Assembly Gas - compile using gcc and ld cuase segmentation fault

Question

I am using Ubuntu 18.04 in a virtual box to run assembly GAS code.

.section    .rodata
input_int:      .string "%d"
.text
.global main
    .type   main, @function 
main:
    movq    %rsp, %rbp #for correct debugging -added by the ide
    pushq   %rbp
    movq    %rsp,%rbp

    subq    $8,%rsp

    movq    $input_int,%rdi
    movq    %rsp,%rsi
    movq    $0,%rax 
    call scanf

    movq    %rsp,%rdi
    movq    $0,%rax    
    call printf

    addq    $8,%rsp
    movq    $0,%rax 
    movq    %rbp,%rsp
    popq    %rbp
    ret

This is a very simple which should get an input integer and print the ascii character which corresponds to this integer.

I compile it using gcc:

gcc -m64 -c ./main.s
gcc -m64 ./main.o -no-pie -fno-pie
./a.out

Once reaching to the call scanf command it throws Segmentation Fault.

Trying to run a very similar code in the same way DO WORK:

    .section .data
    f:  .string "%d"
    x:  .long 0

.text   #the beginnig of the code
.globl  main    #the label "main" is used to state the initial point of this program
    .type   main, @function # the label "main" representing the beginning of a function
main:
movq %rsp, %rbp #for correct debugging  # the main function:
pushq   %rbp        #save the old frame pointer
movq    %rsp,   %rbp    #create the new frame pointer

# Get User Input
movq $0, %rax    # clear rax
movq $f, %rdi    # load format string
movq $x, %rsi    # set storage to address of x
call scanf

# Print Input
movq $x, %rdi
movq $0, %rax
call printf

#return from printf:
movq    $0, %rax    #return value is zero (just like in c - we tell the OS that this program finished seccessfully)
movq    %rbp, %rsp  #restore the old stack pointer - release all used memory.
popq    %rbp        #restore old frame pointer (the caller function frame)
ret         #return to caller function (OS)

Which outputs:

>> 97
a

The main difference between the two codes is that one uses an address from the data section and one is saving the variable to the stack of the program.

What may cause this behavior?

Ubuntu 18.04, x64 intel cpu. Using sasm to debug. Assembly options:$SOURCE$ -o $PROGRAM.OBJ$ --64 -a=$LSTOUTPUT$ . Linking options: $PROGRAM.OBJ$ $MACRO.OBJ$ -g -o $PROGRAM$ -m64 -fno-pie -no-pie — Montoya, Nov 17 '18 at 11:25
Your stack is misaligned. Let me quickly find the relevant question to link to. To fix your first example, try to only allocate stack space in increments of 16 bytes at a time. — fuz, Nov 17 '18 at 11:36
Hm... No really good duplicate. Read [this one](https://stackoverflow.com/q/49391001/417501) perhaps. — fuz, Nov 17 '18 at 11:41
Wow, Changing 8 to 16 bytes on moving the rsp did worked. Can you please explain why is it working like that? — Montoya, Nov 17 '18 at 12:01
@DannyHambourg It's because `rsp` has to be a multiple of 16 bytes when calling a function. After you `push %rbp`, this is the case, but when you subtract 8 bytes you destroy the alignment again. By subtracting 16 bytes, you keep the alignment and thus `scanf` does not crash. — fuz, Nov 18 '18 at 12:36

Assembly Gas - compile using gcc and ld cuase segmentation fault

0 Answers0