How does an assembler work?（simple one）

From stackoverflow

Q：

I am looking for a brief description of the use of an assembler in producing machine code.

So I know that assembly is a 1:1 translation of machine code. But I am getting confused about object code and linkers and how they place into it.

A：

Both an assembler and a compiler translate source files into object files.

Object files are effectively an intermediate step before the final executable output (generated by the linker).

The linker takes the specified object files and libraries (which are packages of object files) and resolves relocation (or 'fixup') records.

These relocation records are made when the compiler/assembler doesn't know the address of a function or variable used in the source code, and generates a reference for it by name, which can be resolved by the linker.

For example, say you want a program to print a message to the screen, seperated into two source files, and you want to assemble them seperately and link them (example using Linux x86-64 syscalls) -

main.asm :

bits 64
section .text
extern do_message
global _start
_start:
    call do_message
    mov rax, 1
    int 0x80 

message.asm :

bits 64
section .text
global do_message
do_message:
    mov rdi, message
    mov rcx, dword -1
    xor rax, rax
    repnz scasb
    sub rdi, message
    mov rax, 4
    mov rbx, 1
    mov rcx, message
    mov rdx, rdi
    int 0x80
    ret

section .data
message: db "hello world",10,0

If you assemble these and look at the object file output of main.asm (eg, objdump -d main.o), you will notice the 'call do_message' has an address of 00 00 00 00 - which is invalid.

0000000000000000 <_start>:
   0:   e8 00 00 00 00          callq  5 <_start+0x5>
   5:   48 c7 c0 01 00 00 00    mov    $0x1,%rax
   c:   cd 80                   int    $0x80

But, a relocation record is made for the 4 bytes of the address :

$ objdump -r main.o
main.o:     file format elf64-x86-64
RELOCATION RECORDS FOR [.text]:
OFFSET           TYPE              VALUE 
0000000000000001 R_X86_64_PC32     do_message+0xfffffffffffffffc
000000000000000d R_X86_64_32       .data

The offset is '1' and the type is 'R_X86_64_PC32' which tells the linker to resolve this reference, and put the resolved address into the specified offset.

When you link the final program with 'ld -o program main.o message.o', the relocations are all resolved, and if nothing is unresolved, you are left with an executable.

When we 'objdump -d' the executable, we can see the resolved address :

00000000004000f0 <_start>:
  4000f0:   e8 0b 00 00 00          callq  400100 <do_message>
  4000f5:   48 c7 c0 01 00 00 00    mov    $0x1,%rax
  4000fc:   cd 80                   int    $0x80

The same kind of relocations are used for variables as well as functions. The same process happens when you link your program against multiple large libraries, such as libc - you define a function called 'main' which libc has an external reference to - then libc is started before your program, and calls your 'main' function when you run the executable.