Example, Volatile in GCC

Below shows how GCC optimized the code with a non-volatile variable, (which caused a bug ).
Lesson: do not forget put volatile to your hardware register variables.

// In below code "stat" and "tbr" are registers could be updated by hardware
static void helx_serial_putc(const char c)
{
 int done = 0;

 if ( c == 0xFF ) {
  return;
 }

 while ( done == 0 ) {                           
  if ( stat->bits.txEmpty ) {
   tbr->bits.data = c;            
   done = 1;
  }
 }
}


// When "stat" is defined as volatile
volatile union uart_stat *stat = (void *)rUART_STAT;

00000000 :
   0: e35000ff  cmp r0, #255 ; 0xff
   4: 012fff1e  bxeq lr
   8: e59f3024  ldr r3, [pc, #36] ; 34 
   c: e5931000  ldr r1, [r3]
  10: e59f3020  ldr r3, [pc, #32] ; 38 
  14: e5933000  ldr r3, [r3]
  18: e5d12000  ldrb r2, [r1]
  1c: e3120040  tst r2, #64 ; 0x40
  20: 0afffffc  beq 18 
  24: e5d32000  ldrb r2, [r3]
  28: e7c72010  bfi r2, r0, #0, #8
  2c: e5c32000  strb r2, [r3]
  30: e12fff1e  bx lr


// When "stat" is defined as none volatile, 
// dead loop assembly code is generated, see offset 0x28
union uart_stat *stat = (void *)rUART_STAT;

00000000 :
   0: e35000ff  cmp r0, #255 ; 0xff
   4: 012fff1e  bxeq lr
   8: e59f201c  ldr r2, [pc, #28] ; 2c 
   c: e59f301c  ldr r3, [pc, #28] ; 30 
  10: e5922000  ldr r2, [r2]
  14: e5933000  ldr r3, [r3]
  18: e5d22000  ldrb r2, [r2]
  1c: e3120040  tst r2, #64 ; 0x40
  20: 15c30000  strbne r0, [r3]
  24: 112fff1e  bxne lr
  28: eafffffe  b 28