/*
* ProcWakeup -- wake up a process by releasing its private semaphore.
*
* Also remove the process from the wait queue and set its links invalid.
* RETURN: the next process in the wait queue.
*
* The appropriate lock partition lock must be held by caller.
*
* XXX: presently, this code is only used for the "success" case, and only
* works correctly for that case. To clean up in failure case, would need
* to twiddle the lock's request counts too --- see RemoveFromWaitQueue.
* Hence, in practice the waitStatus parameter must be STATUS_OK.
*/
PGPROC *
ProcWakeup(PGPROC *proc, int waitStatus)
{
PGPROC *retProc;
/* Proc should be sleeping ... */
if (proc->links.prev == NULL ||
proc->links.next == NULL)
return NULL;
Assert(proc->waitStatus == STATUS_WAITING);
/* Save next process before we zap the list link */
retProc = (PGPROC *) proc->links.next;
/* Remove process from wait queue */
SHMQueueDelete(&(proc->links));
(proc->waitLock->waitProcs.size)--;
/* Clean up process' state and pass it the ok/fail signal */
proc->waitLock = NULL;
proc->waitProcLock = NULL;
proc->waitStatus = waitStatus;
/* And awaken it */
PGSemaphoreUnlock(&proc->sem);
return retProc;
}
/*
* ProcSendSignal - send a signal to a backend identified by PID
*/
void
ProcSendSignal(int pid)
{
PGPROC *proc = NULL;
if (RecoveryInProgress())
{
/* use volatile pointer to prevent code rearrangement */
volatile PROC_HDR *procglobal = ProcGlobal;
SpinLockAcquire(ProcStructLock);
/*
* Check to see whether it is the Startup process we wish to signal.
* This call is made by the buffer manager when it wishes to wake up a
* process that has been waiting for a pin in so it can obtain a
* cleanup lock using LockBufferForCleanup(). Startup is not a normal
* backend, so BackendPidGetProc() will not return any pid at all. So
* we remember the information for this special case.
*/
if (pid == procglobal->startupProcPid)
proc = procglobal->startupProc;
SpinLockRelease(ProcStructLock);
}
if (proc == NULL)
proc = BackendPidGetProc(pid);
if (proc != NULL)
PGSemaphoreUnlock(&proc->sem);
}
/*
* ProcSendSignal - send a signal to a backend identified by PID
*/
void
ProcSendSignal(int pid)
{
PGPROC *proc = BackendPidGetProc(pid);
if (proc != NULL)
PGSemaphoreUnlock(&proc->sem);
}
void
s_unlock_sema(volatile slock_t *lock)
{
int lockndx = *lock;
if (lockndx <= 0 || lockndx > NUM_SPINLOCK_SEMAPHORES)
elog(ERROR, "invalid spinlock number: %d", lockndx);
PGSemaphoreUnlock(&SpinlockSemaArray[lockndx - 1]);
}
/*
* LWLockUpdateVar - Update a variable and wake up waiters atomically
*
* Sets *valptr to 'val', and wakes up all processes waiting for us with
* LWLockWaitForVar(). Setting the value and waking up the processes happen
* atomically so that any process calling LWLockWaitForVar() on the same lock
* is guaranteed to see the new value, and act accordingly.
*
* The caller must be holding the lock in exclusive mode.
*/
void
LWLockUpdateVar(LWLock *lock, uint64 *valptr, uint64 val)
{
PGPROC *head;
PGPROC *proc;
PGPROC *next;
/* Acquire mutex. Time spent holding mutex should be short! */
SpinLockAcquire(&lock->mutex);
/* we should hold the lock */
Assert(lock->exclusive == 1);
/* Update the lock's value */
*valptr = val;
/*
* See if there are any LW_WAIT_UNTIL_FREE waiters that need to be woken
* up. They are always in the front of the queue.
*/
head = lock->head;
if (head != NULL && head->lwWaitMode == LW_WAIT_UNTIL_FREE)
{
proc = head;
next = proc->lwWaitLink;
while (next && next->lwWaitMode == LW_WAIT_UNTIL_FREE)
{
proc = next;
next = next->lwWaitLink;
}
/* proc is now the last PGPROC to be released */
lock->head = next;
proc->lwWaitLink = NULL;
}
else
head = NULL;
/* We are done updating shared state of the lock itself. */
SpinLockRelease(&lock->mutex);
/*
* Awaken any waiters I removed from the queue.
*/
while (head != NULL)
{
proc = head;
head = proc->lwWaitLink;
proc->lwWaitLink = NULL;
/* check comment in LWLockRelease() about this barrier */
pg_write_barrier();
proc->lwWaiting = false;
PGSemaphoreUnlock(&proc->sem);
}
}
static void BgwPoolMainLoop(Datum arg)
{
BgwPoolExecutorCtx* ctx = (BgwPoolExecutorCtx*)arg;
int id = ctx->id;
BgwPool* pool = ctx->constructor();
int size;
void* work;
BackgroundWorkerUnblockSignals();
BackgroundWorkerInitializeConnection(pool->dbname, NULL);
while(true) {
PGSemaphoreLock(&pool->available);
SpinLockAcquire(&pool->lock);
size = *(int*)&pool->queue[pool->head];
Assert(size < pool->size);
work = malloc(size);
pool->pending -= 1;
pool->active += 1;
if (pool->lastPeakTime == 0 && pool->active == pool->nWorkers && pool->pending != 0) {
pool->lastPeakTime = MtmGetSystemTime();
}
if (pool->head + size + 4 > pool->size) {
memcpy(work, pool->queue, size);
pool->head = INTALIGN(size);
} else {
memcpy(work, &pool->queue[pool->head+4], size);
pool->head += 4 + INTALIGN(size);
}
if (pool->size == pool->head) {
pool->head = 0;
}
if (pool->producerBlocked) {
pool->producerBlocked = false;
PGSemaphoreUnlock(&pool->overflow);
pool->lastPeakTime = 0;
}
SpinLockRelease(&pool->lock);
pool->executor(id, work, size);
free(work);
SpinLockAcquire(&pool->lock);
pool->active -= 1;
pool->lastPeakTime = 0;
SpinLockRelease(&pool->lock);
}
}
void BgwPoolExecute(BgwPool* pool, void* work, size_t size)
{
if (size+4 > pool->size) {
/*
* Size of work is larger than size of shared buffer:
* run it immediately
*/
pool->executor(0, work, size);
return;
}
SpinLockAcquire(&pool->lock);
while (true) {
if ((pool->head <= pool->tail && pool->size - pool->tail < size + 4 && pool->head < size)
|| (pool->head > pool->tail && pool->head - pool->tail < size + 4))
{
if (pool->lastPeakTime == 0) {
pool->lastPeakTime = MtmGetSystemTime();
}
pool->producerBlocked = true;
SpinLockRelease(&pool->lock);
PGSemaphoreLock(&pool->overflow);
SpinLockAcquire(&pool->lock);
} else {
pool->pending += 1;
if (pool->lastPeakTime == 0 && pool->active == pool->nWorkers && pool->pending != 0) {
pool->lastPeakTime = MtmGetSystemTime();
}
*(int*)&pool->queue[pool->tail] = size;
if (pool->size - pool->tail >= size + 4) {
memcpy(&pool->queue[pool->tail+4], work, size);
pool->tail += 4 + INTALIGN(size);
} else {
memcpy(pool->queue, work, size);
pool->tail = INTALIGN(size);
}
if (pool->tail == pool->size) {
pool->tail = 0;
}
PGSemaphoreUnlock(&pool->available);
break;
}
}
SpinLockRelease(&pool->lock);
}
/*
* LWLockRelease - release a previously acquired lock
*/
void
LWLockRelease(LWLockId lockid)
{
volatile LWLock *lock = &(LWLockArray[lockid].lock);
PGPROC *head;
PGPROC *proc;
int i;
PRINT_LWDEBUG("LWLockRelease", lockid, lock);
/*
* Remove lock from list of locks held. Usually, but not always, it will
* be the latest-acquired lock; so search array backwards.
*/
for (i = num_held_lwlocks; --i >= 0;)
{
if (lockid == held_lwlocks[i])
break;
}
if (i < 0)
elog(ERROR, "lock %d is not held", (int) lockid);
num_held_lwlocks--;
for (; i < num_held_lwlocks; i++)
held_lwlocks[i] = held_lwlocks[i + 1];
/* Acquire mutex. Time spent holding mutex should be short! */
SpinLockAcquire(&lock->mutex);
/* Release my hold on lock */
if (lock->exclusive > 0)
lock->exclusive--;
else
{
Assert(lock->shared > 0);
lock->shared--;
}
/*
* See if I need to awaken any waiters. If I released a non-last shared
* hold, there cannot be anything to do. Also, do not awaken any waiters
* if someone has already awakened waiters that haven't yet acquired the
* lock.
*/
head = lock->head;
if (head != NULL)
{
if (lock->exclusive == 0 && lock->shared == 0 && lock->releaseOK)
{
/*
* Remove the to-be-awakened PGPROCs from the queue. If the front
* waiter wants exclusive lock, awaken him only. Otherwise awaken
* as many waiters as want shared access.
*/
proc = head;
if (!proc->lwExclusive)
{
while (proc->lwWaitLink != NULL &&
!proc->lwWaitLink->lwExclusive)
proc = proc->lwWaitLink;
}
/* proc is now the last PGPROC to be released */
lock->head = proc->lwWaitLink;
proc->lwWaitLink = NULL;
/* prevent additional wakeups until retryer gets to run */
lock->releaseOK = false;
}
else
{
/* lock is still held, can't awaken anything */
head = NULL;
}
}
/* We are done updating shared state of the lock itself. */
SpinLockRelease(&lock->mutex);
TRACE_POSTGRESQL_LWLOCK_RELEASE(lockid);
/*
* Awaken any waiters I removed from the queue.
*/
while (head != NULL)
{
LOG_LWDEBUG("LWLockRelease", lockid, "release waiter");
proc = head;
head = proc->lwWaitLink;
proc->lwWaitLink = NULL;
proc->lwWaiting = false;
PGSemaphoreUnlock(&proc->sem);
}
/*
* Now okay to allow cancel/die interrupts.
*/
RESUME_INTERRUPTS();
}
/*
* LWLockAcquire - acquire a lightweight lock in the specified mode
*
* If the lock is not available, sleep until it is.
*
* Side effect: cancel/die interrupts are held off until lock release.
*/
void
LWLockAcquire(LWLockId lockid, LWLockMode mode)
{
volatile LWLock *lock = &(LWLockArray[lockid].lock);
PGPROC *proc = MyProc;
bool retry = false;
int extraWaits = 0;
PRINT_LWDEBUG("LWLockAcquire", lockid, lock);
#ifdef LWLOCK_STATS
/* Set up local count state first time through in a given process */
if (counts_for_pid != MyProcPid)
{
int *LWLockCounter = (int *) ((char *) LWLockArray - 2 * sizeof(int));
int numLocks = LWLockCounter[1];
sh_acquire_counts = calloc(numLocks, sizeof(int));
ex_acquire_counts = calloc(numLocks, sizeof(int));
block_counts = calloc(numLocks, sizeof(int));
counts_for_pid = MyProcPid;
on_shmem_exit(print_lwlock_stats, 0);
}
/* Count lock acquisition attempts */
if (mode == LW_EXCLUSIVE)
ex_acquire_counts[lockid]++;
else
sh_acquire_counts[lockid]++;
#endif /* LWLOCK_STATS */
/*
* We can't wait if we haven't got a PGPROC. This should only occur
* during bootstrap or shared memory initialization. Put an Assert here
* to catch unsafe coding practices.
*/
Assert(!(proc == NULL && IsUnderPostmaster));
/* Ensure we will have room to remember the lock */
if (num_held_lwlocks >= MAX_SIMUL_LWLOCKS)
elog(ERROR, "too many LWLocks taken");
/*
* Lock out cancel/die interrupts until we exit the code section protected
* by the LWLock. This ensures that interrupts will not interfere with
* manipulations of data structures in shared memory.
*/
HOLD_INTERRUPTS();
/*
* Loop here to try to acquire lock after each time we are signaled by
* LWLockRelease.
*
* NOTE: it might seem better to have LWLockRelease actually grant us the
* lock, rather than retrying and possibly having to go back to sleep. But
* in practice that is no good because it means a process swap for every
* lock acquisition when two or more processes are contending for the same
* lock. Since LWLocks are normally used to protect not-very-long
* sections of computation, a process needs to be able to acquire and
* release the same lock many times during a single CPU time slice, even
* in the presence of contention. The efficiency of being able to do that
* outweighs the inefficiency of sometimes wasting a process dispatch
* cycle because the lock is not free when a released waiter finally gets
* to run. See pgsql-hackers archives for 29-Dec-01.
*/
for (;;)
{
bool mustwait;
/* Acquire mutex. Time spent holding mutex should be short! */
SpinLockAcquire(&lock->mutex);
/* If retrying, allow LWLockRelease to release waiters again */
if (retry)
lock->releaseOK = true;
/* If I can get the lock, do so quickly. */
if (mode == LW_EXCLUSIVE)
{
if (lock->exclusive == 0 && lock->shared == 0)
{
lock->exclusive++;
mustwait = false;
}
else
mustwait = true;
}
else
{
if (lock->exclusive == 0)
{
lock->shared++;
mustwait = false;
}
else
mustwait = true;
}
if (!mustwait)
break; /* got the lock */
/*
* Add myself to wait queue.
*
* If we don't have a PGPROC structure, there's no way to wait. This
* should never occur, since MyProc should only be null during shared
* memory initialization.
*/
if (proc == NULL)
elog(PANIC, "cannot wait without a PGPROC structure");
proc->lwWaiting = true;
proc->lwExclusive = (mode == LW_EXCLUSIVE);
proc->lwWaitLink = NULL;
if (lock->head == NULL)
lock->head = proc;
else
lock->tail->lwWaitLink = proc;
lock->tail = proc;
/* Can release the mutex now */
SpinLockRelease(&lock->mutex);
/*
* Wait until awakened.
*
* Since we share the process wait semaphore with the regular lock
* manager and ProcWaitForSignal, and we may need to acquire an LWLock
* while one of those is pending, it is possible that we get awakened
* for a reason other than being signaled by LWLockRelease. If so,
* loop back and wait again. Once we've gotten the LWLock,
* re-increment the sema by the number of additional signals received,
* so that the lock manager or signal manager will see the received
* signal when it next waits.
*/
LOG_LWDEBUG("LWLockAcquire", lockid, "waiting");
#ifdef LWLOCK_STATS
block_counts[lockid]++;
#endif
TRACE_POSTGRESQL_LWLOCK_WAIT_START(lockid, mode);
for (;;)
{
/* "false" means cannot accept cancel/die interrupt here. */
PGSemaphoreLock(&proc->sem, false);
if (!proc->lwWaiting)
break;
extraWaits++;
}
TRACE_POSTGRESQL_LWLOCK_WAIT_DONE(lockid, mode);
LOG_LWDEBUG("LWLockAcquire", lockid, "awakened");
/* Now loop back and try to acquire lock again. */
retry = true;
}
/* We are done updating shared state of the lock itself. */
SpinLockRelease(&lock->mutex);
TRACE_POSTGRESQL_LWLOCK_ACQUIRE(lockid, mode);
/* Add lock to list of locks held by this backend */
held_lwlocks[num_held_lwlocks++] = lockid;
/*
* Fix the process wait semaphore's count for any absorbed wakeups.
*/
while (extraWaits-- > 0)
PGSemaphoreUnlock(&proc->sem);
}
/*
* LWLockWaitForVar - Wait until lock is free, or a variable is updated.
*
* If the lock is held and *valptr equals oldval, waits until the lock is
* either freed, or the lock holder updates *valptr by calling
* LWLockUpdateVar. If the lock is free on exit (immediately or after
* waiting), returns true. If the lock is still held, but *valptr no longer
* matches oldval, returns false and sets *newval to the current value in
* *valptr.
*
* It's possible that the lock holder releases the lock, but another backend
* acquires it again before we get a chance to observe that the lock was
* momentarily released. We wouldn't need to wait for the new lock holder,
* but we cannot distinguish that case, so we will have to wait.
*
* Note: this function ignores shared lock holders; if the lock is held
* in shared mode, returns 'true'.
*/
bool
LWLockWaitForVar(LWLock *lock, uint64 *valptr, uint64 oldval, uint64 *newval)
{
PGPROC *proc = MyProc;
int extraWaits = 0;
bool result = false;
#ifdef LWLOCK_STATS
lwlock_stats *lwstats;
#endif
PRINT_LWDEBUG("LWLockWaitForVar", lock);
#ifdef LWLOCK_STATS
lwstats = get_lwlock_stats_entry(lock);
#endif /* LWLOCK_STATS */
/*
* Quick test first to see if it the slot is free right now.
*
* XXX: the caller uses a spinlock before this, so we don't need a memory
* barrier here as far as the current usage is concerned. But that might
* not be safe in general.
*/
if (lock->exclusive == 0)
return true;
/*
* Lock out cancel/die interrupts while we sleep on the lock. There is no
* cleanup mechanism to remove us from the wait queue if we got
* interrupted.
*/
HOLD_INTERRUPTS();
/*
* Loop here to check the lock's status after each time we are signaled.
*/
for (;;)
{
bool mustwait;
uint64 value;
/* Acquire mutex. Time spent holding mutex should be short! */
#ifdef LWLOCK_STATS
lwstats->spin_delay_count += SpinLockAcquire(&lock->mutex);
#else
SpinLockAcquire(&lock->mutex);
#endif
/* Is the lock now free, and if not, does the value match? */
if (lock->exclusive == 0)
{
result = true;
mustwait = false;
}
else
{
value = *valptr;
if (value != oldval)
{
result = false;
mustwait = false;
*newval = value;
}
else
mustwait = true;
}
if (!mustwait)
break; /* the lock was free or value didn't match */
/*
* Add myself to wait queue.
*/
proc->lwWaiting = true;
proc->lwWaitMode = LW_WAIT_UNTIL_FREE;
/* waiters are added to the front of the queue */
proc->lwWaitLink = lock->head;
if (lock->head == NULL)
lock->tail = proc;
lock->head = proc;
/*
* Set releaseOK, to make sure we get woken up as soon as the lock is
* released.
*/
lock->releaseOK = true;
/* Can release the mutex now */
SpinLockRelease(&lock->mutex);
/*
* Wait until awakened.
*
* Since we share the process wait semaphore with the regular lock
* manager and ProcWaitForSignal, and we may need to acquire an LWLock
* while one of those is pending, it is possible that we get awakened
* for a reason other than being signaled by LWLockRelease. If so,
* loop back and wait again. Once we've gotten the LWLock,
* re-increment the sema by the number of additional signals received,
* so that the lock manager or signal manager will see the received
* signal when it next waits.
*/
LOG_LWDEBUG("LWLockWaitForVar", T_NAME(lock), T_ID(lock), "waiting");
#ifdef LWLOCK_STATS
lwstats->block_count++;
#endif
TRACE_POSTGRESQL_LWLOCK_WAIT_START(T_NAME(lock), T_ID(lock),
LW_EXCLUSIVE);
for (;;)
{
/* "false" means cannot accept cancel/die interrupt here. */
PGSemaphoreLock(&proc->sem, false);
if (!proc->lwWaiting)
break;
extraWaits++;
}
TRACE_POSTGRESQL_LWLOCK_WAIT_DONE(T_NAME(lock), T_ID(lock),
LW_EXCLUSIVE);
LOG_LWDEBUG("LWLockWaitForVar", T_NAME(lock), T_ID(lock), "awakened");
/* Now loop back and check the status of the lock again. */
}
/* We are done updating shared state of the lock itself. */
SpinLockRelease(&lock->mutex);
TRACE_POSTGRESQL_LWLOCK_ACQUIRE(T_NAME(lock), T_ID(lock), LW_EXCLUSIVE);
/*
* Fix the process wait semaphore's count for any absorbed wakeups.
*/
while (extraWaits-- > 0)
PGSemaphoreUnlock(&proc->sem);
/*
* Now okay to allow cancel/die interrupts.
*/
RESUME_INTERRUPTS();
return result;
}
/*
* LWLockAcquireOrWait - Acquire lock, or wait until it's free
*
* The semantics of this function are a bit funky. If the lock is currently
* free, it is acquired in the given mode, and the function returns true. If
* the lock isn't immediately free, the function waits until it is released
* and returns false, but does not acquire the lock.
*
* This is currently used for WALWriteLock: when a backend flushes the WAL,
* holding WALWriteLock, it can flush the commit records of many other
* backends as a side-effect. Those other backends need to wait until the
* flush finishes, but don't need to acquire the lock anymore. They can just
* wake up, observe that their records have already been flushed, and return.
*/
bool
LWLockAcquireOrWait(LWLock *lock, LWLockMode mode)
{
PGPROC *proc = MyProc;
bool mustwait;
int extraWaits = 0;
#ifdef LWLOCK_STATS
lwlock_stats *lwstats;
#endif
PRINT_LWDEBUG("LWLockAcquireOrWait", lock);
#ifdef LWLOCK_STATS
lwstats = get_lwlock_stats_entry(lock);
#endif
/* Ensure we will have room to remember the lock */
if (num_held_lwlocks >= MAX_SIMUL_LWLOCKS)
elog(ERROR, "too many LWLocks taken");
/*
* Lock out cancel/die interrupts until we exit the code section protected
* by the LWLock. This ensures that interrupts will not interfere with
* manipulations of data structures in shared memory.
*/
HOLD_INTERRUPTS();
/* Acquire mutex. Time spent holding mutex should be short! */
SpinLockAcquire(&lock->mutex);
/* If I can get the lock, do so quickly. */
if (mode == LW_EXCLUSIVE)
{
if (lock->exclusive == 0 && lock->shared == 0)
{
lock->exclusive++;
mustwait = false;
}
else
mustwait = true;
}
else
{
if (lock->exclusive == 0)
{
lock->shared++;
mustwait = false;
}
else
mustwait = true;
}
if (mustwait)
{
/*
* Add myself to wait queue.
*
* If we don't have a PGPROC structure, there's no way to wait. This
* should never occur, since MyProc should only be null during shared
* memory initialization.
*/
if (proc == NULL)
elog(PANIC, "cannot wait without a PGPROC structure");
proc->lwWaiting = true;
proc->lwWaitMode = LW_WAIT_UNTIL_FREE;
proc->lwWaitLink = NULL;
if (lock->head == NULL)
lock->head = proc;
else
lock->tail->lwWaitLink = proc;
lock->tail = proc;
/* Can release the mutex now */
SpinLockRelease(&lock->mutex);
/*
* Wait until awakened. Like in LWLockAcquire, be prepared for bogus
* wakups, because we share the semaphore with ProcWaitForSignal.
*/
LOG_LWDEBUG("LWLockAcquireOrWait", T_NAME(lock), T_ID(lock),
"waiting");
#ifdef LWLOCK_STATS
lwstats->block_count++;
#endif
TRACE_POSTGRESQL_LWLOCK_WAIT_START(T_NAME(lock), T_ID(lock), mode);
for (;;)
{
/* "false" means cannot accept cancel/die interrupt here. */
PGSemaphoreLock(&proc->sem, false);
if (!proc->lwWaiting)
break;
extraWaits++;
}
TRACE_POSTGRESQL_LWLOCK_WAIT_DONE(T_NAME(lock), T_ID(lock), mode);
LOG_LWDEBUG("LWLockAcquireOrWait", T_NAME(lock), T_ID(lock),
"awakened");
}
else
{
/* We are done updating shared state of the lock itself. */
SpinLockRelease(&lock->mutex);
}
/*
* Fix the process wait semaphore's count for any absorbed wakeups.
*/
while (extraWaits-- > 0)
PGSemaphoreUnlock(&proc->sem);
if (mustwait)
{
/* Failed to get lock, so release interrupt holdoff */
RESUME_INTERRUPTS();
LOG_LWDEBUG("LWLockAcquireOrWait", T_NAME(lock), T_ID(lock), "failed");
TRACE_POSTGRESQL_LWLOCK_ACQUIRE_OR_WAIT_FAIL(T_NAME(lock), T_ID(lock),
mode);
}
else
{
/* Add lock to list of locks held by this backend */
held_lwlocks[num_held_lwlocks++] = lock;
TRACE_POSTGRESQL_LWLOCK_ACQUIRE_OR_WAIT(T_NAME(lock), T_ID(lock),
mode);
}
return !mustwait;
}
/*
* LWLockRelease - release a previously acquired lock
*/
void
LWLockRelease(LWLock *lock)
{
PGPROC *head;
PGPROC *proc;
int i;
PRINT_LWDEBUG("LWLockRelease", lock);
/*
* Remove lock from list of locks held. Usually, but not always, it will
* be the latest-acquired lock; so search array backwards.
*/
for (i = num_held_lwlocks; --i >= 0;)
{
if (lock == held_lwlocks[i])
break;
}
if (i < 0)
elog(ERROR, "lock %s %d is not held", T_NAME(lock), T_ID(lock));
num_held_lwlocks--;
for (; i < num_held_lwlocks; i++)
held_lwlocks[i] = held_lwlocks[i + 1];
/* Acquire mutex. Time spent holding mutex should be short! */
SpinLockAcquire(&lock->mutex);
/* Release my hold on lock */
if (lock->exclusive > 0)
lock->exclusive--;
else
{
Assert(lock->shared > 0);
lock->shared--;
}
/*
* See if I need to awaken any waiters. If I released a non-last shared
* hold, there cannot be anything to do. Also, do not awaken any waiters
* if someone has already awakened waiters that haven't yet acquired the
* lock.
*/
head = lock->head;
if (head != NULL)
{
if (lock->exclusive == 0 && lock->shared == 0 && lock->releaseOK)
{
/*
* Remove the to-be-awakened PGPROCs from the queue.
*/
bool releaseOK = true;
proc = head;
/*
* First wake up any backends that want to be woken up without
* acquiring the lock.
*/
while (proc->lwWaitMode == LW_WAIT_UNTIL_FREE && proc->lwWaitLink)
proc = proc->lwWaitLink;
/*
* If the front waiter wants exclusive lock, awaken him only.
* Otherwise awaken as many waiters as want shared access.
*/
if (proc->lwWaitMode != LW_EXCLUSIVE)
{
while (proc->lwWaitLink != NULL &&
proc->lwWaitLink->lwWaitMode != LW_EXCLUSIVE)
{
if (proc->lwWaitMode != LW_WAIT_UNTIL_FREE)
releaseOK = false;
proc = proc->lwWaitLink;
}
}
/* proc is now the last PGPROC to be released */
lock->head = proc->lwWaitLink;
proc->lwWaitLink = NULL;
/*
* Prevent additional wakeups until retryer gets to run. Backends
* that are just waiting for the lock to become free don't retry
* automatically.
*/
if (proc->lwWaitMode != LW_WAIT_UNTIL_FREE)
releaseOK = false;
lock->releaseOK = releaseOK;
}
else
{
/* lock is still held, can't awaken anything */
head = NULL;
}
}
/* We are done updating shared state of the lock itself. */
SpinLockRelease(&lock->mutex);
TRACE_POSTGRESQL_LWLOCK_RELEASE(T_NAME(lock), T_ID(lock));
/*
* Awaken any waiters I removed from the queue.
*/
while (head != NULL)
{
LOG_LWDEBUG("LWLockRelease", T_NAME(lock), T_ID(lock),
"release waiter");
proc = head;
head = proc->lwWaitLink;
proc->lwWaitLink = NULL;
/*
* Guarantee that lwWaiting being unset only becomes visible once the
* unlink from the link has completed. Otherwise the target backend
* could be woken up for other reason and enqueue for a new lock - if
* that happens before the list unlink happens, the list would end up
* being corrupted.
*
* The barrier pairs with the SpinLockAcquire() when enqueing for
* another lock.
*/
pg_write_barrier();
proc->lwWaiting = false;
PGSemaphoreUnlock(&proc->sem);
}
/*
* Now okay to allow cancel/die interrupts.
*/
RESUME_INTERRUPTS();
}
int
main(int argc, char **argv)
{
MyStorage *storage;
int cpid;
printf("Creating shared memory ... ");
fflush(stdout);
storage = (MyStorage *) PGSharedMemoryCreate(8192, false, 5433);
storage->flag = 1234;
printf("OK\n");
printf("Creating semaphores ... ");
fflush(stdout);
PGReserveSemaphores(2, 5433);
PGSemaphoreCreate(&storage->sem);
printf("OK\n");
/* sema initial value is 1, so lock should work */
printf("Testing Lock ... ");
fflush(stdout);
PGSemaphoreLock(&storage->sem, false);
printf("OK\n");
/* now sema value is 0, so trylock should fail */
printf("Testing TryLock ... ");
fflush(stdout);
if (PGSemaphoreTryLock(&storage->sem))
printf("unexpected result!\n");
else
printf("OK\n");
/* unlocking twice and then locking twice should work... */
printf("Testing Multiple Lock ... ");
fflush(stdout);
PGSemaphoreUnlock(&storage->sem);
PGSemaphoreUnlock(&storage->sem);
PGSemaphoreLock(&storage->sem, false);
PGSemaphoreLock(&storage->sem, false);
printf("OK\n");
/* check Reset too */
printf("Testing Reset ... ");
fflush(stdout);
PGSemaphoreUnlock(&storage->sem);
PGSemaphoreReset(&storage->sem);
if (PGSemaphoreTryLock(&storage->sem))
printf("unexpected result!\n");
else
printf("OK\n");
/* Fork a child process and see if it can communicate */
printf("Forking child process ... ");
fflush(stdout);
cpid = fork();
if (cpid == 0)
{
/* In child */
on_exit_reset();
sleep(3);
storage->flag++;
PGSemaphoreUnlock(&storage->sem);
proc_exit(0);
}
if (cpid < 0)
{
/* Fork failed */
printf("failed: %s\n", strerror(errno));
proc_exit(1);
}
printf("forked child pid %d OK\n", cpid);
if (storage->flag != 1234)
printf("Wrong value found in shared memory!\n");
printf("Waiting for child (should wait 3 sec here) ... ");
fflush(stdout);
PGSemaphoreLock(&storage->sem, false);
printf("OK\n");
if (storage->flag != 1235)
printf("Wrong value found in shared memory!\n");
/* Test shutdown */
printf("Running shmem_exit processing ... ");
fflush(stdout);
shmem_exit(0);
printf("OK\n");
printf("Tests complete.\n");
proc_exit(0);
return 0; /* not reached */
}
/*
* LWLockAcquire - acquire a lightweight lock in the specified mode
*
* If the lock is not available, sleep until it is.
*
* Side effect: cancel/die interrupts are held off until lock release.
*/
void
LWLockAcquire(LWLockId lockid, LWLockMode mode)
{
volatile LWLock *lock = &(LWLockArray[lockid].lock);
#if LWLOCK_LOCK_PARTS > 1
volatile LWLockPart *part = LWLOCK_PART(lock, lockid, MyBackendId);
#endif
PGPROC *proc = MyProc;
bool retry = false;
int extraWaits = 0;
PRINT_LWDEBUG("LWLockAcquire", lockid, lock);
#ifdef LWLOCK_STATS
/* Set up local count state first time through in a given process */
if (counts_for_pid != MyProcPid)
{
int *LWLockCounter = (int *) ((char *) LWLockArray - 2 * sizeof(int));
int numLocks = LWLockCounter[1];
sh_acquire_counts = calloc(numLocks, sizeof(int));
ex_acquire_counts = calloc(numLocks, sizeof(int));
block_counts = calloc(numLocks, sizeof(int));
counts_for_pid = MyProcPid;
on_shmem_exit(print_lwlock_stats, 0);
}
/* Count lock acquisition attempts */
if (mode == LW_EXCLUSIVE)
ex_acquire_counts[lockid]++;
else
sh_acquire_counts[lockid]++;
#endif /* LWLOCK_STATS */
/*
* We can't wait if we haven't got a PGPROC. This should only occur
* during bootstrap or shared memory initialization. Put an Assert here
* to catch unsafe coding practices.
*/
Assert(!(proc == NULL && IsUnderPostmaster));
/* Ensure we will have room to remember the lock */
if (num_held_lwlocks >= MAX_SIMUL_LWLOCKS)
elog(ERROR, "too many LWLocks taken");
/*
* Lock out cancel/die interrupts until we exit the code section protected
* by the LWLock. This ensures that interrupts will not interfere with
* manipulations of data structures in shared memory.
*/
HOLD_INTERRUPTS();
/*
* Loop here to try to acquire lock after each time we are signaled by
* LWLockRelease.
*
* NOTE: it might seem better to have LWLockRelease actually grant us the
* lock, rather than retrying and possibly having to go back to sleep. But
* in practice that is no good because it means a process swap for every
* lock acquisition when two or more processes are contending for the same
* lock. Since LWLocks are normally used to protect not-very-long
* sections of computation, a process needs to be able to acquire and
* release the same lock many times during a single CPU time slice, even
* in the presence of contention. The efficiency of being able to do that
* outweighs the inefficiency of sometimes wasting a process dispatch
* cycle because the lock is not free when a released waiter finally gets
* to run. See pgsql-hackers archives for 29-Dec-01.
*/
for (;;)
{
bool mustwait;
if (mode == LW_SHARED)
{
#ifdef LWLOCK_PART_SHARED_OPS_ATOMIC
/* Increment shared counter partition. If there's no contention,
* this is sufficient to take the lock
*/
LWLOCK_PART_SHARED_POSTINC_ATOMIC(lock, lockid, part, MyBackendId);
LWLOCK_PART_SHARED_FENCE();
/* A concurrent exclusive locking attempt does the following
* three steps
* 1) Acquire mutex
* 2) Check shared counter partitions for readers.
* 3a) If found add proc to wait queue, block, restart at (1)
* 3b) If not found, set exclusive flag, continue with (4)
* 4) Enter protected section
* The fence after the atomic add above ensures that no further
* such attempt can proceed to (3b) or beyond. There may be
* pre-existing exclusive locking attempts at step (3b) or beyond,
* but we can recognize those by either the mutex being taken, or
* the exclusive flag being set. Conversely, if we see neither, we
* may proceed and enter the protected section.
*
* FIXME: This doesn't work if slock_t is a struct or doesn't
* use 0 for state "unlocked".
*/
if ((lock->mutex == 0) && (lock->exclusive == 0)) {
/* If retrying, allow LWLockRelease to release waiters again.
* Usually this happens after we acquired the mutex, but if
* we skip that, we still need to set releaseOK.
*
* Acquiring the mutex here is not really an option - if many
* reader are awoken simultaneously by an exclusive unlock,
* that would be a source of considerable contention.
*
* Fotunately, this is safe even without the mutex. First,
* there actually cannot be any non-fast path unlocking
* attempt in progress, because we'd then either still see
* the exclusive flag set or the mutex being taken. And
* even if there was, and such an attempt cleared the flag
* immediately after we set it, it'd also wake up some waiter
* who'd then re-set the flag.
*
* The only reason to do this here, and not directly
* after returning from PGSemaphoreLock(), is that it seems
* benefical to make SpinLockAcquire() the first thing to
* touch the lock if possible, in case we acquire the spin
* lock at all. That way, the cache line doesn't go through
* a possible shared state, but instead directly to exclusive.
* On Opterons at least, there seems to be a difference, c.f.
* the comment above tas() for x86_64 in s_lock.h
*/
if (retry && !lock->releaseOK)
lock->releaseOK = true;
goto lock_acquired;
}
/* At this point, we don't know if the concurrent exclusive locker
* has proceeded to (3b) or blocked. We must take the mutex and
* re-check
*/
#endif /* LWLOCK_PART_SHARED_OPS_ATOMIC */
/* Acquire mutex. Time spent holding mutex should be short! */
SpinLockAcquire(&lock->mutex);
if (lock->exclusive == 0)
{
#ifdef LWLOCK_PART_SHARED_OPS_ATOMIC
/* Already incremented the shared counter partition above */
#else
lock->shared++;
#endif
mustwait = false;
}
else
{
#ifdef LWLOCK_PART_SHARED_OPS_ATOMIC
/* Must undo shared counter partition increment. Note that
* we *need* to do that while holding the mutex. Otherwise,
* the exclusive lock could be released and attempted to be
* re-acquired before we undo the increment. That attempt
* would then block, even though there'd be no lock holder
* left
*/
LWLOCK_PART_SHARED_POSTDEC_ATOMIC(lock, lockid, part, MyBackendId);
#endif
mustwait = true;
}
}
else
{
/* Step (1). Acquire mutex. Time spent holding mutex should be
* short!
*/
SpinLockAcquire(&lock->mutex);
if (lock->exclusive == 0)
{
/* Step (2). Check for shared lockers. This surely happens
* after (1), otherwise SpinLockAcquire() is broken. Lock
* acquire semantics demand that no load must be re-ordered
* from after a lock acquisition to before, for obvious
* reasons.
*/
LWLOCK_IS_SHARED(mustwait, lock, lockid);
if (!mustwait) {
/* Step (3a). Set exclusive flag. This surely happens
* after (2) because it depends on the result of (2),
* no matter how much reordering is going on here.
*/
lock->exclusive++;
}
}
else
mustwait = true;
}
/* If retrying, allow LWLockRelease to release waiters again.
* This is also separately done in the LW_SHARED early exit case
* above, and in contrast to there we don't hold the mutex there.
* See the comment there for why this is safe
*/
if (retry)
lock->releaseOK = true;
if (!mustwait)
break; /* got the lock */
/*
* Step (3b). Add myself to wait queue.
*
* If we don't have a PGPROC structure, there's no way to wait. This
* should never occur, since MyProc should only be null during shared
* memory initialization.
*/
if (proc == NULL)
elog(PANIC, "cannot wait without a PGPROC structure");
proc->lwWaiting = true;
proc->lwExclusive = (mode == LW_EXCLUSIVE);
proc->lwWaitLink = NULL;
if (lock->head == NULL)
lock->head = proc;
else
lock->tail->lwWaitLink = proc;
lock->tail = proc;
/* Can release the mutex now */
SpinLockRelease(&lock->mutex);
/*
* Wait until awakened.
*
* Since we share the process wait semaphore with the regular lock
* manager and ProcWaitForSignal, and we may need to acquire an LWLock
* while one of those is pending, it is possible that we get awakened
* for a reason other than being signaled by LWLockRelease. If so,
* loop back and wait again. Once we've gotten the LWLock,
* re-increment the sema by the number of additional signals received,
* so that the lock manager or signal manager will see the received
* signal when it next waits.
*/
LOG_LWDEBUG("LWLockAcquire", lockid, "waiting");
#ifdef LWLOCK_STATS
block_counts[lockid]++;
#endif
TRACE_POSTGRESQL_LWLOCK_WAIT_START(lockid, mode);
for (;;)
{
/* "false" means cannot accept cancel/die interrupt here. */
PGSemaphoreLock(&proc->sem, false);
if (!proc->lwWaiting)
break;
extraWaits++;
}
TRACE_POSTGRESQL_LWLOCK_WAIT_DONE(lockid, mode);
LOG_LWDEBUG("LWLockAcquire", lockid, "awakened");
/* Now loop back and try to acquire lock again. */
retry = true;
}
/* We are done updating shared state of the lock itself. */
SpinLockRelease(&lock->mutex);
/* Step 4. Enter protected section. This surely happens after (3),
* this time because lock release semantics demand that no store
* must be moved from before a lock release to after the release,
* again for obvious reasons
*/
#ifdef LWLOCK_PART_SHARED_OPS_ATOMIC
lock_acquired:
#endif
TRACE_POSTGRESQL_LWLOCK_ACQUIRE(lockid, mode);
/* Add lock to list of locks held by this backend */
held_lwlocks[num_held_lwlocks] = lockid;
held_lwlocks_mode[num_held_lwlocks] = mode;
++num_held_lwlocks;
/*
* Fix the process wait semaphore's count for any absorbed wakeups.
*/
while (extraWaits-- > 0)
PGSemaphoreUnlock(&proc->sem);
}
/*
* CheckDeadLock
*
* We only get to this routine if the DEADLOCK_TIMEOUT fired
* while waiting for a lock to be released by some other process. Look
* to see if there's a deadlock; if not, just return and continue waiting.
* (But signal ProcSleep to log a message, if log_lock_waits is true.)
* If we have a real deadlock, remove ourselves from the lock's wait queue
* and signal an error to ProcSleep.
*
* NB: this is run inside a signal handler, so be very wary about what is done
* here or in called routines.
*/
void
CheckDeadLock(void)
{
int i;
/*
* Acquire exclusive lock on the entire shared lock data structures. Must
* grab LWLocks in partition-number order to avoid LWLock deadlock.
*
* Note that the deadlock check interrupt had better not be enabled
* anywhere that this process itself holds lock partition locks, else this
* will wait forever. Also note that LWLockAcquire creates a critical
* section, so that this routine cannot be interrupted by cancel/die
* interrupts.
*/
for (i = 0; i < NUM_LOCK_PARTITIONS; i++)
LWLockAcquire(FirstLockMgrLock + i, LW_EXCLUSIVE);
/*
* Check to see if we've been awoken by anyone in the interim.
*
* If we have, we can return and resume our transaction -- happy day.
* Before we are awoken the process releasing the lock grants it to us so
* we know that we don't have to wait anymore.
*
* We check by looking to see if we've been unlinked from the wait queue.
* This is quicker than checking our semaphore's state, since no kernel
* call is needed, and it is safe because we hold the lock partition lock.
*/
if (MyProc->links.prev == NULL ||
MyProc->links.next == NULL)
goto check_done;
#ifdef LOCK_DEBUG
if (Debug_deadlocks)
DumpAllLocks();
#endif
/* Run the deadlock check, and set deadlock_state for use by ProcSleep */
deadlock_state = DeadLockCheck(MyProc);
if (deadlock_state == DS_HARD_DEADLOCK)
{
/*
* Oops. We have a deadlock.
*
* Get this process out of wait state. (Note: we could do this more
* efficiently by relying on lockAwaited, but use this coding to
* preserve the flexibility to kill some other transaction than the
* one detecting the deadlock.)
*
* RemoveFromWaitQueue sets MyProc->waitStatus to STATUS_ERROR, so
* ProcSleep will report an error after we return from the signal
* handler.
*/
Assert(MyProc->waitLock != NULL);
RemoveFromWaitQueue(MyProc, LockTagHashCode(&(MyProc->waitLock->tag)));
/*
* Unlock my semaphore so that the interrupted ProcSleep() call can
* finish.
*/
PGSemaphoreUnlock(&MyProc->sem);
/*
* We're done here. Transaction abort caused by the error that
* ProcSleep will raise will cause any other locks we hold to be
* released, thus allowing other processes to wake up; we don't need
* to do that here. NOTE: an exception is that releasing locks we
* hold doesn't consider the possibility of waiters that were blocked
* behind us on the lock we just failed to get, and might now be
* wakable because we're not in front of them anymore. However,
* RemoveFromWaitQueue took care of waking up any such processes.
*/
}
else if (log_lock_waits || deadlock_state == DS_BLOCKED_BY_AUTOVACUUM)
{
/*
* Unlock my semaphore so that the interrupted ProcSleep() call can
* print the log message (we daren't do it here because we are inside
* a signal handler). It will then sleep again until someone releases
* the lock.
*
* If blocked by autovacuum, this wakeup will enable ProcSleep to send
* the canceling signal to the autovacuum worker.
*/
PGSemaphoreUnlock(&MyProc->sem);
}
/*
* And release locks. We do this in reverse order for two reasons: (1)
* Anyone else who needs more than one of the locks will be trying to lock
* them in increasing order; we don't want to release the other process
* until it can get all the locks it needs. (2) This avoids O(N^2)
* behavior inside LWLockRelease.
*/
check_done:
for (i = NUM_LOCK_PARTITIONS; --i >= 0;)
LWLockRelease(FirstLockMgrLock + i);
}
/*
* LWLockRelease - release a previously acquired lock
*/
void
LWLockRelease(LWLockId lockid)
{
volatile LWLock *lock = &(LWLockArray[lockid].lock);
PGPROC *head;
PGPROC *proc;
int i;
bool saveExclusive;
PRINT_LWDEBUG("LWLockRelease", lockid, lock);
/*
* Remove lock from list of locks held. Usually, but not always, it will
* be the latest-acquired lock; so search array backwards.
*/
for (i = num_held_lwlocks; --i >= 0;)
{
if (lockid == held_lwlocks[i])
break;
}
if (i < 0)
elog(ERROR, "lock %d is not held", (int) lockid);
saveExclusive = held_lwlocks_exclusive[i];
if (InterruptHoldoffCount <= 0)
elog(PANIC, "upon entering lock release, the interrupt holdoff count is bad (%d) for release of lock %d (%s)",
InterruptHoldoffCount,
(int)lockid,
(saveExclusive ? "Exclusive" : "Shared"));
#ifdef LWLOCK_TRACE_MIRROREDLOCK
if (lockid == MirroredLock)
elog(LOG,
"LWLockRelease: release for MirroredLock by PID %u in held_lwlocks[%d] %s",
MyProcPid,
i,
(held_lwlocks_exclusive[i] ? "Exclusive" : "Shared"));
#endif
num_held_lwlocks--;
for (; i < num_held_lwlocks; i++)
{
held_lwlocks_exclusive[i] = held_lwlocks_exclusive[i + 1];
held_lwlocks[i] = held_lwlocks[i + 1];
#ifdef USE_TEST_UTILS_X86
/* shift stack traces */
held_lwlocks_depth[i] = held_lwlocks_depth[i + 1];
memcpy
(
held_lwlocks_addresses[i],
held_lwlocks_addresses[i + 1],
held_lwlocks_depth[i] * sizeof(*held_lwlocks_depth)
)
;
#endif /* USE_TEST_UTILS_X86 */
}
// Clear out old last entry.
held_lwlocks_exclusive[num_held_lwlocks] = false;
held_lwlocks[num_held_lwlocks] = 0;
#ifdef USE_TEST_UTILS_X86
held_lwlocks_depth[num_held_lwlocks] = 0;
#endif /* USE_TEST_UTILS_X86 */
/* Acquire mutex. Time spent holding mutex should be short! */
SpinLockAcquire(&lock->mutex);
/* Release my hold on lock */
if (lock->exclusive > 0)
{
lock->exclusive--;
lock->exclusivePid = 0;
}
else
{
Assert(lock->shared > 0);
lock->shared--;
}
/*
* See if I need to awaken any waiters. If I released a non-last shared
* hold, there cannot be anything to do. Also, do not awaken any waiters
* if someone has already awakened waiters that haven't yet acquired the
* lock.
*/
head = lock->head;
if (head != NULL)
{
if (lock->exclusive == 0 && lock->shared == 0 && lock->releaseOK)
{
/*
* Remove the to-be-awakened PGPROCs from the queue. If the front
* waiter wants exclusive lock, awaken him only. Otherwise awaken
* as many waiters as want shared access.
*/
proc = head;
if (!proc->lwExclusive)
{
while (proc->lwWaitLink != NULL &&
!proc->lwWaitLink->lwExclusive)
{
proc = proc->lwWaitLink;
if (proc->pid != 0)
{
lock->releaseOK = false;
}
}
}
/* proc is now the last PGPROC to be released */
lock->head = proc->lwWaitLink;
proc->lwWaitLink = NULL;
/* proc->pid can be 0 if process exited while waiting for lock */
if (proc->pid != 0)
{
/* prevent additional wakeups until retryer gets to run */
lock->releaseOK = false;
}
}
else
{
/* lock is still held, can't awaken anything */
head = NULL;
}
}
/* We are done updating shared state of the lock itself. */
SpinLockRelease(&lock->mutex);
PG_TRACE1(lwlock__release, lockid);
/*
* Awaken any waiters I removed from the queue.
*/
while (head != NULL)
{
#ifdef LWLOCK_TRACE_MIRROREDLOCK
if (lockid == MirroredLock)
elog(LOG, "LWLockRelease: release waiter for MirroredLock (this PID %u", MyProcPid);
#endif
LOG_LWDEBUG("LWLockRelease", lockid, "release waiter");
proc = head;
head = proc->lwWaitLink;
proc->lwWaitLink = NULL;
pg_write_barrier();
proc->lwWaiting = false;
PGSemaphoreUnlock(&proc->sem);
}
/*
* Now okay to allow cancel/die interrupts.
*/
if (InterruptHoldoffCount <= 0)
elog(PANIC, "upon exiting lock release, the interrupt holdoff count is bad (%d) for release of lock %d (%s)",
InterruptHoldoffCount,
(int)lockid,
(saveExclusive ? "Exclusive" : "Shared"));
RESUME_INTERRUPTS();
}
/*
* LWLockAcquire - acquire a lightweight lock in the specified mode
*
* If the lock is not available, sleep until it is.
*
* Side effect: cancel/die interrupts are held off until lock release.
*/
void
LWLockAcquire(LWLockId lockid, LWLockMode mode)
{
volatile LWLock *lock = LWLockArray + lockid;
PGPROC *proc = MyProc;
bool retry = false;
int extraWaits = 0;
PRINT_LWDEBUG("LWLockAcquire", lockid, lock);
/*
* We can't wait if we haven't got a PGPROC. This should only occur
* during bootstrap or shared memory initialization. Put an Assert
* here to catch unsafe coding practices.
*/
Assert(!(proc == NULL && IsUnderPostmaster));
/*
* Lock out cancel/die interrupts until we exit the code section
* protected by the LWLock. This ensures that interrupts will not
* interfere with manipulations of data structures in shared memory.
*/
HOLD_INTERRUPTS();
/*
* Loop here to try to acquire lock after each time we are signaled by
* LWLockRelease.
*
* NOTE: it might seem better to have LWLockRelease actually grant us the
* lock, rather than retrying and possibly having to go back to sleep.
* But in practice that is no good because it means a process swap for
* every lock acquisition when two or more processes are contending
* for the same lock. Since LWLocks are normally used to protect
* not-very-long sections of computation, a process needs to be able
* to acquire and release the same lock many times during a single CPU
* time slice, even in the presence of contention. The efficiency of
* being able to do that outweighs the inefficiency of sometimes
* wasting a process dispatch cycle because the lock is not free when
* a released waiter finally gets to run. See pgsql-hackers archives
* for 29-Dec-01.
*/
for (;;)
{
bool mustwait;
/* Acquire mutex. Time spent holding mutex should be short! */
SpinLockAcquire_NoHoldoff(&lock->mutex);
/* If retrying, allow LWLockRelease to release waiters again */
if (retry)
lock->releaseOK = true;
/* If I can get the lock, do so quickly. */
if (mode == LW_EXCLUSIVE)
{
if (lock->exclusive == 0 && lock->shared == 0)
{
lock->exclusive++;
mustwait = false;
}
else
mustwait = true;
}
else
{
if (lock->exclusive == 0)
{
lock->shared++;
mustwait = false;
}
else
mustwait = true;
}
if (!mustwait)
break; /* got the lock */
/*
* Add myself to wait queue.
*
* If we don't have a PGPROC structure, there's no way to wait. This
* should never occur, since MyProc should only be null during
* shared memory initialization.
*/
if (proc == NULL)
elog(FATAL, "cannot wait without a PGPROC structure");
proc->lwWaiting = true;
proc->lwExclusive = (mode == LW_EXCLUSIVE);
proc->lwWaitLink = NULL;
if (lock->head == NULL)
lock->head = proc;
else
lock->tail->lwWaitLink = proc;
lock->tail = proc;
/* Can release the mutex now */
SpinLockRelease_NoHoldoff(&lock->mutex);
/*
* Wait until awakened.
*
* Since we share the process wait semaphore with the regular lock
* manager and ProcWaitForSignal, and we may need to acquire an
* LWLock while one of those is pending, it is possible that we
* get awakened for a reason other than being signaled by
* LWLockRelease. If so, loop back and wait again. Once we've
* gotten the LWLock, re-increment the sema by the number of
* additional signals received, so that the lock manager or signal
* manager will see the received signal when it next waits.
*/
LOG_LWDEBUG("LWLockAcquire", lockid, "waiting");
for (;;)
{
/* "false" means cannot accept cancel/die interrupt here. */
PGSemaphoreLock(&proc->sem, false);
if (!proc->lwWaiting)
break;
extraWaits++;
}
LOG_LWDEBUG("LWLockAcquire", lockid, "awakened");
/* Now loop back and try to acquire lock again. */
retry = true;
}
/* We are done updating shared state of the lock itself. */
SpinLockRelease_NoHoldoff(&lock->mutex);
/* Add lock to list of locks held by this backend */
Assert(num_held_lwlocks < MAX_SIMUL_LWLOCKS);
held_lwlocks[num_held_lwlocks++] = lockid;
/*
* Fix the process wait semaphore's count for any absorbed wakeups.
*/
while (extraWaits-- > 0)
PGSemaphoreUnlock(&proc->sem);
}
/*
* CheckDeadLock
*
* We only get to this routine if we got SIGALRM after DeadlockTimeout
* while waiting for a lock to be released by some other process. Look
* to see if there's a deadlock; if not, just return and continue waiting.
* (But signal ProcSleep to log a message, if log_lock_waits is true.)
* If we have a real deadlock, remove ourselves from the lock's wait queue
* and signal an error to ProcSleep.
*
* NB: this is run inside a signal handler, so be very wary about what is done
* here or in called routines.
*/
static void
CheckDeadLock(void)
{
int i;
/*
* Acquire exclusive lock on the entire shared lock data structures. Must
* grab LWLocks in partition-number order to avoid LWLock deadlock.
*
* Note that the deadlock check interrupt had better not be enabled
* anywhere that this process itself holds lock partition locks, else this
* will wait forever. Also note that LWLockAcquire creates a critical
* section, so that this routine cannot be interrupted by cancel/die
* interrupts.
*/
for (i = 0; i < NUM_LOCK_PARTITIONS; i++)
LWLockAcquire(FirstLockMgrLock + i, LW_EXCLUSIVE);
/*
* Check to see if we've been awoken by anyone in the interim.
*
* If we have, we can return and resume our transaction -- happy day.
* Before we are awoken the process releasing the lock grants it to us
* so we know that we don't have to wait anymore.
*
* We check by looking to see if we've been unlinked from the wait queue.
* This is quicker than checking our semaphore's state, since no kernel
* call is needed, and it is safe because we hold the lock partition lock.
*/
if (MyProc->links.prev == INVALID_OFFSET ||
MyProc->links.next == INVALID_OFFSET)
goto check_done;
#ifdef LOCK_DEBUG
if (Debug_deadlocks)
DumpAllLocks();
#endif
if (!DeadLockCheck(MyProc))
{
/* No deadlock, so keep waiting */
goto check_done;
}
/*
* Unlock my semaphore so that the interrupted ProcSleep() call can
* finish.
*/
PGSemaphoreUnlock(&MyProc->sem);
/*
* We're done here. Transaction abort caused by the error that ProcSleep
* will raise will cause any other locks we hold to be released, thus
* allowing other processes to wake up; we don't need to do that here.
* NOTE: an exception is that releasing locks we hold doesn't consider the
* possibility of waiters that were blocked behind us on the lock we just
* failed to get, and might now be wakable because we're not in front of
* them anymore. However, RemoveFromWaitQueue took care of waking up any
* such processes.
*/
/*
* Release locks acquired at head of routine. Order is not critical, so
* do it back-to-front to avoid waking another CheckDeadLock instance
* before it can get all the locks.
*/
check_done:
for (i = NUM_LOCK_PARTITIONS; --i >= 0;)
LWLockRelease(FirstLockMgrLock + i);
}
/*
* LWLockRelease - release a previously acquired lock
*/
void
LWLockRelease(LWLockId lockid)
{
volatile LWLock *lock = &(LWLockArray[lockid].lock);
#if LWLOCK_LOCK_PARTS > 1
volatile LWLockPart *part = LWLOCK_PART(lock, lockid, MyBackendId);
#endif
PGPROC *head = NULL;
PGPROC *proc;
LWLockMode mode;
int i;
PRINT_LWDEBUG("LWLockRelease", lockid, lock);
/*
* Remove lock from list of locks held. Usually, but not always, it will
* be the latest-acquired lock; so search array backwards.
*/
for (i = num_held_lwlocks; --i >= 0;)
{
if (lockid == held_lwlocks[i])
break;
}
if (i < 0)
elog(ERROR, "lock %d is not held", (int) lockid);
mode = held_lwlocks_mode[i];
num_held_lwlocks--;
for (; i < num_held_lwlocks; i++) {
held_lwlocks[i] = held_lwlocks[i + 1];
held_lwlocks_mode[i] = held_lwlocks_mode[i + 1];
}
if (mode == LW_SHARED) {
#ifdef LWLOCK_PART_SHARED_OPS_ATOMIC
int shared_pre;
/* Release my hold on lock */
Assert(lock->exclusive == 0);
shared_pre = LWLOCK_PART_SHARED_POSTDEC_ATOMIC(lock, lockid, part, MyBackendId);
Assert(shared_pre > 0);
/* If the count didn't drop to zero (i.e., there are more lockers
* using the same shared counter partition), we can leave waiting
* up blocked exclusive locking attempts to them. Note that there
* may also be shared lockers using a *different* partition, so
* we're not necessarily the last share lockers, even if we continue.
* Still, it's an easy optimization, so we got for it
*/
if (shared_pre > 1)
goto lock_released;
LWLOCK_PART_SHARED_FENCE();
/* A concurrent exclusive locking attempt does the following
* three steps
* 1) Acquire mutex
* 2) Check shared counter partitions for readers.
* 3a) If found add proc to wait queue, block, restart at (1)
* 3b) If not found, set exclusive flag, continue with (4)
* 4) Enter protected section
* Assume now that we're that last share lock holder. Then, the
* fence after the atomic add above ensures that no further such
* concurrent exclusive locking attempts will proceed to (3a) and
* thus block. There may be such attempts currently blocking or
* about to block, but we can recognize those by either wait queue
* being non-empty or the mutex being taken. Conversely, if we see
* neither, we may assume that nobody needs to be signalled.
*
* Note that if two shared lockers release their lock while
* an exclusive locking attempt is in progress, both may decide
* they need to signal here. Taking the mutex below will sort that
* out, but it's a bit unfortunate that they have to race for the
* mutex here. Also, taking the mutex will force *other* shared
* lockers to take the mutex also in their release path.
* XXX: We may be able to improve that if we could dinstinguish
* been mutexed held for the purpose of unlocking and mutexes
* held for the purpose of locking.
*
* FIXME: This doesn't work if slock_t is a struct.
*/
if ((lock->mutex == 0) && (lock->head == NULL))
goto lock_released;
/* At this point, we don't know if the concurrent exclusive locker
* has seen on-zero in our shared counter partition in his step
* (2) or not. We must thus take the mutex and re-check.
*/
#endif /* LWLOCK_PART_SHARED_OPS_ATOMIC */
/* Acquire mutex. Time spent holding mutex should be short! */
SpinLockAcquire(&lock->mutex);
#ifdef LWLOCK_PART_SHARED_OPS_ATOMIC
/* Already decremented the shared counter partition above */
#else
/* Release my hold on lock */
lock->shared--;
#endif
}
else {
/* Acquire mutex. Time spent holding mutex should be short! */
SpinLockAcquire(&lock->mutex);
/* Release my hold on lock */
lock->exclusive--;
Assert(lock->exclusive == 0);
}
/*
* See if I need to awaken any waiters. If I released a non-last shared
* hold, there cannot be anything to do. Also, do not awaken any waiters
* if someone has already awakened waiters that haven't yet acquired the
* lock.
*/
head = lock->head;
if (head != NULL)
{
bool is_shared;
if (mode == LW_SHARED)
LWLOCK_IS_SHARED(is_shared, lock, lockid);
else
is_shared = false;
if (lock->exclusive == 0 && !is_shared && lock->releaseOK)
{
/*
* Remove the to-be-awakened PGPROCs from the queue. If the front
* waiter wants exclusive lock, awaken him only. Otherwise awaken
* as many waiters as want shared access.
*/
proc = head;
if (!proc->lwExclusive)
{
while (proc->lwWaitLink != NULL &&
!proc->lwWaitLink->lwExclusive)
proc = proc->lwWaitLink;
}
/* proc is now the last PGPROC to be released */
lock->head = proc->lwWaitLink;
proc->lwWaitLink = NULL;
/* prevent additional wakeups until retryer gets to run */
lock->releaseOK = false;
}
else
{
/* lock is still held, can't awaken anything */
head = NULL;
}
}
/* We are done updating shared state of the lock itself. */
SpinLockRelease(&lock->mutex);
#ifdef LWLOCK_PART_SHARED_OPS_ATOMIC
lock_released:
#endif
TRACE_POSTGRESQL_LWLOCK_RELEASE(lockid);
/*
* Awaken any waiters I removed from the queue.
*/
while (head != NULL)
{
LOG_LWDEBUG("LWLockRelease", lockid, "release waiter");
proc = head;
head = proc->lwWaitLink;
proc->lwWaitLink = NULL;
proc->lwWaiting = false;
PGSemaphoreUnlock(&proc->sem);
}
/*
* Now okay to allow cancel/die interrupts.
*/
RESUME_INTERRUPTS();
}
void
s_unlock_sema(volatile slock_t *lock)
{
PGSemaphoreUnlock((PGSemaphore) lock);
}
/*
* Create a semaphore set with the given number of useful semaphores
* (an additional sema is actually allocated to serve as identifier).
* Dead Postgres sema sets are recycled if found, but we do not fail
* upon collision with non-Postgres sema sets.
*
* The idea here is to detect and re-use keys that may have been assigned
* by a crashed postmaster or backend.
*/
static IpcSemaphoreId
IpcSemaphoreCreate(int numSems)
{
IpcSemaphoreId semId;
union semun semun;
PGSemaphoreData mysema;
/* Loop till we find a free IPC key */
for (nextSemaKey++;; nextSemaKey++)
{
pid_t creatorPID;
/* Try to create new semaphore set */
semId = InternalIpcSemaphoreCreate(nextSemaKey, numSems + 1);
if (semId >= 0)
break; /* successful create */
/* See if it looks to be leftover from a dead Postgres process */
semId = semget(nextSemaKey, numSems + 1, 0);
if (semId < 0)
continue; /* failed: must be some other app's */
if (IpcSemaphoreGetValue(semId, numSems) != PGSemaMagic)
continue; /* sema belongs to a non-Postgres app */
/*
* If the creator PID is my own PID or does not belong to any extant
* process, it's safe to zap it.
*/
creatorPID = IpcSemaphoreGetLastPID(semId, numSems);
if (creatorPID <= 0)
continue; /* oops, GETPID failed */
if (creatorPID != getpid())
{
if (kill(creatorPID, 0) == 0 || errno != ESRCH)
continue; /* sema belongs to a live process */
}
/*
* The sema set appears to be from a dead Postgres process, or from a
* previous cycle of life in this same process. Zap it, if possible.
* This probably shouldn't fail, but if it does, assume the sema set
* belongs to someone else after all, and continue quietly.
*/
semun.val = 0; /* unused, but keep compiler quiet */
if (semctl(semId, 0, IPC_RMID, semun) < 0)
continue;
/*
* Now try again to create the sema set.
*/
semId = InternalIpcSemaphoreCreate(nextSemaKey, numSems + 1);
if (semId >= 0)
break; /* successful create */
/*
* Can only get here if some other process managed to create the same
* sema key before we did. Let him have that one, loop around to try
* next key.
*/
}
/*
* OK, we created a new sema set. Mark it as created by this process. We
* do this by setting the spare semaphore to PGSemaMagic-1 and then
* incrementing it with semop(). That leaves it with value PGSemaMagic
* and sempid referencing this process.
*/
IpcSemaphoreInitialize(semId, numSems, PGSemaMagic - 1);
mysema.semId = semId;
mysema.semNum = numSems;
PGSemaphoreUnlock(&mysema);
return semId;
}
void
s_unlock_sema(volatile slock_t *lock)
{
PGSemaphoreUnlock(&SpinlockSemaArray[*lock]);
}