plan9fox/sys/man/2/lock

.TH LOCK 2
.SH NAME
lock, canlock, unlock,
qlock, canqlock, qunlock,
rlock, canrlock, runlock,
wlock, canwlock, wunlock,
rsleep, rwakeup, rwakeupall,
incref, decref
\- spin locks, queueing rendezvous locks, reader-writer locks, rendezvous points, and reference counts
.SH SYNOPSIS
.ft L
.nf
#include <u.h>
#include <libc.h>
.PP
.ft L
.nf
void	lock(Lock *l)
int	canlock(Lock *l)
void	unlock(Lock *l)
.PP
.ft L
.nf
void	qlock(QLock *l)
int	canqlock(QLock *l)
void	qunlock(QLock *l)
.PP
.ft L
.nf
void	rlock(RWLock *l)
int	canrlock(RWLock *l)
void	runlock(RWLock *l)
.PP
.ft L
.nf
void	wlock(RWLock *l)
int	canwlock(RWLock *l)
void	wunlock(RWLock *l)
.PP
.ft L
.nf
typedef struct Rendez {
	QLock *l;
	\fI...\fP
} Rendez;
.PP
.ft L
.nf
void	rsleep(Rendez *r)
int	rwakeup(Rendez *r)
int	rwakeupall(Rendez *r)
.PP
.ft L
#include <thread.h>
.PP
.ft L
.nf
typedef struct Ref {
	long ref;
} Ref;
.PP
.ft L
.nf
void incref(Ref*)
long decref(Ref*)
.fi
.SH DESCRIPTION
These routines are used  to synchronize processes sharing memory.
.PP
.B Locks
are spin locks,
.B QLocks
and
.B RWLocks
are different types of queueing rendezvous locks,
and
.B Rendezes
are rendezvous points.
.PP
Locks and rendezvous points work in regular programs as
well as programs that use the thread library
(see
.IR thread (2)).
The thread library replaces the
.IR rendezvous (2)
system call
with its own implementation,
.IR threadrendezvous ,
so that threads as well as processes may be synchronized by locking calls
in threaded programs.
.PP
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.
.PP
.I Lock
blocks until the lock has been obtained.
.I Canlock
is non-blocking.
It tries to obtain a lock and returns a non-zero value if it
was successful, 0 otherwise.
.I Unlock
releases a lock.
.PP
.B QLocks
have the same interface but are not spin locks; instead if the lock is taken
.I qlock
will suspend execution of the calling task until it is released.
.PP
Although
.B Locks
are the more primitive lock, they have limitations; for example,
they cannot synchronize between tasks in the same
.IR proc .
Use
.B QLocks
instead.
.PP
.B RWLocks
manage access to a data structure that has distinct readers and writers.
.I Rlock
grants read access;
.I runlock
releases it.
.I Wlock
grants write access;
.I wunlock
releases it.
.I Canrlock
and
.I 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.
.PP
All types of lock should be initialized to all zeros before use; this
puts them in the unlocked state.
.PP
.B Rendezes
are rendezvous points.  Each
.B Rendez
.I r
is protected by a
.B QLock
.IB r -> l \fR,
which must be held by the callers of
.IR rsleep ,
.IR rwakeup ,
and
.IR rwakeupall .
.I Rsleep
atomically releases
.IB r -> l
and suspends execution of the calling task.
After resuming execution,
.I rsleep
will reacquire
.IB r -> l
before returning.
If any processes are sleeping on
.IR r ,
.I rwakeup
wakes one of them.
it returns 1 if a process was awakened, 0 if not.
.I Rwakeupall
wakes all processes sleeping on
.IR r ,
returning the number of processes awakened.
.I Rwakeup
and
.I rwakeupall
do not release
.IB r -> l
and do not suspend execution of the current task.
.PP
Before use,
.B Rendezes
should be initialized to all zeros except for
.IB r -> l
pointer, which should point at the
.B QLock
that will guard
.IR r .
It is important that this
.B QLock
is the same one that protects the rendezvous condition; see the example.
.PP
A
.B Ref
contains a
.B long
that can be incremented and decremented atomically:
.I Incref
increments the
.I Ref
in one atomic operation.
.I Decref
atomically decrements the
.B Ref
and returns zero if the resulting value is zero, non-zero otherwise.
.SH EXAMPLE
Implement a buffered single-element channel using 
.I rsleep
and
.IR rwakeup :
.IP
.EX
.ta +4n +4n +4n
typedef struct Chan
{
	QLock l;
	Rendez full, empty;
	int val, haveval;
} Chan;
.EE
.IP
.EX
.ta +4n +4n +4n
Chan*
mkchan(void)
{
	Chan *c;

	c = mallocz(sizeof *c, 1);
	c->full.l = &c->l;
	c->empty.l = &c->l;
	return c;
}
.EE
.IP
.EX
.ta +4n +4n +4n
void
send(Chan *c, int val)
{
	qlock(&c->l);
	while(c->haveval)
		rsleep(&c->full);
	c->haveval = 1;
	c->val = val;
	rwakeup(&c->empty);  /* no longer empty */
	qunlock(&c->l);
}
.EE
.IP
.EX
.ta +4n +4n +4n
int
recv(Chan *c)
{
	int v;

	qlock(&c->l);
	while(!c->haveval)
		rsleep(&c->empty);
	c->haveval = 0;
	v = c->val;
	rwakeup(&c->full);  /* no longer full */
	qunlock(&c->l);
	return v;
}
.EE
.LP
Note that the 
.B QLock
protecting the
.B Chan
is the same
.B QLock
used for the 
.BR Rendez ;
this ensures that wakeups are not missed.
.SH SOURCE
.B /sys/src/libc/port/lock.c
.br
.B /sys/src/libc/9sys/qlock.c
.br
.B /sys/src/libthread/ref.c
.SH SEE ALSO
.I rfork
in
.IR fork (2)
.SH BUGS
.B Locks
are not strictly spin locks.
After each unsuccessful attempt,
.I lock
calls
.B sleep(0)
to yield the CPU; this handles the common case
where some other process holds the lock.
After a thousand unsuccessful attempts,
.I lock
sleeps for 100ms between attempts.
After another thousand unsuccessful attempts,
.I lock
sleeps for a full second between attempts.
.B 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.
.PP
It is an error for a program to
.I fork
when it holds a lock in shared memory, since this will result
in two processes holding the same lock at the same time,
which should not happen.