer 1

Mach-O 64-bit does not support 32-bit absolute addressing because the image base is greater than 2^32.

Normally you should use RIP relative addressing for accessing a single memory element. In your case however you're accessing a static array (arrays allocated in the data section/bss section) and
as explained in the the section Addressing static arrays in 64 bit mode in Agner Fog's Optimizing Assembly manual.

It is not possible to access static arrays with RIP-relative addressing and an index register.

So when NASM processes your code

mov rax, [data_items+rbx*4]

it can't do RIP relative addressing so it tries to 32-bit absolute + index address which is not allow with Mach-O 64-bit which causes NASM to report the error.

Exampels 3.11b-3.11d In Agner's manual presents three ways to access static arrays. However, since 64-bit OSX does not allow 32bit absolute addressing (though it's possible in Linux) the first example 3.11b is not possible.

Example 3.11c uses the image base reference point __mh_execute_header. I have not looked into this but 3.11d is easy to understand. Use lea to load the RIP+offset into a register like this:

lea rsi, [rel data_items]

And then change your code using mov rax, [data_items+rbx*4] to

mov rax, [rsi+rbx*4]

Since you have delcared DEFAULT REL you should be able to ommit the rel in [rel data_items].