Linux initializes all general purpose registers to zero. It's not documented AFAIK, but should be reliable - it has to init them to some value anyway to avoid leaking kernel state. So you can get away with:

    mov     al,1       ;write
    mov     edi,eax    ;handle=stdout
    mov     esi,msg    ;assumes load address below 4G
    mov     dl,msg.len
    syscall
    mov     al,60      ;assuming syscall succeeded, EAX was bytes written
    xor     edi,edi
    syscall

The load address stays constant unless there's some magic GNU extension header to enable ASLR. If we could get the code loaded below 64K, we could save another byte by using SI instead of ESI; however this doesn't work by default, you'd have to run 'echo 0 > /proc/sys/vm/mmap_min_addr' as root first.