How to link a C object file with a Assembly Language object file?

Question

I am having trouble linking 2 object files one of which was generated from an Assembly Language Source File and another that was generated from a C Source file.

C source code:

//main2.c
extern int strlength(char *);
int main(){
    char * test = "hello";
    int num = strlength(test);
    return num;
}

Assembly source code:

#strlength.s
.include "Linux32.s"

.section .text
.globl strlength
.type strlength, @function
strlength:
 pushl %ebp
 movl %esp, %ebp
 movl $0, %ecx
 movl 8(%ebp), %edx
read_next_byte:
 movb (%edx), %al
 cmpb $END_OF_FILE, %al
 jle end
 incl %edx
 incl %ecx
 jmp read_next_byte
end:
 movl %ecx, %eax
 popl %ebp
 ret

When I compile and run using 'gcc' like this:

gcc main2.c strlength.s -m32 -o test
./test
echo $?

I get 5 which is correct. However when I compile/assemble separately and then link with 'ld' like this:

as strlength.s --32 -o strlength.o
cc main2.c -m32 -o main2.o
ld -melf_i386 -e main main2.o strlength.o -o test
./test

I get a segmentation fault. What is causing this? Am I not following the C calling convention 100% correctly?

Answer 1

ld -melf_i386 -e main main2.o strlength.o -o test

Don't do that. Do this instead:

gcc -m32 main2.o strlength.o -o test

(You should probably not call your test exectuable test, as it may conflict with the /bin/test, standard on most UNIX systems.)

Explanation: UNIX binaries do not generally start executing at main. They start executing in a function called _start, which comes from crt1.o or similar ("C Runtime startup"). That file is part of libc, and it arranges for various initializations, required for the proper start up of your application.

Your program actually doesn't require anything from libc, which is why you were able to link it with ld.

However, consider what happens after your main returns. Normally, the code in crt1.o will execute (equivalent of) exit(main(argc, argv));. Since you linked without crt1.o, there is nobody to do that final exit for you, so the code returns to ... undefined location and promptly crashes.

Answer 2

You also need to link the crt1.o (might have a different name, contains the necessary code up until main can be called) and necessary libraries. GCC also usually needs to link to libgcc.so which contains necessary helper functions (for example, when doing 64 bit calculations on a 32 bit system) plus other system libraries. For example, on my Mac, it also needs to link to libSystem which also contain the usual C functions like printf. On Linux, that's usually libc.

Note that your program cannot directly start with main (as you're trying to do with with ld .. -e main), the entry point needs to set up a few things prior to calling the C function main. That's what the previously mentioned crt1.o is doing. I guess the segmentation fault is a result of this missing setup.

To see what GCC is exactly doing on your system, call:

gcc main2.c strlength.s -m32 -o test -v