0

I tested two successfully exiting minimal programs, first uses libc, second does not use libc.

First:

    segment .text
    global main
main:
    xor eax, eax
    ret

Build and check size:

yasm -f elf64 main.s; clang -o main main.o; stat ./main
...
Size: 8552
...

Second:

    segment .text
    global _start
_start:
    mov eax, 60
    xor edi, edi
    syscall

Build and check size:

yasm -f elf64 main.s; ld -o main main.o; stat ./main
...
Size: 704
---

Obviously program using main is larger in size. Why linkers does not strip main routine from binaries to optimise for size? I mean linkers that get object files after compilation of C program.

too honest for this site
  • 12,050
  • 4
  • 30
  • 52
Bulat M.
  • 680
  • 9
  • 25
  • What do you mean? I was asking about the possibility to "cut" code from main and paste it to _start, so that main will not be called. – Bulat M. Sep 18 '16 at 13:25
  • It was meant as a joke. Sry, for having missed the ";-)". – alk Sep 18 '16 at 13:30
  • I don't see any C code. Did you inspect the **contents**? Concluding from the size of a binary what happens internally is a really bad approach. There is no function/"routine" in your Assembler code! Just a label which generates a symbol. – too honest for this site Sep 18 '16 at 14:23
  • 1
    See also http://stackoverflow.com/questions/9827157/what-does-keep-mean-in-a-linker-script/9830185#9830185 why main is kept for the --gc-sections case – Marco van de Voort Sep 19 '16 at 14:51

1 Answers1

1

why linker does not strip main routine from executable?

Because linkers in general do not strip any routines from the object files they are given. They are simply not in that business.

Linkers do not operate on routines, they operate on sections. Some linkers can do limited removal of unreferenced sections, e.g. GNU BFD ld and Gold have --gc-sections flag, but in your example there are no unreferenced sections, so nothing would be GC'd anyway.

Now, the difference between your two programs is that the one with main links in C runtime startup (usually crt0.o), and sets up argc and argv arguments the way main is defined to expect them.

Your main doesn't care about this, but it would take a really smart linker to deduce that. Also, your main does care about calling sys_exit after returning, so it actually needs crt0.o for proper shutdown.

Is is that argc/argv setting-up process, that takes up few kilobytes?

If you are using GNU ld, you can ask the linker to tell you exactly what's using how much space with the linker --print-map argument.

On ELF platforms, you can also use nm main to see which symbols are taking up space in the final binary, and use ld -y <symbol> ... to see where these symbols are getting linked from.

Employed Russian
  • 199,314
  • 34
  • 295
  • 362