NAME

SYNOPSIS

#include <u.h>
#include <libc.h>

void	lock(Lock *l)
int	canlock(Lock *l)
void	unlock(Lock *l)

void	qlock(QLock *l)
int	canqlock(QLock *l)
void	qunlock(QLock *l)

void	rlock(RWLock *l)
int	canrlock(RWLock *l)
void	runlock(RWLock *l)

void	wlock(RWLock *l)
int	canwlock(RWLock *l)
void	wunlock(RWLock *l)

typedef struct Rendez {
	QLock *l;
	...
} Rendez;

void	rsleep(Rendez *r)
int	rwakeup(Rendez *r)
int	rwakeupall(Rendez *r)

#include <thread.h>

typedef struct Ref {
	long ref;
} Ref;

void incref(Ref*)
long decref(Ref*)

DESCRIPTION

Locks are spin locks, QLocks and RWLocks are different types of queueing locks, and Rendezes are rendezvous points.

Locks and rendezvous points have trivial implementations in programs not using the thread library (see since such programs have no concurrency.

Used carelessly, spin locks can be expensive and can easily generate deadlocks. Their use is discouraged, especially in programs that use the thread library because they prevent context switches between threads.

Lock blocks until the lock has been obtained. Canlock is non-blocking. It tries to obtain a lock and returns a non-zero value if it was successful, 0 otherwise. Unlock releases a lock.

QLocks have the same interface but are not spin locks; instead if the lock is taken qlock will suspend execution of the calling thread until it is released.

Although Locks are the more primitive lock, they have limitations; for example, they cannot synchronize between tasks in the same proc. Use QLocks instead.

RWLocks manage access to a data structure that has distinct readers and writers. Rlock grants read access; runlock releases it. Wlock grants write access; wunlock releases it. Canrlock and canwlock are the non-blocking versions. There may be any number of simultaneous readers, but only one writer. Moreover, if write access is granted no one may have read access until write access is released.

All types of lock should be initialized to all zeros before use; this puts them in the unlocked state.

Rendezes are rendezvous points. Each Rendez r is protected by a QLock r->l, which must be held by the callers of rsleep, rwakeup, and rwakeupall. Rsleep atomically releases r->l and suspends execution of the calling task. After resuming execution, rsleep will reacquire r->l before returning. If any processes are sleeping on r, rwakeup wakes one of them. It returns 1 if a process was awakened, 0 if not. Rwakeupall wakes all processes sleeping on r, returning the number of processes awakened. Rwakeup and rwakeupall do not release r->l and do not suspend execution of the current task.

Before use, Rendezes should be initialized to all zeros except for r->l pointer, which should point at the QLock that will guard r.

A Ref contains a long that can be incremented and decremented atomically: Incref increments the Ref in one atomic operation. Decref atomically decrements the Ref and returns zero if the resulting value is zero, non-zero otherwise.

SOURCE

BUGS

On pthreads-based systems, the implementation of Lock never calls pthread_mutex_destroy to free the pthread_mutex_t's. This leads to resource leaks on FreeBSD 5 (though not on Linux 2.6, where pthread_mutex_destroy is a no-op).

On systems that do not have a usable pthreads implementation, the Lock implementation provided by libthread is still not exactly a spin lock. After each unsuccessful attempt, lock calls sleep(0) to yield the CPU; this handles the common case where some other process holds the lock. After a thousand unsuccessful attempts, lock sleeps for 100ms between attempts. Another another thousand unsuccessful attempts, lock sleeps for a full second between attempts. Locks are not intended to be held for long periods of time. The 100ms and full second sleeps are only heuristics to avoid tying up the CPU when a process deadlocks. As discussed above, if a lock is to be held for much more than a few instructions, the queueing lock types should be almost always be used.