I am trying to print all doublewords in an array. However, there was an error in execution "Segmentation fault". After trying so many things, I finally made it work by storing and loading registers x8, x9, and x10 (I tested with many registers and kept eliminating registers until the program worked with these three and I can't miss a single one of these).

I have no clue why I even need to store and load these registers, isn't printf supposed to do that? And why does printf even mess these registers up? I am very confused because the parameter and return registers are x0-x7. And if printf uses other registers it should store and load.

Here's my code. (This is armv8 aarch64)

.text
    .equ ELEM, 10
    .extern printf
    .global _start
_start:
    .global selec_start
selec_start:
    ldr x0, =stack
    mov sp, x0
    sub sp, sp, #16

    ldr x8, =vec    // address of vec (first element)

    mov x9, #ELEM   // index of last element in vec
    sub x9, x9, #1

    mov x10, #0     // increment i
print:
    cmp x10, x9     // compare i with index of last element in vec
    bgt end         // branch to end if we hit end of loop

    lsl x11, x10, #3    // multiply i by 8
    add x12, x8, x11    // x12 = address of vec   i*8
    // x12 is address of index i in vec

    str x8, [sp, #-8]!
    str x9, [sp, #-8]!
    str x10, [sp, #-8]!

    ldr x0, =string
    ldr x1, [x12]
    bl printf

    ldr x10, [sp], #8
    ldr x9, [sp], #8
    ldr x8, [sp], #8

    add x10, x10, #1    // i  
    b print
end:
  mov w8, #93
  svc #0

    .data
string:
    .ascii "%d\n\0"
vec:
    .quad 1,2,3,4,5,6,7,8,9,10
    .bss
    .align 8
    .space 4096
stack:
    .space 16
.end

CodePudding user response：

x8, x9, and x10 are all caller-save registers on AARCH64, so if you care about the values in them, you need to save and restore them around any function call (such as to printf), as the called function might clobber them.

Only x19-x29 and x31 (sp) are callee-save and will(must) be preserved by any function call.