illumos: manual page: rw

RWLOCK(9F) Kernel Functions for Drivers RWLOCK(9F)

NAME

rwlock, rw_init, rw_destroy, rw_enter, rw_exit, rw_tryenter,
rw_downgrade, rw_tryupgrade, rw_read_locked - readers/writer lock
functions

SYNOPSIS

#include <sys/ksynch.h>

void rw_init(krwlock_t *rwlp, char *name, krw_type_t type, void *arg);

void rw_destroy(krwlock_t *rwlp);

void rw_enter(krwlock_t *rwlp, krw_t enter_type);

void rw_exit(krwlock_t *rwlp);

int rw_tryenter(krwlock_t *rwlp, krw_t enter_type);

void rw_downgrade(krwlock_t *rwlp);

int rw_tryupgrade(krwlock_t *rwlp);

int rw_read_locked(krwlock_t *rwlp);

INTERFACE LEVEL

illumos DDI specific (illumos DDI).

PARAMETERS

rwlp
Pointer to a krwlock_t readers/writer lock.

name
Descriptive string. This is obsolete and should be NULL.
(Non-null strings are legal, but they're a waste of kernel
memory.)

type
Type of readers/writer lock.

arg
Type-specific argument for initialization function.

enter_type
One of the values RW_WRITER, RW_READER or
RW_READER_STARVEWRITER, indicating whether the lock is to
be acquired exclusively (RW_WRITER), non-exclusively
(RW_READER) or non-exclusively without regard to any
threads that may be blocked on exclusive access
(RW_READER_STARVEWRITER).

DESCRIPTION

A multiple-readers, single-writer lock is represented by the krwlock_t
data type. This type of lock will allow many threads to have simultaneous
read-only access to an object. Only one thread may have write access at
any one time. An object that is searched more frequently than it is
changed is a good candidate for a readers/writer lock.

Readers/writer locks are slightly more expensive than mutex locks, and
the advantage of multiple read access may not occur if the lock will only
be held for a short time.

The rw_init() function initializes a readers/writer lock. It is an error
to initialize a lock more than once. The type argument should be set to
RW_DRIVER. If the lock is used by the interrupt handler, the type-
specific argument, arg, should be the interrupt priority returned from
ddi_intr_get_pri(9F) or ddi_intr_get_softint_pri(9F). Note that arg
should be the value of the interrupt priority cast by calling the
DDI_INTR_PRI macro. If the lock is not used by any interrupt handler, the
argument should be NULL.

The rw_destroy() function releases any resources that might have been
allocated by rw_init(). It should be called before freeing the memory
containing the lock. The lock must not be held by any thread when it is
destroyed.

The rw_enter() function acquires the lock, and blocks if necessary. If
enter_type is RW_WRITER, the caller blocks if any thread holds the lock.
If enter_type is RW_READER, the caller blocks if there is a writer or a
thread attempting to enter for writing. If enter_type is
RW_READER_STARVEWRITER, the caller blocks only if there is a writer; if
the lock is held for reading and a thread is blocked attempting to enter
for writing, the caller will acquire the lock as a reader instead of
blocking on the pending writer.

NOTE: It is a programming error for any thread to acquire an rwlock as
RW_READER that it already holds. Doing so can deadlock the system: if
thread R acquires the lock as RW_READER, then thread W tries to acquire
the lock as a writer, W will set write-wanted and block. When R tries to
get its second read hold on the lock, it will honor the write-wanted bit
and block waiting for W; but W cannot run until R drops the lock. Thus
threads R and W deadlock. To opt out of this behavior -- that is, to
safely allow a lock to be grabbed recursively as a reader -- the lock
should be acquired as RW_READER_STARVEWRITER, which will allow R to get
its second read hold without regard for the write-wanted bit set by W.
Note that the RW_READER_STARVEWRITER behavior will starve writers in the
presence of infinite readers; it should be used with care, and only where
the default RW_READER behavior is unacceptable.

The rw_exit() function releases the lock and may wake up one or more
threads waiting on the lock.

The rw_tryenter() function attempts to enter the lock, like rw_enter(),
but never blocks. It returns a non-zero value if the lock was
successfully entered, and zero otherwise.

A thread that holds the lock exclusively (entered with RW_WRITER), may
call rw_downgrade() to convert to holding the lock non-exclusively (as if
entered with RW_READER). One or more waiting readers may be unblocked.

The rw_tryupgrade() function can be called by a thread that holds the
lock for reading to attempt to convert to holding it for writing. This
upgrade can only succeed if no other thread is holding the lock and no
other thread is blocked waiting to acquire the lock for writing.

The rw_read_locked() function returns non-zero if the calling thread
holds the lock for read, and zero if the caller holds the lock for write.
The caller must hold the lock. The system may panic if rw_read_locked()
is called for a lock that isn't held by the caller.

RETURN VALUES

0
rw_tryenter() could not obtain the lock without blocking.

0
rw_tryupgrade() was unable to perform the upgrade because of
other threads holding or waiting to hold the lock.

0
rw_read_locked() returns 0 if the lock is held by the caller
for write.

non-zero
from rw_read_locked() if the lock is held by the caller for
read.

non-zero
successful return from rw_tryenter() or rw_tryupgrade().

CONTEXT

These functions can be called from user, interrupt, or kernel context,
except for rw_init() and rw_destroy(), which can be called from user
context only.

NOTES

Compiling with _LOCKTEST or _MPSTATS defined no longer has any effect. To
gather lock statistics, see lockstat(8).

September 19, 2013 RWLOCK(9F)