Machine Mirage in a Silicon Whisper

The Genesis

It all began with a tweet from Shafik Yaghmour, sharing a "cursed Hello World" where an array of integers is passed to `main`, and somehow, against all conventions, it surprisingly prints a hello message on x86_64 machines. This compelled me to wonder: Why not perform this magic trick on my AArch64 MacBook?

The Process

Although, technically, the provided code doesn't align with the C standard's defined prototypes for the main function

int main(void);

or

int main(int argc, char *argv[]);

many compilers permit such definitions as they basically look for a symbol named main to assign it as the program's entry point—even if its signature doesn't match the recognised prototypes.

Understanding that a function is essentially a pointer to the start location of its code, it's evident that, to put it simply, when a function gets invoked, the processor begins executing the machine code at that memory address. Thus, when we designate an array of integers with the intent of invoking it as a function, since those integers should follow the associated memory address in sequence, upon invocation it is anticipated to be understood as machine code. And since main represents the global entry point (termed "_main" on macOS or "_start" on most UNIX-like operating systems,) it led me to a simple hypothesis: the data should represent raw machine code.

To validate this, I wrote a basic assembly program that prints "Hello, world!" to the console. Here's a glance at the code:

    .section    __TEXT,__text,regular,pure_instructions
    .global     _main
    .p2align    2
_main:
    mov x0, #1          ; stdout descriptor
    adr x1, hello_world ; address of the message
    mov x2, #14         ; message length (14 chars)
    mov x16, #4         ; syscall: write
    svc #0x80           ; invoke syscall

    mov x0, #0          ; return 0 (success)
    mov x16, #1         ; syscall: exit
    svc #0x80           ; invoke syscall

hello_world: .ascii "Hello, world!\n"

After compiling and delving into the code with `lldb` (though `objdump` could suffice as a quicker and simpler route,) a fascinating series of discoveries unfolded and I ensured this is how it works.

My investigation consisted of the following steps to inspect how the machine code is laid out in memory:

1. Locating the address of the main function

   (lldb) image lookup --name main
   1 match found in /tmp/hello-world:
           Address: hello-world[0x0000000100003f88] (hello-world.__TEXT.__text + 0)
           Summary: hello-world`main

2. Reading from memory to obtain the machine code

   Starting address is retrieved, now it's time to calculate the ending address. Since 0x100003f88 is where main starts, and the assembly code consists of 8 instructions, since ARM specifies that instructions are 4 byte each here, adding 32 gives 0x100003fa8. Then, knowing that the length of the message is 14 bytes, adding 14 gives 0x100003fb6. Depending on whether an array of char or int is desired, we either pass 0x100003fb6 or 0x100003fb8 respectively, since for char we read 1 byte at a time but for int we do 4 (and therefore the address needs to be aligned to a 4-byte boundary.)

   Below, you'll find how memory read could be performed for either char or int:

   (lldb) memory read --size 1 --format x 0x100003f88 0x100003fb6
   0x100003f88: 0x20 0x00 0x80 0xd2 0xe1 0x00 0x00 0x10
   0x100003f90: 0xc2 0x01 0x80 0xd2 0x90 0x00 0x80 0xd2
   0x100003f98: 0x01 0x10 0x00 0xd4 0x00 0x00 0x80 0xd2
   0x100003fa0: 0x30 0x00 0x80 0xd2 0x01 0x10 0x00 0xd4
   0x100003fa8: 0x48 0x65 0x6c 0x6c 0x6f 0x2c 0x20 0x77
   0x100003fb0: 0x6f 0x72 0x6c 0x64 0x21 0x0a
   (lldb) memory read --size 4 --format x 0x100003f88 0x100003fba
   0x100003f88: 0xd2800020 0x100000e1 0xd28001c2 0xd2800090
   0x100003f98: 0xd4001001 0xd2800000 0xd2800030 0xd4001001
   0x100003fa8: 0x6c6c6548 0x77202c6f 0x646c726f 0x00000a21

3. Crafting the enigmatic code

   With having the machine code at hand, a new form of the “Hello, world!” program emerged. Let's begin with the int array of hexadecimal numbers to better understand what is done exactly:

   const int main[] = {
   	0xd2800020,     // mov x0, #1
   	0x100000e1,     // adr x1, hello_world
   	0xd28001c2,     // mov x2, #14
   	0xd2800090,     // mov x16, #4
   	0xd4001001,     // svc #0x80
   
   	0xd2800000,     // mov x0, #0
   	0xd2800030,     // mov x16, #1
   	0xd4001001,     // svc #0x80
   
   	// ASCII encoded "Hello, world!\n" in little-endian format
   	0x6c6c6548,     // "lleH"
   	0x77202c6f,     // "w ,o"
   	0x646c726f,     // "dlro""
   	0x00000a21      // "\n!"
   };

   Now if one prefers to write the same thing in decimal numbers, nothing differs but the look:

   const int main[] = {
   	-763363296,
   	268435681,
   	-763362878,
   	-763363184,
   	-738193407,
   	-763363328,
   	-763363280,
   	-738193407,
   	1819043144,
   	1998597231,
   	1684828783,
   	2593
   };

   For the char array, it's essential to consider the little-endianness. Neglecting this detail can result in illegal hardware instructions or similar errors.

   const char main[] = {
       0x20, 0x00, 0x80, 0xd2,
       0xe1, 0x00, 0x00, 0x10,
       0xc2, 0x01, 0x80, 0xd2,
       0x90, 0x00, 0x80, 0xd2,
       0x01, 0x10, 0x00, 0xd4,
       0x00, 0x00, 0x80, 0xd2,
       0x30, 0x00, 0x80, 0xd2,
       0x01, 0x10, 0x00, 0xd4,
       0x48, 0x65, 0x6c, 0x6c,
       0x6f, 0x2c, 0x20, 0x77,
       0x6f, 0x72, 0x6c, 0x64,
       0x21, 0x0a, 0x00, 0x00
   };

   Overall, once we get the machine code and we understand how the assembler, linker, or our C compiler in general, work, the rest is just a personalised game with data.

The Importance

Why, you may ask, would a seasoned engineer, well-versed in modern languages, dig this deep into such low-level trickery, when it has no practicality and therefore no place in real world?

You're right! But understanding the inner workings of our programming language, compiler, interpreter, or even the machine code and the hardware is more than just an academic exercise. It brings forth the following benefits:

1. Enhanced Troubleshooting & Debugging Skills: Recognising how instructions are executed at the machine level sharpens your debugging prowess.
2. Expanded Horizon: It broadens your perspective, making you a deeper and more versatile engineer.
3. Mastering Through Exploration: Pushing boundaries, abusing the type system, understanding standards and noticing limitations are just means to master and respect the art of programming. The more we explore, the deeper our understanding becomes.

So, the essence is not to promote abusing undefined behaviours in C, but to appreciate the science and art behind every line of code we write. This exploration, reminiscent of our first "Hello, world!", reminds us that every professional was once a beginner and that continuous learning keeps that spark alive, and that even a simple "Hello, world!" is not that simple if seen from the other way around.

Thus, this journey is not merely a trip down the rabbit hole; it is a testament to the wonders of computer science and the beauty of understanding the machine. Every time you write "Hello, world!", remember: there's always more to discover, understand, and appreciate.

Keep your curiosity alive, explore the depths of what you know, and happy coding!