/*
* Attempt to do a non-blocking upgrade from a read lock to a write
* lock. This will only succeed if this thread holds a single read
* lock. Returns true if the upgrade succeeded and false otherwise.
*/
int
_rw_try_upgrade(struct rwlock *rw, const char *file, int line)
{
uintptr_t v, x, tid;
struct turnstile *ts;
int success;
if (SCHEDULER_STOPPED())
return (1);
KASSERT(rw->rw_lock != RW_DESTROYED,
("rw_try_upgrade() of destroyed rwlock @ %s:%d", file, line));
_rw_assert(rw, RA_RLOCKED, file, line);
/*
* Attempt to switch from one reader to a writer. If there
* are any write waiters, then we will have to lock the
* turnstile first to prevent races with another writer
* calling turnstile_wait() before we have claimed this
* turnstile. So, do the simple case of no waiters first.
*/
tid = (uintptr_t)curthread;
success = 0;
for (;;) {
v = rw->rw_lock;
if (RW_READERS(v) > 1)
break;
if (!(v & RW_LOCK_WAITERS)) {
success = atomic_cmpset_ptr(&rw->rw_lock, v, tid);
if (!success)
continue;
break;
}
/*
* Ok, we think we have waiters, so lock the turnstile.
*/
ts = turnstile_trywait(&rw->lock_object);
v = rw->rw_lock;
if (RW_READERS(v) > 1) {
turnstile_cancel(ts);
break;
}
/*
* Try to switch from one reader to a writer again. This time
* we honor the current state of the waiters flags.
* If we obtain the lock with the flags set, then claim
* ownership of the turnstile.
*/
x = rw->rw_lock & RW_LOCK_WAITERS;
success = atomic_cmpset_ptr(&rw->rw_lock, v, tid | x);
if (success) {
if (x)
turnstile_claim(ts);
else
turnstile_cancel(ts);
break;
}
turnstile_cancel(ts);
}
LOCK_LOG_TRY("WUPGRADE", &rw->lock_object, 0, success, file, line);
if (success) {
curthread->td_rw_rlocks--;
WITNESS_UPGRADE(&rw->lock_object, LOP_EXCLUSIVE | LOP_TRYLOCK,
file, line);
LOCKSTAT_RECORD0(LS_RW_TRYUPGRADE_UPGRADE, rw);
}
return (success);
}
/*
* Downgrade a write lock into a single read lock.
*/
void
_rw_downgrade(struct rwlock *rw, const char *file, int line)
{
struct turnstile *ts;
uintptr_t tid, v;
int rwait, wwait;
if (SCHEDULER_STOPPED())
return;
KASSERT(rw->rw_lock != RW_DESTROYED,
("rw_downgrade() of destroyed rwlock @ %s:%d", file, line));
_rw_assert(rw, 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(LS_RW_DOWNGRADE_DOWNGRADE, rw);
}
void
_rw_runlock(struct rwlock *rw, const char *file, int line)
{
struct turnstile *ts;
uintptr_t x, v, queue;
if (SCHEDULER_STOPPED())
return;
KASSERT(rw->rw_lock != RW_DESTROYED,
("rw_runlock() of destroyed rwlock @ %s:%d", file, line));
_rw_assert(rw, RA_RLOCKED, file, line);
curthread->td_locks--;
curthread->td_rw_rlocks--;
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_LOCK(LS_RW_RUNLOCK_RELEASE, rw);
}
/*
* 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(struct rwlock *rw, uintptr_t tid, const char *file, int line)
{
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
uint64_t spin_cnt = 0;
uint64_t sleep_cnt = 0;
int64_t sleep_time = 0;
#endif
if (SCHEDULER_STOPPED())
return;
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);
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);
while ((struct thread*)RW_OWNER(rw->rw_lock) == owner &&
TD_IS_RUNNING(owner)) {
cpu_spinwait();
#ifdef KDTRACE_HOOKS
spin_cnt++;
#endif
}
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++;
for (i = 0; i < ROWNER_LOOPS; i++) {
if ((rw->rw_lock & RW_LOCK_WRITE_SPINNER) == 0)
break;
cpu_spinwait();
}
#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();
#endif
turnstile_wait(ts, rw_owner(rw), TS_EXCLUSIVE_QUEUE);
#ifdef KDTRACE_HOOKS
sleep_time += lockstat_nsecs();
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
}
LOCKSTAT_PROFILE_OBTAIN_LOCK_SUCCESS(LS_RW_WLOCK_ACQUIRE, rw, contested,
waittime, file, line);
#ifdef KDTRACE_HOOKS
if (sleep_time)
LOCKSTAT_RECORD1(LS_RW_WLOCK_BLOCK, rw, sleep_time);
/*
* Record only the loops spinning and not sleeping.
*/
if (spin_cnt > sleep_cnt)
LOCKSTAT_RECORD1(LS_RW_WLOCK_SPIN, rw, (spin_cnt - sleep_cnt));
#endif
}
/*
* 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(volatile uintptr_t *c, uintptr_t tid, const char *file,
int line)
{
struct rwlock *rw;
struct turnstile *ts;
uintptr_t v;
int queue;
if (SCHEDULER_STOPPED())
return;
rw = rwlock2rw(c);
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);
}
void
_rw_rlock(struct rwlock *rw, const char *file, int line)
{
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
uint64_t spin_cnt = 0;
uint64_t sleep_cnt = 0;
int64_t sleep_time = 0;
#endif
if (SCHEDULER_STOPPED())
return;
KASSERT(rw->rw_lock != RW_DESTROYED,
("rw_rlock() of destroyed rwlock @ %s:%d", file, line));
KASSERT(rw_wowner(rw) != curthread,
("%s (%s): wlock already held @ %s:%d", __func__,
rw->lock_object.lo_name, file, line));
WITNESS_CHECKORDER(&rw->lock_object, LOP_NEWORDER, file, line, NULL);
for (;;) {
#ifdef KDTRACE_HOOKS
spin_cnt++;
#endif
/*
* 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 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);
while ((struct thread*)RW_OWNER(rw->rw_lock) ==
owner && TD_IS_RUNNING(owner)) {
cpu_spinwait();
#ifdef KDTRACE_HOOKS
spin_cnt++;
#endif
}
continue;
}
} else if (spintries < ROWNER_RETRIES) {
spintries++;
for (i = 0; i < ROWNER_LOOPS; i++) {
v = rw->rw_lock;
if ((v & RW_LOCK_READ) == 0 || RW_CAN_READ(v))
break;
cpu_spinwait();
}
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();
#endif
turnstile_wait(ts, rw_owner(rw), TS_SHARED_QUEUE);
#ifdef KDTRACE_HOOKS
sleep_time += lockstat_nsecs();
sleep_cnt++;
#endif
if (LOCK_LOG_TEST(&rw->lock_object, 0))
CTR2(KTR_LOCK, "%s: %p resuming from turnstile",
__func__, rw);
}
/*
* 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_LOCK_SUCCESS(LS_RW_RLOCK_ACQUIRE, rw, contested,
waittime, file, line);
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++;
#ifdef KDTRACE_HOOKS
if (sleep_time)
LOCKSTAT_RECORD1(LS_RW_RLOCK_BLOCK, rw, sleep_time);
/*
* Record only the loops spinning and not sleeping.
*/
if (spin_cnt > sleep_cnt)
LOCKSTAT_RECORD1(LS_RW_RLOCK_SPIN, rw, (spin_cnt - sleep_cnt));
#endif
}
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
}
/*
* _mtx_lock_spin_cookie: the tougher part of acquiring an MTX_SPIN lock.
*
* This is only called if we need to actually spin for the lock. Recursion
* is handled inline.
*/
void
_mtx_lock_spin_cookie(volatile uintptr_t *c, uintptr_t tid, int opts,
const char *file, int line)
{
struct mtx *m;
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 (SCHEDULER_STOPPED())
return;
lock_delay_arg_init(&lda, &mtx_spin_delay);
m = mtxlock2mtx(c);
if (LOCK_LOG_TEST(&m->lock_object, opts))
CTR1(KTR_LOCK, "_mtx_lock_spin: %p spinning", m);
KTR_STATE1(KTR_SCHED, "thread", sched_tdname((struct thread *)tid),
"spinning", "lockname:\"%s\"", m->lock_object.lo_name);
#ifdef HWPMC_HOOKS
PMC_SOFT_CALL( , , lock, failed);
#endif
lock_profile_obtain_lock_failed(&m->lock_object, &contested, &waittime);
#ifdef KDTRACE_HOOKS
spin_time -= lockstat_nsecs(&m->lock_object);
#endif
for (;;) {
if (m->mtx_lock == MTX_UNOWNED && _mtx_obtain_lock(m, tid))
break;
/* Give interrupts a chance while we spin. */
spinlock_exit();
while (m->mtx_lock != MTX_UNOWNED) {
if (lda.spin_cnt < 10000000) {
lock_delay(&lda);
continue;
}
lda.spin_cnt++;
if (lda.spin_cnt < 60000000 || kdb_active ||
panicstr != NULL)
DELAY(1);
else
_mtx_lock_spin_failed(m);
cpu_spinwait();
}
spinlock_enter();
}
#ifdef KDTRACE_HOOKS
spin_time += lockstat_nsecs(&m->lock_object);
#endif
if (LOCK_LOG_TEST(&m->lock_object, opts))
CTR1(KTR_LOCK, "_mtx_lock_spin: %p spin done", m);
KTR_STATE0(KTR_SCHED, "thread", sched_tdname((struct thread *)tid),
"running");
#ifdef KDTRACE_HOOKS
LOCKSTAT_PROFILE_OBTAIN_LOCK_SUCCESS(spin__acquire, m,
contested, waittime, file, line);
if (spin_time != 0)
LOCKSTAT_RECORD1(spin__spin, m, spin_time);
#endif
}
/*
* __mtx_lock_sleep: the tougher part of acquiring an MTX_DEF lock.
*
* We call this if the lock is either contested (i.e. we need to go to
* sleep waiting for it), or if we need to recurse on it.
*/
void
__mtx_lock_sleep(volatile uintptr_t *c, uintptr_t tid, int opts,
const char *file, int line)
{
struct mtx *m;
struct turnstile *ts;
uintptr_t v;
#ifdef ADAPTIVE_MUTEXES
volatile struct thread *owner;
#endif
#ifdef KTR
int cont_logged = 0;
#endif
#ifdef LOCK_PROFILING
int contested = 0;
uint64_t waittime = 0;
#endif
#if defined(ADAPTIVE_MUTEXES) || defined(KDTRACE_HOOKS)
struct lock_delay_arg lda;
#endif
#ifdef KDTRACE_HOOKS
u_int sleep_cnt = 0;
int64_t sleep_time = 0;
int64_t all_time = 0;
#endif
if (SCHEDULER_STOPPED())
return;
#if defined(ADAPTIVE_MUTEXES)
lock_delay_arg_init(&lda, &mtx_delay);
#elif defined(KDTRACE_HOOKS)
lock_delay_arg_init(&lda, NULL);
#endif
m = mtxlock2mtx(c);
if (mtx_owned(m)) {
KASSERT((m->lock_object.lo_flags & LO_RECURSABLE) != 0 ||
(opts & MTX_RECURSE) != 0,
("_mtx_lock_sleep: recursed on non-recursive mutex %s @ %s:%d\n",
m->lock_object.lo_name, file, line));
opts &= ~MTX_RECURSE;
m->mtx_recurse++;
atomic_set_ptr(&m->mtx_lock, MTX_RECURSED);
if (LOCK_LOG_TEST(&m->lock_object, opts))
CTR1(KTR_LOCK, "_mtx_lock_sleep: %p recursing", m);
return;
}
opts &= ~MTX_RECURSE;
#ifdef HWPMC_HOOKS
PMC_SOFT_CALL( , , lock, failed);
#endif
lock_profile_obtain_lock_failed(&m->lock_object,
&contested, &waittime);
if (LOCK_LOG_TEST(&m->lock_object, opts))
CTR4(KTR_LOCK,
"_mtx_lock_sleep: %s contested (lock=%p) at %s:%d",
m->lock_object.lo_name, (void *)m->mtx_lock, file, line);
#ifdef KDTRACE_HOOKS
all_time -= lockstat_nsecs(&m->lock_object);
#endif
for (;;) {
if (m->mtx_lock == MTX_UNOWNED && _mtx_obtain_lock(m, tid))
break;
#ifdef KDTRACE_HOOKS
lda.spin_cnt++;
#endif
#ifdef ADAPTIVE_MUTEXES
/*
* If the owner is running on another CPU, spin until the
* owner stops running or the state of the lock changes.
*/
v = m->mtx_lock;
if (v != MTX_UNOWNED) {
owner = (struct thread *)(v & ~MTX_FLAGMASK);
if (TD_IS_RUNNING(owner)) {
if (LOCK_LOG_TEST(&m->lock_object, 0))
CTR3(KTR_LOCK,
"%s: spinning on %p held by %p",
__func__, m, owner);
KTR_STATE1(KTR_SCHED, "thread",
sched_tdname((struct thread *)tid),
"spinning", "lockname:\"%s\"",
m->lock_object.lo_name);
while (mtx_owner(m) == owner &&
TD_IS_RUNNING(owner))
lock_delay(&lda);
KTR_STATE0(KTR_SCHED, "thread",
sched_tdname((struct thread *)tid),
"running");
continue;
}
}
#endif
ts = turnstile_trywait(&m->lock_object);
v = m->mtx_lock;
/*
* Check if the lock has been released while spinning for
* the turnstile chain lock.
*/
if (v == MTX_UNOWNED) {
turnstile_cancel(ts);
continue;
}
#ifdef ADAPTIVE_MUTEXES
/*
* 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.
*/
owner = (struct thread *)(v & ~MTX_FLAGMASK);
if (TD_IS_RUNNING(owner)) {
turnstile_cancel(ts);
continue;
}
#endif
/*
* If the mutex isn't already contested and a failure occurs
* setting the contested bit, the mutex was either released
* or the state of the MTX_RECURSED bit changed.
*/
if ((v & MTX_CONTESTED) == 0 &&
!atomic_cmpset_ptr(&m->mtx_lock, v, v | MTX_CONTESTED)) {
turnstile_cancel(ts);
continue;
}
/*
* We definitely must sleep for this lock.
*/
mtx_assert(m, MA_NOTOWNED);
#ifdef KTR
if (!cont_logged) {
CTR6(KTR_CONTENTION,
"contention: %p at %s:%d wants %s, taken by %s:%d",
(void *)tid, file, line, m->lock_object.lo_name,
WITNESS_FILE(&m->lock_object),
WITNESS_LINE(&m->lock_object));
cont_logged = 1;
}
#endif
/*
* Block on the turnstile.
*/
#ifdef KDTRACE_HOOKS
sleep_time -= lockstat_nsecs(&m->lock_object);
#endif
turnstile_wait(ts, mtx_owner(m), TS_EXCLUSIVE_QUEUE);
#ifdef KDTRACE_HOOKS
sleep_time += lockstat_nsecs(&m->lock_object);
sleep_cnt++;
#endif
}
#ifdef KDTRACE_HOOKS
all_time += lockstat_nsecs(&m->lock_object);
#endif
#ifdef KTR
if (cont_logged) {
CTR4(KTR_CONTENTION,
"contention end: %s acquired by %p at %s:%d",
m->lock_object.lo_name, (void *)tid, file, line);
}
#endif
LOCKSTAT_PROFILE_OBTAIN_LOCK_SUCCESS(adaptive__acquire, m, contested,
waittime, file, line);
#ifdef KDTRACE_HOOKS
if (sleep_time)
LOCKSTAT_RECORD1(adaptive__block, m, sleep_time);
/*
* Only record the loops spinning and not sleeping.
*/
if (lda.spin_cnt > sleep_cnt)
LOCKSTAT_RECORD1(adaptive__spin, m, all_time - sleep_time);
#endif
}
/*
* Shutdown the system cleanly to prepare for reboot, halt, or power off.
*/
void
kern_reboot(int howto)
{
static int first_buf_printf = 1;
#if defined(SMP)
/*
* Bind us to CPU 0 so that all shutdown code runs there. Some
* systems don't shutdown properly (i.e., ACPI power off) if we
* run on another processor.
*/
if (!SCHEDULER_STOPPED()) {
thread_lock(curthread);
sched_bind(curthread, 0);
thread_unlock(curthread);
KASSERT(PCPU_GET(cpuid) == 0, ("boot: not running on cpu 0"));
}
#endif
/* We're in the process of rebooting. */
rebooting = 1;
/* collect extra flags that shutdown_nice might have set */
howto |= shutdown_howto;
/* We are out of the debugger now. */
kdb_active = 0;
/*
* Do any callouts that should be done BEFORE syncing the filesystems.
*/
EVENTHANDLER_INVOKE(shutdown_pre_sync, howto);
/*
* Now sync filesystems
*/
if (!cold && (howto & RB_NOSYNC) == 0 && waittime < 0) {
register struct buf *bp;
int iter, nbusy, pbusy;
#ifndef PREEMPTION
int subiter;
#endif
waittime = 0;
wdog_kern_pat(WD_LASTVAL);
sys_sync(curthread, NULL);
/*
* With soft updates, some buffers that are
* written will be remarked as dirty until other
* buffers are written.
*/
for (iter = pbusy = 0; iter < 20; iter++) {
nbusy = 0;
for (bp = &buf[nbuf]; --bp >= buf; )
if (isbufbusy(bp))
nbusy++;
if (nbusy == 0) {
if (first_buf_printf)
printf("All buffers synced.");
break;
}
if (first_buf_printf) {
printf("Syncing disks, buffers remaining... ");
first_buf_printf = 0;
}
printf("%d ", nbusy);
if (nbusy < pbusy)
iter = 0;
pbusy = nbusy;
wdog_kern_pat(WD_LASTVAL);
sys_sync(curthread, NULL);
#ifdef PREEMPTION
/*
* Drop Giant and spin for a while to allow
* interrupt threads to run.
*/
DROP_GIANT();
DELAY(50000 * iter);
PICKUP_GIANT();
#else
/*
* Drop Giant and context switch several times to
* allow interrupt threads to run.
*/
DROP_GIANT();
for (subiter = 0; subiter < 50 * iter; subiter++) {
thread_lock(curthread);
mi_switch(SW_VOL, NULL);
thread_unlock(curthread);
DELAY(1000);
}
PICKUP_GIANT();
#endif
}
printf("\n");
/*
* Count only busy local buffers to prevent forcing
* a fsck if we're just a client of a wedged NFS server
*/
nbusy = 0;
for (bp = &buf[nbuf]; --bp >= buf; ) {
if (isbufbusy(bp)) {
#if 0
/* XXX: This is bogus. We should probably have a BO_REMOTE flag instead */
if (bp->b_dev == NULL) {
TAILQ_REMOVE(&mountlist,
bp->b_vp->v_mount, mnt_list);
continue;
}
#endif
nbusy++;
if (show_busybufs > 0) {
printf(
"%d: buf:%p, vnode:%p, flags:%0x, blkno:%jd, lblkno:%jd, buflock:",
nbusy, bp, bp->b_vp, bp->b_flags,
(intmax_t)bp->b_blkno,
(intmax_t)bp->b_lblkno);
BUF_LOCKPRINTINFO(bp);
if (show_busybufs > 1)
vn_printf(bp->b_vp,
"vnode content: ");
}
}
}
if (nbusy) {
/*
* Failed to sync all blocks. Indicate this and don't
* unmount filesystems (thus forcing an fsck on reboot).
*/
printf("Giving up on %d buffers\n", nbusy);
DELAY(5000000); /* 5 seconds */
} else {
if (!first_buf_printf)
printf("Final sync complete\n");
/*
* Unmount filesystems
*/
if (panicstr == 0)
vfs_unmountall();
}
swapoff_all();
DELAY(100000); /* wait for console output to finish */
}
print_uptime();
cngrab();
/*
* Ok, now do things that assume all filesystem activity has
* been completed.
*/
EVENTHANDLER_INVOKE(shutdown_post_sync, howto);
if ((howto & (RB_HALT|RB_DUMP)) == RB_DUMP && !cold && !dumping)
doadump(TRUE);
/* Now that we're going to really halt the system... */
EVENTHANDLER_INVOKE(shutdown_final, howto);
for(;;) ; /* safety against shutdown_reset not working */
/* NOTREACHED */
}
/*
* malloc:
*
* Allocate a block of memory.
*
* If M_NOWAIT is set, this routine will not block and return NULL if
* the allocation fails.
*/
void *
malloc(unsigned long size, struct malloc_type *mtp, int flags)
{
int indx;
struct malloc_type_internal *mtip;
caddr_t va;
uma_zone_t zone;
#if defined(DIAGNOSTIC) || defined(DEBUG_REDZONE)
unsigned long osize = size;
#endif
#ifdef INVARIANTS
KASSERT(mtp->ks_magic == M_MAGIC, ("malloc: bad malloc type magic"));
/*
* Check that exactly one of M_WAITOK or M_NOWAIT is specified.
*/
indx = flags & (M_WAITOK | M_NOWAIT);
if (indx != M_NOWAIT && indx != M_WAITOK) {
static struct timeval lasterr;
static int curerr, once;
if (once == 0 && ppsratecheck(&lasterr, &curerr, 1)) {
printf("Bad malloc flags: %x\n", indx);
kdb_backtrace();
flags |= M_WAITOK;
once++;
}
}
#endif
#ifdef MALLOC_MAKE_FAILURES
if ((flags & M_NOWAIT) && (malloc_failure_rate != 0)) {
atomic_add_int(&malloc_nowait_count, 1);
if ((malloc_nowait_count % malloc_failure_rate) == 0) {
atomic_add_int(&malloc_failure_count, 1);
t_malloc_fail = time_uptime;
return (NULL);
}
}
#endif
if (flags & M_WAITOK)
KASSERT(curthread->td_intr_nesting_level == 0,
("malloc(M_WAITOK) in interrupt context"));
KASSERT(curthread->td_critnest == 0 || SCHEDULER_STOPPED(),
("malloc: called with spinlock or critical section held"));
#ifdef DEBUG_MEMGUARD
if (memguard_cmp_mtp(mtp, size)) {
va = memguard_alloc(size, flags);
if (va != NULL)
return (va);
/* This is unfortunate but should not be fatal. */
}
#endif
#ifdef DEBUG_REDZONE
size = redzone_size_ntor(size);
#endif
if (size <= kmem_zmax) {
mtip = mtp->ks_handle;
if (size & KMEM_ZMASK)
size = (size & ~KMEM_ZMASK) + KMEM_ZBASE;
indx = kmemsize[size >> KMEM_ZSHIFT];
KASSERT(mtip->mti_zone < numzones,
("mti_zone %u out of range %d",
mtip->mti_zone, numzones));
zone = kmemzones[indx].kz_zone[mtip->mti_zone];
#ifdef MALLOC_PROFILE
krequests[size >> KMEM_ZSHIFT]++;
#endif
va = uma_zalloc(zone, flags);
if (va != NULL)
size = zone->uz_size;
malloc_type_zone_allocated(mtp, va == NULL ? 0 : size, indx);
} else {