Compilers — this critical piece of software acts as the very bridge between imagination and reality, enabling developers to run code on all sorts of hardware — everything from tiny embedded systems to massive data centers.
A compiler not only translates code, but also optimizes it — reordering, simplifying, and even removing instructions to improve performance.
While this is generally beneficial, it can sometimes clash with the programmer’s intent, especially in systems where precise control over hardware and memory is crucial (e.g., embedded systems).
This is where the volatile keyword comes into play, guiding the compiler to respect certain constraints to ensure correct behavior in concurrent and low-level code.
Here we have a simple piece of code where a flag, system_ready, is used to busy wait before starting the actual application code. This flag is accessed by other code files using extern to signal to the main program that it is ready to proceed.
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
#include "main.h"
uint8_t system_ready = 0;
int main() {
/* System Init */
// Busy wait until the flag is set and indicate by LED when ready
while (system_ready == 0);
HAL_GPIO_WritePin(LED_GPIO_Port, LED_Pin, GPIO_PIN_SET);
while (1) {
/* Looping user code */
}
}
On compilation, this is what we expect to see in the disassembly:
1
2
3
4
5
6
7
8
9
10
11
12
13
14
...
while (system_ready == 0);
800023a: 4b09 ldr r3, [pc, #36] ; (8000260 <main+0x38>) <--- Load address of 'system_ready' to R3 <-----+
800023c: 781b ldrb r3, [r3, #0] <--- Load a byte from the address to R3 |
800023e: 2b00 cmp r3, #0 <--- Compare R3 to 0 |
8000240: d0fb beq.n 800023a <main+0x12> <--- If equal, loop back to address 800023a ----+
HAL_GPIO_WritePin(LED_GPIO_Port, LED_Pin, GPIO_PIN_SET);
...
8000260: 20000028 .word 0x20000028 <--- RAM address of 'system_ready' variable
...
Every time, we load the variable system_ready from memory, compare it to 0 and then either loop or break out based on the outcome of the comparison (as intended by the programmer).
But, as we increase the optimization level to -O1 (or even -Os, which is the default for embedded systems), we see this instead
1
2
3
4
5
6
7
8
9
10
11
12
13
14
...
while (system_ready == 0);
8000236: 4b0b ldr r3, [pc, #44] ; (8000264 <main+0x40>) <--- Load address of 'system_ready' to R3
8000238: 781b ldrb r3, [r3, #0] <--- Load a byte from the address to R3
800023a: 2b00 cmp r3, #0 <--- Compare R3 to 0 <--------------------------+
800023c: d0fd beq.n 800023a <main+0x16> <--- If equal, loop back to address 800023a ----+
HAL_GPIO_WritePin(LED_GPIO_Port, LED_Pin, GPIO_PIN_SET);
...
8000264: 20000028 .word 0x20000028 <--- RAM address of 'system_ready' variable
...
Here, the variable is loaded only once from memory, kind of like caching the variable, and the same cached value is repeatedly compared to 0. The compiler assumes the variable won’t change unexpectedly and optimizes the code by removing repeated memory reads. As a result, if the variable hasn’t become non-zero before that initial load, the code gets stuck in an infinite loop, with no way to break out.
Compilation results and outputs may vary depending on the compiler version, target architecture, optimization settings, and maybe even black magic.
This behavior would be entirely unexpected to the programmer, potentially resulting in hours — or even days — of unnecessary and most likely unproductive debugging.
volatile is the magic word that instructs the compiler to be mindful of the variable during optimization. It informs the compiler that its value can change at any time, outside the normal program flow, and hence not to optimize anything related to that variable.
Let’s make a small modification to the example above — we’ll add the volatile type qualifier to the system_ready variable, as shown below:
1
volatile uint8_t system_ready = 0;
This tells the compiler not to optimize access to the variable and to check the actual memory each time, rather than relying on a cached value. The resulting disassembly (compiled with -O3) is shown below:
1
2
3
4
5
6
7
8
9
10
11
12
13
14
...
while (system_ready == 0);
800023a: 4a0a ldr r2, [pc, #40] ; (8000264 <main+0x3c>) <--- Load address of 'system_ready' to R2
800023c: 7813 ldrb r3, [r2, #0] <--- Load a byte from the address to R3 <-------+
800023e: 2b00 cmp r3, #0 <--- Compare R3 to 0 |
8000240: d0fc beq.n 800023c <main+0x14> <--- If equal, loop back to address 800023c ----+
HAL_GPIO_WritePin(LED_GPIO_Port, LED_Pin, GPIO_PIN_SET);
...
8000264: 20000028 .word 0x20000028 <--- RAM address of 'system_ready' variable
...
And just like that, a single keyword saves us from hours of head-scratching. volatile to the rescue!
The volatile keyword tells the compiler not to optimize reads / writes to a variable because its value might change unexpectedly (e.g., due to hardware or another thread).
When using higher optimization levels, compilers may overlook the importance of a seemingly unchanged variable, resulting in incorrect logic in the generated binary.
Until next time, may your pointers never be null.
]]>