Search code examples
assemblylinker-errorsattgnu-assembler

.quad $label in AT&T syntax: undefined reference to [label-name]


I learn assembler(AT&T). I tried to write program to find the length of the biggest string in record. During linking with command

ld person-data-with-names-pointer.o find-longest-name.o -o longest

I got error

ld: person-data-with-names-pointer.o: in function `people':
(.data+0x55): undefined reference to `$john_silver'
ld: (.data+0x85): undefined reference to `$billy_bones'
ld: (.data+0xb5): undefined reference to `$black_beard'
ld: (.data+0xe5): undefined reference to `$dave_jones'
ld: (.data+0x115): undefined reference to `$jack_the_sparrow'
ld: (.data+0x145): undefined reference to `$genry_morgan'

full listing of person-data-with-names-pointer.s

.section .data

.global people, numpeople
numpeople:
    # Calculate the number of people in array
    .quad (endpeople - people)/PERSON_RECORD_SIZE
people:
    .quad $john_silver, 200, 10, 2, 74, 20 
    .quad $billy_bones, 280, 14, 2, 74, 44
    .quad $black_beard, 150, 8, 1, 68, 30
    .quad $dave_jones, 250, 14, 3, 75, 24
    .quad $jack_the_sparrow, 250, 10, 2, 70, 11
    .quad $genry_morgan, 180, 11, 5, 69, 65 
endpeople: # Marks the end of the array for calculation purposes

john_silver:
    .ascii "John Silver\0"
billy_bones: 
    .ascii "Billy Bones\0"
black_beard: 
    .ascii "Black Beard\0"
dave_jones: 
    .ascii "Dave Jones\0"
jack_the_sparrow: 
    .ascii "Jack The Sparrow\0"
genry_morgan:
    .ascii "Genry Morgan\0" 
# Describes the components of the struct
.global NAME_PTR_OFFSET, WEIGHT_OFFSET, SHOE_OFFSET 
.global HAIR_OFFSET, HEIGHT_OFFSET, AGE_OFFSET
.equ NAME_PTR_OFFSET, 0
.equ WEIGHT_OFFSET,  8
.equ SHOE_OFFSET, 16 
.equ HAIR_OFFSET, 24
.equ HEIGHT_OFFSET, 32
.equ AGE_OFFSET, 40
# Total size of the struct 
.global PERSON_RECORD_SIZE
.equ PERSON_RECORD_SIZE, 48

I was tried to move labels but nothing works. :(


Solution

  • .quad isn't a real instruction like mov; a symbol name will always be taken as the symbol address, and a $ will be taken as a literal part of the symbol name. As the linker error shows you, it's looking for symbol names like $john_silver, not john_silver.

        .quad john_silver, 200, 10, 2, 74, 20 
        .quad billy_bones, 280, 14, 2, 74, 44
    

    For the same reason you write .quad 123 not .quad $123.

    Like most assemblers, there's no syntax that will get GAS to copy data from one .quad or .string to another; if you wanted the same data in two places, you'd define a macro or .equ assemble-time constant and emit the same data separately. You can't read back in some bytes emitted into an output section by a different source line.

    So there's nothing equivalent to the immediate vs. memory-operand possibility here, which is what $ distinguishes in instructions.

    mov symbol, %rax is a load of 8 bytes from that address (using 32-bit absolute addressing, not RIP-relative), vs. mov $symbol, %rax is the address as a 32-bit sign-extended immediate.
    In 64-bit code, normally you'd use mov symbol(%rip), %rax or lea symbol(%rip), %rax, or in non-PIE executables, 32-bit absolute mov $symbol, %eax. See How to load address of function or label into register