CODEWITHSUNDEEP
assemblycompiler-constructionx86system-callsstatic-linking
Cris
Cris

Reputation: 185

Calling read syscall from assembly (x86/64) yields segmentation fault (compiler construction)

I am building a compiler for a C-like language, and i'm trying to link a basic "void readString(int, char*)" function implemented in assembly, with assembly generated by my compiler.

The compiled c-like file is

void main () {
    char t[20];     
    readString(7,t); // Read 7 bytes and place them in t buffer
}

After the compilation, The resulting file is: out.s file (notice that the calling convention is: passing arguments through stack. also Integers have 2 bytes size in this language):

.$0:
.globl main
main:
pushq %rbp
movq %rsp,%rbp
subq $20,%rsp

.$1:
movw $7,%ax  # Push first argument in the stack
pushw %ax

.$2:
leaq -20(%rbp),%rax # Push address of the second arg in the stack
pushq %rax

.$3:
subq $8,%rsp   # this is not important, needed for the convention being followed
pushq 16(%rbp) # pushing "access link",
call _readString 
addq $26,%rsp # caller clears the "leftovers" 

.$4:
.$main_0_11:
movq %rbp,%rsp
popq %rbp
ret

Code in reads.asm for the library function:

.intel_syntax noprefix
            .global _readString

_readString     push    rbp
            mov     rbp, rsp
            push    rdi
            push    rsi

            mov     rdi, [rbp+32]           # First argument 
            mov     rsi, [rbp+34]           # Second Argument

            mov     rdx, rdi

doRead:
            mov     byte ptr [rsi], 0x00
            xor     rax, rax
            mov     rdi, rax
            syscall                         # read syscall, reads up to $rdx bytes
            or      rax, rax                # nothing read
            jz      finish                  #
            add     rsi, rax                #
            cmp     byte ptr [rsi-1], 0x0a  # check if last character read was '\n'
            jne     addZero                 #
            sub     rsi, 1                  # if so, replace with '\0'
addZero:
            mov     byte ptr [rsi], 0x00
finish:
            pop     rsi
            pop     rdi
            pop     rbp
            ret

The linking/running follows

$ gcc -c out.s
$ gcc -c reads.s 
$ gcc out.o reads.o 
$ ./a.out
[2] Segmentation fault  ./a.out

A Picture that depicts the calling convention

calling convention

Upvotes: 0

Views: 537

Answers (1)

Cris
Cris

Reputation: 185

Ok, as @David Hoelzer suggested i checked again the way "readString" grabs the arguments from the stack and they were reversed in order!

So the first lines of reads.asm will become

_readString:
            push    rbp
            mov     rbp, rsp
            push    rdi
            push    rsi

            movzx   edi, word ptr [rbp+40] # Get first argument, check comments for more info
            mov     rsi, [rbp+32]  # Second Argument


            mov     rdx, rdi

Upvotes: 1

Related Questions