Hardware Specific Synchronization Instructions

Implement a mutex atomically

No other thread can change its value between testing and setting the mutex

Test-and-Set

while (*mutex == true) {}; //  Test if mutex is held
*mutex = true; // set it to true

Testing to setting the mutex should be done in one atomic step

Example

mutex_lock(bool *mutex) {
	while (xchg(mutex, true) == true) {}; //test and set
}

mutex_unlock(bool *mutex) {
	*mutex = false;
}

xchg(mutex, true) returns the old value of mutex

If returns true, then mutex was previously held
If returns false, then mutex was not held and we are now holding it
This is known as a Spinlock

A lot of variation for multithreaded code on different processors

x64 has Complex Instruction Set with total store order consistency

A thread can read its own write and becomes visible to all other threads
ARM has Reduces Instruction Set with relaxed consistency
Reads and writes can occur out of order, possible different order in different threads