linux uppercaser not uppercaser

[yancyribbens]

I tested the example file uppercaser-linux-c.s although it doesn't look like it's designed to uppercaser..

from the comment section in the file uppercaser-linux-c.s

# Will print out command-line arguments and argument count

Also, it seems like while it's designed to print the number of command line arguments, it always prints -1 for every amount of arguments I feed it..

[jessicard]

Oh no! Thank you for catching this. Unfortunately I do not have a computer that was able to test this file when it came in, so I was unable to fully verify it working.

Some solutions:

• Would you have the time to potentially fix it?
• We could fix the arg count and rename the file to what it's actually doing
• We can remove the file entirely (not ideal - I think any working code is useful code for Assembly stuff)

Thoughts @yancyribbens?

[TheOneKevin]

Hi! Could you drop the platform and compiler you used? I remember this file doesn't perform any actual uppercasing but the it should at least print the arguments. I will try to reproduce this as soon as I have time.

[roussov]

You’re getting -1 because the code reads argc from the wrong stack slot and then mangles it with shr r12, 32. Under the System V AMD64 ABI, when main starts (before you push rbp), the stack looks like this: at [rsp + 0x00] you have argc (an int), at [rsp + 0x08] you have argv (a char **), and at [rsp + 0x10] you have envp. If you use the wrong offsets after setting up the frame, you end up treating a pointer as an integer and then shifting it, which is why the printed argument count always comes out as -1.

[rsp + 0x00] = argc (int)
[rsp + 0x08] = argv (char **)
[rsp + 0x10] = envp

Below is a corrected version that actually prints the right argument count and values:

.intel_syntax noprefix
.global main
.global printf
.global exit

.text

main:
    # int 3                 # optional breakpoint

    push    rbp
    mov     rbp, rsp

    # Stack layout after prologue:
    # [rbp+0x08] = old rbp
    # [rbp+0x10] = return address
    # [rbp+0x18] = argc (int)
    # [rbp+0x20] = argv (char **)

    mov     r12d, [rbp+0x18]    # r12d = argc
    mov     r13,  [rbp+0x20]    # r13  = argv

    dec     r12                 # argc-- (skip argv[0])

    lea     rdi, [rip+msg1]     # printf("Number of args: %d\n", argc-1)
    mov     rsi, r12
    xor     eax, eax            # no vector args
    call    printf

    mov     r14, 1              # i = 1
    add     r13, 8              # argv = &argv[1]

Lloop:
    cmp     r14, r12
    jg      Lexit               # if (i > argc-1) break

    lea     rdi, [rip+msg2]     # printf("Arg %d is \"%s\"\n", i, argv[i-1])
    mov     rsi, r14
    mov     rdx, [r13]
    xor     eax, eax
    call    printf

    inc     r14                 # i++
    add     r13, 8              # argv++
    jmp     Lloop

Lexit:
    xor     edi, edi            # exit(0)
    call    exit

.data

msg1: .asciz "Number of args: %d\n"
msg2: .asciz "Arg %d is \"%s\"\n"

[TheOneKevin]

i had forgotten this existed until i got a notif 😆 !! you're right, the stack slots for argc and argv are incorrectly swapped.

[yancyribbens]

@jessicard sorry to not respond, somehow I missed the notification. And 3 years later someone else seems to have fixed the issue :)

It's not longer fresh in my memory, but @roussov if you want to create a PR, I will try to review it. It sounds like you're correct though. Thanks!!