Where do I find the assembly that creates a static variable in the .data section of my C program?

Question

First time poster. 2nd year CS student.

I am exploring the creation of static variables in the .data section of the Virtual Address Space in the context of a C sourc

Answer 1

Just to add that on bare-metal hardware (ie. microcontrollers) (where you don't have the benefits of binary loaders), you will see the code that zeroes the .bbs and copies .data section from RO FLASH into RAM.

That code will be something similar to http://repo.or.cz/w/cbaos.git/blob/HEAD:/kernel/init.c#l23 .

Answer 2

The initializers for staticVar is stored in the .data section of the executable. Using objdump (e.g. How can I examine contents of a data section of an ELF file on Linux?) should reveal something like this for your file:

./test:     file format elf64-x86-64

Contents of section .data:
 00d2c0 00000000 00000000 00000000 00000000  ................
 00d2d0 00000000 00000000 00000000 00000000  ................
 00d2e0 01000000 02000000 03000000 04000000  ................
 00d2f0 05000000 06000000 07000000 08000000  ................
 00d300 09000000 ffffffff 00000000 00000000  ................
 00d310 00000000 00000000 00000000 00000000  ................

That content from the executable is directly mapped into the address space of your process, so there is no need for any code to create the data. Codes that operate on staticVar will refer to the content directly using memory pointers; e.g. for the loop you posted, gcc -S gave me this:

  18                .L5:
  19 0013 90            nop
  20                .L2:
  21 0014 4863C3        movslq  %ebx, %rax
  22 0017 8B148500      movl    staticVar.1707(,%rax,4), %edx
  22      000000
  23 001e 8B45F4        movl    -12(%rbp), %eax
  24 0021 01D0          addl    %edx, %eax
  25 0023 8945F4        movl    %eax, -12(%rbp)
  26 0026 83C301        addl    $1, %ebx
  27 0029 83FB0A        cmpl    $10, %ebx
  28 002c 75E5          jne .L5

Lifetime of this static array would be the lifetime of your process, similar to a global variable. In any case, there is no code that builds it. It's just some data in memory.

P/S: You may need to add volatile to sum like such: volatile int sum = 0; Otherwise gcc would probably optimize it away since the resulting value of sum is never used.

Answer 3

My executable format is limited to old formats that aren't used anymore, but I'm pretty sure the code you're looking for does not exist in the ELF executable itself.

Executable files define sections that are copied/mapped to your process's memory verbatim. Just like your executable doesn't include code to populate the instruction stream of the functions it defines, it does not include code to populate static non-executable data. To that end, remember that there isn't any fundamental difference between data symbols and executable symbols: as far as storage is concerned, they're both data.

Some languages, like C++, will allow you to use dynamic initializers. Dynamic initializers are run before the entry point of your executable and populate data symbols with information that couldn't be inferred at compile-time. In that case, yes, there will be code for it. However, statically-initialized symbols don't need that, they can be copied or mapped straight to your process's address space.

Look at the data for staticVar closer. Forget the instructions for a bit; what happens if you put all the bytes next to one another?

01 00 00 00
02 00 00 00
03 00 00 00
04 00 00 00
05 00 00 00
06 00 00 00
07 00 00 00
08 00 00 00
09 00 00 00
ff ff ff ff

This is the hexadecimal representation of the sequence of little-endian integers {1, 2, 3, 4, 5, 6, 7, 8, 9, -1}, laid out in group of 4 bytes to make it easier to spot.

Answer 4

As zneak answered in the object file there is no code initializing the staticVar variable. The actual bytes of the .data section are directly in the test.o file. If you define a const variable gcc will put it into .rodata section by default.

The disassembly you obtained most probably by objdump --disassemble-all file.o. Maybe the disassembly of data confused you into thinking that it is a real code. For normal use I would recommend objdump --disassemble file.o which disassembles just sections containing actual machine code.

You can get detailed relevant information by running: objdump -xdst test.o

Shortly about the life of staticVar

1. gcc puts the variable (including its value) to the .data section as it is a static initialized variable. The size of the section is just for the variable, the address is not determined yet. See objdump -xdst test.o.
2. Linker determines the address of the .data section (and others) as the object files are known. In the .data section other variables from other object files appears. See objdump -xdst test.
3. When the binary test is executed the contained segments are directly mapped (mmap()) to the address space of the process so the value of staticVar is directly read from the test binary (possibly when needed). Then the dynamic linked ld-linux maps shared libraries like libc. See cat /proc/$PID/maps or pmap $PID when the process with $PID is loaded in memory.

There could be a variable initialization code in C

Just try to change the staticVar to a local non-static variable (i.e. stored on the stack by default):

   int staticVar[10] = {1,2,3,4,5,6,7,8,9,-1};

By default gcc will generate an initialization code directly in the main() function. Just see objdump -xdst test.o. This is because the stack variables are allocated (so their addresses are determined) at run-time.