Why do I have to reserve space on stack when calling puts?

Question

I have a simple questions but couldn't find the answer to it. I am working on a programming language which translates to NASM and of course I need to have compatibility with the x64 calling convention Microsoft uses. My test codes just calls puts() with the string "Hello", sets RAX to 0 and returns. I know that xor rax, rax would be smaller but optimization is a job for later on.

extern puts

global main

section .data

section .rdata
    constp_main:
        .c0: db 72,97,108,108,111,0

section .text
    main:
        push rbp
        mov rbp, rsp

        lea rcx, [constp_main.c0]
        call puts
        mov dword eax, dword 0

    .return:
        pop rbp
        ret

This code does not work and produces a crash. If I reserve 24 bytes or more on stack, code like this works:

extern puts

global main

section .data

section .rdata
    constp_main:
        .c0: db 72,97,108,108,111,0

section .text
    main:
        push rbp
        mov rbp, rsp
        sub rsp, 24

        lea rcx, [constp_main.c0]
        call puts
        mov dword eax, dword 0

    .return:
        mov rsp, rbp
        pop rbp
        ret

Can anybody tell me why? I thought the callee has to do all that stuff?

(Editor's note: these examples aren't actually safe because not enough stack space is reserved, even if they happen to work in some cases. See comments.)