Software Engineering Talk

Guest · Posted: Mon Jun 23, 2008 12:12 am Post subject: mutex implementation

I am implementing a kernel mutex which will block a thread if it
cannot acquire the mutex. Below is my attempt at this. I have just
wrapped the entire down() and up() functions within a common spinlock,
to stop race conditions between the actual mutex modification and the
wait queue operations. Is this code correct? And if so, is it
efficient (ie. is there a better way of doing this). I looked to the
Linux source for guidance but I found it too hard to follow.

void down(semaphore * sem)
{
spinlock_acquire(sem->spinlock);
if(atomic_dec(sem->count) >= 0)
{
spinlock_release(sem->spinlock);
return;
}
/*adds current thread t queue tail and marks as blocked */
add_waitq(current_thread, sem->waitq);
spinlock_release(sem->spinlock);
schedule();
}

void up(semaphore * sem)
{
spinlock_acquire(sem->spinlock);
if(atomic_inc(sem->count) <= 0)
{
/* removes thread from head of queue and marks runnable */
remove_waitq(sem->waitq);
}
spinlock_release(sem->spinlock);
}

Alexei A. Frounze · Guest

On Jun 22, 5:12 pm, boroph...@gmail.com wrote:

DG.Ward · Guest

Alexei A. Frounze wrote:

Maxim S. Shatskih · Guest