int
__rw_try_wlock(volatile uintptr_t *c, const char *file, int line)
{
struct rwlock *rw;
int rval;
if (SCHEDULER_STOPPED())
return (1);
rw = rwlock2rw(c);
KASSERT(kdb_active != 0 || !TD_IS_IDLETHREAD(curthread),
("rw_try_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_try_wlock() of destroyed rwlock @ %s:%d", file, line));
if (rw_wlocked(rw) &&
(rw->lock_object.lo_flags & LO_RECURSABLE) != 0) {
rw->rw_recurse++;
rval = 1;
} else
rval = atomic_cmpset_acq_ptr(&rw->rw_lock, RW_UNLOCKED,
(uintptr_t)curthread);
LOCK_LOG_TRY("WLOCK", &rw->lock_object, 0, rval, file, line);
if (rval) {
WITNESS_LOCK(&rw->lock_object, LOP_EXCLUSIVE | LOP_TRYLOCK,
file, line);
curthread->td_locks++;
}
return (rval);
}
int
_rw_try_wlock(struct rwlock *rw, const char *file, int line)
{
int rval;
if (SCHEDULER_STOPPED())
return (1);
KASSERT(rw->rw_lock != RW_DESTROYED,
("rw_try_wlock() of destroyed rwlock @ %s:%d", file, line));
if (rw_wlocked(rw) &&
(rw->lock_object.lo_flags & LO_RECURSABLE) != 0) {
rw->rw_recurse++;
rval = 1;
} else
rval = atomic_cmpset_acq_ptr(&rw->rw_lock, RW_UNLOCKED,
(uintptr_t)curthread);
LOCK_LOG_TRY("WLOCK", &rw->lock_object, 0, rval, file, line);
if (rval) {
WITNESS_LOCK(&rw->lock_object, LOP_EXCLUSIVE | LOP_TRYLOCK,
file, line);
curthread->td_locks++;
}
return (rval);
}
/*
* This function is called if the first try at releasing a write lock failed.
* This means that one of the 2 waiter bits must be set indicating that at
* least one thread is waiting on this lock.
*/
void
_rw_wunlock_hard(struct rwlock *rw, uintptr_t tid, const char *file, int line)
{
struct turnstile *ts;
uintptr_t v;
int queue;
if (SCHEDULER_STOPPED())
return;
if (rw_wlocked(rw) && rw_recursed(rw)) {
rw->rw_recurse--;
if (LOCK_LOG_TEST(&rw->lock_object, 0))
CTR2(KTR_LOCK, "%s: %p unrecursing", __func__, rw);
return;
}
KASSERT(rw->rw_lock & (RW_LOCK_READ_WAITERS | RW_LOCK_WRITE_WAITERS),
("%s: neither of the waiter flags are set", __func__));
if (LOCK_LOG_TEST(&rw->lock_object, 0))
CTR2(KTR_LOCK, "%s: %p contested", __func__, rw);
turnstile_chain_lock(&rw->lock_object);
ts = turnstile_lookup(&rw->lock_object);
MPASS(ts != NULL);
/*
* Use the same algo as sx locks for now. Prefer waking up shared
* waiters if we have any over writers. This is probably not ideal.
*
* 'v' is the value we are going to write back to rw_lock. If we
* have waiters on both queues, we need to preserve the state of
* the waiter flag for the queue we don't wake up. For now this is
* hardcoded for the algorithm mentioned above.
*
* In the case of both readers and writers waiting we wakeup the
* readers but leave the RW_LOCK_WRITE_WAITERS flag set. If a
* new writer comes in before a reader it will claim the lock up
* above. There is probably a potential priority inversion in
* there that could be worked around either by waking both queues
* of waiters or doing some complicated lock handoff gymnastics.
*/
v = RW_UNLOCKED;
if (rw->rw_lock & RW_LOCK_WRITE_WAITERS) {
queue = TS_EXCLUSIVE_QUEUE;
v |= (rw->rw_lock & RW_LOCK_READ_WAITERS);
} else
queue = TS_SHARED_QUEUE;
/* Wake up all waiters for the specific queue. */
if (LOCK_LOG_TEST(&rw->lock_object, 0))
CTR3(KTR_LOCK, "%s: %p waking up %s waiters", __func__, rw,
queue == TS_SHARED_QUEUE ? "read" : "write");
turnstile_broadcast(ts, queue);
atomic_store_rel_ptr(&rw->rw_lock, v);
turnstile_unpend(ts, TS_EXCLUSIVE_LOCK);
turnstile_chain_unlock(&rw->lock_object);
}
/*
* This function is called when we are unable to obtain a write lock on the
* first try. This means that at least one other thread holds either a
* read or write lock.
*/
void
__rw_wlock_hard(volatile uintptr_t *c, uintptr_t tid, const char *file,
int line)
{
struct rwlock *rw;
struct turnstile *ts;
#ifdef ADAPTIVE_RWLOCKS
volatile struct thread *owner;
int spintries = 0;
int i;
#endif
uintptr_t v, x;
#ifdef LOCK_PROFILING
uint64_t waittime = 0;
int contested = 0;
#endif
#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);
if (rw_wlocked(rw)) {
KASSERT(rw->lock_object.lo_flags & LO_RECURSABLE,
("%s: recursing but non-recursive rw %s @ %s:%d\n",
__func__, rw->lock_object.lo_name, file, line));
rw->rw_recurse++;
if (LOCK_LOG_TEST(&rw->lock_object, 0))
CTR2(KTR_LOCK, "%s: %p recursing", __func__, rw);
return;
}
if (LOCK_LOG_TEST(&rw->lock_object, 0))
CTR5(KTR_LOCK, "%s: %s contested (lock=%p) at %s:%d", __func__,
rw->lock_object.lo_name, (void *)rw->rw_lock, file, line);
#ifdef KDTRACE_HOOKS
all_time -= lockstat_nsecs(&rw->lock_object);
state = rw->rw_lock;
#endif
while (!_rw_write_lock(rw, tid)) {
#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 lock is write locked and the owner is
* running on another CPU, spin until the owner stops
* running or the state of the lock changes.
*/
v = rw->rw_lock;
owner = (struct thread *)RW_OWNER(v);
if (!(v & RW_LOCK_READ) && 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;
}
if ((v & RW_LOCK_READ) && RW_READERS(v) &&
spintries < rowner_retries) {
if (!(v & RW_LOCK_WRITE_SPINNER)) {
if (!atomic_cmpset_ptr(&rw->rw_lock, v,
v | RW_LOCK_WRITE_SPINNER)) {
continue;
}
}
spintries++;
KTR_STATE1(KTR_SCHED, "thread", sched_tdname(curthread),
"spinning", "lockname:\"%s\"",
rw->lock_object.lo_name);
for (i = 0; i < rowner_loops; i++) {
if ((rw->rw_lock & RW_LOCK_WRITE_SPINNER) == 0)
break;
cpu_spinwait();
}
KTR_STATE0(KTR_SCHED, "thread", sched_tdname(curthread),
"running");
#ifdef KDTRACE_HOOKS
spin_cnt += rowner_loops - i;
#endif
if (i != rowner_loops)
continue;
}
#endif
ts = turnstile_trywait(&rw->lock_object);
v = rw->rw_lock;
#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)) {
owner = (struct thread *)RW_OWNER(v);
if (TD_IS_RUNNING(owner)) {
turnstile_cancel(ts);
continue;
}
}
#endif
/*
* Check for the waiters flags about this rwlock.
* If the lock was released, without maintain any pending
* waiters queue, simply try to acquire it.
* If a pending waiters queue is present, claim the lock
* ownership and maintain the pending queue.
*/
x = v & (RW_LOCK_WAITERS | RW_LOCK_WRITE_SPINNER);
if ((v & ~x) == RW_UNLOCKED) {
x &= ~RW_LOCK_WRITE_SPINNER;
if (atomic_cmpset_acq_ptr(&rw->rw_lock, v, tid | x)) {
if (x)
turnstile_claim(ts);
else
turnstile_cancel(ts);
break;
}
turnstile_cancel(ts);
continue;
}
/*
* If the RW_LOCK_WRITE_WAITERS flag isn't set, then try to
* set it. If we fail to set it, then loop back and try
* again.
*/
if (!(v & RW_LOCK_WRITE_WAITERS)) {
if (!atomic_cmpset_ptr(&rw->rw_lock, v,
v | RW_LOCK_WRITE_WAITERS)) {
turnstile_cancel(ts);
continue;
}
if (LOCK_LOG_TEST(&rw->lock_object, 0))
CTR2(KTR_LOCK, "%s: %p set write waiters flag",
__func__, rw);
}
/*
* We were unable to acquire the lock and the write 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_EXCLUSIVE_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 ADAPTIVE_RWLOCKS
spintries = 0;
#endif
}
#ifdef KDTRACE_HOOKS
all_time += lockstat_nsecs(&rw->lock_object);
if (sleep_time)
LOCKSTAT_RECORD4(rw__block, rw, sleep_time,
LOCKSTAT_WRITER, (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
LOCKSTAT_PROFILE_OBTAIN_RWLOCK_SUCCESS(rw__acquire, rw, contested,
waittime, file, line, LOCKSTAT_WRITER);
}