Question

I wrote a small program to test GCC’s options.

int main()
{
    int a=0;
    __asm__("movl %0,%%ecx\n"
            "jmp jmpsection\n"
            "inc %%ecx\n"
            "jmpsection: movl $1,%%eax\n"
            "movl $0,%%ebx\n"
            "int $0x80\n"::"a"(a):"ecx","ebx");
}

In order to keep var a equal to 1, skip the inc instruction.
I want to force GCC to generate the jmp instruction using IP relative addressing method.
I have searched the GCC manual to find a solution, but I feiled.
Thanks for replay.

Answer 1

For the x86, IP relative addressing is only possible in long mode (64 bit) and you seem to be writing 32 bit code.

Edit: Actually, in 32 bit mode it is possible to make jumps relative to the current IP and I think the compiler actually generates the correct code in your case. Let’s take a look at the relevant part of the generated assembly:

00000000 <main>:
  ...
  13:   eb 01                   jmp    16 <jmpsection>
  15:   41                      inc    %ecx

00000016 <jmpsection>:
  16:   b8 01 00 00 00          mov    $0x1,%eax
  ...

So, although it looks like the instruction at address 13 makes a direct jump to address 16, it actually is an IP relative jump. Take a look at the machine code:

eb 01

eb is the opcode for a short jmp with an 8-bit displacement. 01 is this displacement. The result of this instruction is a jump to %eip + 0x01 and, as the IP points to the next instruction to be executed, this will jump to address 16.