Assembling functions from gas api

Assembling functions from gas api

[Philip Herron]

Hey all,

I am trying to assemble this function:

.code32
        .section .data
format_string:   .string "val = <%i>\n"
        .section .text
        .globl test
test:
        push $5
        lea format_string, %edi
        push %edi
        call printf
        pop %eax
        pop %eax
        ret

When i compile it with:

$ gcc -m32 -c t2.c -o t2.o
$ gcc -m32 -c t1.s -o t1.o
$ gcc -m32 -o test t1.o t2.o

The t2.c is simply calling test ().

This works 100% and the t1.o file is:608 bytes in size. I've been
following the gas api o assemble an input file and come up with this
so far:

int main (int argc, char **argv)
{
    char * out_file_name = "./ninja";
    need_pass_2 = 0;

    bfd_init ();
    symbol_begin ();
    frag_init ();
    subsegs_begin ();
    read_begin ();
    input_scrub_begin ();
    expr_begin ();

    output_file_create (out_file_name);
    dot_symbol_init ();

    need_pass_2 = 0;
    listing = 1;

    text_section = subseg_new (TEXT_SECTION_NAME, 0);
    data_section = subseg_new (DATA_SECTION_NAME, 0);
    bss_section = subseg_new (BSS_SECTION_NAME, 0);

    flagword applicable = bfd_applicable_section_flags (stdoutput);
    bfd_set_section_flags (stdoutput, text_section,
                           applicable & (SEC_ALLOC | SEC_LOAD | SEC_RELOC
                                         | SEC_CODE | SEC_READONLY));
    bfd_set_section_flags (stdoutput, data_section,
                           applicable & (SEC_ALLOC | SEC_LOAD | SEC_RELOC
                                         | SEC_DATA));
    bfd_set_section_flags (stdoutput, bss_section, applicable & SEC_ALLOC);
    seg_info (bss_section)->bss = 1;
    subseg_new (BFD_ABS_SECTION_NAME, 0);
    subseg_new (BFD_UND_SECTION_NAME, 0);
    reg_section = subseg_new ("*GAS `reg' section*", 0);
    expr_section = subseg_new ("*GAS `expr' section*", 0);
    subseg_set (text_section, 0);

    md_begin ();
    elf_begin ();
    read_a_source_file ("./t1.s");
    cond_finish_check (-1);

    subsegs_finish ();
    write_object_file ();
    input_scrub_end ();

    return 0;
}

And the resulting file i called ninja has stuff in it but i don't
think its correct. I realise more that i am playing with stuff i don't
understand at all i was expecting this to output the exact same output
as when i invoke the assembler but it doesn't. It outputs something
very similar but not the same i think its missing the ELF stuff at the
start of the output must be missing something. Maybe you have some
help on how i can debug this.

Thanks and sorry for the very vague question because i am messing with
the internals of something i don't understand that well quite yet.

--Phil

Re: Assembling functions from gas api

[Richard Sandiford]

Philip Herron <redbrain@gcc.gnu.org> writes:
> Hey all,
>
> I am trying to assemble this function:
>
> .code32
>         .section .data
> format_string:   .string "val = <%i>\n"
>         .section .text
>         .globl test
> test:
>         push $5
>         lea format_string, %edi
>         push %edi
>         call printf
>         pop %eax
>         pop %eax
>         ret
>
> When i compile it with:
>
> $ gcc -m32 -c t2.c -o t2.o
> $ gcc -m32 -c t1.s -o t1.o
> $ gcc -m32 -o test t1.o t2.o
>
> The t2.c is simply calling test ().
>
> This works 100% and the t1.o file is:608 bytes in size. I've been
> following the gas api o assemble an input file and come up with this
> so far:
>
> int main (int argc, char **argv)
> {
>     char * out_file_name = "./ninja";
>     need_pass_2 = 0;
>
>     bfd_init ();
>     symbol_begin ();
>     frag_init ();
>     subsegs_begin ();
>     read_begin ();
>     input_scrub_begin ();
>     expr_begin ();
>
>     output_file_create (out_file_name);
>     dot_symbol_init ();
>
>     need_pass_2 = 0;
>     listing = 1;
>
>     text_section = subseg_new (TEXT_SECTION_NAME, 0);
>     data_section = subseg_new (DATA_SECTION_NAME, 0);
>     bss_section = subseg_new (BSS_SECTION_NAME, 0);
>
>     flagword applicable = bfd_applicable_section_flags (stdoutput);
>     bfd_set_section_flags (stdoutput, text_section,
>                            applicable & (SEC_ALLOC | SEC_LOAD | SEC_RELOC
>                                          | SEC_CODE | SEC_READONLY));
>     bfd_set_section_flags (stdoutput, data_section,
>                            applicable & (SEC_ALLOC | SEC_LOAD | SEC_RELOC
>                                          | SEC_DATA));
>     bfd_set_section_flags (stdoutput, bss_section, applicable & SEC_ALLOC);
>     seg_info (bss_section)->bss = 1;
>     subseg_new (BFD_ABS_SECTION_NAME, 0);
>     subseg_new (BFD_UND_SECTION_NAME, 0);
>     reg_section = subseg_new ("*GAS `reg' section*", 0);
>     expr_section = subseg_new ("*GAS `expr' section*", 0);
>     subseg_set (text_section, 0);
>
>     md_begin ();
>     elf_begin ();
>     read_a_source_file ("./t1.s");
>     cond_finish_check (-1);
>
>     subsegs_finish ();
>     write_object_file ();
>     input_scrub_end ();
>
>     return 0;
> }
>
> And the resulting file i called ninja has stuff in it but i don't
> think its correct. I realise more that i am playing with stuff i don't
> understand at all i was expecting this to output the exact same output
> as when i invoke the assembler but it doesn't. It outputs something
> very similar but not the same i think its missing the ELF stuff at the
> start of the output must be missing something. Maybe you have some
> help on how i can debug this.

You also need:

  output_file_close (out_file_name);

or:

  keep_it = 1;
  ...
  close_output_file ();

This is normally run as an (x)atexit hook.

Thanks,
Richard