I know the posting is old, but I found a minor problem that might plague beginners. I believe helloworld.s should read something like this… ideally without the branches:

1 .data ; section declaration – variables only
2
3 msg:
4 .ascii “Hello, world!\n\0″
5 len = . – msg ; length of our dear string
6
7 .text ; section declaration – begin code
8
9 .globl _main
10 _main:
11
12 # write our string to stdout
13
14 li r0, 4 ; syscall number (sys_write)
15 li r3, 1 ; first argument: file descriptor (stdout)
16 ; second argument: pointer to message to write
17 lis r4, ha16(msg); load top 16 bits of &msg
18 addi r4, r4,lo16(msg) ; load bottom 16 bits
19 li r5, len ; third argument: message length
20 sc ; call kernel
21
22 ; exit with error code for success of sys_write
23 b _err
24 b _success
25
26 _exit: ; assume r3 has been preloaded
27 li r0, 1 ; syscall number (sys_exit)
28 sc ; call kernel
29
30 _success:
31 li r3, 0
32 b _exit ; exit(0)
33
34 _err:
35 li r3, 1
36 b _exit ; exit(1)
37

On OS X / Darwin (not sure if this applies to all PowerPC environments), “sc” calls don’t resume to the next immediate instruction unless there’s an error. By convention, the instruction immediately after a “sc” (srr + 0) is reserved for error handling. Upon success, the kernel will jump to (srr + 4), over the instruction immediately after the “sc”.

If you comment out line 23 above, On OS X, the exit code is 1, since it enters _exit right away and r3 already had a value set on line 15.

In your original example, r0 was still 4… so a duplicate “li r0, 1″ would have done the trick and given the intended exit code, as it wasn’t actually exiting cleanly… it was calling sys_write again.

p