58 KiB
Week 6: Static Variables in Embedded Systems: Debugging and Hacking Static Variables w/ GPIO Input Basics
What You'll Learn This Week
By the end of this tutorial, you will be able to:
- Understand the difference between regular (automatic) variables and static variables
- Know where different types of variables are stored (stack vs static storage)
- Configure GPIO pins as inputs and use internal pull-up resistors
- Read button states using
gpio_get()and control LEDs based on input - Use GDB to examine how the compiler handles static vs automatic variables
- Identify compiler optimizations by stepping through assembly
- Hack variable values and invert GPIO input/output logic using a hex editor
- Convert patched binaries to UF2 format for flashing
Part 1: Understanding Static Variables
What is a Static Variable?
A static variable is a special kind of variable that "remembers" its value between function calls or loop iterations. Unlike regular variables that get created and destroyed each time, static variables persist for the entire lifetime of your program.
Think of it like this:
- Regular variable: Like writing on a whiteboard that gets erased after each class
- Static variable: Like writing in a notebook that you keep forever
+-----------------------------------------------------------------+
| Regular vs Static Variables |
| |
| REGULAR (automatic): |
| +------------------------------------------------------------+ |
| | Loop 1: Create -> Set to 42 -> Increment to 43 -> Destroy |
| | Loop 2: Create -> Set to 42 -> Increment to 43 -> Destroy |
| | Loop 3: Create -> Set to 42 -> Increment to 43 -> Destroy |
| | Result: Always appears as 42! |
| +------------------------------------------------------------+ |
| |
| STATIC: |
| +------------------------------------------------------------+ |
| | Loop 1: Already exists -> Read 42 -> Increment -> Store 43 |
| | Loop 2: Already exists -> Read 43 -> Increment -> Store 44 |
| | Loop 3: Already exists -> Read 44 -> Increment -> Store 45 |
| | Result: Keeps incrementing! |
| +------------------------------------------------------------+ |
| |
+-----------------------------------------------------------------+
The static Keyword
In C, you declare a static variable by adding the static keyword:
uint8_t regular_fav_num = 42; // Regular - recreated each time
static uint8_t static_fav_num = 42; // Static - persists forever
Where Do Variables Live in Memory?
Different types of variables are stored in different memory locations:
| Variable Type | Storage Location | Lifetime | Example |
|---|---|---|---|
| Automatic (local) | Stack | Until function/block ends | uint8_t x = 5; |
| Static | Static Storage | Entire program lifetime | static uint8_t x = 5; |
| Global | Static Storage | Entire program lifetime | uint8_t x = 5; (outside functions) |
| Dynamic (heap) | Heap | Until free() is called |
malloc(sizeof(int)) |
Stack vs Static Storage vs Heap
+-----------------------------------------------------------------+
| Memory Layout |
| |
| +-------------------+ High Address (0x20082000) |
| | STACK | ?? Automatic/local variables |
| | (grows down) | Created/destroyed per function |
| +-------------------+ |
| | | |
| | (free space) | |
| | | |
| +-------------------+ |
| | HEAP | ?? Dynamic allocation (malloc/free) |
| | (grows up) | |
| +-------------------+ |
| | .bss section | ?? Uninitialized static/global vars |
| +-------------------+ |
| | .data section | ?? Initialized static/global vars |
| +-------------------+ Low Address (0x20000000) |
| |
+-----------------------------------------------------------------+
Key Point: Static variables are NOT on the heap! They live in a fixed location in the .data section (if initialized) or .bss section (if uninitialized). This is different from heap memory which is dynamically allocated at runtime.
What Happens with Overflow?
Since static_fav_num is a uint8_t (unsigned 8-bit), it can only hold values 0-255. What happens when it reaches 255 and we add 1?
255 + 1 = 256... but that doesn't fit in 8 bits!
Binary: 11111111 + 1 = 100000000 (9 bits)
The 9th bit is lost, so we get: 00000000 = 0
This is called overflow or wrap-around. The value "wraps" back to 0 and starts counting again!
Part 2: Understanding GPIO Inputs
Input vs Output
So far, we've used GPIO pins as outputs to control LEDs. Now we'll learn to use them as inputs to read button states!
+-----------------------------------------------------------------+
| GPIO Direction |
| |
| OUTPUT (what we've done before): |
| +---------+ |
| | Pico | -------? LED |
| | GPIO 16 | (We control the LED) |
| +---------+ |
| |
| INPUT (new this week): |
| +---------+ |
| | Pico | ?------- Button |
| | GPIO 15 | (We read the button state) |
| +---------+ |
| |
+-----------------------------------------------------------------+
The Floating Input Problem
When a GPIO pin is set as an input but nothing is connected, it's called a floating input. The voltage on the pin is undefined and can randomly read as HIGH (1) or LOW (0) due to electrical noise.
+-----------------------------------------------------------------+
| Floating Input = Random Values! |
| |
| GPIO Pin (no connection): |
| Reading 1: HIGH |
| Reading 2: LOW |
| Reading 3: HIGH |
| Reading 4: HIGH |
| Reading 5: LOW |
| (Completely unpredictable!) |
| |
+-----------------------------------------------------------------+
Pull-Up and Pull-Down Resistors
To solve the floating input problem, we use pull resistors:
| Resistor Type | Default State | When Button Pressed |
|---|---|---|
| Pull-Up | HIGH (1) | LOW (0) |
| Pull-Down | LOW (0) | HIGH (1) |
The Pico 2 has internal pull resistors that you can enable with software - no external components needed!
+-----------------------------------------------------------------+
| Pull-Up Resistor (what we're using) |
| |
| 3.3V |
| | |
| + (internal pull-up resistor) |
| | |
| +------? GPIO 15 (reads HIGH normally) |
| | |
| +-+-+ |
| |BTN| ?? Button connects GPIO to GND when pressed |
| +-+-+ |
| | |
| GND |
| |
| Button NOT pressed: GPIO reads 1 (HIGH) |
| Button PRESSED: GPIO reads 0 (LOW) |
| |
+-----------------------------------------------------------------+
GPIO Input Functions
| Function | Purpose |
|---|---|
gpio_init(pin) |
Initialize a GPIO pin for use |
gpio_set_dir(pin, GPIO_IN) |
Set pin as INPUT |
gpio_pull_up(pin) |
Enable internal pull-up resistor |
gpio_pull_down(pin) |
Enable internal pull-down resistor |
gpio_get(pin) |
Read the current state (returns 0 or 1) |
The Ternary Operator
The code uses a ternary operator to control the LED based on button state:
gpio_put(LED_GPIO, pressed ? 0 : 1);
This is a compact if-else statement:
- If
pressedis true (1): output0(LED OFF... wait, that seems backwards!) - If
pressedis false (0): output1(LED ON)
Why is it inverted? Because of the pull-up resistor!
- Button released -> GPIO reads
1->pressed = 1-> output0-> LED OFF - Button pressed -> GPIO reads
0->pressed = 0-> output1-> LED ON
A clearer way to write this:
gpio_put(LED_GPIO, !gpio_get(BUTTON_GPIO));
Part 3: Understanding Compiler Optimizations
Why Does Code Disappear?
When you compile code, the compiler tries to make it faster and smaller. This is called optimization. Sometimes the compiler removes code that it thinks has no effect!
Example from our code:
while (true) {
uint8_t regular_fav_num = 42; // Created
regular_fav_num++; // Incremented to 43
// But then it's destroyed and recreated as 42 next loop!
}
The compiler sees that incrementing regular_fav_num has no lasting effect (because it's recreated as 42 each loop), so it may optimize away the increment operation entirely!
Function Inlining
Sometimes the compiler inlines functions, meaning it replaces a function call with the function's code directly.
Original code:
gpio_pull_up(BUTTON_GPIO);
What the compiler might do:
// Instead of calling gpio_pull_up, it calls the underlying function:
gpio_set_pulls(BUTTON_GPIO, true, false);
This is why when you look for gpio_pull_up in the binary, you might find gpio_set_pulls instead!
Part 4: Setting Up Your Environment
Prerequisites
Before we start, make sure you have:
- A Raspberry Pi Pico 2 board
- A Raspberry Pi Pico Debug Probe
- OpenOCD installed and configured
- GDB (
arm-none-eabi-gdb) installed - Python installed (for UF2 conversion)
- A serial monitor (PuTTY, minicom, or screen)
- A push button connected to GPIO 15
- An LED connected to GPIO 16 (or use the breadboard LED)
- A hex editor (HxD, ImHex, or similar)
- The sample project:
0x0014_static-variables
Hardware Setup
Connect your button like this:
- One side of button -> GPIO 15
- Other side of button -> GND
The internal pull-up resistor provides the 3.3V connection, so you only need to connect to GND!
+-----------------------------------------------------------------+
| Breadboard Wiring |
| |
| Pico 2 |
| +----------+ |
| | | |
| | GPIO 15 |--------+ |
| | | | |
| | GPIO 16 |--------+---? LED (with resistor to GND) |
| | | | |
| | GND |--------+---+ |
| | | | | |
| +----------+ +-+-+ | |
| |BTN|-+ |
| +---+ |
| |
+-----------------------------------------------------------------+
Project Structure
Embedded-Hacking/
+-- 0x0014_static-variables/
| +-- build/
| | +-- 0x0014_static-variables.uf2
| | +-- 0x0014_static-variables.elf
| +-- 0x0014_static-variables.c
+-- uf2conv.py
Part 5: Hands-On Tutorial - Static Variables and GPIO Input
Step 1: Review the Source Code
Let's examine the static variables code:
File: 0x0014_static-variables.c
#include <stdio.h>
#include "pico/stdlib.h"
int main(void) {
stdio_init_all();
const uint BUTTON_GPIO = 15;
const uint LED_GPIO = 16;
bool pressed = 0;
gpio_init(BUTTON_GPIO);
gpio_set_dir(BUTTON_GPIO, GPIO_IN);
gpio_pull_up(BUTTON_GPIO);
gpio_init(LED_GPIO);
gpio_set_dir(LED_GPIO, GPIO_OUT);
while (true) {
uint8_t regular_fav_num = 42;
static uint8_t static_fav_num = 42;
printf("regular_fav_num: %d\r\n", regular_fav_num);
printf("static_fav_num: %d\r\n", static_fav_num);
regular_fav_num++;
static_fav_num++;
pressed = gpio_get(BUTTON_GPIO);
gpio_put(LED_GPIO, pressed ? 0 : 1);
}
}
What this code does:
- Line 6-8: Defines constants for button (GPIO 15) and LED (GPIO 16) pins
- Line 10-12: Sets up GPIO 15 as input with internal pull-up resistor
- Line 14-15: Sets up GPIO 16 as output for the LED
- Line 18-19: Creates two variables:
regular_fav_num- a normal local variable (recreated each loop)static_fav_num- a static variable (persists across loops)
- Line 21-22: Prints both values to the serial terminal
- Line 24-25: Increments both values
- Line 27-28: Reads button and controls LED accordingly
Step 2: Flash the Binary to Your Pico 2
- Hold the BOOTSEL button on your Pico 2
- Plug in the USB cable (while holding BOOTSEL)
- Release BOOTSEL - a drive called "RPI-RP2" appears
- Drag and drop
0x0014_static-variables.uf2onto the drive - The Pico will reboot and start running!
Step 3: Open Your Serial Monitor
Open PuTTY, minicom, or screen and connect to your Pico's serial port.
You should see output like this:
...
regular_fav_num: 42
static_fav_num: 42
regular_fav_num: 42
static_fav_num: 43
regular_fav_num: 42
static_fav_num: 44
regular_fav_num: 42
static_fav_num: 45
...
Notice the difference:
regular_fav_numstays at 42 every time (it's recreated each loop)static_fav_numincreases each time (it persists and remembers its value)
Step 4: Test the Button
Now test the button behavior:
- Button NOT pressed: LED should be OFF
- Button PRESSED: LED should turn ON
That may feel backwards at first glance, but it matches the pull-up + ternary logic in the code. We'll hack this later to make the behavior more intuitive.
Step 5: Watch for Overflow
Keep the program running and watch static_fav_num. After 255, you'll see:
static_fav_num: 254
static_fav_num: 255
static_fav_num: 0 ?? Wrapped around!
static_fav_num: 1
static_fav_num: 2
...
This demonstrates unsigned integer overflow!
Part 6: Debugging with GDB (Dynamic Analysis)
? REVIEW: This setup is identical to previous weeks. If you need a refresher on OpenOCD and GDB connection, refer back to Week 3 Part 6.
Starting the Debug Session
Terminal 1 - Start OpenOCD:
openocd -s "%USERPROFILE%\.pico-sdk\openocd\0.12.0+dev\scripts" -f interface/cmsis-dap.cfg -f target/rp2350.cfg -c "adapter speed 5000"
Terminal 2 - Start GDB:
arm-none-eabi-gdb build\0x0014_static-variables.elf
Connect to target:
(gdb) target extended-remote :3333
(gdb) monitor reset halt
Step 6: Examine Main Function
Let's examine the main function at its entry point. First, disassemble from the start:
x/38i 0x10000234
You should see output like:
(gdb) x/38i 0x10000234
0x10000234 <main>: push {r4, lr}
0x10000236 <main+2>: bl 0x10003014 <stdio_init_all>
0x1000023a <main+6>: movs r0, #15
0x1000023c <main+8>: bl 0x10000300 <gpio_init>
0x10000240 <main+12>: movs r0, #15
0x10000242 <main+14>: mov.w r3, #0
0x10000246 <main+18>: mcrr 0, 4, r0, r3, cr4
0x1000024a <main+22>: movs r2, #0
0x1000024c <main+24>: movs r1, #1
0x1000024e <main+26>: bl 0x100002d8 <gpio_set_pulls>
0x10000252 <main+30>: movs r0, #16
0x10000254 <main+32>: bl 0x10000300 <gpio_init>
0x10000258 <main+36>: movs r3, #16
0x1000025a <main+38>: mov.w r2, #1
0x1000025e <main+42>: mcrr 0, 4, r3, r2, cr4
0x10000262 <main+46>: ldr r4, [pc, #44] @ (0x10000290 <main+92>)
0x10000264 <main+48>: movs r1, #42 @ 0x2a
0x10000266 <main+50>: ldr r0, [pc, #44] @ (0x10000294 <main+96>)
0x10000268 <main+52>: bl 0x100031a4 <__wrap_printf>
0x1000026c <main+56>: ldrb r1, [r4, #0]
0x1000026e <main+58>: ldr r0, [pc, #40] @ (0x10000298 <main+100>)
0x10000270 <main+60>: bl 0x100031a4 <__wrap_printf>
0x10000274 <main+64>: mov.w r1, #3489660928 @ 0xd0000000
0x10000278 <main+68>: ldrb r3, [r4, #0]
0x1000027a <main+70>: movs r2, #16
0x1000027c <main+72>: adds r3, #1
0x1000027e <main+74>: strb r3, [r4, #0]
0x10000280 <main+76>: ldr r3, [r1, #4]
0x10000282 <main+78>: ubfx r3, r3, #15, #1
0x10000286 <main+82>: eor.w r3, r3, #1
0x1000028a <main+86>: mcrr 0, 4, r2, r3, cr0
0x1000028e <main+90>: b.n 0x10000264 <main+48>
0x10000290 <main+92>: lsls r0, r5, #22
0x10000292 <main+94>: movs r0, #0
0x10000294 <main+96>: adds r5, #96 @ 0x60
0x10000296 <main+98>: asrs r0, r0, #32
0x10000298 <main+100>: adds r5, #120 @ 0x78
0x1000029a <main+102>: asrs r0, r0, #32
Step 7: Set a Breakpoint at Main
b *0x10000234
c
Step 8: Examine the Static Variable Location
Static variables live at fixed RAM addresses. But how do we find that address? Look at 0x10000262 in the disassembly from Step 6:
(gdb) x/16i 0x10000234
=> 0x10000234 <main>: push {r4, lr}
0x10000236 <main+2>: bl 0x10003014 <stdio_init_all>
0x1000023a <main+6>: movs r0, #15
0x1000023c <main+8>: bl 0x10000300 <gpio_init>
0x10000240 <main+12>: movs r0, #15
0x10000242 <main+14>: mov.w r3, #0
0x10000246 <main+18>: mcrr 0, 4, r0, r3, cr4
0x1000024a <main+22>: movs r2, #0
0x1000024c <main+24>: movs r1, #1
0x1000024e <main+26>: bl 0x100002d8 <gpio_set_pulls>
0x10000252 <main+30>: movs r0, #16
0x10000254 <main+32>: bl 0x10000300 <gpio_init>
0x10000258 <main+36>: movs r3, #16
0x1000025a <main+38>: mov.w r2, #1
0x1000025e <main+42>: mcrr 0, 4, r3, r2, cr4
0x10000262 <main+46>: ldr r4, [pc, #44] @ (0x10000290 <main+92>)
This loads r4 from the literal pool at address 0x10000290. The literal pool stores constants that are too large for immediate encoding - in this case, a 32-bit RAM address. Let's examine what's stored there:
(gdb) x/1wx 0x10000290
0x10000290 <main+92>: 0x200005a8
That's 0x200005a8 - the RAM address of static_fav_num! The compiler placed this address in the literal pool because it can't encode a full 32-bit address in a single Thumb instruction.
Common confusion three things that trip people up here:
1. GDB shows
<striped_spin_lock_num>instead of<static_fav_num>.static_fav_numis a function-local static, so it does not appear in GDB's global symbol table. GDB labels the address with the nearest named global symbol it can find in this case an SDK-internal spin-lock variable that happens to be nearby. The address0x200005a8is still correct; the label is just GDB's best guess at a name.
(gdb) x/x 0x200005a8
0x200005a8 <striped_spin_lock_num>: 0x0000122a
2.
x/x 0x200005a8shows0x0000122a, not0x0000002a(42).x/xreads a 4-byte word (32 bits).static_fav_numis auint8_tonly 1 byte. The low byte of0x0000122ais0x2a= 42. That is the variable. The upper bytes belong to whatever happens to be in adjacent RAM. To read exactly one byte and see42, use:
(gdb) x/1ub 0x200005a8
0x200005a8 <striped_spin_lock_num>: 42
3.
x/x *0x200005a8shows junk at address0x122a. The*is a pointer dereference. GDB first reads the 4-byte word at0x200005a8(=0x0000122a), then treats that value as a new address and reads from0x0000122a. You are asking GDB to follow a pointer that was never meant to be a pointer the result is meaningless. Drop the*: usex/1ub 0x200005a8to read the variable's value directly.
Tip: Why did the disassembly at
0x10000290showlsls r0, r5, #22instead? Becausex/i(disassemble) interprets raw data as instructions. The bytesA8 05 00 20at that address are the little-endian encoding of0x200005A8, but GDB's disassembler doesn't know it's data - it tries to decode it as a Thumb instruction. Usingx/wx(examine as word) shows the actual value.
Step 9: Step Through the Loop
The loop body starts at 0x10000264. Set a breakpoint there and continue:
(gdb) b *0x10000264
(gdb) c
(gdb) x/i $pc
=> 0x10000264 <main+48>: movs r1, #42 @ 0x2a
GDB will stop at movs r1, #42 that is the compiler's constant for regular_fav_num. Let's set a breakpoint and continue at 0x10000278.
(gdb) b *0x10000278
(gdb) c
(gdb) x/i $pc
=> 0x10000278 <main+68>: ldrb r3, [r4, #0]
This is where the static variable is actually manipulated:
(gdb) x/4i $pc
=> 0x10000278 <main+68>: ldrb r3, [r4, #0] <- load static_fav_num from RAM into r3
0x1000027a <main+70>: movs r2, #16 <- load LED pin number (16) into r2 for later
0x1000027c <main+72>: adds r3, #1 <- increment r3 by 1
0x1000027e <main+74>: strb r3, [r4, #0] <- store incremented value back to RAM
Confirm what r4 actually contains right now:
(gdb) x/x $r4
0x200005a8 <striped_spin_lock_num>: 0x2a
r4 is 0x200005a8, a RAM address. The value at that address is 0x2a (42 in decimal), which is the current value of static_fav_num. After adds r3, #1 and strb r3, [r4, #0] execute, examining r4 again will show 0x0000002b (43).
What is the literal pool?
A Thumb ldr instruction is only 16 or 32 bits wide. There is no room inside those bits to encode a full 32-bit constant like 0x200005a8. To solve this, the compiler collects all such large constants and writes them into a small block of raw data embedded directly in the flash image this block is called a literal pool (also called a constant pool).
There is not one literal pool per flash, there is (at least) one per function. Every function that references large constants gets its own pool appended to its code. A large function can have several pools scattered through it wherever the compiler decides to emit them. Each pool is purely data; the CPU must never try to execute it.
Where exactly is a literal pool placed? Always after an unconditional branch, never before or in the middle of a straight-line code path. This is by design: the CPU falls through code sequentially, so the pool must be unreachable by fall-through. In our case the pool sits at 0x10000290, directly after the b.n 0x10000264 branch at 0x1000028e that closes the while (true) loop:
0x1000028e: b.n 0x10000264 <- unconditional branch back to loop top
0x10000290: [literal pool] <- CPU never reaches here by fall-through
Why does the pool have to be nearby? The ldr rN, [pc, #offset] instruction encodes the offset in a limited number of bits. In standard Thumb the maximum reach is 1020 bytes from the instruction. If the function is long enough that the end is out of range, the assembler emits an intermediate pool mid-function after the next unconditional branch it can find. This is in the ARM v8-M Architecture Reference Manual p. 672.
How the load works in our function: At 0x10000262 the compiler emits:
ldr r4, [pc, #44] @ (0x10000290)
On ARM, PC reads as the address of the instruction plus 4 during execution. So the effective address is 0x10000262 + 4 + 44 = 0x10000290. That word in the pool contains 0x200005a8, the RAM address of static_fav_num. The pool is also where the printf format-string pointers live (the other ldr instructions you saw in the disassembly).
Flash (read-only)
+-------------------------------------------+ RAM
| 0x10000262: ldr r4, [pc, #44] | +------------------+
| ...loop body... | | 0x200005a8: 42 | <- static_fav_num
| 0x1000028e: b.n 0x10000264 (end of loop) | +------------------+
+------- literal pool (data, not code) -----+ ^
| 0x10000290: 0x200005a8 --------------------+----------------+
| 0x10000294: ptr to "regular_fav_num: %d" | r4 holds this address
| 0x10000298: ptr to "static_fav_num: %d" |
+------- next function ---------------------+
| 0x1000029c: push {r4} |
+-------------------------------------------+
You can see this directly. Run x/10i 0x1000028e in GDB and you will get exactly this:
(gdb) x/10i 0x1000028e
(gdb) x/10i 0x1000028e
0x1000028e <main+90>: b.n 0x10000264 <main+48> <- unconditional branch, loop top
0x10000290 <main+92>: lsls r0, r5, #22 <-+
0x10000292 <main+94>: movs r0, #0 <-+-- word 1: 0x200005a8 (fixed RAM addr static_fav_num)
0x10000294 <main+96>: adds r5, #96 @ 0x60 <-+
0x10000296 <main+98>: asrs r0, r0, #32 <-+-- word 2: 0x10004b60 (flash addr regular_fav_num text)
0x10000298 <main+100>: adds r5, #120 <-+
0x1000029a <main+102>: asrs r0, r0, #32 <-+-- word 3: 0x10004b78 (flash addr static_fav_num text)
0x10000290 - 0x1000029b is the literal pool 12 bytes, 3 32-bit words. GDB's x/i (disassemble) has no idea those bytes are data, so it blindly decodes them as Thumb instructions and produces nonsense (lsls r0, r5, #22, etc.). The giveaway that you have entered pool territory is always the same: garbled, implausible instructions immediately after an unconditional branch. Use x/wx to read those addresses as data and you get the real values:
(gdb) x/3wx 0x10000290
0x10000290 <main+92>: 0x200005a8 0x10003560 0x10003578
(gdb) x/1ub 0x200005a8
0x200005a8 <striped_spin_lock_num>: 42
0x200005a8RAM address ofstatic_fav_num, lives in the.datasection (initialized static RAM,0x20000000``0x200xxxxx). It is.dataand not.bssbecause the variable has an explicit initializer (= 42)..bssis for variables that are zero-initialized or have no initializer at all.
(gdb) x/s 0x10003560
0x10003560: "regular_fav_num: %d\r\n"
0x10003560flash address of the"regular_fav_num: %d\r\n"string literal, lives in the.rodatasection (read-only data) inside flash (0x10000000+). String literals are constants baked into the binary at compile time they never change and never move, so they stay in flash.
(gdb) x/s 0x10003578
0x10003578: "static_fav_num: %d\r\n"
- `0x10003578` flash address of the `"static_fav_num: %d\r\n"` string literal, also in **`.rodata`** in flash for the same reason.
So the pool contains addresses into three different regions: RAM `.data` (the static variable), and flash `.rodata` (both format strings). Only the RAM address needed to be in the pool, the flash addresses could in principle be reached other ways, but they are also too large to encode as 16-bit immediates, so they go in the pool too.
**Why `.data` and not `.bss`?** The distinction matters:
| Section | What goes here | Example declarations | Initial value stored in flash? |
|---------|---------------|----------------------|-------------------------------|
| `.data` | Any static-duration variable with a **non-zero initializer** | `static uint8_t x = 42;` (inside a function) | Yes the initial value lives in flash; startup copies it to RAM |
| `.data` | Same rule applies to **global** variables with non-zero initializers | `uint8_t g = 10;` (outside all functions) | Yes same startup copy |
| `.bss` | Any static-duration variable with **no initializer or zero initializer** | `static uint8_t x;` or `static uint8_t x = 0;` | No startup just zeroes the RAM range; no flash copy needed |
| `.bss` | Same rule for **global** variables that are zero/uninitialized | `uint8_t g;` or `uint8_t g = 0;` (outside functions) | No same zero-fill at startup |
The rule is not about `static` specifically it is about **lifetime and initial value**. Any variable whose lifetime spans the whole program (static locals, globals) goes into `.data` or `.bss`. The `static` keyword inside a function just forces that lifetime. A global variable without `static` has the same lifetime and follows the same rule.
`static_fav_num = 42` has an explicit non-zero initializer, so the linker puts it in `.data`. The startup code copies the initial value `42` from flash into RAM address `0x200005a8` before `main()` runs. If it had been `static uint8_t static_fav_num;` (no initializer), the linker would put it in `.bss` instead and the startup code would zero that RAM region no flash copy needed.
**Why is there no pool entry for the address of `regular_fav_num`?**
Because `regular_fav_num` is an **automatic (stack) variable** it has no fixed address. Automatic variables are allocated on the call stack at runtime, relative to the current stack pointer (`sp`). Their address changes every time the function runs and is different on every call. There is no constant 32-bit address to put in a pool.
In this specific case the compiler went even further: it **never gave `regular_fav_num` a stack slot at all**. The compiler saw that the value is always `42` when printed, so it baked the constant `42` directly into the `movs r1, #42` instruction. The variable lives only in a register at the moment it is needed — no stack space reserved, no pool entry, no load-store cycle.
**You can prove this with three GDB checks:**
**Proof 1 `info locals` shows the value came from a register, not a stack slot:**
```gdb
(gdb) b *0x10000264
(gdb) c
(gdb) info locals
regular_fav_num = 42 '*'
static_fav_num = 42 '*'
BUTTON_GPIO = 15
LED_GPIO = 16
pressed = <optimized out>
The '*' suffix after 42 means GDB retrieved the value from a register (specifically r1, which holds 42 at this point), not from a stack slot. The variable has no fixed memory address — GDB can report its current value only because the right register happens to hold it right now. Notice that pressed is the one showing <optimized out> — the compiler removed it from tracking entirely because its value is never needed after the ternary expression.
Proof 2 The function prologue allocates no stack space for locals:
(gdb) x/2i 0x10000234
0x10000234 <main>: push {r4, lr}
0x10000236 <main+2>: bl 0x10003014 <stdio_init_all>
The very first instruction is push {r4, lr}, that saves r4 (a callee-saved register) and lr (return address). There is no sub sp, #N instruction after it. On ARM, allocating stack space for local variables requires subtracting from sp to reserve room. If regular_fav_num lived on the stack, there would be a sub sp, #4 (or similar) here. There isn't, so no stack slot was ever created.
Proof 3 The loop body contains no store to the stack for regular_fav_num:
(gdb) x/6i 0x10000264
=> 0x10000264 <main+48>: movs r1, #42 @ 0x2a
// constant 42 baked directly into instruction
0x10000266 <main+50>: ldr r0, [pc, #44] @ (0x10000294 <main+96>)
// load format string address from pool
0x10000268 <main+52>: bl 0x100031a4 <__wrap_printf>
// call printf
0x1000026c <main+56>: ldrb r1, [r4, #0]
// load static_fav_num from RAM
0x1000026e <main+58>: ldr r0, [pc, #40] @ (0x10000298 <main+100>)
// load format string address from pool
0x10000270 <main+60>: bl 0x100031a4 <__wrap_printf>
call printf
For regular_fav_num, the value 42 goes straight into r1 via movs, it is never written to memory and never read back from memory. There is no strb r1, [sp, #N] storing it to the stack, and no ldrb r1, [sp, #N] loading it back. The variable exists only in the source code. In the binary it is just a number in an instruction encoding.
static_fav_num is the exact opposite. It lives in the .data section of RAM, a fixed, known address (0x200005a8) that is set at link time and never changes for the life of the program. That fixed address is exactly the kind of value that cannot fit in a 16-bit Thumb instruction, so the linker writes it into the literal pool and the CPU fetches it with ldr r4, [pc, #44].
IMPORTANT: Static variables are not on the heap. The heap is for dynamic memory (
malloc/free), memory whose lifetime is controlled explicitly at runtime. Static and global variables live in the.datasection (if initialized) or.bsssection (if zero/uninitialized), a region of RAM with a fixed layout determined at link time, not at runtime.
Memory regions for our two variables:
+------------------------------------------+
| FLASH (read-only, 0x10000000+) |
| - Code (instructions) |
| - Literal pool (our 3 words) |
| - String literals ("regular_fav_num...")|
+------------------------------------------+
| RAM .data section (0x20000000+) |
| - static_fav_num @ 0x200005a8 <- pool |
+------------------------------------------+
| STACK (grows down from 0x20082000) |
| - regular_fav_num (if not optimized) |
| frame-relative, no fixed address |
+------------------------------------------+
| HEAP (grows up, between stack and .bss) |
| - malloc/free only, nothing here today |
+------------------------------------------+
The pool ends at 0x1000029b. At 0x1000029d a new function begins, specifically gpio_set_function. The first two instructions, push {r4} (save caller's r4 to the stack) and mov.w r4, #256 (load a working constant), are a standard SDK function prologue, nothing to do with main. The ldr r3, [pc, #48]i r/. at 0x100002a3 (gpio_set_function+6) is that next function loading from its own literal pool further ahead in flash — proof that every function manages its own pool.
Step 10: Examine Register Values
After breaking at 0x10000264, check the registers:
(gdb) i r
What you will actually see:
r4=0x200005a8— the static variable's fixed RAM address. This was loaded from the literal pool in the function prologue and never changes across loop iterations.r1= leftover value from the previousprintfcall (e.g.0x40039044). The breakpoint is stopped atmovs r1, #42— that instruction has not executed yet, sor1does not yet hold42.r3= leftover from prior operations (e.g.0x10). It will be used later in the loop body to load, increment, and storestatic_fav_num, but not yet.pc=0x10000264— the loop top, pointing at themovs r1, #42instruction.
You can confirm r4 holds the right address and value:
(gdb) x/1db $r4
0x200005a8 <striped_spin_lock_num>: 42
The <striped_spin_lock_num> label is just the nearest named global symbol in the SDK that GDB finds at that address — the actual variable stored there is our static_fav_num.
Step 11: Watch the Static Variable Change
Now that we know the static variable lives at 0x200005a8, examine it directly:
(gdb) x/1db 0x200005a8
0x200005a8 <striped_spin_lock_num>: 42
Step through a full loop iteration (back to 0x10000264) and re-examine:
(gdb) c
(gdb) x/1db 0x200005a8
0x200005a8 <striped_spin_lock_num>: 43
The value incremented from 42 to 43! Each loop iteration, the adds r3, #1 at 0x1000027c bumps it by 1, and strb r3, [r4, #0] at 0x1000027e writes it back to RAM.
Step 12: Examine GPIO State
Read the GPIO input register to see the button state:
(gdb) x/1wx 0xd0000004
0xd0000004: 0x00008003
The SIO GPIO input register at 0xd0000004 shows the current state of all GPIO pins. Bit 15 corresponds to our button on GPIO 15. To extract just bit 15:
(gdb) p/x (*(unsigned int *)0xd0000004 >> 15) & 1
$1 = 0x1
- Returns
1when button is not pressed (pull-up holds it HIGH) - Returns
0when button is pressed (connected to GND)
TRY IT!
Part 7: Understanding the Assembly
Now that we've explored the binary in GDB, let's make sense of the key patterns.
Step 13: Analyze the Regular Variable
In GDB, examine the code at the start of the loop:
(gdb) x/5i 0x10000262
0x10000262 <main+46>: ldr r4, [pc, #44] @ (0x10000290 <main+92>)
=> 0x10000264 <main+48>: movs r1, #42 @ 0x2a
0x10000266 <main+50>: ldr r0, [pc, #44] @ (0x10000294 <main+96>)
0x10000268 <main+52>: bl 0x100031a4 <__wrap_printf>
0x1000026c <main+56>: ldrb r1, [r4, #0]
Look for this instruction:
0x10000264 <main+48>: movs r1, #42 @ 0x2a
This loads the value 0x2a (42 in decimal) directly into register r1 for the first printf call.
Key insight: The compiler never allocated stack space for regular_fav_num. Since it is always 42 when printed, the compiler bakes the constant directly into the movs r1, #42 instruction. The regular_fav_num++ after the print is also removed because it has no observable effect — the variable is recreated as 42 on the next loop iteration anyway.
Step 14: Analyze the Static Variable
Examine the static variable operations in the second half of the loop body:
(gdb) x/10i 0x10000274
0x10000274 <main+64>: mov.w r1, #3489660928 @ 0xd0000000
0x10000278 <main+68>: ldrb r3, [r4, #0]
0x1000027a <main+70>: movs r2, #16
0x1000027c <main+72>: adds r3, #1
0x1000027e <main+74>: strb r3, [r4, #0]
0x10000280 <main+76>: ldr r3, [r1, #4]
0x10000282 <main+78>: ubfx r3, r3, #15, #1
0x10000286 <main+82>: eor.w r3, r3, #1
0x1000028a <main+86>: mcrr 0, 4, r2, r3, cr0
0x1000028e <main+90>: b.n 0x10000264 <main+48>
Look for the load-increment-store pattern using r4 (which holds the static variable's RAM address):
...
0x10000278 <main+68>: ldrb r3, [r4, #0]
0x1000027a <main+70>: movs r2, #16
0x1000027c <main+72>: adds r3, #1
0x1000027e <main+74>: strb r3, [r4, #0]
...
Note that r4 was loaded earlier at 0x10000262 via ldr r4, [pc, #44] - this pulled the static variable's RAM address (0x200005a8) from the literal pool at 0x10000290.
Key insight: The static variable lives at a fixed RAM address (0x200005a8). It's loaded, incremented, and stored back. The regular variable, by contrast, never gets a stack slot — the compiler holds it in a register and bakes the constant 42 directly into the instruction, so there is nothing to load or store.
Verify the static variable value which should be 43:
(gdb) x/1db 0x200005a8
0x200005a8 <striped_spin_lock_num>: 43
Step 15: Analyze the GPIO Logic
Examine the GPIO input/output code:
(gdb) x/10i 0x10000274
0x10000274 <main+64>: mov.w r1, #3489660928 @ 0xd0000000
0x10000278 <main+68>: ldrb r3, [r4, #0]
0x1000027a <main+70>: movs r2, #16
0x1000027c <main+72>: adds r3, #1
0x1000027e <main+74>: strb r3, [r4, #0]
0x10000280 <main+76>: ldr r3, [r1, #4]
0x10000282 <main+78>: ubfx r3, r3, #15, #1
0x10000286 <main+82>: eor.w r3, r3, #1
0x1000028a <main+86>: mcrr 0, 4, r2, r3, cr0
0x1000028e <main+90>: b.n 0x10000264 <main+48>
Breaking this down:
| Address | Instruction | Purpose |
|---|---|---|
0x10000274 |
mov.w r1, #0xd0000000 |
Load SIO (Single-cycle I/O) base address into r1 |
0x10000278 |
ldrb r3, [r4, #0] |
Load static_fav_num from RAM into r3 |
0x1000027a |
movs r2, #16 |
Load LED pin number (16) into r2 for later |
0x1000027c |
adds r3, #1 |
Increment static_fav_num by 1 |
0x1000027e |
strb r3, [r4, #0] |
Store incremented value back to RAM |
0x10000280 |
ldr r3, [r1, #4] |
Read GPIO input state (SIO_GPIO_IN at offset 0x04) |
0x10000282 |
ubfx r3, r3, #15, #1 |
Extract bit 15 (GPIO 15 = button) |
0x10000286 |
eor.w r3, r3, #1 |
XOR with 1 to invert (implements ? 0 : 1) |
0x1000028a |
mcrr 0, 4, r2, r3, cr0 |
Write r3 (button) and r2 (pin 16) to GPIO output |
0x1000028e |
b.n 0x10000264 |
Loop back to start (while (true)) |
Tip: Notice how the compiler interleaves the static variable increment with the GPIO logic. It loads the SIO base address (
r1) before doing the increment, and sets upr2 = 16(LED pin) in between. This is called instruction scheduling - the compiler reorders instructions to avoid pipeline stalls while waiting for memory reads.
Step 16: Find the Infinite Loop
The last instruction at 0x1000028e is already covered in the table above:
0x1000028e <main+90>: b.n 0x10000264 <main+48>
This is an unconditional branch back to 0x10000264 (the movs r1, #42 at the top of the loop) - this is the while (true) in our code! There is no pop or bx lr to return from main because the loop never exits.
Part 8: Hacking the Binary with a Hex Editor
Now for the fun part - we'll patch the .bin file directly using a hex editor!
Tip: Why a hex editor? GDB cannot write to flash memory - the
0x10000000+address range where program instructions live. Tryingset *(char *)0x10000264 = 0x2bin GDB givesWriting to flash memory forbidden in this context. To make permanent patches that survive a power cycle, we edit the.binfile directly with a hex editor and re-flash it.
Step 17: Open the Binary in a Hex Editor
- Open HxD (or your preferred hex editor: ImHex, 010 Editor, etc.)
- Click File -> Open
- Navigate to
C:\Users\flare-vm\Desktop\Embedded-Hacking-main\0x0014_static-variables\build\ - Open
0x0014_static-variables.bin
Step 18: Calculate the File Offset
The binary is loaded at base address 0x10000000. To find the file offset of any address:
file_offset = address - 0x10000000
For example:
- Address
0x10000264-> file offset0x264(612 in decimal) - Address
0x10000286-> file offset0x286(646 in decimal)
Step 19: Hack #1 - Change regular_fav_num from 42 to 43
From our GDB analysis, we know the instruction at 0x10000264 is:
movs r1, #0x2a -> bytes: 2a 21
To change the value from 42 (0x2a) to 43 (0x2b):
- In HxD, open
C:\Users\flare-vm\Desktop\Embedded-Hacking-main\0x0014_static-variables\build\0x0014_static-variables.bin - Press Ctrl+G (Go to offset)
- Enter offset:
264 - You should see the byte
2Aat this position - Change
2Ato2B - The instruction is now
movs r1, #0x2b(43 in decimal)
?? How Thumb encoding works: In
movs r1, #imm8, the immediate value is the first byte, and the opcode21is the second byte. So the bytes2a 21encodemovs r1, #0x2a.
Step 20: Hack #2 - Invert the Button Logic
Understand the Encoding
From GDB, we found the eor.w r3, r3, #1 instruction at 0x10000286 that inverts the button value. Examine the exact bytes:
(gdb) x/4bx 0x10000286
0x10000286 <main+82>: 0x83 0xf0 0x01 0x03
This is the 32-bit Thumb-2 encoding of eor.w r3, r3, #1. The bytes break down as:
+-----------------------------------------------------------------+
| eor.w r3, r3, #1 -> bytes: 83 F0 01 03 |
| |
| Byte 0: 0x83 -?? |
| Byte 1: 0xF0 -+ First halfword (opcode + source register) |
| Byte 2: 0x01 ---- Immediate value (#1) ?? CHANGE THIS |
| Byte 3: 0x03 ---- Destination register (r3) |
| |
+-----------------------------------------------------------------+
To change eor.w r3, r3, #1 to eor.w r3, r3, #0 (making XOR do nothing):
The file offset is 0x10000286 - 0x10000000 = 0x286. The immediate byte is the 3rd byte of the instruction, so: 0x286 + 2 = 0x288.
To change eor.w r3, r3, #1 to eor.w r3, r3, #0:
- In HxD, press Ctrl+G (Go to offset)
- Enter offset:
288(the third byte of the 4-byte instruction) - You should see the byte
01at this position - Change
01to00
?? Why offset
0x288and not0x286? The immediate value#1is in the third byte of the 4-byte instruction. The instruction starts at file offset0x286, so the immediate byte is at0x286 + 2 = 0x288.
Now the logic is permanently changed:
- Button released (input = 1):
1 XOR 0 = 1-> LED ON - Button pressed (input = 0):
0 XOR 0 = 0-> LED OFF
This is the opposite of the original behavior!
Step 21: Save the Patched Binary
- Click File -> Save As
- Save as
0x0014_static-variables-h.binin the build directory - Close the hex editor
Part 9: Converting and Flashing the Hacked Binary
Step 22: Convert to UF2 Format
Open a terminal and navigate to your project directory:
cd C:\Users\flare-vm\Desktop\Embedded-Hacking-main\0x0014_static-variables
Run the conversion command:
python ..\uf2conv.py build\0x0014_static-variables-h.bin --base 0x10000000 --family 0xe48bff59 --output build\hacked.uf2
What this command means:
uf2conv.py= the conversion script (in the parentEmbedded-Hackingdirectory)--base 0x10000000= the XIP base address where code runs from--family 0xe48bff59= the RP2350 family ID--output build\hacked.uf2= the output filename
Step 23: Flash the Hacked Binary
- Hold BOOTSEL and plug in your Pico 2
- Drag and drop
hacked.uf2onto the RPI-RP2 drive - Open your serial monitor
Step 24: Verify the Hacks
Check the serial output:
regular_fav_num: 43 ?? Changed from 42!
static_fav_num: 42
regular_fav_num: 43
static_fav_num: 43
...
Check the LED behavior:
- LED should now be ON by default (when button is NOT pressed)
- LED should turn OFF when you press the button
BOOM! We successfully:
- Changed the printed value from 42 to 43
- Inverted the LED/button logic
Part 10: Summary and Review
What We Accomplished
- Learned about static variables - How they persist across function calls and loop iterations
- Understood memory layout - Stack vs static storage vs heap
- Configured GPIO inputs - Using pull-up resistors and reading button states
- Analyzed compiled code in GDB - Saw how the compiler optimizes code
- Discovered function inlining -
gpio_pull_upbecamegpio_set_pulls - Hacked variable values - Changed 42 to 43 using a hex editor
- Inverted GPIO logic - Made LED behavior opposite
Static vs Automatic Variables
| Aspect | Automatic (Regular) | Static |
|---|---|---|
| Storage | Stack | Static storage (.data/.bss) |
| Lifetime | Block/function scope | Entire program |
| Initialization | Every time block entered | Once at program start |
| Persistence | Lost when scope exits | Retained between calls |
| Compiler view | May be optimized away | Always has memory location |
GPIO Input Configuration
+-----------------------------------------------------------------+
| GPIO Input Setup Steps |
| |
| 1. gpio_init(pin) - Initialize the pin |
| 2. gpio_set_dir(pin, GPIO_IN) - Set as input |
| 3. gpio_pull_up(pin) - Enable pull-up |
| OR gpio_pull_down(pin) - OR enable pull-down |
| 4. gpio_get(pin) - Read the state |
| |
+-----------------------------------------------------------------+
The Binary Hacking Workflow
+-----------------------------------------------------------------+
| 1. Analyze the binary with GDB |
| - Disassemble functions with x/Ni |
| - Identify key instructions and addresses |
+-----------------------------------------------------------------+
| 2. Understand compiler optimizations |
| - Some functions get inlined (gpio_pull_up -> gpio_set_pulls)|
| - Some variables are optimized away |
+-----------------------------------------------------------------+
| 3. Calculate file offsets |
| - file_offset = address - 0x10000000 |
+-----------------------------------------------------------------+
| 4. Patch the .bin file with a hex editor |
| - Open the .bin file in HxD / ImHex |
| - Go to the calculated offset |
| - Change the target byte(s) |
+-----------------------------------------------------------------+
| 5. Convert to UF2 |
| python uf2conv.py file.bin --base 0x10000000 |
| --family 0xe48bff59 --output hacked.uf2 |
+-----------------------------------------------------------------+
| 6. Flash and verify |
| - Hold BOOTSEL, plug in, drag UF2 |
| - Check serial output and button/LED behavior |
+-----------------------------------------------------------------+
Key Memory Addresses
| Address | Description |
|---|---|
0x10000234 |
Typical main() entry point |
0x10003014 |
stdio_init_all() function |
0x200005a8 |
Static variable storage (example) |
0xd0000000 |
SIO (Single-cycle I/O) base address |
Key Takeaways
-
Static variables persist - They keep their value between function calls and loop iterations.
-
Static storage ? heap - Static variables are in a fixed location, not dynamically allocated.
-
Compilers optimize aggressively - Regular variables may be optimized away if the compiler sees no effect.
-
Function inlining is common -
gpio_pull_upbecomesgpio_set_pullsin the binary. -
Pull-up resistors invert logic - Button pressed = LOW, button released = HIGH.
-
XOR is useful for inverting -
eor r3,r3,#0x1flips a bit between 0 and 1. -
Static variables have fixed addresses - You can find them in the .data section at known RAM addresses.
-
Overflow wraps around - A
uint8_tat 255 becomes 0 when incremented. -
UBFX extracts bits - Used to read a single GPIO pin from a register.
-
Binary patching is powerful - Change values and logic without source code!
Glossary
| Term | Definition |
|---|---|
| Automatic | Variable that's created and destroyed automatically (local vars) |
| eor/XOR | Exclusive OR - flips bits where operands differ |
| Floating Input | GPIO input with undefined voltage (reads random values) |
| Function Inlining | Compiler replaces function call with the function's code |
| gpio_get | Function to read the current state of a GPIO pin |
| Heap | Memory area for dynamic allocation (malloc/free) |
| Overflow | When a value exceeds its type's maximum and wraps around |
| Pull-Down | Resistor that holds a pin LOW when nothing drives it |
| Pull-Up | Resistor that holds a pin HIGH when nothing drives it |
| SIO | Single-cycle I/O - fast GPIO access on RP2350 |
| Stack | Memory area for local variables and function call frames |
| Static Storage | Fixed memory area for static and global variables |
| Static Variable | Variable declared with static that persists across calls |
| Ternary Operator | condition ? value_if_true : value_if_false |
| UBFX | Unsigned Bit Field Extract - extracts bits from a register |
| Varargs | Variable arguments - functions that take unlimited parameters |
Additional Resources
GPIO Input Reference
| Function | Purpose |
|---|---|
gpio_init(pin) |
Initialize GPIO pin |
gpio_set_dir(pin, GPIO_IN) |
Set pin as input |
gpio_set_dir(pin, GPIO_OUT) |
Set pin as output |
gpio_pull_up(pin) |
Enable internal pull-up |
gpio_pull_down(pin) |
Enable internal pull-down |
gpio_disable_pulls(pin) |
Disable all pull resistors |
gpio_get(pin) |
Read pin state (0 or 1) |
gpio_put(pin, value) |
Set pin output (0 or 1) |
Key Assembly Instructions
| Instruction | Description |
|---|---|
movs rN, #imm |
Move immediate value to register |
ldrb rN, [rM, #off] |
Load byte from memory |
strb rN, [rM, #off] |
Store byte to memory |
adds rN, #imm |
Add immediate value to register |
eor rN, rM, #imm |
Exclusive OR (XOR) with immediate |
ubfx rN, rM, #lsb, #w |
Extract unsigned bit field |
mcrr p0, ... |
Move to coprocessor (GPIO control on RP2350) |
b LABEL |
Unconditional branch (jump) |
Memory Map Quick Reference
| Address Range | Description |
|---|---|
0x10000000 |
XIP Flash (code execution) |
0x20000000-200005xx |
SRAM (.data section) |
0x20082000 |
Stack top (initial SP) |
0x40038000 |
PADS_BANK0 (pad configuration) |
0xd0000000 |
SIO (single-cycle I/O) |
Remember: Static variables are your friends when you need to remember values across function calls. But they also make your program's behavior more complex to analyze - which is exactly why we practice reverse engineering!
Happy hacking! ?