For debugging purposes it would be nice if mutex_t::val stored the current thread id, not a boolean. The test if the mutex is held would remain the same: mutex_t::val == 0 → mutex_t::val == KERNEL_PID_UNDEF (KERNEL_PID_UNDEF == 0), so I see no extra overhead. The locking would be no more difficult either: atomic_set_to_one(&mutex.val)(which uses atomic_cas) → atomic_cas(&mutex.val, KERNEL_PID_UNDEF, sched_active_pid).