int
_sx_slock(struct sx *sx, int opts, const char *file, int line)
{
int error = 0;
MPASS(curthread != NULL);
KASSERT(sx->sx_lock != SX_LOCK_DESTROYED,
("sx_slock() of destroyed sx @ %s:%d", file, line));
WITNESS_CHECKORDER(&sx->lock_object, LOP_NEWORDER, file, line, NULL);
error = __sx_slock(sx, opts, file, line);
if (!error) {
LOCK_LOG_LOCK("SLOCK", &sx->lock_object, 0, 0, file, line);
WITNESS_LOCK(&sx->lock_object, 0, file, line);
curthread->td_locks++;
}
return (error);
}
void
_rw_wunlock(struct rwlock *rw, const char *file, int line)
{
if (SCHEDULER_STOPPED())
return;
MPASS(curthread != NULL);
KASSERT(rw->rw_lock != RW_DESTROYED,
("rw_wunlock() of destroyed rwlock @ %s:%d", file, line));
_rw_assert(rw, RA_WLOCKED, file, line);
curthread->td_locks--;
WITNESS_UNLOCK(&rw->lock_object, LOP_EXCLUSIVE, file, line);
LOCK_LOG_LOCK("WUNLOCK", &rw->lock_object, 0, rw->rw_recurse, file,
line);
if (!rw_recursed(rw))
LOCKSTAT_PROFILE_RELEASE_LOCK(LS_RW_WUNLOCK_RELEASE, rw);
__rw_wunlock(rw, curthread, file, line);
}
void
_rw_wlock(struct rwlock *rw, const char *file, int line)
{
if (SCHEDULER_STOPPED())
return;
KASSERT(!TD_IS_IDLETHREAD(curthread),
("rw_wlock() by idle thread %p on rwlock %s @ %s:%d",
curthread, rw->lock_object.lo_name, file, line));
KASSERT(rw->rw_lock != RW_DESTROYED,
("rw_wlock() of destroyed rwlock @ %s:%d", file, line));
WITNESS_CHECKORDER(&rw->lock_object, LOP_NEWORDER | LOP_EXCLUSIVE, file,
line, NULL);
__rw_wlock(rw, curthread, file, line);
LOCK_LOG_LOCK("WLOCK", &rw->lock_object, 0, rw->rw_recurse, file, line);
WITNESS_LOCK(&rw->lock_object, LOP_EXCLUSIVE, file, line);
curthread->td_locks++;
}
/*
* Function versions of the inlined __mtx_* macros. These are used by
* modules and can also be called from assembly language if needed.
*/
void
_mtx_lock_flags(struct mtx *m, int opts, const char *file, int line)
{
MPASS(curthread != NULL);
KASSERT(m->mtx_lock != MTX_DESTROYED,
("mtx_lock() of destroyed mutex @ %s:%d", file, line));
KASSERT(LOCK_CLASS(&m->lock_object) == &lock_class_mtx_sleep,
("mtx_lock() of spin mutex %s @ %s:%d", m->lock_object.lo_name,
file, line));
WITNESS_CHECKORDER(&m->lock_object, opts | LOP_NEWORDER | LOP_EXCLUSIVE,
file, line, NULL);
_get_sleep_lock(m, curthread, opts, file, line);
LOCK_LOG_LOCK("LOCK", &m->lock_object, opts, m->mtx_recurse, file,
line);
WITNESS_LOCK(&m->lock_object, opts | LOP_EXCLUSIVE, file, line);
curthread->td_locks++;
}
void
_mtx_unlock_flags(struct mtx *m, int opts, const char *file, int line)
{
MPASS(curthread != NULL);
KASSERT(m->mtx_lock != MTX_DESTROYED,
("mtx_unlock() of destroyed mutex @ %s:%d", file, line));
KASSERT(LOCK_CLASS(&m->lock_object) == &lock_class_mtx_sleep,
("mtx_unlock() of spin mutex %s @ %s:%d", m->lock_object.lo_name,
file, line));
curthread->td_locks--;
WITNESS_UNLOCK(&m->lock_object, opts | LOP_EXCLUSIVE, file, line);
LOCK_LOG_LOCK("UNLOCK", &m->lock_object, opts, m->mtx_recurse, file,
line);
mtx_assert(m, MA_OWNED);
if (m->mtx_recurse == 0)
LOCKSTAT_PROFILE_RELEASE_LOCK(LS_MTX_UNLOCK_RELEASE, m);
_rel_sleep_lock(m, curthread, opts, file, line);
}
void
_sx_downgrade(struct sx *sx, const char *file, int line)
{
_sx_assert(sx, SX_XLOCKED, file, line);
mtx_lock(sx->sx_lock);
MPASS(sx->sx_cnt == -1);
WITNESS_DOWNGRADE(&sx->sx_object, 0, file, line);
sx->sx_cnt = 1;
sx->sx_xholder = NULL;
if (sx->sx_shrd_wcnt > 0)
cv_broadcast(&sx->sx_shrd_cv);
LOCK_LOG_LOCK("XDOWNGRADE", &sx->sx_object, 0, 0, file, line);
mtx_unlock(sx->sx_lock);
}
void
_rw_wunlock_cookie(volatile uintptr_t *c, const char *file, int line)
{
struct rwlock *rw;
if (SCHEDULER_STOPPED())
return;
rw = rwlock2rw(c);
KASSERT(rw->rw_lock != RW_DESTROYED,
("rw_wunlock() of destroyed rwlock @ %s:%d", file, line));
__rw_assert(c, RA_WLOCKED, file, line);
WITNESS_UNLOCK(&rw->lock_object, LOP_EXCLUSIVE, file, line);
LOCK_LOG_LOCK("WUNLOCK", &rw->lock_object, 0, rw->rw_recurse, file,
line);
__rw_wunlock(rw, curthread, file, line);
curthread->td_locks--;
}
void
__mtx_unlock_spin_flags(volatile uintptr_t *c, int opts, const char *file,
int line)
{
struct mtx *m;
m = mtxlock2mtx(c);
KASSERT(m->mtx_lock != MTX_DESTROYED,
("mtx_unlock_spin() of destroyed mutex @ %s:%d", file, line));
KASSERT(LOCK_CLASS(&m->lock_object) == &lock_class_mtx_spin,
("mtx_unlock_spin() of sleep mutex %s @ %s:%d",
m->lock_object.lo_name, file, line));
WITNESS_UNLOCK(&m->lock_object, opts | LOP_EXCLUSIVE, file, line);
LOCK_LOG_LOCK("UNLOCK", &m->lock_object, opts, m->mtx_recurse, file,
line);
mtx_assert(m, MA_OWNED);
__mtx_unlock_spin(m);
}
void
_mtx_lock_spin_flags(struct mtx *m, int opts, const char *file, int line)
{
MPASS(curthread != NULL);
KASSERT(m->mtx_lock != MTX_DESTROYED,
("mtx_lock_spin() of destroyed mutex @ %s:%d", file, line));
KASSERT(LOCK_CLASS(&m->lock_object) == &lock_class_mtx_spin,
("mtx_lock_spin() of sleep mutex %s @ %s:%d",
m->lock_object.lo_name, file, line));
if (mtx_owned(m))
KASSERT((m->lock_object.lo_flags & LO_RECURSABLE) != 0,
("mtx_lock_spin: recursed on non-recursive mutex %s @ %s:%d\n",
m->lock_object.lo_name, file, line));
WITNESS_CHECKORDER(&m->lock_object, opts | LOP_NEWORDER | LOP_EXCLUSIVE,
file, line, NULL);
_get_spin_lock(m, curthread, opts, file, line);
LOCK_LOG_LOCK("LOCK", &m->lock_object, opts, m->mtx_recurse, file,
line);
WITNESS_LOCK(&m->lock_object, opts | LOP_EXCLUSIVE, file, line);
}
void
__mtx_lock_spin_flags(volatile uintptr_t *c, int opts, const char *file,
int line)
{
struct mtx *m;
#ifdef SMP
uintptr_t tid, v;
#endif
m = mtxlock2mtx(c);
KASSERT(m->mtx_lock != MTX_DESTROYED,
("mtx_lock_spin() of destroyed mutex @ %s:%d", file, line));
KASSERT(LOCK_CLASS(&m->lock_object) == &lock_class_mtx_spin,
("mtx_lock_spin() of sleep mutex %s @ %s:%d",
m->lock_object.lo_name, file, line));
if (mtx_owned(m))
KASSERT((m->lock_object.lo_flags & LO_RECURSABLE) != 0 ||
(opts & MTX_RECURSE) != 0,
("mtx_lock_spin: recursed on non-recursive mutex %s @ %s:%d\n",
m->lock_object.lo_name, file, line));
opts &= ~MTX_RECURSE;
WITNESS_CHECKORDER(&m->lock_object, opts | LOP_NEWORDER | LOP_EXCLUSIVE,
file, line, NULL);
#ifdef SMP
spinlock_enter();
tid = (uintptr_t)curthread;
v = MTX_UNOWNED;
if (!_mtx_obtain_lock_fetch(m, &v, tid))
_mtx_lock_spin(m, v, opts, file, line);
else
LOCKSTAT_PROFILE_OBTAIN_LOCK_SUCCESS(spin__acquire,
m, 0, 0, file, line);
#else
__mtx_lock_spin(m, curthread, opts, file, line);
#endif
LOCK_LOG_LOCK("LOCK", &m->lock_object, opts, m->mtx_recurse, file,
line);
WITNESS_LOCK(&m->lock_object, opts | LOP_EXCLUSIVE, file, line);
}
int
_sx_slock(struct sx *sx, int opts, const char *file, int line)
{
int error = 0;
if (SCHEDULER_STOPPED())
return (0);
KASSERT(kdb_active != 0 || !TD_IS_IDLETHREAD(curthread),
("sx_slock() by idle thread %p on sx %s @ %s:%d",
curthread, sx->lock_object.lo_name, file, line));
KASSERT(sx->sx_lock != SX_LOCK_DESTROYED,
("sx_slock() of destroyed sx @ %s:%d", file, line));
WITNESS_CHECKORDER(&sx->lock_object, LOP_NEWORDER, file, line, NULL);
error = __sx_slock(sx, opts, file, line);
if (!error) {
LOCK_LOG_LOCK("SLOCK", &sx->lock_object, 0, 0, file, line);
WITNESS_LOCK(&sx->lock_object, 0, file, line);
curthread->td_locks++;
}
return (error);
}
int
_sx_xlock(struct sx *sx, int opts, const char *file, int line)
{
int error = 0;
if (SCHEDULER_STOPPED())
return (0);
MPASS(curthread != NULL);
KASSERT(sx->sx_lock != SX_LOCK_DESTROYED,
("sx_xlock() of destroyed sx @ %s:%d", file, line));
WITNESS_CHECKORDER(&sx->lock_object, LOP_NEWORDER | LOP_EXCLUSIVE, file,
line, NULL);
error = __sx_xlock(sx, curthread, opts, file, line);
if (!error) {
LOCK_LOG_LOCK("XLOCK", &sx->lock_object, 0, sx->sx_recurse,
file, line);
WITNESS_LOCK(&sx->lock_object, LOP_EXCLUSIVE, file, line);
curthread->td_locks++;
}
return (error);
}
void
__mtx_unlock_flags(volatile uintptr_t *c, int opts, const char *file, int line)
{
struct mtx *m;
if (SCHEDULER_STOPPED())
return;
m = mtxlock2mtx(c);
KASSERT(m->mtx_lock != MTX_DESTROYED,
("mtx_unlock() of destroyed mutex @ %s:%d", file, line));
KASSERT(LOCK_CLASS(&m->lock_object) == &lock_class_mtx_sleep,
("mtx_unlock() of spin mutex %s @ %s:%d", m->lock_object.lo_name,
file, line));
WITNESS_UNLOCK(&m->lock_object, opts | LOP_EXCLUSIVE, file, line);
LOCK_LOG_LOCK("UNLOCK", &m->lock_object, opts, m->mtx_recurse, file,
line);
mtx_assert(m, MA_OWNED);
__mtx_unlock(m, curthread, opts, file, line);
TD_LOCKS_DEC(curthread);
}
void
_rm_wlock_debug(struct rmlock *rm, const char *file, int line)
{
if (SCHEDULER_STOPPED())
return;
KASSERT(kdb_active != 0 || !TD_IS_IDLETHREAD(curthread),
("rm_wlock() by idle thread %p on rmlock %s @ %s:%d",
curthread, rm->lock_object.lo_name, file, line));
KASSERT(!rm_destroyed(rm),
("rm_wlock() of destroyed rmlock @ %s:%d", file, line));
_rm_assert(rm, RA_UNLOCKED, file, line);
WITNESS_CHECKORDER(&rm->lock_object, LOP_NEWORDER | LOP_EXCLUSIVE,
file, line, NULL);
_rm_wlock(rm);
LOCK_LOG_LOCK("RMWLOCK", &rm->lock_object, 0, 0, file, line);
WITNESS_LOCK(&rm->lock_object, LOP_EXCLUSIVE, file, line);
TD_LOCKS_INC(curthread);
}
void
_thread_lock_flags(struct thread *td, int opts, const char *file, int line)
{
struct mtx *m;
uintptr_t tid;
int i;
#ifdef LOCK_PROFILING
int contested = 0;
uint64_t waittime = 0;
#endif
#ifdef KDTRACE_HOOKS
uint64_t spin_cnt = 0;
#endif
i = 0;
tid = (uintptr_t)curthread;
for (;;) {
retry:
spinlock_enter();
m = td->td_lock;
KASSERT(m->mtx_lock != MTX_DESTROYED,
("thread_lock() of destroyed mutex @ %s:%d", file, line));
KASSERT(LOCK_CLASS(&m->lock_object) == &lock_class_mtx_spin,
("thread_lock() of sleep mutex %s @ %s:%d",
m->lock_object.lo_name, file, line));
if (mtx_owned(m))
KASSERT((m->lock_object.lo_flags & LO_RECURSABLE) != 0,
("thread_lock: recursed on non-recursive mutex %s @ %s:%d\n",
m->lock_object.lo_name, file, line));
WITNESS_CHECKORDER(&m->lock_object,
opts | LOP_NEWORDER | LOP_EXCLUSIVE, file, line, NULL);
while (!_obtain_lock(m, tid)) {
#ifdef KDTRACE_HOOKS
spin_cnt++;
#endif
if (m->mtx_lock == tid) {
m->mtx_recurse++;
break;
}
lock_profile_obtain_lock_failed(&m->lock_object,
&contested, &waittime);
/* Give interrupts a chance while we spin. */
spinlock_exit();
while (m->mtx_lock != MTX_UNOWNED) {
if (i++ < 10000000)
cpu_spinwait();
else if (i < 60000000 ||
kdb_active || panicstr != NULL)
DELAY(1);
else
_mtx_lock_spin_failed(m);
cpu_spinwait();
if (m != td->td_lock)
goto retry;
}
spinlock_enter();
}
if (m == td->td_lock)
break;
_rel_spin_lock(m); /* does spinlock_exit() */
#ifdef KDTRACE_HOOKS
spin_cnt++;
#endif
}
if (m->mtx_recurse == 0)
LOCKSTAT_PROFILE_OBTAIN_LOCK_SUCCESS(LS_MTX_SPIN_LOCK_ACQUIRE,
m, contested, waittime, (file), (line));
LOCK_LOG_LOCK("LOCK", &m->lock_object, opts, m->mtx_recurse, file,
line);
WITNESS_LOCK(&m->lock_object, opts | LOP_EXCLUSIVE, file, line);
LOCKSTAT_RECORD1(LS_THREAD_LOCK_SPIN, m, spin_cnt);
}
void
thread_lock_flags_(struct thread *td, int opts, const char *file, int line)
{
struct mtx *m;
uintptr_t tid;
struct lock_delay_arg lda;
#ifdef LOCK_PROFILING
int contested = 0;
uint64_t waittime = 0;
#endif
#ifdef KDTRACE_HOOKS
int64_t spin_time = 0;
#endif
tid = (uintptr_t)curthread;
if (SCHEDULER_STOPPED()) {
/*
* Ensure that spinlock sections are balanced even when the
* scheduler is stopped, since we may otherwise inadvertently
* re-enable interrupts while dumping core.
*/
spinlock_enter();
return;
}
lock_delay_arg_init(&lda, &mtx_spin_delay);
#ifdef KDTRACE_HOOKS
spin_time -= lockstat_nsecs(&td->td_lock->lock_object);
#endif
for (;;) {
retry:
spinlock_enter();
m = td->td_lock;
KASSERT(m->mtx_lock != MTX_DESTROYED,
("thread_lock() of destroyed mutex @ %s:%d", file, line));
KASSERT(LOCK_CLASS(&m->lock_object) == &lock_class_mtx_spin,
("thread_lock() of sleep mutex %s @ %s:%d",
m->lock_object.lo_name, file, line));
if (mtx_owned(m))
KASSERT((m->lock_object.lo_flags & LO_RECURSABLE) != 0,
("thread_lock: recursed on non-recursive mutex %s @ %s:%d\n",
m->lock_object.lo_name, file, line));
WITNESS_CHECKORDER(&m->lock_object,
opts | LOP_NEWORDER | LOP_EXCLUSIVE, file, line, NULL);
for (;;) {
if (m->mtx_lock == MTX_UNOWNED && _mtx_obtain_lock(m, tid))
break;
if (m->mtx_lock == tid) {
m->mtx_recurse++;
break;
}
#ifdef HWPMC_HOOKS
PMC_SOFT_CALL( , , lock, failed);
#endif
lock_profile_obtain_lock_failed(&m->lock_object,
&contested, &waittime);
/* Give interrupts a chance while we spin. */
spinlock_exit();
while (m->mtx_lock != MTX_UNOWNED) {
if (lda.spin_cnt < 10000000) {
lock_delay(&lda);
} else {
lda.spin_cnt++;
if (lda.spin_cnt < 60000000 ||
kdb_active || panicstr != NULL)
DELAY(1);
else
_mtx_lock_spin_failed(m);
cpu_spinwait();
}
if (m != td->td_lock)
goto retry;
}
spinlock_enter();
}
if (m == td->td_lock)
break;
__mtx_unlock_spin(m); /* does spinlock_exit() */
}
#ifdef KDTRACE_HOOKS
spin_time += lockstat_nsecs(&m->lock_object);
#endif
if (m->mtx_recurse == 0)
LOCKSTAT_PROFILE_OBTAIN_LOCK_SUCCESS(spin__acquire, m,
contested, waittime, file, line);
LOCK_LOG_LOCK("LOCK", &m->lock_object, opts, m->mtx_recurse, file,
line);
WITNESS_LOCK(&m->lock_object, opts | LOP_EXCLUSIVE, file, line);
#ifdef KDTRACE_HOOKS
if (spin_time != 0)
LOCKSTAT_RECORD1(thread__spin, m, spin_time);
#endif
}
void
thread_lock_flags_(struct thread *td, int opts, const char *file, int line)
{
struct mtx *m;
uintptr_t tid, v;
struct lock_delay_arg lda;
#ifdef LOCK_PROFILING
int contested = 0;
uint64_t waittime = 0;
#endif
#ifdef KDTRACE_HOOKS
int64_t spin_time = 0;
#endif
#if defined(KDTRACE_HOOKS) || defined(LOCK_PROFILING)
int doing_lockprof = 1;
#endif
tid = (uintptr_t)curthread;
if (SCHEDULER_STOPPED()) {
/*
* Ensure that spinlock sections are balanced even when the
* scheduler is stopped, since we may otherwise inadvertently
* re-enable interrupts while dumping core.
*/
spinlock_enter();
return;
}
lock_delay_arg_init(&lda, &mtx_spin_delay);
#ifdef LOCK_PROFILING
doing_lockprof = 1;
#elif defined(KDTRACE_HOOKS)
doing_lockprof = lockstat_enabled;
if (__predict_false(doing_lockprof))
spin_time -= lockstat_nsecs(&td->td_lock->lock_object);
#endif
for (;;) {
retry:
v = MTX_UNOWNED;
spinlock_enter();
m = td->td_lock;
thread_lock_validate(m, opts, file, line);
for (;;) {
if (_mtx_obtain_lock_fetch(m, &v, tid))
break;
if (v == MTX_UNOWNED)
continue;
if (v == tid) {
m->mtx_recurse++;
break;
}
#ifdef HWPMC_HOOKS
PMC_SOFT_CALL( , , lock, failed);
#endif
lock_profile_obtain_lock_failed(&m->lock_object,
&contested, &waittime);
/* Give interrupts a chance while we spin. */
spinlock_exit();
do {
if (lda.spin_cnt < 10000000) {
lock_delay(&lda);
} else {
lda.spin_cnt++;
if (lda.spin_cnt < 60000000 ||
kdb_active || panicstr != NULL)
DELAY(1);
else
_mtx_lock_spin_failed(m);
cpu_spinwait();
}
if (m != td->td_lock)
goto retry;
v = MTX_READ_VALUE(m);
} while (v != MTX_UNOWNED);
spinlock_enter();
}
if (m == td->td_lock)
break;
__mtx_unlock_spin(m); /* does spinlock_exit() */
}
LOCK_LOG_LOCK("LOCK", &m->lock_object, opts, m->mtx_recurse, file,
line);
WITNESS_LOCK(&m->lock_object, opts | LOP_EXCLUSIVE, file, line);
#if defined(KDTRACE_HOOKS) || defined(LOCK_PROFILING)
if (__predict_true(!doing_lockprof))
return;
#endif
#ifdef KDTRACE_HOOKS
spin_time += lockstat_nsecs(&m->lock_object);
#endif
if (m->mtx_recurse == 0)
LOCKSTAT_PROFILE_OBTAIN_LOCK_SUCCESS(spin__acquire, m,
contested, waittime, file, line);
#ifdef KDTRACE_HOOKS
if (lda.spin_cnt != 0)
LOCKSTAT_RECORD1(thread__spin, m, spin_time);
#endif
}
void
thread_lock_flags_(struct thread *td, int opts, const char *file, int line)
{
struct mtx *m;
uintptr_t tid;
int i;
#ifdef LOCK_PROFILING
int contested = 0;
uint64_t waittime = 0;
#endif
#ifdef KDTRACE_HOOKS
int64_t spin_time = 0;
#endif
i = 0;
tid = (uintptr_t)curthread;
if (SCHEDULER_STOPPED())
return;
#ifdef KDTRACE_HOOKS
spin_time -= lockstat_nsecs(&td->td_lock->lock_object);
#endif
for (;;) {
retry:
spinlock_enter();
m = td->td_lock;
KASSERT(m->mtx_lock != MTX_DESTROYED,
("thread_lock() of destroyed mutex @ %s:%d", file, line));
KASSERT(LOCK_CLASS(&m->lock_object) == &lock_class_mtx_spin,
("thread_lock() of sleep mutex %s @ %s:%d",
m->lock_object.lo_name, file, line));
if (mtx_owned(m))
KASSERT((m->lock_object.lo_flags & LO_RECURSABLE) != 0,
("thread_lock: recursed on non-recursive mutex %s @ %s:%d\n",
m->lock_object.lo_name, file, line));
WITNESS_CHECKORDER(&m->lock_object,
opts | LOP_NEWORDER | LOP_EXCLUSIVE, file, line, NULL);
while (!_mtx_obtain_lock(m, tid)) {
if (m->mtx_lock == tid) {
m->mtx_recurse++;
break;
}
#ifdef HWPMC_HOOKS
PMC_SOFT_CALL( , , lock, failed);
#endif
lock_profile_obtain_lock_failed(&m->lock_object,
&contested, &waittime);
/* Give interrupts a chance while we spin. */
spinlock_exit();
while (m->mtx_lock != MTX_UNOWNED) {
if (i++ < 10000000)
cpu_spinwait();
else if (i < 60000000 ||
kdb_active || panicstr != NULL)
DELAY(1);
else
_mtx_lock_spin_failed(m);
cpu_spinwait();
if (m != td->td_lock)
goto retry;
}
spinlock_enter();
}
if (m == td->td_lock)
break;
__mtx_unlock_spin(m); /* does spinlock_exit() */
}
#ifdef KDTRACE_HOOKS
spin_time += lockstat_nsecs(&m->lock_object);
#endif
if (m->mtx_recurse == 0)
LOCKSTAT_PROFILE_OBTAIN_LOCK_SUCCESS(spin__acquire, m,
contested, waittime, file, line);
LOCK_LOG_LOCK("LOCK", &m->lock_object, opts, m->mtx_recurse, file,
line);
WITNESS_LOCK(&m->lock_object, opts | LOP_EXCLUSIVE, file, line);
LOCKSTAT_RECORD1(thread__spin, m, spin_time);
}
void
_rw_runlock_cookie(volatile uintptr_t *c, const char *file, int line)
{
struct rwlock *rw;
struct turnstile *ts;
uintptr_t x, v, queue;
if (SCHEDULER_STOPPED())
return;
rw = rwlock2rw(c);
KASSERT(rw->rw_lock != RW_DESTROYED,
("rw_runlock() of destroyed rwlock @ %s:%d", file, line));
__rw_assert(c, RA_RLOCKED, file, line);
WITNESS_UNLOCK(&rw->lock_object, 0, file, line);
LOCK_LOG_LOCK("RUNLOCK", &rw->lock_object, 0, 0, file, line);
/* TODO: drop "owner of record" here. */
for (;;) {
/*
* See if there is more than one read lock held. If so,
* just drop one and return.
*/
x = rw->rw_lock;
if (RW_READERS(x) > 1) {
if (atomic_cmpset_rel_ptr(&rw->rw_lock, x,
x - RW_ONE_READER)) {
if (LOCK_LOG_TEST(&rw->lock_object, 0))
CTR4(KTR_LOCK,
"%s: %p succeeded %p -> %p",
__func__, rw, (void *)x,
(void *)(x - RW_ONE_READER));
break;
}
continue;
}
/*
* If there aren't any waiters for a write lock, then try
* to drop it quickly.
*/
if (!(x & RW_LOCK_WAITERS)) {
MPASS((x & ~RW_LOCK_WRITE_SPINNER) ==
RW_READERS_LOCK(1));
if (atomic_cmpset_rel_ptr(&rw->rw_lock, x,
RW_UNLOCKED)) {
if (LOCK_LOG_TEST(&rw->lock_object, 0))
CTR2(KTR_LOCK, "%s: %p last succeeded",
__func__, rw);
break;
}
continue;
}
/*
* Ok, we know we have waiters and we think we are the
* last reader, so grab the turnstile lock.
*/
turnstile_chain_lock(&rw->lock_object);
v = rw->rw_lock & (RW_LOCK_WAITERS | RW_LOCK_WRITE_SPINNER);
MPASS(v & RW_LOCK_WAITERS);
/*
* Try to drop our lock leaving the lock in a unlocked
* state.
*
* If you wanted to do explicit lock handoff you'd have to
* do it here. You'd also want to use turnstile_signal()
* and you'd have to handle the race where a higher
* priority thread blocks on the write lock before the
* thread you wakeup actually runs and have the new thread
* "steal" the lock. For now it's a lot simpler to just
* wakeup all of the waiters.
*
* As above, if we fail, then another thread might have
* acquired a read lock, so drop the turnstile lock and
* restart.
*/
x = RW_UNLOCKED;
if (v & RW_LOCK_WRITE_WAITERS) {
queue = TS_EXCLUSIVE_QUEUE;
x |= (v & RW_LOCK_READ_WAITERS);
} else
queue = TS_SHARED_QUEUE;
if (!atomic_cmpset_rel_ptr(&rw->rw_lock, RW_READERS_LOCK(1) | v,
x)) {
turnstile_chain_unlock(&rw->lock_object);
continue;
}
if (LOCK_LOG_TEST(&rw->lock_object, 0))
CTR2(KTR_LOCK, "%s: %p last succeeded with waiters",
__func__, rw);
/*
* Ok. The lock is released and all that's left is to
* wake up the waiters. Note that the lock might not be
* free anymore, but in that case the writers will just
* block again if they run before the new lock holder(s)
* release the lock.
*/
ts = turnstile_lookup(&rw->lock_object);
MPASS(ts != NULL);
turnstile_broadcast(ts, queue);
turnstile_unpend(ts, TS_SHARED_LOCK);
turnstile_chain_unlock(&rw->lock_object);
break;
}
LOCKSTAT_PROFILE_RELEASE_RWLOCK(rw__release, rw, LOCKSTAT_READER);
curthread->td_locks--;
curthread->td_rw_rlocks--;
}
void
__rw_rlock(volatile uintptr_t *c, const char *file, int line)
{
struct rwlock *rw;
struct turnstile *ts;
#ifdef ADAPTIVE_RWLOCKS
volatile struct thread *owner;
int spintries = 0;
int i;
#endif
#ifdef LOCK_PROFILING
uint64_t waittime = 0;
int contested = 0;
#endif
uintptr_t v;
#ifdef KDTRACE_HOOKS
uintptr_t state;
uint64_t spin_cnt = 0;
uint64_t sleep_cnt = 0;
int64_t sleep_time = 0;
int64_t all_time = 0;
#endif
if (SCHEDULER_STOPPED())
return;
rw = rwlock2rw(c);
KASSERT(kdb_active != 0 || !TD_IS_IDLETHREAD(curthread),
("rw_rlock() by idle thread %p on rwlock %s @ %s:%d",
curthread, rw->lock_object.lo_name, file, line));
KASSERT(rw->rw_lock != RW_DESTROYED,
("rw_rlock() of destroyed rwlock @ %s:%d", file, line));
KASSERT(rw_wowner(rw) != curthread,
("rw_rlock: wlock already held for %s @ %s:%d",
rw->lock_object.lo_name, file, line));
WITNESS_CHECKORDER(&rw->lock_object, LOP_NEWORDER, file, line, NULL);
#ifdef KDTRACE_HOOKS
all_time -= lockstat_nsecs(&rw->lock_object);
state = rw->rw_lock;
#endif
for (;;) {
/*
* Handle the easy case. If no other thread has a write
* lock, then try to bump up the count of read locks. Note
* that we have to preserve the current state of the
* RW_LOCK_WRITE_WAITERS flag. If we fail to acquire a
* read lock, then rw_lock must have changed, so restart
* the loop. Note that this handles the case of a
* completely unlocked rwlock since such a lock is encoded
* as a read lock with no waiters.
*/
v = rw->rw_lock;
if (RW_CAN_READ(v)) {
/*
* The RW_LOCK_READ_WAITERS flag should only be set
* if the lock has been unlocked and write waiters
* were present.
*/
if (atomic_cmpset_acq_ptr(&rw->rw_lock, v,
v + RW_ONE_READER)) {
if (LOCK_LOG_TEST(&rw->lock_object, 0))
CTR4(KTR_LOCK,
"%s: %p succeed %p -> %p", __func__,
rw, (void *)v,
(void *)(v + RW_ONE_READER));
break;
}
continue;
}
#ifdef KDTRACE_HOOKS
spin_cnt++;
#endif
#ifdef HWPMC_HOOKS
PMC_SOFT_CALL( , , lock, failed);
#endif
lock_profile_obtain_lock_failed(&rw->lock_object,
&contested, &waittime);
#ifdef ADAPTIVE_RWLOCKS
/*
* If the owner is running on another CPU, spin until
* the owner stops running or the state of the lock
* changes.
*/
if ((v & RW_LOCK_READ) == 0) {
owner = (struct thread *)RW_OWNER(v);
if (TD_IS_RUNNING(owner)) {
if (LOCK_LOG_TEST(&rw->lock_object, 0))
CTR3(KTR_LOCK,
"%s: spinning on %p held by %p",
__func__, rw, owner);
KTR_STATE1(KTR_SCHED, "thread",
sched_tdname(curthread), "spinning",
"lockname:\"%s\"", rw->lock_object.lo_name);
while ((struct thread*)RW_OWNER(rw->rw_lock) ==
owner && TD_IS_RUNNING(owner)) {
cpu_spinwait();
#ifdef KDTRACE_HOOKS
spin_cnt++;
#endif
}
KTR_STATE0(KTR_SCHED, "thread",
sched_tdname(curthread), "running");
continue;
}
} else if (spintries < rowner_retries) {
spintries++;
KTR_STATE1(KTR_SCHED, "thread", sched_tdname(curthread),
"spinning", "lockname:\"%s\"",
rw->lock_object.lo_name);
for (i = 0; i < rowner_loops; i++) {
v = rw->rw_lock;
if ((v & RW_LOCK_READ) == 0 || RW_CAN_READ(v))
break;
cpu_spinwait();
}
#ifdef KDTRACE_HOOKS
spin_cnt += rowner_loops - i;
#endif
KTR_STATE0(KTR_SCHED, "thread", sched_tdname(curthread),
"running");
if (i != rowner_loops)
continue;
}
#endif
/*
* Okay, now it's the hard case. Some other thread already
* has a write lock or there are write waiters present,
* acquire the turnstile lock so we can begin the process
* of blocking.
*/
ts = turnstile_trywait(&rw->lock_object);
/*
* The lock might have been released while we spun, so
* recheck its state and restart the loop if needed.
*/
v = rw->rw_lock;
if (RW_CAN_READ(v)) {
turnstile_cancel(ts);
continue;
}
#ifdef ADAPTIVE_RWLOCKS
/*
* The current lock owner might have started executing
* on another CPU (or the lock could have changed
* owners) while we were waiting on the turnstile
* chain lock. If so, drop the turnstile lock and try
* again.
*/
if ((v & RW_LOCK_READ) == 0) {
owner = (struct thread *)RW_OWNER(v);
if (TD_IS_RUNNING(owner)) {
turnstile_cancel(ts);
continue;
}
}
#endif
/*
* The lock is held in write mode or it already has waiters.
*/
MPASS(!RW_CAN_READ(v));
/*
* If the RW_LOCK_READ_WAITERS flag is already set, then
* we can go ahead and block. If it is not set then try
* to set it. If we fail to set it drop the turnstile
* lock and restart the loop.
*/
if (!(v & RW_LOCK_READ_WAITERS)) {
if (!atomic_cmpset_ptr(&rw->rw_lock, v,
v | RW_LOCK_READ_WAITERS)) {
turnstile_cancel(ts);
continue;
}
if (LOCK_LOG_TEST(&rw->lock_object, 0))
CTR2(KTR_LOCK, "%s: %p set read waiters flag",
__func__, rw);
}
/*
* We were unable to acquire the lock and the read waiters
* flag is set, so we must block on the turnstile.
*/
if (LOCK_LOG_TEST(&rw->lock_object, 0))
CTR2(KTR_LOCK, "%s: %p blocking on turnstile", __func__,
rw);
#ifdef KDTRACE_HOOKS
sleep_time -= lockstat_nsecs(&rw->lock_object);
#endif
turnstile_wait(ts, rw_owner(rw), TS_SHARED_QUEUE);
#ifdef KDTRACE_HOOKS
sleep_time += lockstat_nsecs(&rw->lock_object);
sleep_cnt++;
#endif
if (LOCK_LOG_TEST(&rw->lock_object, 0))
CTR2(KTR_LOCK, "%s: %p resuming from turnstile",
__func__, rw);
}
#ifdef KDTRACE_HOOKS
all_time += lockstat_nsecs(&rw->lock_object);
if (sleep_time)
LOCKSTAT_RECORD4(rw__block, rw, sleep_time,
LOCKSTAT_READER, (state & RW_LOCK_READ) == 0,
(state & RW_LOCK_READ) == 0 ? 0 : RW_READERS(state));
/* Record only the loops spinning and not sleeping. */
if (spin_cnt > sleep_cnt)
LOCKSTAT_RECORD4(rw__spin, rw, all_time - sleep_time,
LOCKSTAT_READER, (state & RW_LOCK_READ) == 0,
(state & RW_LOCK_READ) == 0 ? 0 : RW_READERS(state));
#endif
/*
* TODO: acquire "owner of record" here. Here be turnstile dragons
* however. turnstiles don't like owners changing between calls to
* turnstile_wait() currently.
*/
LOCKSTAT_PROFILE_OBTAIN_RWLOCK_SUCCESS(rw__acquire, rw, contested,
waittime, file, line, LOCKSTAT_READER);
LOCK_LOG_LOCK("RLOCK", &rw->lock_object, 0, 0, file, line);
WITNESS_LOCK(&rw->lock_object, 0, file, line);
curthread->td_locks++;
curthread->td_rw_rlocks++;
}
/*
* Downgrade an unrecursed exclusive lock into a single shared lock.
*/
void
sx_downgrade_(struct sx *sx, const char *file, int line)
{
uintptr_t x;
int wakeup_swapper;
if (SCHEDULER_STOPPED())
return;
KASSERT(sx->sx_lock != SX_LOCK_DESTROYED,
("sx_downgrade() of destroyed sx @ %s:%d", file, line));
_sx_assert(sx, SA_XLOCKED | SA_NOTRECURSED, file, line);
#ifndef INVARIANTS
if (sx_recursed(sx))
panic("downgrade of a recursed lock");
#endif
WITNESS_DOWNGRADE(&sx->lock_object, 0, file, line);
/*
* Try to switch from an exclusive lock with no shared waiters
* to one sharer with no shared waiters. If there are
* exclusive waiters, we don't need to lock the sleep queue so
* long as we preserve the flag. We do one quick try and if
* that fails we grab the sleepq lock to keep the flags from
* changing and do it the slow way.
*
* We have to lock the sleep queue if there are shared waiters
* so we can wake them up.
*/
x = sx->sx_lock;
if (!(x & SX_LOCK_SHARED_WAITERS) &&
atomic_cmpset_rel_ptr(&sx->sx_lock, x, SX_SHARERS_LOCK(1) |
(x & SX_LOCK_EXCLUSIVE_WAITERS))) {
LOCK_LOG_LOCK("XDOWNGRADE", &sx->lock_object, 0, 0, file, line);
return;
}
/*
* Lock the sleep queue so we can read the waiters bits
* without any races and wakeup any shared waiters.
*/
sleepq_lock(&sx->lock_object);
/*
* Preserve SX_LOCK_EXCLUSIVE_WAITERS while downgraded to a single
* shared lock. If there are any shared waiters, wake them up.
*/
wakeup_swapper = 0;
x = sx->sx_lock;
atomic_store_rel_ptr(&sx->sx_lock, SX_SHARERS_LOCK(1) |
(x & SX_LOCK_EXCLUSIVE_WAITERS));
if (x & SX_LOCK_SHARED_WAITERS)
wakeup_swapper = sleepq_broadcast(&sx->lock_object, SLEEPQ_SX,
0, SQ_SHARED_QUEUE);
sleepq_release(&sx->lock_object);
LOCK_LOG_LOCK("XDOWNGRADE", &sx->lock_object, 0, 0, file, line);
LOCKSTAT_RECORD0(sx__downgrade, sx);
if (wakeup_swapper)
kick_proc0();
}
/*
* Downgrade a write lock into a single read lock.
*/
void
__rw_downgrade(volatile uintptr_t *c, const char *file, int line)
{
struct rwlock *rw;
struct turnstile *ts;
uintptr_t tid, v;
int rwait, wwait;
if (SCHEDULER_STOPPED())
return;
rw = rwlock2rw(c);
KASSERT(rw->rw_lock != RW_DESTROYED,
("rw_downgrade() of destroyed rwlock @ %s:%d", file, line));
__rw_assert(c, RA_WLOCKED | RA_NOTRECURSED, file, line);
#ifndef INVARIANTS
if (rw_recursed(rw))
panic("downgrade of a recursed lock");
#endif
WITNESS_DOWNGRADE(&rw->lock_object, 0, file, line);
/*
* Convert from a writer to a single reader. First we handle
* the easy case with no waiters. If there are any waiters, we
* lock the turnstile and "disown" the lock.
*/
tid = (uintptr_t)curthread;
if (atomic_cmpset_rel_ptr(&rw->rw_lock, tid, RW_READERS_LOCK(1)))
goto out;
/*
* Ok, we think we have waiters, so lock the turnstile so we can
* read the waiter flags without any races.
*/
turnstile_chain_lock(&rw->lock_object);
v = rw->rw_lock & RW_LOCK_WAITERS;
rwait = v & RW_LOCK_READ_WAITERS;
wwait = v & RW_LOCK_WRITE_WAITERS;
MPASS(rwait | wwait);
/*
* Downgrade from a write lock while preserving waiters flag
* and give up ownership of the turnstile.
*/
ts = turnstile_lookup(&rw->lock_object);
MPASS(ts != NULL);
if (!wwait)
v &= ~RW_LOCK_READ_WAITERS;
atomic_store_rel_ptr(&rw->rw_lock, RW_READERS_LOCK(1) | v);
/*
* Wake other readers if there are no writers pending. Otherwise they
* won't be able to acquire the lock anyway.
*/
if (rwait && !wwait) {
turnstile_broadcast(ts, TS_SHARED_QUEUE);
turnstile_unpend(ts, TS_EXCLUSIVE_LOCK);
} else
turnstile_disown(ts);
turnstile_chain_unlock(&rw->lock_object);
out:
curthread->td_rw_rlocks++;
LOCK_LOG_LOCK("WDOWNGRADE", &rw->lock_object, 0, 0, file, line);
LOCKSTAT_RECORD0(rw__downgrade, rw);
}
