FlexLocks

diff --git a/contrib/pg_stat_statements/pg_stat_statements.c b/contrib/pg_stat_statements/pg_stat_statements.c
index 8dc3054..51b24d0 100644
--- a/contrib/pg_stat_statements/pg_stat_statements.c
+++ b/contrib/pg_stat_statements/pg_stat_statements.c
@@ -105,7 +105,7 @@ typedef struct pgssEntry
  */
 typedef struct pgssSharedState
 {
-	LWLockId	lock;			/* protects hashtable search/modification */
+	FlexLockId	lock;			/* protects hashtable search/modification */
 	int			query_size;		/* max query length in bytes */
 } pgssSharedState;
 
diff --git a/doc/src/sgml/config.sgml b/doc/src/sgml/config.sgml
index d1e628f..8517b36 100644
--- a/doc/src/sgml/config.sgml
+++ b/doc/src/sgml/config.sgml
@@ -6199,14 +6199,14 @@ LOG:  CleanUpLock: deleting: lock(0xb7acd844) id(24688,24696,0,0,0,1)
      </varlistentry>
 
      <varlistentry>
-      <term><varname>trace_lwlocks</varname> (<type>boolean</type>)</term>
+      <term><varname>trace_flexlocks</varname> (<type>boolean</type>)</term>
       <indexterm>
-       <primary><varname>trace_lwlocks</> configuration parameter</primary>
+       <primary><varname>trace_flexlocks</> configuration parameter</primary>
       </indexterm>
       <listitem>
        <para>
-        If on, emit information about lightweight lock usage.  Lightweight
-        locks are intended primarily to provide mutual exclusion of access
+        If on, emit information about FlexLock usage.  FlexLocks
+        are intended primarily to provide mutual exclusion of access
         to shared-memory data structures.
        </para>
        <para>
diff --git a/doc/src/sgml/monitoring.sgml b/doc/src/sgml/monitoring.sgml
index b9dc1d2..98ed0d3 100644
--- a/doc/src/sgml/monitoring.sgml
+++ b/doc/src/sgml/monitoring.sgml
@@ -1724,49 +1724,49 @@ SELECT pg_stat_get_backend_pid(s.backendid) AS procpid,
       or kilobytes of memory used for an internal sort.</entry>
     </row>
     <row>
-     <entry>lwlock-acquire</entry>
-     <entry>(LWLockId, LWLockMode)</entry>
-     <entry>Probe that fires when an LWLock has been acquired.
-      arg0 is the LWLock's ID.
-      arg1 is the requested lock mode, either exclusive or shared.</entry>
+     <entry>flexlock-acquire</entry>
+     <entry>(FlexLockId, FlexLockMode)</entry>
+     <entry>Probe that fires when an FlexLock has been acquired.
+      arg0 is the FlexLock's ID.
+      arg1 is the requested lock mode.</entry>
     </row>
     <row>
-     <entry>lwlock-release</entry>
-     <entry>(LWLockId)</entry>
-     <entry>Probe that fires when an LWLock has been released (but note
+     <entry>flexlock-release</entry>
+     <entry>(FlexLockId)</entry>
+     <entry>Probe that fires when a FlexLock has been released (but note
       that any released waiters have not yet been awakened).
-      arg0 is the LWLock's ID.</entry>
+      arg0 is the FlexLock's ID.</entry>
     </row>
     <row>
-     <entry>lwlock-wait-start</entry>
-     <entry>(LWLockId, LWLockMode)</entry>
-     <entry>Probe that fires when an LWLock was not immediately available and
+     <entry>flexlock-wait-start</entry>
+     <entry>(FlexLockId, FlexLockMode)</entry>
+     <entry>Probe that fires when an FlexLock was not immediately available and
       a server process has begun to wait for the lock to become available.
-      arg0 is the LWLock's ID.
+      arg0 is the FlexLock's ID.
       arg1 is the requested lock mode, either exclusive or shared.</entry>
     </row>
     <row>
-     <entry>lwlock-wait-done</entry>
-     <entry>(LWLockId, LWLockMode)</entry>
+     <entry>flexlock-wait-done</entry>
+     <entry>(FlexLockId, FlexLockMode)</entry>
      <entry>Probe that fires when a server process has been released from its
-      wait for an LWLock (it does not actually have the lock yet).
-      arg0 is the LWLock's ID.
+      wait for an FlexLock (it does not actually have the lock yet).
+      arg0 is the FlexLock's ID.
       arg1 is the requested lock mode, either exclusive or shared.</entry>
     </row>
     <row>
-     <entry>lwlock-condacquire</entry>
-     <entry>(LWLockId, LWLockMode)</entry>
-     <entry>Probe that fires when an LWLock was successfully acquired when the
-      caller specified no waiting.
-      arg0 is the LWLock's ID.
+     <entry>flexlock-condacquire</entry>
+     <entry>(FlexLockId, FlexLockMode)</entry>
+     <entry>Probe that fires when an FlexLock was successfully acquired when
+      the caller specified no waiting.
+      arg0 is the FlexLock's ID.
       arg1 is the requested lock mode, either exclusive or shared.</entry>
     </row>
     <row>
-     <entry>lwlock-condacquire-fail</entry>
-     <entry>(LWLockId, LWLockMode)</entry>
-     <entry>Probe that fires when an LWLock was not successfully acquired when
-      the caller specified no waiting.
-      arg0 is the LWLock's ID.
+     <entry>flexlock-condacquire-fail</entry>
+     <entry>(FlexLockId, FlexLockMode)</entry>
+     <entry>Probe that fires when an FlexLock was not successfully acquired
+      when the caller specified no waiting.
+      arg0 is the FlexLock's ID.
       arg1 is the requested lock mode, either exclusive or shared.</entry>
     </row>
     <row>
@@ -1813,11 +1813,11 @@ SELECT pg_stat_get_backend_pid(s.backendid) AS procpid,
      <entry>unsigned int</entry>
     </row>
     <row>
-     <entry>LWLockId</entry>
+     <entry>FlexLockId</entry>
      <entry>int</entry>
     </row>
     <row>
-     <entry>LWLockMode</entry>
+     <entry>FlexLockMode</entry>
      <entry>int</entry>
     </row>
     <row>
diff --git a/src/backend/access/transam/slru.c b/src/backend/access/transam/slru.c
index f7caa34..09d5862 100644
--- a/src/backend/access/transam/slru.c
+++ b/src/backend/access/transam/slru.c
@@ -151,7 +151,7 @@ SimpleLruShmemSize(int nslots, int nlsns)
 	sz += MAXALIGN(nslots * sizeof(bool));		/* page_dirty[] */
 	sz += MAXALIGN(nslots * sizeof(int));		/* page_number[] */
 	sz += MAXALIGN(nslots * sizeof(int));		/* page_lru_count[] */
-	sz += MAXALIGN(nslots * sizeof(LWLockId));	/* buffer_locks[] */
+	sz += MAXALIGN(nslots * sizeof(FlexLockId));		/* buffer_locks[] */
 
 	if (nlsns > 0)
 		sz += MAXALIGN(nslots * nlsns * sizeof(XLogRecPtr));	/* group_lsn[] */
@@ -161,7 +161,7 @@ SimpleLruShmemSize(int nslots, int nlsns)
 
 void
 SimpleLruInit(SlruCtl ctl, const char *name, int nslots, int nlsns,
-			  LWLockId ctllock, const char *subdir)
+			  FlexLockId ctllock, const char *subdir)
 {
 	SlruShared	shared;
 	bool		found;
@@ -202,8 +202,8 @@ SimpleLruInit(SlruCtl ctl, const char *name, int nslots, int nlsns,
 		offset += MAXALIGN(nslots * sizeof(int));
 		shared->page_lru_count = (int *) (ptr + offset);
 		offset += MAXALIGN(nslots * sizeof(int));
-		shared->buffer_locks = (LWLockId *) (ptr + offset);
-		offset += MAXALIGN(nslots * sizeof(LWLockId));
+		shared->buffer_locks = (FlexLockId *) (ptr + offset);
+		offset += MAXALIGN(nslots * sizeof(FlexLockId));
 
 		if (nlsns > 0)
 		{
diff --git a/src/backend/access/transam/twophase.c b/src/backend/access/transam/twophase.c
index 477982d..d5d1ee9 100644
--- a/src/backend/access/transam/twophase.c
+++ b/src/backend/access/transam/twophase.c
@@ -318,9 +318,9 @@ MarkAsPreparing(TransactionId xid, const char *gid,
 	gxact->proc.roleId = owner;
 	gxact->proc.inCommit = false;
 	gxact->proc.vacuumFlags = 0;
-	gxact->proc.lwWaiting = false;
-	gxact->proc.lwExclusive = false;
-	gxact->proc.lwWaitLink = NULL;
+	gxact->proc.flWaitResult = 0;
+	gxact->proc.flWaitMode = 0;
+	gxact->proc.flWaitLink = NULL;
 	gxact->proc.waitLock = NULL;
 	gxact->proc.waitProcLock = NULL;
 	for (i = 0; i < NUM_LOCK_PARTITIONS; i++)
diff --git a/src/backend/access/transam/xact.c b/src/backend/access/transam/xact.c
index c151d3b..19b708c 100644
--- a/src/backend/access/transam/xact.c
+++ b/src/backend/access/transam/xact.c
@@ -2248,7 +2248,7 @@ AbortTransaction(void)
 	 * Releasing LW locks is critical since we might try to grab them again
 	 * while cleaning up!
 	 */
-	LWLockReleaseAll();
+	FlexLockReleaseAll();
 
 	/* Clean up buffer I/O and buffer context locks, too */
 	AbortBufferIO();
@@ -4138,7 +4138,7 @@ AbortSubTransaction(void)
 	 * FIXME This may be incorrect --- Are there some locks we should keep?
 	 * Buffer locks, for example?  I don't think so but I'm not sure.
 	 */
-	LWLockReleaseAll();
+	FlexLockReleaseAll();
 
 	AbortBufferIO();
 	UnlockBuffers();
diff --git a/src/backend/bootstrap/bootstrap.c b/src/backend/bootstrap/bootstrap.c
index 6bf2421..9ceee91 100644
--- a/src/backend/bootstrap/bootstrap.c
+++ b/src/backend/bootstrap/bootstrap.c
@@ -562,13 +562,13 @@ bootstrap_signals(void)
  * Begin shutdown of an auxiliary process.	This is approximately the equivalent
  * of ShutdownPostgres() in postinit.c.  We can't run transactions in an
  * auxiliary process, so most of the work of AbortTransaction() is not needed,
- * but we do need to make sure we've released any LWLocks we are holding.
+ * but we do need to make sure we've released any flex locks we are holding.
  * (This is only critical during an error exit.)
  */
 static void
 ShutdownAuxiliaryProcess(int code, Datum arg)
 {
-	LWLockReleaseAll();
+	FlexLockReleaseAll();
 }
 
 /* ----------------------------------------------------------------
diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c
index cacedab..f33f573 100644
--- a/src/backend/postmaster/bgwriter.c
+++ b/src/backend/postmaster/bgwriter.c
@@ -176,9 +176,10 @@ BackgroundWriterMain(void)
 		/*
 		 * These operations are really just a minimal subset of
 		 * AbortTransaction().	We don't have very many resources to worry
-		 * about in bgwriter, but we do have LWLocks, buffers, and temp files.
+		 * about in bgwriter, but we do have flex locks, buffers, and temp
+		 * files.
 		 */
-		LWLockReleaseAll();
+		FlexLockReleaseAll();
 		AbortBufferIO();
 		UnlockBuffers();
 		/* buffer pins are released here: */
diff --git a/src/backend/postmaster/checkpointer.c b/src/backend/postmaster/checkpointer.c
index e9ae1e8..49f07a7 100644
--- a/src/backend/postmaster/checkpointer.c
+++ b/src/backend/postmaster/checkpointer.c
@@ -281,9 +281,10 @@ CheckpointerMain(void)
 		/*
 		 * These operations are really just a minimal subset of
 		 * AbortTransaction().	We don't have very many resources to worry
-		 * about in checkpointer, but we do have LWLocks, buffers, and temp files.
+		 * about in checkpointer, but we do have flex locks, buffers, and temp
+		 * files.
 		 */
-		LWLockReleaseAll();
+		FlexLockReleaseAll();
 		AbortBufferIO();
 		UnlockBuffers();
 		/* buffer pins are released here: */
diff --git a/src/backend/postmaster/postmaster.c b/src/backend/postmaster/postmaster.c
index 6758083..14b4368 100644
--- a/src/backend/postmaster/postmaster.c
+++ b/src/backend/postmaster/postmaster.c
@@ -109,6 +109,7 @@
 #include "postmaster/syslogger.h"
 #include "replication/walsender.h"
 #include "storage/fd.h"
+#include "storage/flexlock_internals.h"
 #include "storage/ipc.h"
 #include "storage/pg_shmem.h"
 #include "storage/pmsignal.h"
@@ -404,8 +405,6 @@ typedef struct
 typedef int InheritableSocket;
 #endif
 
-typedef struct LWLock LWLock;	/* ugly kluge */
-
 /*
  * Structure contains all variables passed to exec:ed backends
  */
@@ -426,7 +425,7 @@ typedef struct
 	slock_t    *ShmemLock;
 	VariableCache ShmemVariableCache;
 	Backend    *ShmemBackendArray;
-	LWLock	   *LWLockArray;
+	FlexLock   *FlexLockArray;
 	slock_t    *ProcStructLock;
 	PROC_HDR   *ProcGlobal;
 	PGPROC	   *AuxiliaryProcs;
@@ -4675,7 +4674,6 @@ MaxLivePostmasterChildren(void)
  * functions
  */
 extern slock_t *ShmemLock;
-extern LWLock *LWLockArray;
 extern slock_t *ProcStructLock;
 extern PGPROC *AuxiliaryProcs;
 extern PMSignalData *PMSignalState;
@@ -4720,7 +4718,7 @@ save_backend_variables(BackendParameters *param, Port *port,
 	param->ShmemVariableCache = ShmemVariableCache;
 	param->ShmemBackendArray = ShmemBackendArray;
 
-	param->LWLockArray = LWLockArray;
+	param->FlexLockArray = FlexLockArray;
 	param->ProcStructLock = ProcStructLock;
 	param->ProcGlobal = ProcGlobal;
 	param->AuxiliaryProcs = AuxiliaryProcs;
@@ -4943,7 +4941,7 @@ restore_backend_variables(BackendParameters *param, Port *port)
 	ShmemVariableCache = param->ShmemVariableCache;
 	ShmemBackendArray = param->ShmemBackendArray;
 
-	LWLockArray = param->LWLockArray;
+	FlexLockArray = param->FlexLockArray;
 	ProcStructLock = param->ProcStructLock;
 	ProcGlobal = param->ProcGlobal;
 	AuxiliaryProcs = param->AuxiliaryProcs;
diff --git a/src/backend/postmaster/walwriter.c b/src/backend/postmaster/walwriter.c
index 157728e..587443d 100644
--- a/src/backend/postmaster/walwriter.c
+++ b/src/backend/postmaster/walwriter.c
@@ -167,9 +167,9 @@ WalWriterMain(void)
 		/*
 		 * These operations are really just a minimal subset of
 		 * AbortTransaction().	We don't have very many resources to worry
-		 * about in walwriter, but we do have LWLocks, and perhaps buffers?
+		 * about in walwriter, but we do have flex locks, and perhaps buffers?
 		 */
-		LWLockReleaseAll();
+		FlexLockReleaseAll();
 		AbortBufferIO();
 		UnlockBuffers();
 		/* buffer pins are released here: */
diff --git a/src/backend/storage/buffer/bufmgr.c b/src/backend/storage/buffer/bufmgr.c
index e59af33..07356ec 100644
--- a/src/backend/storage/buffer/bufmgr.c
+++ b/src/backend/storage/buffer/bufmgr.c
@@ -141,7 +141,7 @@ PrefetchBuffer(Relation reln, ForkNumber forkNum, BlockNumber blockNum)
 	{
 		BufferTag	newTag;		/* identity of requested block */
 		uint32		newHash;	/* hash value for newTag */
-		LWLockId	newPartitionLock;	/* buffer partition lock for it */
+		FlexLockId	newPartitionLock;	/* buffer partition lock for it */
 		int			buf_id;
 
 		/* create a tag so we can lookup the buffer */
@@ -512,10 +512,10 @@ BufferAlloc(SMgrRelation smgr, char relpersistence, ForkNumber forkNum,
 {
 	BufferTag	newTag;			/* identity of requested block */
 	uint32		newHash;		/* hash value for newTag */
-	LWLockId	newPartitionLock;		/* buffer partition lock for it */
+	FlexLockId	newPartitionLock;		/* buffer partition lock for it */
 	BufferTag	oldTag;			/* previous identity of selected buffer */
 	uint32		oldHash;		/* hash value for oldTag */
-	LWLockId	oldPartitionLock;		/* buffer partition lock for it */
+	FlexLockId	oldPartitionLock;		/* buffer partition lock for it */
 	BufFlags	oldFlags;
 	int			buf_id;
 	volatile BufferDesc *buf;
@@ -855,7 +855,7 @@ InvalidateBuffer(volatile BufferDesc *buf)
 {
 	BufferTag	oldTag;
 	uint32		oldHash;		/* hash value for oldTag */
-	LWLockId	oldPartitionLock;		/* buffer partition lock for it */
+	FlexLockId	oldPartitionLock;		/* buffer partition lock for it */
 	BufFlags	oldFlags;
 
 	/* Save the original buffer tag before dropping the spinlock */
diff --git a/src/backend/storage/ipc/ipci.c b/src/backend/storage/ipc/ipci.c
index 56c0bd8..02ee8d8 100644
--- a/src/backend/storage/ipc/ipci.c
+++ b/src/backend/storage/ipc/ipci.c
@@ -113,7 +113,7 @@ CreateSharedMemoryAndSemaphores(bool makePrivate, int port)
 		size = add_size(size, SUBTRANSShmemSize());
 		size = add_size(size, TwoPhaseShmemSize());
 		size = add_size(size, MultiXactShmemSize());
-		size = add_size(size, LWLockShmemSize());
+		size = add_size(size, FlexLockShmemSize());
 		size = add_size(size, ProcArrayShmemSize());
 		size = add_size(size, BackendStatusShmemSize());
 		size = add_size(size, SInvalShmemSize());
@@ -179,7 +179,7 @@ CreateSharedMemoryAndSemaphores(bool makePrivate, int port)
 	 * needed for InitShmemIndex.
 	 */
 	if (!IsUnderPostmaster)
-		CreateLWLocks();
+		CreateFlexLocks();
 
 	/*
 	 * Set up shmem.c index hashtable
diff --git a/src/backend/storage/lmgr/Makefile b/src/backend/storage/lmgr/Makefile
index e12a854..3730e51 100644
--- a/src/backend/storage/lmgr/Makefile
+++ b/src/backend/storage/lmgr/Makefile
@@ -12,7 +12,8 @@ subdir = src/backend/storage/lmgr
 top_builddir = ../../../..
 include $(top_builddir)/src/Makefile.global
 
-OBJS = lmgr.o lock.o proc.o deadlock.o lwlock.o spin.o s_lock.o predicate.o
+OBJS = flexlock.o lmgr.o lock.o proc.o deadlock.o lwlock.o spin.o s_lock.o \
+	predicate.o
 
 include $(top_srcdir)/src/backend/common.mk
 
diff --git a/src/backend/storage/lmgr/flexlock.c b/src/backend/storage/lmgr/flexlock.c
new file mode 100644
index 0000000..7f657b3
--- /dev/null
+++ b/src/backend/storage/lmgr/flexlock.c
@@ -0,0 +1,353 @@
+/*-------------------------------------------------------------------------
+ *
+ * flexlock.c
+ *	  Low-level routines for managing flex locks.
+ *
+ * Flex locks are intended primarily to provide mutual exclusion of access
+ * to shared-memory data structures.  Most, but not all, flex locks are
+ * lightweight locks (LWLocks).  This file contains support routines that
+ * are used for all types of flex locks, including lwlocks.  User-level
+ * locking should be done with the full lock manager --- which depends on
+ * LWLocks to protect its shared state.
+ *
+ * Portions Copyright (c) 1996-2011, PostgreSQL Global Development Group
+ * Portions Copyright (c) 1994, Regents of the University of California
+ *
+ * IDENTIFICATION
+ *	  src/backend/storage/lmgr/flexlock.c
+ *
+ *-------------------------------------------------------------------------
+ */
+
+#include "postgres.h"
+
+#include "miscadmin.h"
+#include "access/clog.h"
+#include "access/multixact.h"
+#include "access/subtrans.h"
+#include "commands/async.h"
+#include "storage/flexlock_internals.h"
+#include "storage/lwlock.h"
+#include "storage/predicate.h"
+#include "storage/proc.h"
+#include "storage/spin.h"
+#include "utils/elog.h"
+
+/*
+ * We use this structure to keep track of flex locks held, for release
+ * during error recovery.  The maximum size could be determined at runtime
+ * if necessary, but it seems unlikely that more than a few locks could
+ * ever be held simultaneously.
+ */
+#define MAX_SIMUL_FLEXLOCKS	100
+
+int	num_held_flexlocks = 0;
+FlexLockId held_flexlocks[MAX_SIMUL_FLEXLOCKS];
+
+static int	lock_addin_request = 0;
+static bool lock_addin_request_allowed = true;
+
+#ifdef LOCK_DEBUG
+bool		Trace_flexlocks = false;
+#endif
+
+/*
+ * This points to the array of FlexLocks in shared memory.  Backends inherit
+ * the pointer by fork from the postmaster (except in the EXEC_BACKEND case,
+ * where we have special measures to pass it down).
+ */
+FlexLockPadded *FlexLockArray = NULL;
+
+/* We use the ShmemLock spinlock to protect LWLockAssign */
+extern slock_t *ShmemLock;
+
+static void FlexLockInit(FlexLock *flex, char locktype);
+
+/*
+ * Compute number of FlexLocks to allocate.
+ */
+int
+NumFlexLocks(void)
+{
+	int			numLocks;
+
+	/*
+	 * Possibly this logic should be spread out among the affected modules,
+	 * the same way that shmem space estimation is done.  But for now, there
+	 * are few enough users of FlexLocks that we can get away with just keeping
+	 * the knowledge here.
+	 */
+
+	/* Predefined FlexLocks */
+	numLocks = (int) NumFixedFlexLocks;
+
+	/* bufmgr.c needs two for each shared buffer */
+	numLocks += 2 * NBuffers;
+
+	/* proc.c needs one for each backend or auxiliary process */
+	numLocks += MaxBackends + NUM_AUXILIARY_PROCS;
+
+	/* clog.c needs one per CLOG buffer */
+	numLocks += NUM_CLOG_BUFFERS;
+
+	/* subtrans.c needs one per SubTrans buffer */
+	numLocks += NUM_SUBTRANS_BUFFERS;
+
+	/* multixact.c needs two SLRU areas */
+	numLocks += NUM_MXACTOFFSET_BUFFERS + NUM_MXACTMEMBER_BUFFERS;
+
+	/* async.c needs one per Async buffer */
+	numLocks += NUM_ASYNC_BUFFERS;
+
+	/* predicate.c needs one per old serializable xid buffer */
+	numLocks += NUM_OLDSERXID_BUFFERS;
+
+	/*
+	 * Add any requested by loadable modules; for backwards-compatibility
+	 * reasons, allocate at least NUM_USER_DEFINED_FLEXLOCKS of them even if
+	 * there are no explicit requests.
+	 */
+	lock_addin_request_allowed = false;
+	numLocks += Max(lock_addin_request, NUM_USER_DEFINED_FLEXLOCKS);
+
+	return numLocks;
+}
+
+
+/*
+ * RequestAddinFlexLocks
+ *		Request that extra FlexLocks be allocated for use by
+ *		a loadable module.
+ *
+ * This is only useful if called from the _PG_init hook of a library that
+ * is loaded into the postmaster via shared_preload_libraries.	Once
+ * shared memory has been allocated, calls will be ignored.  (We could
+ * raise an error, but it seems better to make it a no-op, so that
+ * libraries containing such calls can be reloaded if needed.)
+ */
+void
+RequestAddinFlexLocks(int n)
+{
+	if (IsUnderPostmaster || !lock_addin_request_allowed)
+		return;					/* too late */
+	lock_addin_request += n;
+}
+
+
+/*
+ * Compute shmem space needed for FlexLocks.
+ */
+Size
+FlexLockShmemSize(void)
+{
+	Size		size;
+	int			numLocks = NumFlexLocks();
+
+	/* Space for the FlexLock array. */
+	size = mul_size(numLocks, FLEX_LOCK_BYTES);
+
+	/* Space for dynamic allocation counter, plus room for alignment. */
+	size = add_size(size, 2 * sizeof(int) + FLEX_LOCK_BYTES);
+
+	return size;
+}
+
+/*
+ * Allocate shmem space for FlexLocks and initialize the locks.
+ */
+void
+CreateFlexLocks(void)
+{
+	int			numLocks = NumFlexLocks();
+	Size		spaceLocks = FlexLockShmemSize();
+	FlexLockPadded *lock;
+	int		   *FlexLockCounter;
+	char	   *ptr;
+	int			id;
+
+	/* Allocate and zero space */
+	ptr = (char *) ShmemAlloc(spaceLocks);
+	memset(ptr, 0, spaceLocks);
+
+	/* Leave room for dynamic allocation counter */
+	ptr += 2 * sizeof(int);
+
+	/* Ensure desired alignment of FlexLock array */
+	ptr += FLEX_LOCK_BYTES - ((uintptr_t) ptr) % FLEX_LOCK_BYTES;
+
+	FlexLockArray = (FlexLockPadded *) ptr;
+
+	/* All of the "fixed" FlexLocks are LWLocks. */
+	for (id = 0, lock = FlexLockArray; id < NumFixedFlexLocks; id++, lock++)
+		FlexLockInit(&lock->flex, FLEXLOCK_TYPE_LWLOCK);
+
+	/*
+	 * Initialize the dynamic-allocation counter, which is stored just before
+	 * the first FlexLock.
+	 */
+	FlexLockCounter = (int *) ((char *) FlexLockArray - 2 * sizeof(int));
+	FlexLockCounter[0] = (int) NumFixedFlexLocks;
+	FlexLockCounter[1] = numLocks;
+}
+
+/*
+ * FlexLockAssign - assign a dynamically-allocated FlexLock number
+ *
+ * We interlock this using the same spinlock that is used to protect
+ * ShmemAlloc().  Interlocking is not really necessary during postmaster
+ * startup, but it is needed if any user-defined code tries to allocate
+ * LWLocks after startup.
+ */
+FlexLockId
+FlexLockAssign(char locktype)
+{
+	FlexLockId	result;
+
+	/* use volatile pointer to prevent code rearrangement */
+	volatile int *FlexLockCounter;
+
+	FlexLockCounter = (int *) ((char *) FlexLockArray - 2 * sizeof(int));
+	SpinLockAcquire(ShmemLock);
+	if (FlexLockCounter[0] >= FlexLockCounter[1])
+	{
+		SpinLockRelease(ShmemLock);
+		elog(ERROR, "no more FlexLockIds available");
+	}
+	result = (FlexLockId) (FlexLockCounter[0]++);
+	SpinLockRelease(ShmemLock);
+
+	FlexLockInit(&FlexLockArray[result].flex, locktype);
+
+	return result;
+}
+
+/*
+ * Initialize a FlexLock.
+ */
+static void
+FlexLockInit(FlexLock *flex, char locktype)
+{
+	SpinLockInit(&flex->mutex);
+	flex->releaseOK = true;
+	flex->locktype = locktype;
+	/*
+	 * We might need to think a little harder about what should happen here
+	 * if some future type of FlexLock requires more initialization than this.
+	 * For now, this will suffice.
+	 */
+}
+
+/*
+ * Remove lock from list of locks held.  Usually, but not always, it will
+ * be the latest-acquired lock; so search array backwards.
+ */
+void
+FlexLockRemember(FlexLockId id)
+{
+	if (num_held_flexlocks >= MAX_SIMUL_FLEXLOCKS)
+		elog(PANIC, "too many FlexLocks taken");
+	held_flexlocks[num_held_flexlocks++] = id;
+}
+
+/*
+ * Remove lock from list of locks held.  Usually, but not always, it will
+ * be the latest-acquired lock; so search array backwards.
+ */
+void
+FlexLockForget(FlexLockId id)
+{
+	int			i;
+
+	for (i = num_held_flexlocks; --i >= 0;)
+	{
+		if (id == held_flexlocks[i])
+			break;
+	}
+	if (i < 0)
+		elog(ERROR, "lock %d is not held", (int) id);
+	num_held_flexlocks--;
+	for (; i < num_held_flexlocks; i++)
+		held_flexlocks[i] = held_flexlocks[i + 1];
+}
+
+/*
+ * FlexLockWait - wait until awakened
+ *
+ * Since we share the process wait semaphore with the regular lock manager
+ * and ProcWaitForSignal, and we may need to acquire a FlexLock while one of
+ * those is pending, it is possible that we get awakened for a reason other
+ * than being signaled by a FlexLock release.  If so, loop back and wait again.
+ *
+ * Returns the number of "extra" waits absorbed so that, once we've gotten the
+ * FlexLock, we can 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.
+ */
+int
+FlexLockWait(FlexLockId id, int mode)
+{
+	int		extraWaits = 0;
+
+	FlexLockDebug("LWLockAcquire", id, "waiting");
+	TRACE_POSTGRESQL_FLEXLOCK_WAIT_START(id, mode);
+
+	for (;;)
+   	{
+		/* "false" means cannot accept cancel/die interrupt here. */
+		PGSemaphoreLock(&MyProc->sem, false);
+		/*
+		 * FLEXTODO: I think we should return this, instead of ignoring it.
+		 * Any non-zero value means "wake up".
+		 */
+		if (MyProc->flWaitResult)
+			break;
+		extraWaits++;
+   	}
+
+	TRACE_POSTGRESQL_FLEXLOCK_WAIT_DONE(id, mode);
+	FlexLockDebug("LWLockAcquire", id, "awakened");
+
+	return extraWaits;
+}
+
+/*
+ * FlexLockReleaseAll - release all currently-held locks
+ *
+ * Used to clean up after ereport(ERROR). An important difference between this
+ * function and retail LWLockRelease calls is that InterruptHoldoffCount is
+ * unchanged by this operation.  This is necessary since InterruptHoldoffCount
+ * has been set to an appropriate level earlier in error recovery. We could
+ * decrement it below zero if we allow it to drop for each released lock!
+ */
+void
+FlexLockReleaseAll(void)
+{
+	while (num_held_flexlocks > 0)
+	{
+		HOLD_INTERRUPTS();		/* match the upcoming RESUME_INTERRUPTS */
+
+		/*
+		 * FLEXTODO: When we have multiple types of flex locks, this will
+		 * need to call the appropriate release function for each lock type.
+		 */
+		LWLockRelease(held_flexlocks[num_held_flexlocks - 1]);
+	}
+}
+
+/*
+ * FlexLockHeldByMe - test whether my process currently holds a lock
+ *
+ * This is meant as debug support only.  We do not consider the lock mode.
+ */
+bool
+FlexLockHeldByMe(FlexLockId id)
+{
+	int			i;
+
+	for (i = 0; i < num_held_flexlocks; i++)
+	{
+		if (held_flexlocks[i] == id)
+			return true;
+	}
+	return false;
+}
diff --git a/src/backend/storage/lmgr/lock.c b/src/backend/storage/lmgr/lock.c
index 905502f..adc5fd9 100644
--- a/src/backend/storage/lmgr/lock.c
+++ b/src/backend/storage/lmgr/lock.c
@@ -591,7 +591,7 @@ LockAcquireExtended(const LOCKTAG *locktag,
 	bool		found;
 	ResourceOwner owner;
 	uint32		hashcode;
-	LWLockId	partitionLock;
+	FlexLockId	partitionLock;
 	int			status;
 	bool		log_lock = false;
 
@@ -1546,7 +1546,7 @@ LockRelease(const LOCKTAG *locktag, LOCKMODE lockmode, bool sessionLock)
 	LOCALLOCK  *locallock;
 	LOCK	   *lock;
 	PROCLOCK   *proclock;
-	LWLockId	partitionLock;
+	FlexLockId	partitionLock;
 	bool		wakeupNeeded;
 
 	if (lockmethodid <= 0 || lockmethodid >= lengthof(LockMethods))
@@ -1912,7 +1912,7 @@ LockReleaseAll(LOCKMETHODID lockmethodid, bool allLocks)
 	 */
 	for (partition = 0; partition < NUM_LOCK_PARTITIONS; partition++)
 	{
-		LWLockId	partitionLock = FirstLockMgrLock + partition;
+		FlexLockId	partitionLock = FirstLockMgrLock + partition;
 		SHM_QUEUE  *procLocks = &(MyProc->myProcLocks[partition]);
 
 		proclock = (PROCLOCK *) SHMQueueNext(procLocks, procLocks,
@@ -2197,7 +2197,7 @@ static bool
 FastPathTransferRelationLocks(LockMethod lockMethodTable, const LOCKTAG *locktag,
 					  uint32 hashcode)
 {
-	LWLockId		partitionLock = LockHashPartitionLock(hashcode);
+	FlexLockId		partitionLock = LockHashPartitionLock(hashcode);
 	Oid				relid = locktag->locktag_field2;
 	uint32			i;
 
@@ -2281,7 +2281,7 @@ FastPathGetRelationLockEntry(LOCALLOCK *locallock)
 	LockMethod		lockMethodTable = LockMethods[DEFAULT_LOCKMETHOD];
 	LOCKTAG		   *locktag = &locallock->tag.lock;
 	PROCLOCK	   *proclock = NULL;
-	LWLockId		partitionLock = LockHashPartitionLock(locallock->hashcode);
+	FlexLockId		partitionLock = LockHashPartitionLock(locallock->hashcode);
 	Oid				relid = locktag->locktag_field2;
 	uint32			f;
 
@@ -2382,7 +2382,7 @@ GetLockConflicts(const LOCKTAG *locktag, LOCKMODE lockmode)
 	SHM_QUEUE  *procLocks;
 	PROCLOCK   *proclock;
 	uint32		hashcode;
-	LWLockId	partitionLock;
+	FlexLockId	partitionLock;
 	int			count = 0;
 	int			fast_count = 0;
 
@@ -2593,7 +2593,7 @@ LockRefindAndRelease(LockMethod lockMethodTable, PGPROC *proc,
 	PROCLOCKTAG proclocktag;
 	uint32		hashcode;
 	uint32		proclock_hashcode;
-	LWLockId	partitionLock;
+	FlexLockId	partitionLock;
 	bool		wakeupNeeded;
 
 	hashcode = LockTagHashCode(locktag);
@@ -2827,7 +2827,7 @@ PostPrepare_Locks(TransactionId xid)
 	 */
 	for (partition = 0; partition < NUM_LOCK_PARTITIONS; partition++)
 	{
-		LWLockId	partitionLock = FirstLockMgrLock + partition;
+		FlexLockId	partitionLock = FirstLockMgrLock + partition;
 		SHM_QUEUE  *procLocks = &(MyProc->myProcLocks[partition]);
 
 		proclock = (PROCLOCK *) SHMQueueNext(procLocks, procLocks,
@@ -3342,7 +3342,7 @@ lock_twophase_recover(TransactionId xid, uint16 info,
 	uint32		hashcode;
 	uint32		proclock_hashcode;
 	int			partition;
-	LWLockId	partitionLock;
+	FlexLockId	partitionLock;
 	LockMethod	lockMethodTable;
 
 	Assert(len == sizeof(TwoPhaseLockRecord));
diff --git a/src/backend/storage/lmgr/lwlock.c b/src/backend/storage/lmgr/lwlock.c
index 079eb29..ce6c931 100644
--- a/src/backend/storage/lmgr/lwlock.c
+++ b/src/backend/storage/lmgr/lwlock.c
@@ -21,74 +21,23 @@
  */
 #include "postgres.h"
 
-#include "access/clog.h"
-#include "access/multixact.h"
-#include "access/subtrans.h"
-#include "commands/async.h"
 #include "miscadmin.h"
 #include "pg_trace.h"
+#include "storage/flexlock_internals.h"
 #include "storage/ipc.h"
-#include "storage/predicate.h"
 #include "storage/proc.h"
 #include "storage/spin.h"
 
-
-/* We use the ShmemLock spinlock to protect LWLockAssign */
-extern slock_t *ShmemLock;
-
-
 typedef struct LWLock
 {
-	slock_t		mutex;			/* Protects LWLock and queue of PGPROCs */
-	bool		releaseOK;		/* T if ok to release waiters */
+	FlexLock	flex;			/* common FlexLock infrastructure */
 	char		exclusive;		/* # of exclusive holders (0 or 1) */
 	int			shared;			/* # of shared holders (0..MaxBackends) */
-	PGPROC	   *head;			/* head of list of waiting PGPROCs */
-	PGPROC	   *tail;			/* tail of list of waiting PGPROCs */
-	/* tail is undefined when head is NULL */
 } LWLock;
 
-/*
- * All the LWLock structs are allocated as an array in shared memory.
- * (LWLockIds are indexes into the array.)	We force the array stride to
- * be a power of 2, which saves a few cycles in indexing, but more
- * importantly also ensures that individual LWLocks don't cross cache line
- * boundaries.	This reduces cache contention problems, especially on AMD
- * Opterons.  (Of course, we have to also ensure that the array start
- * address is suitably aligned.)
- *
- * LWLock is between 16 and 32 bytes on all known platforms, so these two
- * cases are sufficient.
- */
-#define LWLOCK_PADDED_SIZE	(sizeof(LWLock) <= 16 ? 16 : 32)
-
-typedef union LWLockPadded
-{
-	LWLock		lock;
-	char		pad[LWLOCK_PADDED_SIZE];
-} LWLockPadded;
-
-/*
- * This points to the array of LWLocks in shared memory.  Backends inherit
- * the pointer by fork from the postmaster (except in the EXEC_BACKEND case,
- * where we have special measures to pass it down).
- */
-NON_EXEC_STATIC LWLockPadded *LWLockArray = NULL;
-
-
-/*
- * We use this structure to keep track of locked LWLocks for release
- * during error recovery.  The maximum size could be determined at runtime
- * if necessary, but it seems unlikely that more than a few locks could
- * ever be held simultaneously.
- */
-#define MAX_SIMUL_LWLOCKS	100
-
-static int	num_held_lwlocks = 0;
-static LWLockId held_lwlocks[MAX_SIMUL_LWLOCKS];
-
-static int	lock_addin_request = 0;
-static bool lock_addin_request_allowed = true;
+#define	LWLockPointer(lockid) \
+	(AssertMacro(FlexLockArray[lockid].flex.locktype == FLEXLOCK_TYPE_LWLOCK), \
+	 (volatile LWLock *) &FlexLockArray[lockid])
 
 #ifdef LWLOCK_STATS
 static int	counts_for_pid = 0;
@@ -98,27 +47,17 @@ static int *block_counts;
 #endif
 
 #ifdef LOCK_DEBUG
-bool		Trace_lwlocks = false;
-
 inline static void
-PRINT_LWDEBUG(const char *where, LWLockId lockid, const volatile LWLock *lock)
+PRINT_LWDEBUG(const char *where, FlexLockId lockid, const volatile LWLock *lock)
 {
-	if (Trace_lwlocks)
+	if (Trace_flexlocks)
 		elog(LOG, "%s(%d): excl %d shared %d head %p rOK %d",
 			 where, (int) lockid,
-			 (int) lock->exclusive, lock->shared, lock->head,
-			 (int) lock->releaseOK);
-}
-
-inline static void
-LOG_LWDEBUG(const char *where, LWLockId lockid, const char *msg)
-{
-	if (Trace_lwlocks)
-		elog(LOG, "%s(%d): %s", where, (int) lockid, msg);
+			 (int) lock->exclusive, lock->shared, lock->flex.head,
+			 (int) lock->flex.releaseOK);
 }
 #else							/* not LOCK_DEBUG */
 #define PRINT_LWDEBUG(a,b,c)
-#define LOG_LWDEBUG(a,b,c)
 #endif   /* LOCK_DEBUG */
 
 #ifdef LWLOCK_STATS
@@ -127,8 +66,8 @@ static void
 print_lwlock_stats(int code, Datum arg)
 {
 	int			i;
-	int		   *LWLockCounter = (int *) ((char *) LWLockArray - 2 * sizeof(int));
-	int			numLocks = LWLockCounter[1];
+	int		   *FlexLockCounter = (int *) ((char *) FlexLockArray - 2 * sizeof(int));
+	int			numLocks = FlexLockCounter[1];
 
 	/* Grab an LWLock to keep different backends from mixing reports */
 	LWLockAcquire(0, LW_EXCLUSIVE);
@@ -145,173 +84,15 @@ print_lwlock_stats(int code, Datum arg)
 }
 #endif   /* LWLOCK_STATS */
 
-
 /*
- * Compute number of LWLocks to allocate.
+ * LWLockAssign - initialize a new lwlock and return its ID
  */
-int
-NumLWLocks(void)
-{
-	int			numLocks;
-
-	/*
-	 * Possibly this logic should be spread out among the affected modules,
-	 * the same way that shmem space estimation is done.  But for now, there
-	 * are few enough users of LWLocks that we can get away with just keeping
-	 * the knowledge here.
-	 */
-
-	/* Predefined LWLocks */
-	numLocks = (int) NumFixedLWLocks;
-
-	/* bufmgr.c needs two for each shared buffer */
-	numLocks += 2 * NBuffers;
-
-	/* proc.c needs one for each backend or auxiliary process */
-	numLocks += MaxBackends + NUM_AUXILIARY_PROCS;
-
-	/* clog.c needs one per CLOG buffer */
-	numLocks += NUM_CLOG_BUFFERS;
-
-	/* subtrans.c needs one per SubTrans buffer */
-	numLocks += NUM_SUBTRANS_BUFFERS;
-
-	/* multixact.c needs two SLRU areas */
-	numLocks += NUM_MXACTOFFSET_BUFFERS + NUM_MXACTMEMBER_BUFFERS;
-
-	/* async.c needs one per Async buffer */
-	numLocks += NUM_ASYNC_BUFFERS;
-
-	/* predicate.c needs one per old serializable xid buffer */
-	numLocks += NUM_OLDSERXID_BUFFERS;
-
-	/*
-	 * Add any requested by loadable modules; for backwards-compatibility
-	 * reasons, allocate at least NUM_USER_DEFINED_LWLOCKS of them even if
-	 * there are no explicit requests.
-	 */
-	lock_addin_request_allowed = false;
-	numLocks += Max(lock_addin_request, NUM_USER_DEFINED_LWLOCKS);
-
-	return numLocks;
-}
-
-
-/*
- * RequestAddinLWLocks
- *		Request that extra LWLocks be allocated for use by
- *		a loadable module.
- *
- * This is only useful if called from the _PG_init hook of a library that
- * is loaded into the postmaster via shared_preload_libraries.	Once
- * shared memory has been allocated, calls will be ignored.  (We could
- * raise an error, but it seems better to make it a no-op, so that
- * libraries containing such calls can be reloaded if needed.)
- */
-void
-RequestAddinLWLocks(int n)
-{
-	if (IsUnderPostmaster || !lock_addin_request_allowed)
-		return;					/* too late */
-	lock_addin_request += n;
-}
-
-
-/*
- * Compute shmem space needed for LWLocks.
- */
-Size
-LWLockShmemSize(void)
-{
-	Size		size;
-	int			numLocks = NumLWLocks();
-
-	/* Space for the LWLock array. */
-	size = mul_size(numLocks, sizeof(LWLockPadded));
-
-	/* Space for dynamic allocation counter, plus room for alignment. */
-	size = add_size(size, 2 * sizeof(int) + LWLOCK_PADDED_SIZE);
-
-	return size;
-}
-
-
-/*
- * Allocate shmem space for LWLocks and initialize the locks.
- */
-void
-CreateLWLocks(void)
-{
-	int			numLocks = NumLWLocks();
-	Size		spaceLocks = LWLockShmemSize();
-	LWLockPadded *lock;
-	int		   *LWLockCounter;
-	char	   *ptr;
-	int			id;
-
-	/* Allocate space */
-	ptr = (char *) ShmemAlloc(spaceLocks);
-
-	/* Leave room for dynamic allocation counter */
-	ptr += 2 * sizeof(int);
-
-	/* Ensure desired alignment of LWLock array */
-	ptr += LWLOCK_PADDED_SIZE - ((uintptr_t) ptr) % LWLOCK_PADDED_SIZE;
-
-	LWLockArray = (LWLockPadded *) ptr;
-
-	/*
-	 * Initialize all LWLocks to "unlocked" state
-	 */
-	for (id = 0, lock = LWLockArray; id < numLocks; id++, lock++)
-	{
-		SpinLockInit(&lock->lock.mutex);
-		lock->lock.releaseOK = true;
-		lock->lock.exclusive = 0;
-		lock->lock.shared = 0;
-		lock->lock.head = NULL;
-		lock->lock.tail = NULL;
-	}
-
-	/*
-	 * Initialize the dynamic-allocation counter, which is stored just before
-	 * the first LWLock.
-	 */
-	LWLockCounter = (int *) ((char *) LWLockArray - 2 * sizeof(int));
-	LWLockCounter[0] = (int) NumFixedLWLocks;
-	LWLockCounter[1] = numLocks;
-}
-
-
-/*
- * LWLockAssign - assign a dynamically-allocated LWLock number
- *
- * We interlock this using the same spinlock that is used to protect
- * ShmemAlloc().  Interlocking is not really necessary during postmaster
- * startup, but it is needed if any user-defined code tries to allocate
- * LWLocks after startup.
- */
-LWLockId
+FlexLockId
 LWLockAssign(void)
 {
-	LWLockId	result;
-
-	/* use volatile pointer to prevent code rearrangement */
-	volatile int *LWLockCounter;
-
-	LWLockCounter = (int *) ((char *) LWLockArray - 2 * sizeof(int));
-	SpinLockAcquire(ShmemLock);
-	if (LWLockCounter[0] >= LWLockCounter[1])
-	{
-		SpinLockRelease(ShmemLock);
-		elog(ERROR, "no more LWLockIds available");
-	}
-	result = (LWLockId) (LWLockCounter[0]++);
-	SpinLockRelease(ShmemLock);
-	return result;
+	return FlexLockAssign(FLEXLOCK_TYPE_LWLOCK);
 }
 
-
 /*
  * LWLockAcquire - acquire a lightweight lock in the specified mode
  *
@@ -320,9 +101,9 @@ LWLockAssign(void)
  * Side effect: cancel/die interrupts are held off until lock release.
  */
 void
-LWLockAcquire(LWLockId lockid, LWLockMode mode)
+LWLockAcquire(FlexLockId lockid, LWLockMode mode)
 {
-	volatile LWLock *lock = &(LWLockArray[lockid].lock);
+	volatile LWLock *lock = LWLockPointer(lockid);
 	PGPROC	   *proc = MyProc;
 	bool		retry = false;
 	int			extraWaits = 0;
@@ -333,8 +114,8 @@ LWLockAcquire(LWLockId lockid, LWLockMode mode)
 	/* 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];
+		int		   *FlexLockCounter = (int *) ((char *) FlexLockArray - 2 * sizeof(int));
+		int			numLocks = FlexLockCounter[1];
 
 		sh_acquire_counts = calloc(numLocks, sizeof(int));
 		ex_acquire_counts = calloc(numLocks, sizeof(int));
@@ -356,10 +137,6 @@ LWLockAcquire(LWLockId lockid, LWLockMode mode)
 	 */
 	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
@@ -388,11 +165,11 @@ LWLockAcquire(LWLockId lockid, LWLockMode mode)
 		bool		mustwait;
 
 		/* Acquire mutex.  Time spent holding mutex should be short! */
-		SpinLockAcquire(&lock->mutex);
+		SpinLockAcquire(&lock->flex.mutex);
 
 		/* If retrying, allow LWLockRelease to release waiters again */
 		if (retry)
-			lock->releaseOK = true;
+			lock->flex.releaseOK = true;
 
 		/* If I can get the lock, do so quickly. */
 		if (mode == LW_EXCLUSIVE)
@@ -419,72 +196,30 @@ LWLockAcquire(LWLockId lockid, LWLockMode mode)
 		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;
+		/* Add myself to wait queue. */
+		FlexLockJoinWaitQueue(lock, (int) mode);
 
 		/* 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");
+		SpinLockRelease(&lock->flex.mutex);
+
+		/* Wait until awakened. */
+		extraWaits += FlexLockWait(lockid, mode);
 
 #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);
+	SpinLockRelease(&lock->flex.mutex);
 
-	TRACE_POSTGRESQL_LWLOCK_ACQUIRE(lockid, mode);
+	TRACE_POSTGRESQL_FLEXLOCK_ACQUIRE(lockid, mode);
 
 	/* Add lock to list of locks held by this backend */
-	held_lwlocks[num_held_lwlocks++] = lockid;
+	FlexLockRemember(lockid);
 
 	/*
 	 * Fix the process wait semaphore's count for any absorbed wakeups.
@@ -501,17 +236,13 @@ LWLockAcquire(LWLockId lockid, LWLockMode mode)
  * If successful, cancel/die interrupts are held off until lock release.
  */
 bool
-LWLockConditionalAcquire(LWLockId lockid, LWLockMode mode)
+LWLockConditionalAcquire(FlexLockId lockid, LWLockMode mode)
 {
-	volatile LWLock *lock = &(LWLockArray[lockid].lock);
+	volatile LWLock *lock = LWLockPointer(lockid);
 	bool		mustwait;
 
 	PRINT_LWDEBUG("LWLockConditionalAcquire", lockid, lock);
 
-	/* 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
@@ -520,7 +251,7 @@ LWLockConditionalAcquire(LWLockId lockid, LWLockMode mode)
 	HOLD_INTERRUPTS();
 
 	/* Acquire mutex.  Time spent holding mutex should be short! */
-	SpinLockAcquire(&lock->mutex);
+	SpinLockAcquire(&lock->flex.mutex);
 
 	/* If I can get the lock, do so quickly. */
 	if (mode == LW_EXCLUSIVE)
@@ -545,20 +276,20 @@ LWLockConditionalAcquire(LWLockId lockid, LWLockMode mode)
 	}
 
 	/* We are done updating shared state of the lock itself. */
-	SpinLockRelease(&lock->mutex);
+	SpinLockRelease(&lock->flex.mutex);
 
 	if (mustwait)
 	{
 		/* Failed to get lock, so release interrupt holdoff */
 		RESUME_INTERRUPTS();
-		LOG_LWDEBUG("LWLockConditionalAcquire", lockid, "failed");
-		TRACE_POSTGRESQL_LWLOCK_CONDACQUIRE_FAIL(lockid, mode);
+		FlexLockDebug("LWLockConditionalAcquire", lockid, "failed");
+		TRACE_POSTGRESQL_FLEXLOCK_CONDACQUIRE_FAIL(lockid, mode);
 	}
 	else
 	{
 		/* Add lock to list of locks held by this backend */
-		held_lwlocks[num_held_lwlocks++] = lockid;
-		TRACE_POSTGRESQL_LWLOCK_CONDACQUIRE(lockid, mode);
+		FlexLockRemember(lockid);
+		TRACE_POSTGRESQL_FLEXLOCK_CONDACQUIRE(lockid, mode);
 	}
 
 	return !mustwait;
@@ -568,32 +299,18 @@ LWLockConditionalAcquire(LWLockId lockid, LWLockMode mode)
  * LWLockRelease - release a previously acquired lock
  */
 void
-LWLockRelease(LWLockId lockid)
+LWLockRelease(FlexLockId lockid)
 {
-	volatile LWLock *lock = &(LWLockArray[lockid].lock);
+	volatile LWLock *lock = LWLockPointer(lockid);
 	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];
+	FlexLockForget(lockid);
 
 	/* Acquire mutex.  Time spent holding mutex should be short! */
-	SpinLockAcquire(&lock->mutex);
+	SpinLockAcquire(&lock->flex.mutex);
 
 	/* Release my hold on lock */
 	if (lock->exclusive > 0)
@@ -610,10 +327,10 @@ LWLockRelease(LWLockId lockid)
 	 * if someone has already awakened waiters that haven't yet acquired the
 	 * lock.
 	 */
-	head = lock->head;
+	head = lock->flex.head;
 	if (head != NULL)
 	{
-		if (lock->exclusive == 0 && lock->shared == 0 && lock->releaseOK)
+		if (lock->exclusive == 0 && lock->shared == 0 && lock->flex.releaseOK)
 		{
 			/*
 			 * Remove the to-be-awakened PGPROCs from the queue.  If the front
@@ -621,17 +338,17 @@ LWLockRelease(LWLockId lockid)
 			 * as many waiters as want shared access.
 			 */
 			proc = head;
-			if (!proc->lwExclusive)
+			if (proc->flWaitMode != LW_EXCLUSIVE)
 			{
-				while (proc->lwWaitLink != NULL &&
-					   !proc->lwWaitLink->lwExclusive)
-					proc = proc->lwWaitLink;
+				while (proc->flWaitLink != NULL &&
+					   proc->flWaitLink->flWaitMode != LW_EXCLUSIVE)
+					proc = proc->flWaitLink;
 			}
 			/* proc is now the last PGPROC to be released */
-			lock->head = proc->lwWaitLink;
-			proc->lwWaitLink = NULL;
+			lock->flex.head = proc->flWaitLink;
+			proc->flWaitLink = NULL;
 			/* prevent additional wakeups until retryer gets to run */
-			lock->releaseOK = false;
+			lock->flex.releaseOK = false;
 		}
 		else
 		{
@@ -641,20 +358,20 @@ LWLockRelease(LWLockId lockid)
 	}
 
 	/* We are done updating shared state of the lock itself. */
-	SpinLockRelease(&lock->mutex);
+	SpinLockRelease(&lock->flex.mutex);
 
-	TRACE_POSTGRESQL_LWLOCK_RELEASE(lockid);
+	TRACE_POSTGRESQL_FLEXLOCK_RELEASE(lockid);
 
 	/*
 	 * Awaken any waiters I removed from the queue.
 	 */
 	while (head != NULL)
 	{
-		LOG_LWDEBUG("LWLockRelease", lockid, "release waiter");
+		FlexLockDebug("LWLockRelease", lockid, "release waiter");
 		proc = head;
-		head = proc->lwWaitLink;
-		proc->lwWaitLink = NULL;
-		proc->lwWaiting = false;
+		head = proc->flWaitLink;
+		proc->flWaitLink = NULL;
+		proc->flWaitResult = 1;		/* any non-zero value will do */
 		PGSemaphoreUnlock(&proc->sem);
 	}
 
@@ -664,43 +381,17 @@ LWLockRelease(LWLockId lockid)
 	RESUME_INTERRUPTS();
 }
 
-
-/*
- * LWLockReleaseAll - release all currently-held locks
- *
- * Used to clean up after ereport(ERROR). An important difference between this
- * function and retail LWLockRelease calls is that InterruptHoldoffCount is
- * unchanged by this operation.  This is necessary since InterruptHoldoffCount
- * has been set to an appropriate level earlier in error recovery. We could
- * decrement it below zero if we allow it to drop for each released lock!
- */
-void
-LWLockReleaseAll(void)
-{
-	while (num_held_lwlocks > 0)
-	{
-		HOLD_INTERRUPTS();		/* match the upcoming RESUME_INTERRUPTS */
-
-		LWLockRelease(held_lwlocks[num_held_lwlocks - 1]);
-	}
-}
-
-
 /*
  * LWLockHeldByMe - test whether my process currently holds a lock
  *
- * This is meant as debug support only.  We do not distinguish whether the
- * lock is held shared or exclusive.
+ * The following convenience routine might not be worthwhile but for the fact
+ * that we've had a function by this name since long before FlexLocks existed.
+ * Callers who want to check whether an arbitrary FlexLock (that may or may not
+ * be an LWLock) is held can use FlexLockHeldByMe directly.
  */
 bool
-LWLockHeldByMe(LWLockId lockid)
+LWLockHeldByMe(FlexLockId lockid)
 {
-	int			i;
-
-	for (i = 0; i < num_held_lwlocks; i++)
-	{
-		if (held_lwlocks[i] == lockid)
-			return true;
-	}
-	return false;
+	AssertMacro(FlexLockArray[lockid].flex.locktype == FLEXLOCK_TYPE_LWLOCK);
+	return FlexLockHeldByMe(lockid);
 }
diff --git a/src/backend/storage/lmgr/predicate.c b/src/backend/storage/lmgr/predicate.c
index 345f6f5..15978a4 100644
--- a/src/backend/storage/lmgr/predicate.c
+++ b/src/backend/storage/lmgr/predicate.c
@@ -239,7 +239,7 @@
 #define PredicateLockHashPartition(hashcode) \
 	((hashcode) % NUM_PREDICATELOCK_PARTITIONS)
 #define PredicateLockHashPartitionLock(hashcode) \
-	((LWLockId) (FirstPredicateLockMgrLock + PredicateLockHashPartition(hashcode)))
+	((FlexLockId) (FirstPredicateLockMgrLock + PredicateLockHashPartition(hashcode)))
 
 #define NPREDICATELOCKTARGETENTS() \
 	mul_size(max_predicate_locks_per_xact, add_size(MaxBackends, max_prepared_xacts))
@@ -1840,7 +1840,7 @@ PageIsPredicateLocked(Relation relation, BlockNumber blkno)
 {
 	PREDICATELOCKTARGETTAG targettag;
 	uint32		targettaghash;
-	LWLockId	partitionLock;
+	FlexLockId	partitionLock;
 	PREDICATELOCKTARGET *target;
 
 	SET_PREDICATELOCKTARGETTAG_PAGE(targettag,
@@ -2073,7 +2073,7 @@ DeleteChildTargetLocks(const PREDICATELOCKTARGETTAG *newtargettag)
 		if (TargetTagIsCoveredBy(oldtargettag, *newtargettag))
 		{
 			uint32		oldtargettaghash;
-			LWLockId	partitionLock;
+			FlexLockId	partitionLock;
 			PREDICATELOCK *rmpredlock;
 
 			oldtargettaghash = PredicateLockTargetTagHashCode(&oldtargettag);
@@ -2285,7 +2285,7 @@ CreatePredicateLock(const PREDICATELOCKTARGETTAG *targettag,
 	PREDICATELOCKTARGET *target;
 	PREDICATELOCKTAG locktag;
 	PREDICATELOCK *lock;
-	LWLockId	partitionLock;
+	FlexLockId	partitionLock;
 	bool		found;
 
 	partitionLock = PredicateLockHashPartitionLock(targettaghash);
@@ -2586,10 +2586,10 @@ TransferPredicateLocksToNewTarget(PREDICATELOCKTARGETTAG oldtargettag,
 								  bool removeOld)
 {
 	uint32		oldtargettaghash;
-	LWLockId	oldpartitionLock;
+	FlexLockId	oldpartitionLock;
 	PREDICATELOCKTARGET *oldtarget;
 	uint32		newtargettaghash;
-	LWLockId	newpartitionLock;
+	FlexLockId	newpartitionLock;
 	bool		found;
 	bool		outOfShmem = false;
 
@@ -3578,7 +3578,7 @@ ClearOldPredicateLocks(void)
 			PREDICATELOCKTARGET *target;
 			PREDICATELOCKTARGETTAG targettag;
 			uint32		targettaghash;
-			LWLockId	partitionLock;
+			FlexLockId	partitionLock;
 
 			tag = predlock->tag;
 			target = tag.myTarget;
@@ -3656,7 +3656,7 @@ ReleaseOneSerializableXact(SERIALIZABLEXACT *sxact, bool partial,
 		PREDICATELOCKTARGET *target;
 		PREDICATELOCKTARGETTAG targettag;
 		uint32		targettaghash;
-		LWLockId	partitionLock;
+		FlexLockId	partitionLock;
 
 		nextpredlock = (PREDICATELOCK *)
 			SHMQueueNext(&(sxact->predicateLocks),
@@ -4034,7 +4034,7 @@ static void
 CheckTargetForConflictsIn(PREDICATELOCKTARGETTAG *targettag)
 {
 	uint32		targettaghash;
-	LWLockId	partitionLock;
+	FlexLockId	partitionLock;
 	PREDICATELOCKTARGET *target;
 	PREDICATELOCK *predlock;
 	PREDICATELOCK *mypredlock = NULL;
diff --git a/src/backend/storage/lmgr/proc.c b/src/backend/storage/lmgr/proc.c
index eda3a98..57da345 100644
--- a/src/backend/storage/lmgr/proc.c
+++ b/src/backend/storage/lmgr/proc.c
@@ -325,9 +325,9 @@ InitProcess(void)
 	/* NB -- autovac launcher intentionally does not set IS_AUTOVACUUM */
 	if (IsAutoVacuumWorkerProcess())
 		MyProc->vacuumFlags |= PROC_IS_AUTOVACUUM;
-	MyProc->lwWaiting = false;
-	MyProc->lwExclusive = false;
-	MyProc->lwWaitLink = NULL;
+	MyProc->flWaitResult = 0;
+	MyProc->flWaitMode = 0;
+	MyProc->flWaitLink = NULL;
 	MyProc->waitLock = NULL;
 	MyProc->waitProcLock = NULL;
 #ifdef USE_ASSERT_CHECKING
@@ -479,9 +479,9 @@ InitAuxiliaryProcess(void)
 	MyProc->roleId = InvalidOid;
 	MyProc->inCommit = false;
 	MyProc->vacuumFlags = 0;
-	MyProc->lwWaiting = false;
-	MyProc->lwExclusive = false;
-	MyProc->lwWaitLink = NULL;
+	MyProc->flWaitMode = 0;
+	MyProc->flWaitResult = 0;
+	MyProc->flWaitLink = NULL;
 	MyProc->waitLock = NULL;
 	MyProc->waitProcLock = NULL;
 #ifdef USE_ASSERT_CHECKING
@@ -607,7 +607,7 @@ IsWaitingForLock(void)
 void
 LockWaitCancel(void)
 {
-	LWLockId	partitionLock;
+	FlexLockId	partitionLock;
 
 	/* Nothing to do if we weren't waiting for a lock */
 	if (lockAwaited == NULL)
@@ -718,11 +718,11 @@ ProcKill(int code, Datum arg)
 #endif
 
 	/*
-	 * Release any LW locks I am holding.  There really shouldn't be any, but
-	 * it's cheap to check again before we cut the knees off the LWLock
+	 * Release any felx locks I am holding.  There really shouldn't be any, but
+	 * it's cheap to check again before we cut the knees off the flex lock
 	 * facility by releasing our PGPROC ...
 	 */
-	LWLockReleaseAll();
+	FlexLockReleaseAll();
 
 	/* Release ownership of the process's latch, too */
 	DisownLatch(&MyProc->procLatch);
@@ -779,8 +779,8 @@ AuxiliaryProcKill(int code, Datum arg)
 
 	Assert(MyProc == auxproc);
 
-	/* Release any LW locks I am holding (see notes above) */
-	LWLockReleaseAll();
+	/* Release any flex locks I am holding (see notes above) */
+	FlexLockReleaseAll();
 
 	/* Release ownership of the process's latch, too */
 	DisownLatch(&MyProc->procLatch);
@@ -865,7 +865,7 @@ ProcSleep(LOCALLOCK *locallock, LockMethod lockMethodTable)
 	LOCK	   *lock = locallock->lock;
 	PROCLOCK   *proclock = locallock->proclock;
 	uint32		hashcode = locallock->hashcode;
-	LWLockId	partitionLock = LockHashPartitionLock(hashcode);
+	FlexLockId	partitionLock = LockHashPartitionLock(hashcode);
 	PROC_QUEUE *waitQueue = &(lock->waitProcs);
 	LOCKMASK	myHeldLocks = MyProc->heldLocks;
 	bool		early_deadlock = false;
diff --git a/src/backend/utils/misc/check_guc b/src/backend/utils/misc/check_guc
index 293fb03..1a19e36 100755
--- a/src/backend/utils/misc/check_guc
+++ b/src/backend/utils/misc/check_guc
@@ -19,7 +19,7 @@
 INTENTIONALLY_NOT_INCLUDED="autocommit debug_deadlocks \
 is_superuser lc_collate lc_ctype lc_messages lc_monetary lc_numeric lc_time \
 pre_auth_delay role seed server_encoding server_version server_version_int \
-session_authorization trace_lock_oidmin trace_lock_table trace_locks trace_lwlocks \
+session_authorization trace_lock_oidmin trace_lock_table trace_locks trace_flexlocks \
 trace_notify trace_userlocks transaction_isolation transaction_read_only \
 zero_damaged_pages"
 
diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c
index da7b6d4..52de233 100644
--- a/src/backend/utils/misc/guc.c
+++ b/src/backend/utils/misc/guc.c
@@ -59,6 +59,7 @@
 #include "replication/walreceiver.h"
 #include "replication/walsender.h"
 #include "storage/bufmgr.h"
+#include "storage/flexlock_internals.h"
 #include "storage/standby.h"
 #include "storage/fd.h"
 #include "storage/predicate.h"
@@ -1071,12 +1072,12 @@ static struct config_bool ConfigureNamesBool[] =
 		NULL, NULL, NULL
 	},
 	{
-		{"trace_lwlocks", PGC_SUSET, DEVELOPER_OPTIONS,
+		{"trace_flexlocks", PGC_SUSET, DEVELOPER_OPTIONS,
 			gettext_noop("No description available."),
 			NULL,
 			GUC_NOT_IN_SAMPLE
 		},
-		&Trace_lwlocks,
+		&Trace_flexlocks,
 		false,
 		NULL, NULL, NULL
 	},
diff --git a/src/backend/utils/probes.d b/src/backend/utils/probes.d
index 71c5ab0..5b9cfe6 100644
--- a/src/backend/utils/probes.d
+++ b/src/backend/utils/probes.d
@@ -15,8 +15,8 @@
  * in probe definitions, as they cause compilation errors on Mac OS X 10.5.
  */
 #define LocalTransactionId unsigned int
-#define LWLockId int
-#define LWLockMode int
+#define FlexLockId int
+#define FlexLockMode int
 #define LOCKMODE int
 #define BlockNumber unsigned int
 #define Oid unsigned int
@@ -29,12 +29,12 @@ provider postgresql {
 	probe transaction__commit(LocalTransactionId);
 	probe transaction__abort(LocalTransactionId);
 
-	probe lwlock__acquire(LWLockId, LWLockMode);
-	probe lwlock__release(LWLockId);
-	probe lwlock__wait__start(LWLockId, LWLockMode);
-	probe lwlock__wait__done(LWLockId, LWLockMode);
-	probe lwlock__condacquire(LWLockId, LWLockMode);
-	probe lwlock__condacquire__fail(LWLockId, LWLockMode);
+	probe flexlock__acquire(FlexLockId, FlexLockMode);
+	probe flexlock__release(FlexLockId);
+	probe flexlock__wait__start(FlexLockId, FlexLockMode);
+	probe flexlock__wait__done(FlexLockId, FlexLockMode);
+	probe flexlock__condacquire(FlexLockId, FlexLockMode);
+	probe flexlock__condacquire__fail(FlexLockId, FlexLockMode);
 
 	probe lock__wait__start(unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, LOCKMODE);
 	probe lock__wait__done(unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, LOCKMODE);
diff --git a/src/include/access/slru.h b/src/include/access/slru.h
index e48743f..680a87f 100644
--- a/src/include/access/slru.h
+++ b/src/include/access/slru.h
@@ -55,7 +55,7 @@ typedef enum
  */
 typedef struct SlruSharedData
 {
-	LWLockId	ControlLock;
+	FlexLockId	ControlLock;
 
 	/* Number of buffers managed by this SLRU structure */
 	int			num_slots;
@@ -69,7 +69,7 @@ typedef struct SlruSharedData
 	bool	   *page_dirty;
 	int		   *page_number;
 	int		   *page_lru_count;
-	LWLockId   *buffer_locks;
+	FlexLockId *buffer_locks;
 
 	/*
 	 * Optional array of WAL flush LSNs associated with entries in the SLRU
@@ -136,7 +136,7 @@ typedef SlruCtlData *SlruCtl;
 
 extern Size SimpleLruShmemSize(int nslots, int nlsns);
 extern void SimpleLruInit(SlruCtl ctl, const char *name, int nslots, int nlsns,
-			  LWLockId ctllock, const char *subdir);
+			  FlexLockId ctllock, const char *subdir);
 extern int	SimpleLruZeroPage(SlruCtl ctl, int pageno);
 extern int SimpleLruReadPage(SlruCtl ctl, int pageno, bool write_ok,
 				  TransactionId xid);
diff --git a/src/include/pg_config_manual.h b/src/include/pg_config_manual.h
index 6c8e312..d3b74db 100644
--- a/src/include/pg_config_manual.h
+++ b/src/include/pg_config_manual.h
@@ -49,9 +49,9 @@
 #define SEQ_MINVALUE	(-SEQ_MAXVALUE)
 
 /*
- * Number of spare LWLocks to allocate for user-defined add-on code.
+ * Number of spare FlexLocks to allocate for user-defined add-on code.
  */
-#define NUM_USER_DEFINED_LWLOCKS	4
+#define NUM_USER_DEFINED_FLEXLOCKS	4
 
 /*
  * Define this if you want to allow the lo_import and lo_export SQL
diff --git a/src/include/storage/buf_internals.h b/src/include/storage/buf_internals.h
index b7d4ea5..ac7f665 100644
--- a/src/include/storage/buf_internals.h
+++ b/src/include/storage/buf_internals.h
@@ -103,7 +103,7 @@ typedef struct buftag
 #define BufTableHashPartition(hashcode) \
 	((hashcode) % NUM_BUFFER_PARTITIONS)
 #define BufMappingPartitionLock(hashcode) \
-	((LWLockId) (FirstBufMappingLock + BufTableHashPartition(hashcode)))
+	((FlexLockId) (FirstBufMappingLock + BufTableHashPartition(hashcode)))
 
 /*
  *	BufferDesc -- shared descriptor/state data for a single shared buffer.
@@ -143,8 +143,8 @@ typedef struct sbufdesc
 	int			buf_id;			/* buffer's index number (from 0) */
 	int			freeNext;		/* link in freelist chain */
 
-	LWLockId	io_in_progress_lock;	/* to wait for I/O to complete */
-	LWLockId	content_lock;	/* to lock access to buffer contents */
+	FlexLockId	io_in_progress_lock;	/* to wait for I/O to complete */
+	FlexLockId	content_lock;	/* to lock access to buffer contents */
 } BufferDesc;
 
 #define BufferDescriptorGetBuffer(bdesc) ((bdesc)->buf_id + 1)
diff --git a/src/include/storage/flexlock.h b/src/include/storage/flexlock.h
new file mode 100644
index 0000000..612c21a
--- /dev/null
+++ b/src/include/storage/flexlock.h
@@ -0,0 +1,102 @@
+/*-------------------------------------------------------------------------
+ *
+ * flexlock.h
+ *	  Flex lock manager
+ *
+ * Portions Copyright (c) 1996-2011, PostgreSQL Global Development Group
+ * Portions Copyright (c) 1994, Regents of the University of California
+ *
+ * src/include/storage/flexlock.h
+ *
+ *-------------------------------------------------------------------------
+ */
+#ifndef FLEXLOCK_H
+#define FLEXLOCK_H
+
+/*
+ * It's a bit odd to declare NUM_BUFFER_PARTITIONS and NUM_LOCK_PARTITIONS
+ * here, but we need them to set up enum FlexLockId correctly, and having
+ * this file include lock.h or bufmgr.h would be backwards.
+ */
+
+/* Number of partitions of the shared buffer mapping hashtable */
+#define NUM_BUFFER_PARTITIONS  16
+
+/* Number of partitions the shared lock tables are divided into */
+#define LOG2_NUM_LOCK_PARTITIONS  4
+#define NUM_LOCK_PARTITIONS  (1 << LOG2_NUM_LOCK_PARTITIONS)
+
+/* Number of partitions the shared predicate lock tables are divided into */
+#define LOG2_NUM_PREDICATELOCK_PARTITIONS  4
+#define NUM_PREDICATELOCK_PARTITIONS  (1 << LOG2_NUM_PREDICATELOCK_PARTITIONS)
+
+/*
+ * We have a number of predefined FlexLocks, plus a bunch of locks that are
+ * dynamically assigned (e.g., for shared buffers).  The FlexLock structures
+ * live in shared memory (since they contain shared data) and are identified
+ * by values of this enumerated type.  We abuse the notion of an enum somewhat
+ * by allowing values not listed in the enum declaration to be assigned.
+ * The extra value MaxDynamicFlexLock is there to keep the compiler from
+ * deciding that the enum can be represented as char or short ...
+ *
+ * If you remove a lock, please replace it with a placeholder. This retains
+ * the lock numbering, which is helpful for DTrace and other external
+ * debugging scripts.
+ */
+typedef enum FlexLockId
+{
+	BufFreelistLock,
+	ShmemIndexLock,
+	OidGenLock,
+	XidGenLock,
+	ProcArrayLock,
+	SInvalReadLock,
+	SInvalWriteLock,
+	WALInsertLock,
+	WALWriteLock,
+	ControlFileLock,
+	CheckpointLock,
+	CLogControlLock,
+	SubtransControlLock,
+	MultiXactGenLock,
+	MultiXactOffsetControlLock,
+	MultiXactMemberControlLock,
+	RelCacheInitLock,
+	BgWriterCommLock,
+	TwoPhaseStateLock,
+	TablespaceCreateLock,
+	BtreeVacuumLock,
+	AddinShmemInitLock,
+	AutovacuumLock,
+	AutovacuumScheduleLock,
+	SyncScanLock,
+	RelationMappingLock,
+	AsyncCtlLock,
+	AsyncQueueLock,
+	SerializableXactHashLock,
+	SerializableFinishedListLock,
+	SerializablePredicateLockListLock,
+	OldSerXidLock,
+	SyncRepLock,
+	/* Individual lock IDs end here */
+	FirstBufMappingLock,
+	FirstLockMgrLock = FirstBufMappingLock + NUM_BUFFER_PARTITIONS,
+	FirstPredicateLockMgrLock = FirstLockMgrLock + NUM_LOCK_PARTITIONS,
+
+	/* must be last except for MaxDynamicFlexLock: */
+	NumFixedFlexLocks = FirstPredicateLockMgrLock + NUM_PREDICATELOCK_PARTITIONS,
+
+	MaxDynamicFlexLock = 1000000000
+} FlexLockId;
+
+/* Shared memory setup. */
+extern int	NumFlexLocks(void);
+extern Size FlexLockShmemSize(void);
+extern void RequestAddinFlexLocks(int n);
+extern void CreateFlexLocks(void);
+
+/* Error recovery and debugging support functions. */
+extern void FlexLockReleaseAll(void);
+extern bool FlexLockHeldByMe(FlexLockId id);
+
+#endif   /* FLEXLOCK_H */
diff --git a/src/include/storage/flexlock_internals.h b/src/include/storage/flexlock_internals.h
new file mode 100644
index 0000000..5f78da7
--- /dev/null
+++ b/src/include/storage/flexlock_internals.h
@@ -0,0 +1,88 @@
+/*-------------------------------------------------------------------------
+ *
+ * flexlock_internals.h
+ *	  Flex lock internals.  Only files which implement a FlexLock
+ *    type should need to include this.  Merging this with flexlock.h
+ *    creates a circular header dependency, but even if it didn't, this
+ *    is cleaner.
+ *
+ * Portions Copyright (c) 1996-2011, PostgreSQL Global Development Group
+ * Portions Copyright (c) 1994, Regents of the University of California
+ *
+ * src/include/storage/flexlock_internals.h
+ *
+ *-------------------------------------------------------------------------
+ */
+#ifndef FLEXLOCK_INTERNALS_H
+#define FLEXLOCK_INTERNALS_H
+
+#include "pg_trace.h"
+#include "storage/flexlock.h"
+#include "storage/proc.h"
+#include "storage/s_lock.h"
+
+/*
+ * Individual FlexLock implementations each get this many bytes to store
+ * its state; of course, a given implementation could also allocate additional
+ * shmem elsewhere, but we provide this many bytes within the array.  The
+ * header fields common to all FlexLock types are included in this number.
+ * A power of two should probably be chosen, to avoid alignment issues and
+ * cache line splitting.  It might be useful to increase this on systems where
+ * a cache line is more than 64 bytes in size.
+ */
+#define FLEX_LOCK_BYTES		64
+
+typedef struct FlexLock
+{
+	char		locktype;		/* see FLEXLOCK_TYPE_* constants */
+	slock_t		mutex;			/* Protects FlexLock state and wait queues */
+	bool		releaseOK;		/* T if ok to release waiters */
+	PGPROC	   *head;			/* head of list of waiting PGPROCs */
+	PGPROC	   *tail;			/* tail of list of waiting PGPROCs */
+	/* tail is undefined when head is NULL */
+} FlexLock;
+
+#define FLEXLOCK_TYPE_LWLOCK			'l'
+
+typedef union FlexLockPadded
+{
+	FlexLock	flex;
+	char		pad[FLEX_LOCK_BYTES];
+} FlexLockPadded;
+
+extern FlexLockPadded *FlexLockArray;
+
+extern FlexLockId FlexLockAssign(char locktype);
+extern void FlexLockRemember(FlexLockId id);
+extern void FlexLockForget(FlexLockId id);
+extern int FlexLockWait(FlexLockId id, int mode);
+
+/*
+ * We must join the wait queue while holding the spinlock, so we define this
+ * as a macro, for speed.
+ */
+#define FlexLockJoinWaitQueue(lock, mode) \
+	do { \
+		Assert(MyProc != NULL); \
+		MyProc->flWaitResult = 0; \
+		MyProc->flWaitMode = mode; \
+		MyProc->flWaitLink = NULL; \
+		if (lock->flex.head == NULL) \
+			lock->flex.head = MyProc; \
+		else \
+			lock->flex.tail->flWaitLink = MyProc; \
+		lock->flex.tail = MyProc; \
+	} while (0)
+
+#ifdef LOCK_DEBUG
+extern bool	Trace_flexlocks;
+#define FlexLockDebug(where, id, msg) \
+	do { \
+		if (Trace_flexlocks) \
+			elog(LOG, "%s(%d): %s", where, (int) id, msg); \
+	} while (0)
+#else
+#define FlexLockDebug(where, id, msg)
+#endif
+
+#endif   /* FLEXLOCK_H */
diff --git a/src/include/storage/lock.h b/src/include/storage/lock.h
index e106ad5..ba87db2 100644
--- a/src/include/storage/lock.h
+++ b/src/include/storage/lock.h
@@ -471,7 +471,7 @@ typedef enum
 #define LockHashPartition(hashcode) \
 	((hashcode) % NUM_LOCK_PARTITIONS)
 #define LockHashPartitionLock(hashcode) \
-	((LWLockId) (FirstLockMgrLock + LockHashPartition(hashcode)))
+	((FlexLockId) (FirstLockMgrLock + LockHashPartition(hashcode)))
 
 
 /*
diff --git a/src/include/storage/lwlock.h b/src/include/storage/lwlock.h
index 438a48d..f68cddc 100644
--- a/src/include/storage/lwlock.h
+++ b/src/include/storage/lwlock.h
@@ -14,82 +14,7 @@
 #ifndef LWLOCK_H
 #define LWLOCK_H
 
-/*
- * It's a bit odd to declare NUM_BUFFER_PARTITIONS and NUM_LOCK_PARTITIONS
- * here, but we need them to set up enum LWLockId correctly, and having
- * this file include lock.h or bufmgr.h would be backwards.
- */
-
-/* Number of partitions of the shared buffer mapping hashtable */
-#define NUM_BUFFER_PARTITIONS  16
-
-/* Number of partitions the shared lock tables are divided into */
-#define LOG2_NUM_LOCK_PARTITIONS  4
-#define NUM_LOCK_PARTITIONS  (1 << LOG2_NUM_LOCK_PARTITIONS)
-
-/* Number of partitions the shared predicate lock tables are divided into */
-#define LOG2_NUM_PREDICATELOCK_PARTITIONS  4
-#define NUM_PREDICATELOCK_PARTITIONS  (1 << LOG2_NUM_PREDICATELOCK_PARTITIONS)
-
-/*
- * We have a number of predefined LWLocks, plus a bunch of LWLocks that are
- * dynamically assigned (e.g., for shared buffers).  The LWLock structures
- * live in shared memory (since they contain shared data) and are identified
- * by values of this enumerated type.  We abuse the notion of an enum somewhat
- * by allowing values not listed in the enum declaration to be assigned.
- * The extra value MaxDynamicLWLock is there to keep the compiler from
- * deciding that the enum can be represented as char or short ...
- *
- * If you remove a lock, please replace it with a placeholder. This retains
- * the lock numbering, which is helpful for DTrace and other external
- * debugging scripts.
- */
-typedef enum LWLockId
-{
-	BufFreelistLock,
-	ShmemIndexLock,
-	OidGenLock,
-	XidGenLock,
-	ProcArrayLock,
-	SInvalReadLock,
-	SInvalWriteLock,
-	WALInsertLock,
-	WALWriteLock,
-	ControlFileLock,
-	CheckpointLock,
-	CLogControlLock,
-	SubtransControlLock,
-	MultiXactGenLock,
-	MultiXactOffsetControlLock,
-	MultiXactMemberControlLock,
-	RelCacheInitLock,
-	BgWriterCommLock,
-	TwoPhaseStateLock,
-	TablespaceCreateLock,
-	BtreeVacuumLock,
-	AddinShmemInitLock,
-	AutovacuumLock,
-	AutovacuumScheduleLock,
-	SyncScanLock,
-	RelationMappingLock,
-	AsyncCtlLock,
-	AsyncQueueLock,
-	SerializableXactHashLock,
-	SerializableFinishedListLock,
-	SerializablePredicateLockListLock,
-	OldSerXidLock,
-	SyncRepLock,
-	/* Individual lock IDs end here */
-	FirstBufMappingLock,
-	FirstLockMgrLock = FirstBufMappingLock + NUM_BUFFER_PARTITIONS,
-	FirstPredicateLockMgrLock = FirstLockMgrLock + NUM_LOCK_PARTITIONS,
-
-	/* must be last except for MaxDynamicLWLock: */
-	NumFixedLWLocks = FirstPredicateLockMgrLock + NUM_PREDICATELOCK_PARTITIONS,
-
-	MaxDynamicLWLock = 1000000000
-} LWLockId;
-
+#include "storage/flexlock.h"
 
 typedef enum LWLockMode
 {
@@ -97,22 +22,10 @@ typedef enum LWLockMode
 	LW_SHARED
 } LWLockMode;
 
-
-#ifdef LOCK_DEBUG
-extern bool Trace_lwlocks;
-#endif
-
-extern LWLockId LWLockAssign(void);
-extern void LWLockAcquire(LWLockId lockid, LWLockMode mode);
-extern bool LWLockConditionalAcquire(LWLockId lockid, LWLockMode mode);
-extern void LWLockRelease(LWLockId lockid);
-extern void LWLockReleaseAll(void);
-extern bool LWLockHeldByMe(LWLockId lockid);
-
-extern int	NumLWLocks(void);
-extern Size LWLockShmemSize(void);
-extern void CreateLWLocks(void);
-
-extern void RequestAddinLWLocks(int n);
+extern FlexLockId LWLockAssign(void);
+extern void LWLockAcquire(FlexLockId lockid, LWLockMode mode);
+extern bool LWLockConditionalAcquire(FlexLockId lockid, LWLockMode mode);
+extern void LWLockRelease(FlexLockId lockid);
+extern bool LWLockHeldByMe(FlexLockId lockid);
 
 #endif   /* LWLOCK_H */
diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h
index 6e798b1..7e8630d 100644
--- a/src/include/storage/proc.h
+++ b/src/include/storage/proc.h
@@ -114,10 +114,10 @@ struct PGPROC
 	 */
 	bool		recoveryConflictPending;
 
-	/* Info about LWLock the process is currently waiting for, if any. */
-	bool		lwWaiting;		/* true if waiting for an LW lock */
-	bool		lwExclusive;	/* true if waiting for exclusive access */
-	struct PGPROC *lwWaitLink;	/* next waiter for same LW lock */
+	/* Info about FlexLock the process is currently waiting for, if any. */
+	int			flWaitResult;	/* result of wait, or 0 if still waiting */
+	int			flWaitMode;		/* lock mode sought */
+	struct PGPROC *flWaitLink;	/* next waiter for same FlexLock */
 
 	/* Info about lock the process is currently waiting for, if any. */
 	/* waitLock and waitProcLock are NULL if not currently waiting. */
@@ -147,7 +147,7 @@ struct PGPROC
 	struct XidCache subxids;	/* cache for subtransaction XIDs */
 
 	/* Per-backend LWLock.  Protects fields below. */
-	LWLockId	backendLock;	/* protects the fields below */
+	FlexLockId	backendLock;	/* protects the fields below */
 
 	/* Lock manager data, recording fast-path locks taken by this backend. */
 	uint64		fpLockBits;		/* lock modes held for each fast-path slot */

diff --git a/src/backend/commands/analyze.c b/src/backend/commands/analyze.c
index 32985a4..d6bba6f 100644
--- a/src/backend/commands/analyze.c
+++ b/src/backend/commands/analyze.c
@@ -40,6 +40,7 @@
 #include "storage/lmgr.h"
 #include "storage/proc.h"
 #include "storage/procarray.h"
+#include "storage/procarraylock.h"
 #include "utils/acl.h"
 #include "utils/attoptcache.h"
 #include "utils/datum.h"
@@ -222,9 +223,9 @@ analyze_rel(Oid relid, VacuumStmt *vacstmt, BufferAccessStrategy bstrategy)
 	/*
 	 * OK, let's do it.  First let other backends know I'm in ANALYZE.
 	 */
-	LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
+	ProcArrayLockAcquire(PAL_EXCLUSIVE);
 	MyProc->vacuumFlags |= PROC_IN_ANALYZE;
-	LWLockRelease(ProcArrayLock);
+	ProcArrayLockRelease();
 
 	/*
 	 * Do the normal non-recursive ANALYZE.
@@ -249,9 +250,9 @@ analyze_rel(Oid relid, VacuumStmt *vacstmt, BufferAccessStrategy bstrategy)
 	 * Reset my PGPROC flag.  Note: we need this here, and not in vacuum_rel,
 	 * because the vacuum flag is cleared by the end-of-xact code.
 	 */
-	LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
+	ProcArrayLockAcquire(PAL_EXCLUSIVE);
 	MyProc->vacuumFlags &= ~PROC_IN_ANALYZE;
-	LWLockRelease(ProcArrayLock);
+	ProcArrayLockRelease();
 }
 
 /*
diff --git a/src/backend/commands/vacuum.c b/src/backend/commands/vacuum.c
index f42504c..823dab9 100644
--- a/src/backend/commands/vacuum.c
+++ b/src/backend/commands/vacuum.c
@@ -39,6 +39,7 @@
 #include "storage/lmgr.h"
 #include "storage/proc.h"
 #include "storage/procarray.h"
+#include "storage/procarraylock.h"
 #include "utils/acl.h"
 #include "utils/fmgroids.h"
 #include "utils/guc.h"
@@ -892,11 +893,11 @@ vacuum_rel(Oid relid, VacuumStmt *vacstmt, bool do_toast, bool for_wraparound)
 		 * MyProc->xid/xmin, else OldestXmin might appear to go backwards,
 		 * which is probably Not Good.
 		 */
-		LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
+		ProcArrayLockAcquire(PAL_EXCLUSIVE);
 		MyProc->vacuumFlags |= PROC_IN_VACUUM;
 		if (for_wraparound)
 			MyProc->vacuumFlags |= PROC_VACUUM_FOR_WRAPAROUND;
-		LWLockRelease(ProcArrayLock);
+		ProcArrayLockRelease();
 	}
 
 	/*
diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c
index 1a48485..39c5080 100644
--- a/src/backend/storage/ipc/procarray.c
+++ b/src/backend/storage/ipc/procarray.c
@@ -52,6 +52,7 @@
 #include "access/twophase.h"
 #include "miscadmin.h"
 #include "storage/procarray.h"
+#include "storage/procarraylock.h"
 #include "storage/spin.h"
 #include "utils/builtins.h"
 #include "utils/snapmgr.h"
@@ -254,7 +255,7 @@ ProcArrayAdd(PGPROC *proc)
 {
 	ProcArrayStruct *arrayP = procArray;
 
-	LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
+	ProcArrayLockAcquire(PAL_EXCLUSIVE);
 
 	if (arrayP->numProcs >= arrayP->maxProcs)
 	{
@@ -263,7 +264,7 @@ ProcArrayAdd(PGPROC *proc)
 		 * fixed supply of PGPROC structs too, and so we should have failed
 		 * earlier.)
 		 */
-		LWLockRelease(ProcArrayLock);
+		ProcArrayLockRelease();
 		ereport(FATAL,
 				(errcode(ERRCODE_TOO_MANY_CONNECTIONS),
 				 errmsg("sorry, too many clients already")));
@@ -272,7 +273,7 @@ ProcArrayAdd(PGPROC *proc)
 	arrayP->procs[arrayP->numProcs] = proc;
 	arrayP->numProcs++;
 
-	LWLockRelease(ProcArrayLock);
+	ProcArrayLockRelease();
 }
 
 /*
@@ -297,7 +298,7 @@ ProcArrayRemove(PGPROC *proc, TransactionId latestXid)
 		DisplayXidCache();
 #endif
 
-	LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
+	ProcArrayLockAcquire(PAL_EXCLUSIVE);
 
 	if (TransactionIdIsValid(latestXid))
 	{
@@ -321,13 +322,13 @@ ProcArrayRemove(PGPROC *proc, TransactionId latestXid)
 			arrayP->procs[index] = arrayP->procs[arrayP->numProcs - 1];
 			arrayP->procs[arrayP->numProcs - 1] = NULL; /* for debugging */
 			arrayP->numProcs--;
-			LWLockRelease(ProcArrayLock);
+			ProcArrayLockRelease();
 			return;
 		}
 	}
 
 	/* Ooops */
-	LWLockRelease(ProcArrayLock);
+	ProcArrayLockRelease();
 
 	elog(LOG, "failed to find proc %p in ProcArray", proc);
 }
@@ -351,54 +352,15 @@ ProcArrayEndTransaction(PGPROC *proc, TransactionId latestXid)
 {
 	if (TransactionIdIsValid(latestXid))
 	{
-		/*
-		 * We must lock ProcArrayLock while clearing proc->xid, so that we do
-		 * not exit the set of "running" transactions while someone else is
-		 * taking a snapshot.  See discussion in
-		 * src/backend/access/transam/README.
-		 */
-		Assert(TransactionIdIsValid(proc->xid));
-
-		LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
-
-		proc->xid = InvalidTransactionId;
-		proc->lxid = InvalidLocalTransactionId;
-		proc->xmin = InvalidTransactionId;
-		/* must be cleared with xid/xmin: */
-		proc->vacuumFlags &= ~PROC_VACUUM_STATE_MASK;
-		proc->inCommit = false; /* be sure this is cleared in abort */
-		proc->recoveryConflictPending = false;
-
-		/* Clear the subtransaction-XID cache too while holding the lock */
-		proc->subxids.nxids = 0;
-		proc->subxids.overflowed = false;
-
-		/* Also advance global latestCompletedXid while holding the lock */
-		if (TransactionIdPrecedes(ShmemVariableCache->latestCompletedXid,
-								  latestXid))
-			ShmemVariableCache->latestCompletedXid = latestXid;
-
-		LWLockRelease(ProcArrayLock);
+		Assert(proc == MyProc);
+		ProcArrayLockClearTransaction(latestXid);		
 	}
 	else
-	{
-		/*
-		 * If we have no XID, we don't need to lock, since we won't affect
-		 * anyone else's calculation of a snapshot.  We might change their
-		 * estimate of global xmin, but that's OK.
-		 */
-		Assert(!TransactionIdIsValid(proc->xid));
-
-		proc->lxid = InvalidLocalTransactionId;
 		proc->xmin = InvalidTransactionId;
-		/* must be cleared with xid/xmin: */
-		proc->vacuumFlags &= ~PROC_VACUUM_STATE_MASK;
-		proc->inCommit = false; /* be sure this is cleared in abort */
-		proc->recoveryConflictPending = false;
 
-		Assert(proc->subxids.nxids == 0);
-		Assert(proc->subxids.overflowed == false);
-	}
+	proc->lxid = InvalidLocalTransactionId;
+	proc->inCommit = false; /* be sure this is cleared in abort */
+	proc->recoveryConflictPending = false;
 }
 
 
@@ -528,7 +490,7 @@ ProcArrayApplyRecoveryInfo(RunningTransactions running)
 	/*
 	 * Nobody else is running yet, but take locks anyhow
 	 */
-	LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
+	ProcArrayLockAcquire(PAL_EXCLUSIVE);
 
 	/*
 	 * KnownAssignedXids is sorted so we cannot just add the xids, we have to
@@ -635,7 +597,7 @@ ProcArrayApplyRecoveryInfo(RunningTransactions running)
 	Assert(TransactionIdIsNormal(ShmemVariableCache->latestCompletedXid));
 	Assert(TransactionIdIsValid(ShmemVariableCache->nextXid));
 
-	LWLockRelease(ProcArrayLock);
+	ProcArrayLockRelease();
 
 	KnownAssignedXidsDisplay(trace_recovery(DEBUG3));
 	if (standbyState == STANDBY_SNAPSHOT_READY)
@@ -690,7 +652,7 @@ ProcArrayApplyXidAssignment(TransactionId topxid,
 	/*
 	 * Uses same locking as transaction commit
 	 */
-	LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
+	ProcArrayLockAcquire(PAL_EXCLUSIVE);
 
 	/*
 	 * Remove subxids from known-assigned-xacts.
@@ -703,7 +665,7 @@ ProcArrayApplyXidAssignment(TransactionId topxid,
 	if (TransactionIdPrecedes(procArray->lastOverflowedXid, max_xid))
 		procArray->lastOverflowedXid = max_xid;
 
-	LWLockRelease(ProcArrayLock);
+	ProcArrayLockRelease();
 }
 
 /*
@@ -795,7 +757,7 @@ TransactionIdIsInProgress(TransactionId xid)
 					 errmsg("out of memory")));
 	}
 
-	LWLockAcquire(ProcArrayLock, LW_SHARED);
+	ProcArrayLockAcquire(PAL_SHARED);
 
 	/*
 	 * Now that we have the lock, we can check latestCompletedXid; if the
@@ -803,7 +765,7 @@ TransactionIdIsInProgress(TransactionId xid)
 	 */
 	if (TransactionIdPrecedes(ShmemVariableCache->latestCompletedXid, xid))
 	{
-		LWLockRelease(ProcArrayLock);
+		ProcArrayLockRelease();
 		xc_by_latest_xid_inc();
 		return true;
 	}
@@ -829,7 +791,7 @@ TransactionIdIsInProgress(TransactionId xid)
 		 */
 		if (TransactionIdEquals(pxid, xid))
 		{
-			LWLockRelease(ProcArrayLock);
+			ProcArrayLockRelease();
 			xc_by_main_xid_inc();
 			return true;
 		}
@@ -851,7 +813,7 @@ TransactionIdIsInProgress(TransactionId xid)
 
 			if (TransactionIdEquals(cxid, xid))
 			{
-				LWLockRelease(ProcArrayLock);
+				ProcArrayLockRelease();
 				xc_by_child_xid_inc();
 				return true;
 			}
@@ -879,7 +841,7 @@ TransactionIdIsInProgress(TransactionId xid)
 
 		if (KnownAssignedXidExists(xid))
 		{
-			LWLockRelease(ProcArrayLock);
+			ProcArrayLockRelease();
 			xc_by_known_assigned_inc();
 			return true;
 		}
@@ -895,7 +857,7 @@ TransactionIdIsInProgress(TransactionId xid)
 			nxids = KnownAssignedXidsGet(xids, xid);
 	}
 
-	LWLockRelease(ProcArrayLock);
+	ProcArrayLockRelease();
 
 	/*
 	 * If none of the relevant caches overflowed, we know the Xid is not
@@ -961,7 +923,7 @@ TransactionIdIsActive(TransactionId xid)
 	if (TransactionIdPrecedes(xid, RecentXmin))
 		return false;
 
-	LWLockAcquire(ProcArrayLock, LW_SHARED);
+	ProcArrayLockAcquire(PAL_SHARED);
 
 	for (i = 0; i < arrayP->numProcs; i++)
 	{
@@ -983,7 +945,7 @@ TransactionIdIsActive(TransactionId xid)
 		}
 	}
 
-	LWLockRelease(ProcArrayLock);
+	ProcArrayLockRelease();
 
 	return result;
 }
@@ -1046,7 +1008,7 @@ GetOldestXmin(bool allDbs, bool ignoreVacuum)
 	/* Cannot look for individual databases during recovery */
 	Assert(allDbs || !RecoveryInProgress());
 
-	LWLockAcquire(ProcArrayLock, LW_SHARED);
+	ProcArrayLockAcquire(PAL_SHARED);
 
 	/*
 	 * We initialize the MIN() calculation with latestCompletedXid + 1. This
@@ -1099,7 +1061,7 @@ GetOldestXmin(bool allDbs, bool ignoreVacuum)
 		 */
 		TransactionId kaxmin = KnownAssignedXidsGetOldestXmin();
 
-		LWLockRelease(ProcArrayLock);
+		ProcArrayLockRelease();
 
 		if (TransactionIdIsNormal(kaxmin) &&
 			TransactionIdPrecedes(kaxmin, result))
@@ -1110,7 +1072,7 @@ GetOldestXmin(bool allDbs, bool ignoreVacuum)
 		/*
 		 * No other information needed, so release the lock immediately.
 		 */
-		LWLockRelease(ProcArrayLock);
+		ProcArrayLockRelease();
 
 		/*
 		 * Compute the cutoff XID by subtracting vacuum_defer_cleanup_age,
@@ -1239,7 +1201,7 @@ GetSnapshotData(Snapshot snapshot)
 	 * It is sufficient to get shared lock on ProcArrayLock, even if we are
 	 * going to set MyProc->xmin.
 	 */
-	LWLockAcquire(ProcArrayLock, LW_SHARED);
+	ProcArrayLockAcquire(PAL_SHARED);
 
 	/* xmax is always latestCompletedXid + 1 */
 	xmax = ShmemVariableCache->latestCompletedXid;
@@ -1375,7 +1337,7 @@ GetSnapshotData(Snapshot snapshot)
 	if (!TransactionIdIsValid(MyProc->xmin))
 		MyProc->xmin = TransactionXmin = xmin;
 
-	LWLockRelease(ProcArrayLock);
+	ProcArrayLockRelease();
 
 	/*
 	 * Update globalxmin to include actual process xids.  This is a slightly
@@ -1432,7 +1394,7 @@ ProcArrayInstallImportedXmin(TransactionId xmin, TransactionId sourcexid)
 		return false;
 
 	/* Get lock so source xact can't end while we're doing this */
-	LWLockAcquire(ProcArrayLock, LW_SHARED);
+	ProcArrayLockAcquire(PAL_SHARED);
 
 	for (index = 0; index < arrayP->numProcs; index++)
 	{
@@ -1476,7 +1438,7 @@ ProcArrayInstallImportedXmin(TransactionId xmin, TransactionId sourcexid)
 		break;
 	}
 
-	LWLockRelease(ProcArrayLock);
+	ProcArrayLockRelease();
 
 	return result;
 }
@@ -1550,7 +1512,7 @@ GetRunningTransactionData(void)
 	 * Ensure that no xids enter or leave the procarray while we obtain
 	 * snapshot.
 	 */
-	LWLockAcquire(ProcArrayLock, LW_SHARED);
+	ProcArrayLockAcquire(PAL_SHARED);
 	LWLockAcquire(XidGenLock, LW_SHARED);
 
 	latestCompletedXid = ShmemVariableCache->latestCompletedXid;
@@ -1611,7 +1573,7 @@ GetRunningTransactionData(void)
 	CurrentRunningXacts->latestCompletedXid = latestCompletedXid;
 
 	/* We don't release XidGenLock here, the caller is responsible for that */
-	LWLockRelease(ProcArrayLock);
+	ProcArrayLockRelease();
 
 	Assert(TransactionIdIsValid(CurrentRunningXacts->nextXid));
 	Assert(TransactionIdIsValid(CurrentRunningXacts->oldestRunningXid));
@@ -1644,7 +1606,7 @@ GetOldestActiveTransactionId(void)
 
 	Assert(!RecoveryInProgress());
 
-	LWLockAcquire(ProcArrayLock, LW_SHARED);
+	ProcArrayLockAcquire(PAL_SHARED);
 
 	oldestRunningXid = ShmemVariableCache->nextXid;
 
@@ -1672,7 +1634,7 @@ GetOldestActiveTransactionId(void)
 		 */
 	}
 
-	LWLockRelease(ProcArrayLock);
+	ProcArrayLockRelease();
 
 	return oldestRunningXid;
 }
@@ -1705,7 +1667,7 @@ GetTransactionsInCommit(TransactionId **xids_p)
 	xids = (TransactionId *) palloc(arrayP->maxProcs * sizeof(TransactionId));
 	nxids = 0;
 
-	LWLockAcquire(ProcArrayLock, LW_SHARED);
+	ProcArrayLockAcquire(PAL_SHARED);
 
 	for (index = 0; index < arrayP->numProcs; index++)
 	{
@@ -1718,7 +1680,7 @@ GetTransactionsInCommit(TransactionId **xids_p)
 			xids[nxids++] = pxid;
 	}
 
-	LWLockRelease(ProcArrayLock);
+	ProcArrayLockRelease();
 
 	*xids_p = xids;
 	return nxids;
@@ -1740,7 +1702,7 @@ HaveTransactionsInCommit(TransactionId *xids, int nxids)
 	ProcArrayStruct *arrayP = procArray;
 	int			index;
 
-	LWLockAcquire(ProcArrayLock, LW_SHARED);
+	ProcArrayLockAcquire(PAL_SHARED);
 
 	for (index = 0; index < arrayP->numProcs; index++)
 	{
@@ -1766,7 +1728,7 @@ HaveTransactionsInCommit(TransactionId *xids, int nxids)
 		}
 	}
 
-	LWLockRelease(ProcArrayLock);
+	ProcArrayLockRelease();
 
 	return result;
 }
@@ -1788,7 +1750,7 @@ BackendPidGetProc(int pid)
 	if (pid == 0)				/* never match dummy PGPROCs */
 		return NULL;
 
-	LWLockAcquire(ProcArrayLock, LW_SHARED);
+	ProcArrayLockAcquire(PAL_SHARED);
 
 	for (index = 0; index < arrayP->numProcs; index++)
 	{
@@ -1801,7 +1763,7 @@ BackendPidGetProc(int pid)
 		}
 	}
 
-	LWLockRelease(ProcArrayLock);
+	ProcArrayLockRelease();
 
 	return result;
 }
@@ -1829,7 +1791,7 @@ BackendXidGetPid(TransactionId xid)
 	if (xid == InvalidTransactionId)	/* never match invalid xid */
 		return 0;
 
-	LWLockAcquire(ProcArrayLock, LW_SHARED);
+	ProcArrayLockAcquire(PAL_SHARED);
 
 	for (index = 0; index < arrayP->numProcs; index++)
 	{
@@ -1842,7 +1804,7 @@ BackendXidGetPid(TransactionId xid)
 		}
 	}
 
-	LWLockRelease(ProcArrayLock);
+	ProcArrayLockRelease();
 
 	return result;
 }
@@ -1897,7 +1859,7 @@ GetCurrentVirtualXIDs(TransactionId limitXmin, bool excludeXmin0,
 	vxids = (VirtualTransactionId *)
 		palloc(sizeof(VirtualTransactionId) * arrayP->maxProcs);
 
-	LWLockAcquire(ProcArrayLock, LW_SHARED);
+	ProcArrayLockAcquire(PAL_SHARED);
 
 	for (index = 0; index < arrayP->numProcs; index++)
 	{
@@ -1933,7 +1895,7 @@ GetCurrentVirtualXIDs(TransactionId limitXmin, bool excludeXmin0,
 		}
 	}
 
-	LWLockRelease(ProcArrayLock);
+	ProcArrayLockRelease();
 
 	*nvxids = count;
 	return vxids;
@@ -1992,7 +1954,7 @@ GetConflictingVirtualXIDs(TransactionId limitXmin, Oid dbOid)
 					 errmsg("out of memory")));
 	}
 
-	LWLockAcquire(ProcArrayLock, LW_SHARED);
+	ProcArrayLockAcquire(PAL_SHARED);
 
 	for (index = 0; index < arrayP->numProcs; index++)
 	{
@@ -2025,7 +1987,7 @@ GetConflictingVirtualXIDs(TransactionId limitXmin, Oid dbOid)
 		}
 	}
 
-	LWLockRelease(ProcArrayLock);
+	ProcArrayLockRelease();
 
 	/* add the terminator */
 	vxids[count].backendId = InvalidBackendId;
@@ -2046,7 +2008,7 @@ CancelVirtualTransaction(VirtualTransactionId vxid, ProcSignalReason sigmode)
 	int			index;
 	pid_t		pid = 0;
 
-	LWLockAcquire(ProcArrayLock, LW_SHARED);
+	ProcArrayLockAcquire(PAL_SHARED);
 
 	for (index = 0; index < arrayP->numProcs; index++)
 	{
@@ -2072,7 +2034,7 @@ CancelVirtualTransaction(VirtualTransactionId vxid, ProcSignalReason sigmode)
 		}
 	}
 
-	LWLockRelease(ProcArrayLock);
+	ProcArrayLockRelease();
 
 	return pid;
 }
@@ -2146,7 +2108,7 @@ CountDBBackends(Oid databaseid)
 	int			count = 0;
 	int			index;
 
-	LWLockAcquire(ProcArrayLock, LW_SHARED);
+	ProcArrayLockAcquire(PAL_SHARED);
 
 	for (index = 0; index < arrayP->numProcs; index++)
 	{
@@ -2159,7 +2121,7 @@ CountDBBackends(Oid databaseid)
 			count++;
 	}
 
-	LWLockRelease(ProcArrayLock);
+	ProcArrayLockRelease();
 
 	return count;
 }
@@ -2175,7 +2137,7 @@ CancelDBBackends(Oid databaseid, ProcSignalReason sigmode, bool conflictPending)
 	pid_t		pid = 0;
 
 	/* tell all backends to die */
-	LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
+	ProcArrayLockAcquire(PAL_EXCLUSIVE);
 
 	for (index = 0; index < arrayP->numProcs; index++)
 	{
@@ -2200,7 +2162,7 @@ CancelDBBackends(Oid databaseid, ProcSignalReason sigmode, bool conflictPending)
 		}
 	}
 
-	LWLockRelease(ProcArrayLock);
+	ProcArrayLockRelease();
 }
 
 /*
@@ -2213,7 +2175,7 @@ CountUserBackends(Oid roleid)
 	int			count = 0;
 	int			index;
 
-	LWLockAcquire(ProcArrayLock, LW_SHARED);
+	ProcArrayLockAcquire(PAL_SHARED);
 
 	for (index = 0; index < arrayP->numProcs; index++)
 	{
@@ -2225,7 +2187,7 @@ CountUserBackends(Oid roleid)
 			count++;
 	}
 
-	LWLockRelease(ProcArrayLock);
+	ProcArrayLockRelease();
 
 	return count;
 }
@@ -2273,7 +2235,7 @@ CountOtherDBBackends(Oid databaseId, int *nbackends, int *nprepared)
 
 		*nbackends = *nprepared = 0;
 
-		LWLockAcquire(ProcArrayLock, LW_SHARED);
+		ProcArrayLockAcquire(PAL_SHARED);
 
 		for (index = 0; index < arrayP->numProcs; index++)
 		{
@@ -2297,7 +2259,7 @@ CountOtherDBBackends(Oid databaseId, int *nbackends, int *nprepared)
 			}
 		}
 
-		LWLockRelease(ProcArrayLock);
+		ProcArrayLockRelease();
 
 		if (!found)
 			return false;		/* no conflicting backends, so done */
@@ -2350,7 +2312,7 @@ XidCacheRemoveRunningXids(TransactionId xid,
 	 * to abort subtransactions, but pending closer analysis we'd best be
 	 * conservative.
 	 */
-	LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
+	ProcArrayLockAcquire(PAL_EXCLUSIVE);
 
 	/*
 	 * Under normal circumstances xid and xids[] will be in increasing order,
@@ -2398,7 +2360,7 @@ XidCacheRemoveRunningXids(TransactionId xid,
 							  latestXid))
 		ShmemVariableCache->latestCompletedXid = latestXid;
 
-	LWLockRelease(ProcArrayLock);
+	ProcArrayLockRelease();
 }
 
 #ifdef XIDCACHE_DEBUG
@@ -2565,7 +2527,7 @@ ExpireTreeKnownAssignedTransactionIds(TransactionId xid, int nsubxids,
 	/*
 	 * Uses same locking as transaction commit
 	 */
-	LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
+	ProcArrayLockAcquire(PAL_EXCLUSIVE);
 
 	KnownAssignedXidsRemoveTree(xid, nsubxids, subxids);
 
@@ -2574,7 +2536,7 @@ ExpireTreeKnownAssignedTransactionIds(TransactionId xid, int nsubxids,
 							  max_xid))
 		ShmemVariableCache->latestCompletedXid = max_xid;
 
-	LWLockRelease(ProcArrayLock);
+	ProcArrayLockRelease();
 }
 
 /*
@@ -2584,9 +2546,9 @@ ExpireTreeKnownAssignedTransactionIds(TransactionId xid, int nsubxids,
 void
 ExpireAllKnownAssignedTransactionIds(void)
 {
-	LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
+	ProcArrayLockAcquire(PAL_EXCLUSIVE);
 	KnownAssignedXidsRemovePreceding(InvalidTransactionId);
-	LWLockRelease(ProcArrayLock);
+	ProcArrayLockRelease();
 }
 
 /*
@@ -2596,9 +2558,9 @@ ExpireAllKnownAssignedTransactionIds(void)
 void
 ExpireOldKnownAssignedTransactionIds(TransactionId xid)
 {
-	LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
+	ProcArrayLockAcquire(PAL_EXCLUSIVE);
 	KnownAssignedXidsRemovePreceding(xid);
-	LWLockRelease(ProcArrayLock);
+	ProcArrayLockRelease();
 }
 
 
@@ -2820,7 +2782,7 @@ KnownAssignedXidsAdd(TransactionId from_xid, TransactionId to_xid,
 	{
 		/* must hold lock to compress */
 		if (!exclusive_lock)
-			LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
+			ProcArrayLockAcquire(PAL_EXCLUSIVE);
 
 		KnownAssignedXidsCompress(true);
 
@@ -2828,7 +2790,7 @@ KnownAssignedXidsAdd(TransactionId from_xid, TransactionId to_xid,
 		/* note: we no longer care about the tail pointer */
 
 		if (!exclusive_lock)
-			LWLockRelease(ProcArrayLock);
+			ProcArrayLockRelease();
 
 		/*
 		 * If it still won't fit then we're out of memory
diff --git a/src/backend/storage/lmgr/Makefile b/src/backend/storage/lmgr/Makefile
index 3730e51..27eaa97 100644
--- a/src/backend/storage/lmgr/Makefile
+++ b/src/backend/storage/lmgr/Makefile
@@ -13,7 +13,7 @@ top_builddir = ../../../..
 include $(top_builddir)/src/Makefile.global
 
 OBJS = flexlock.o lmgr.o lock.o proc.o deadlock.o lwlock.o spin.o s_lock.o \
-	predicate.o
+	procarraylock.o predicate.o
 
 include $(top_srcdir)/src/backend/common.mk
 
diff --git a/src/backend/storage/lmgr/flexlock.c b/src/backend/storage/lmgr/flexlock.c
index 7f657b3..c88bd24 100644
--- a/src/backend/storage/lmgr/flexlock.c
+++ b/src/backend/storage/lmgr/flexlock.c
@@ -30,6 +30,7 @@
 #include "storage/lwlock.h"
 #include "storage/predicate.h"
 #include "storage/proc.h"
+#include "storage/procarraylock.h"
 #include "storage/spin.h"
 #include "utils/elog.h"
 
@@ -177,9 +178,14 @@ CreateFlexLocks(void)
 
 	FlexLockArray = (FlexLockPadded *) ptr;
 
-	/* All of the "fixed" FlexLocks are LWLocks. */
+	/* All of the "fixed" FlexLocks are LWLocks - except ProcArrayLock. */
 	for (id = 0, lock = FlexLockArray; id < NumFixedFlexLocks; id++, lock++)
-		FlexLockInit(&lock->flex, FLEXLOCK_TYPE_LWLOCK);
+	{
+		if (id == ProcArrayLock)
+			FlexLockInit(&lock->flex, FLEXLOCK_TYPE_PROCARRAYLOCK);
+		else
+			FlexLockInit(&lock->flex, FLEXLOCK_TYPE_LWLOCK);
+	}
 
 	/*
 	 * Initialize the dynamic-allocation counter, which is stored just before
@@ -324,13 +330,20 @@ FlexLockReleaseAll(void)
 {
 	while (num_held_flexlocks > 0)
 	{
+		FlexLockId	id;
+		FlexLock   *flex;
+
 		HOLD_INTERRUPTS();		/* match the upcoming RESUME_INTERRUPTS */
 
-		/*
-		 * FLEXTODO: When we have multiple types of flex locks, this will
-		 * need to call the appropriate release function for each lock type.
-		 */
-		LWLockRelease(held_flexlocks[num_held_flexlocks - 1]);
+		id = held_flexlocks[num_held_flexlocks - 1];
+		flex = &FlexLockArray[id].flex;
+		if (flex->locktype == FLEXLOCK_TYPE_LWLOCK)
+			LWLockRelease(id);
+		else
+		{
+			Assert(id == ProcArrayLock);
+			ProcArrayLockRelease();
+		}
 	}
 }
 
diff --git a/src/backend/storage/lmgr/proc.c b/src/backend/storage/lmgr/proc.c
index 57da345..510a4c2 100644
--- a/src/backend/storage/lmgr/proc.c
+++ b/src/backend/storage/lmgr/proc.c
@@ -45,6 +45,7 @@
 #include "storage/pmsignal.h"
 #include "storage/proc.h"
 #include "storage/procarray.h"
+#include "storage/procarraylock.h"
 #include "storage/procsignal.h"
 #include "storage/spin.h"
 #include "utils/timestamp.h"
@@ -1046,7 +1047,7 @@ ProcSleep(LOCALLOCK *locallock, LockMethod lockMethodTable)
 		{
 			PGPROC	   *autovac = GetBlockingAutoVacuumPgproc();
 
-			LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
+			ProcArrayLockAcquire(PAL_EXCLUSIVE);
 
 			/*
 			 * Only do it if the worker is not working to protect against Xid
@@ -1062,7 +1063,7 @@ ProcSleep(LOCALLOCK *locallock, LockMethod lockMethodTable)
 					 pid);
 
 				/* don't hold the lock across the kill() syscall */
-				LWLockRelease(ProcArrayLock);
+				ProcArrayLockRelease();
 
 				/* send the autovacuum worker Back to Old Kent Road */
 				if (kill(pid, SIGINT) < 0)
@@ -1074,7 +1075,7 @@ ProcSleep(LOCALLOCK *locallock, LockMethod lockMethodTable)
 				}
 			}
 			else
-				LWLockRelease(ProcArrayLock);
+				ProcArrayLockRelease();
 
 			/* prevent signal from being resent more than once */
 			allow_autovacuum_cancel = false;
diff --git a/src/backend/storage/lmgr/procarraylock.c b/src/backend/storage/lmgr/procarraylock.c
new file mode 100644
index 0000000..6838ed6
--- /dev/null
+++ b/src/backend/storage/lmgr/procarraylock.c
@@ -0,0 +1,341 @@
+/*-------------------------------------------------------------------------
+ *
+ * procarraylock.c
+ *	  Lock management for the ProcArray
+ *
+ * Because the ProcArray data structure is highly trafficked, it is
+ * critical that mutual exclusion for ProcArray options be as efficient
+ * as possible.  A particular problem is transaction end (commit or abort)
+ * which cannot be done in parallel with snapshot acquisition.  We
+ * therefore include some special hacks to deal with this case efficiently.
+ *
+ * Portions Copyright (c) 1996-2011, PostgreSQL Global Development Group
+ * Portions Copyright (c) 1994, Regents of the University of California
+ *
+ * IDENTIFICATION
+ *	  src/backend/storage/lmgr/procarraylock.c
+ *
+ *-------------------------------------------------------------------------
+ */
+#include "postgres.h"
+
+#include "miscadmin.h"
+#include "pg_trace.h"
+#include "access/transam.h"
+#include "storage/flexlock_internals.h"
+#include "storage/ipc.h"
+#include "storage/procarraylock.h"
+#include "storage/proc.h"
+#include "storage/spin.h"
+
+typedef struct ProcArrayLockStruct
+{
+	FlexLock	flex;			/* common FlexLock infrastructure */
+	char		exclusive;		/* # of exclusive holders (0 or 1) */
+	int			shared;			/* # of shared holders (0..MaxBackends) */
+	PGPROC	   *ending;			/* transactions wishing to clear state */
+	TransactionId	latest_ending_xid;	/* latest ending XID */
+} ProcArrayLockStruct;
+
+/* There is only one ProcArrayLock. */
+#define	ProcArrayLockPointer() \
+	(AssertMacro(FlexLockArray[ProcArrayLock].flex.locktype == \
+		FLEXLOCK_TYPE_PROCARRAYLOCK), \
+	 (volatile ProcArrayLockStruct *) &FlexLockArray[ProcArrayLock])
+
+/*
+ * ProcArrayLockAcquire - 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
+ProcArrayLockAcquire(ProcArrayLockMode mode)
+{
+	volatile ProcArrayLockStruct *lock = ProcArrayLockPointer();
+	PGPROC	   *proc = MyProc;
+	bool		retry = false;
+	int			extraWaits = 0;
+
+	/*
+	 * 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 ProcArrayLock.  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
+	 * ProcArrayLockRelease.  See comments in LWLockAcquire for an explanation
+	 * of why do we not attempt to hand off the lock directly.
+	 */
+	for (;;)
+	{
+		bool		mustwait;
+
+		/* Acquire mutex.  Time spent holding mutex should be short! */
+		SpinLockAcquire(&lock->flex.mutex);
+
+		/* If retrying, allow LWLockRelease to release waiters again */
+		if (retry)
+			lock->flex.releaseOK = true;
+
+		/* If I can get the lock, do so quickly. */
+		if (mode == PAL_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. */
+		FlexLockJoinWaitQueue(lock, (int) mode);
+
+		/* Can release the mutex now */
+		SpinLockRelease(&lock->flex.mutex);
+
+		/* Wait until awakened. */
+		extraWaits += FlexLockWait(ProcArrayLock, mode);
+
+		/* Now loop back and try to acquire lock again. */
+		retry = true;
+	}
+
+	/* We are done updating shared state of the lock itself. */
+	SpinLockRelease(&lock->flex.mutex);
+
+	TRACE_POSTGRESQL_FLEXLOCK_ACQUIRE(lockid, mode);
+
+	/* Add lock to list of locks held by this backend */
+	FlexLockRemember(ProcArrayLock);
+
+	/*
+	 * Fix the process wait semaphore's count for any absorbed wakeups.
+	 */
+	while (extraWaits-- > 0)
+		PGSemaphoreUnlock(&proc->sem);
+}
+
+/*
+ * ProcArrayLockClearTransaction - safely clear transaction details
+ *
+ * This can't be done while ProcArrayLock is held, but it's so fast that
+ * we can afford to do it while holding the spinlock, rather than acquiring
+ * and releasing the lock.
+ */
+void
+ProcArrayLockClearTransaction(TransactionId latestXid)
+{
+	volatile ProcArrayLockStruct *lock = ProcArrayLockPointer();
+	PGPROC	   *proc = MyProc;
+	int			extraWaits = 0;
+	bool		mustwait;
+
+	HOLD_INTERRUPTS();
+
+	/* Acquire mutex.  Time spent holding mutex should be short! */
+	SpinLockAcquire(&lock->flex.mutex);
+
+	if (lock->exclusive == 0 && lock->shared == 0)
+	{
+		{
+			volatile PGPROC *vproc = proc;
+			/* If there are no lockers, clar the critical PGPROC fields. */
+			vproc->xid = InvalidTransactionId;
+	        vproc->xmin = InvalidTransactionId;
+	        /* must be cleared with xid/xmin: */
+	        vproc->vacuumFlags &= ~PROC_VACUUM_STATE_MASK;
+			vproc->subxids.nxids = 0;
+			vproc->subxids.overflowed = false;
+		}
+		mustwait = false;
+
+        /* Also advance global latestCompletedXid while holding the lock */
+        if (TransactionIdPrecedes(ShmemVariableCache->latestCompletedXid,
+                                  latestXid))
+            ShmemVariableCache->latestCompletedXid = latestXid;
+	}
+	else
+	{
+		/* Rats, must wait. */
+		proc->flWaitLink = lock->ending;
+		lock->ending = proc;
+		if (!TransactionIdIsValid(lock->latest_ending_xid) ||
+				TransactionIdPrecedes(lock->latest_ending_xid, latestXid)) 
+			lock->latest_ending_xid = latestXid;
+		mustwait = true;
+	}
+
+	/* Can release the mutex now */
+	SpinLockRelease(&lock->flex.mutex);
+
+	/*
+	 * If we were not able to perfom the operation immediately, we must wait.
+	 * But we need not retry after being awoken, because the last lock holder
+	 * to release the lock will do the work first, on our behalf.
+	 */
+	if (mustwait)
+	{
+		extraWaits += FlexLockWait(ProcArrayLock, 2);
+		while (extraWaits-- > 0)
+			PGSemaphoreUnlock(&proc->sem);
+	}
+
+	RESUME_INTERRUPTS();
+}
+
+/*
+ * ProcArrayLockRelease - release a previously acquired lock
+ */
+void
+ProcArrayLockRelease(void)
+{
+	volatile ProcArrayLockStruct *lock = ProcArrayLockPointer();
+	PGPROC	   *head;
+	PGPROC	   *ending = NULL;
+	PGPROC	   *proc;
+
+	FlexLockForget(ProcArrayLock);
+
+	/* Acquire mutex.  Time spent holding mutex should be short! */
+	SpinLockAcquire(&lock->flex.mutex);
+
+	/* Release my hold on lock */
+	if (lock->exclusive > 0)
+		lock->exclusive--;
+	else
+	{
+		Assert(lock->shared > 0);
+		lock->shared--;
+	}
+
+	/*
+	 * If the lock is now free, but there are some transactions trying to
+	 * end, we must clear the critical PGPROC fields for them, and save a
+	 * list of them so we can wake them up.
+	 */
+	if (lock->exclusive == 0 && lock->shared == 0 && lock->ending != NULL)
+	{
+		volatile PGPROC *vproc;
+
+		ending = lock->ending;
+		vproc = ending;
+
+		while (vproc != NULL)
+		{
+        	vproc->xid = InvalidTransactionId;
+	        vproc->xmin = InvalidTransactionId;
+	        /* must be cleared with xid/xmin: */
+	        vproc->vacuumFlags &= ~PROC_VACUUM_STATE_MASK;
+			vproc->subxids.nxids = 0;
+			vproc->subxids.overflowed = false;
+			vproc = vproc->flWaitLink;
+		}
+
+		/* Also advance global latestCompletedXid */
+		if (TransactionIdPrecedes(ShmemVariableCache->latestCompletedXid,
+								  lock->latest_ending_xid))
+			ShmemVariableCache->latestCompletedXid = lock->latest_ending_xid;
+
+		/* Reset lock state. */
+		lock->ending = NULL;
+		lock->latest_ending_xid = InvalidTransactionId;
+	}
+
+	/*
+	 * 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->flex.head;
+	if (head != NULL)
+	{
+		if (lock->exclusive == 0 && lock->shared == 0 && lock->flex.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->flWaitMode != LW_EXCLUSIVE)
+			{
+				while (proc->flWaitLink != NULL &&
+					   proc->flWaitLink->flWaitMode != LW_EXCLUSIVE)
+					proc = proc->flWaitLink;
+			}
+			/* proc is now the last PGPROC to be released */
+			lock->flex.head = proc->flWaitLink;
+			proc->flWaitLink = NULL;
+			/* prevent additional wakeups until retryer gets to run */
+			lock->flex.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->flex.mutex);
+
+	TRACE_POSTGRESQL_FLEXLOCK_RELEASE(lockid);
+
+	/*
+	 * Awaken any waiters I removed from the queue.
+	 */
+	while (head != NULL)
+	{
+		FlexLockDebug("LWLockRelease", lockid, "release waiter");
+		proc = head;
+		head = proc->flWaitLink;
+		proc->flWaitLink = NULL;
+		proc->flWaitResult = 1;		/* any non-zero value will do */
+		PGSemaphoreUnlock(&proc->sem);
+	}
+
+	/*
+	 * Also awaken any processes whose critical PGPROC fields I cleared
+	 */
+	while (ending != NULL)
+	{
+		FlexLockDebug("LWLockRelease", lockid, "release ending");
+		proc = ending;
+		ending = proc->flWaitLink;
+		proc->flWaitLink = NULL;
+		proc->flWaitResult = 1;		/* any non-zero value will do */
+		PGSemaphoreUnlock(&proc->sem);
+	}
+
+	/*
+	 * Now okay to allow cancel/die interrupts.
+	 */
+	RESUME_INTERRUPTS();
+}
diff --git a/src/include/storage/flexlock_internals.h b/src/include/storage/flexlock_internals.h
index 5f78da7..d1bca45 100644
--- a/src/include/storage/flexlock_internals.h
+++ b/src/include/storage/flexlock_internals.h
@@ -43,6 +43,7 @@ typedef struct FlexLock
 } FlexLock;
 
 #define FLEXLOCK_TYPE_LWLOCK			'l'
+#define FLEXLOCK_TYPE_PROCARRAYLOCK		'p'
 
 typedef union FlexLockPadded
 {
diff --git a/src/include/storage/procarraylock.h b/src/include/storage/procarraylock.h
new file mode 100644
index 0000000..678ca6f
--- /dev/null
+++ b/src/include/storage/procarraylock.h
@@ -0,0 +1,28 @@
+/*-------------------------------------------------------------------------
+ *
+ * procarraylock.h
+ *	  Lock management for the ProcArray
+ *
+ * Portions Copyright (c) 1996-2011, PostgreSQL Global Development Group
+ * Portions Copyright (c) 1994, Regents of the University of California
+ *
+ * src/include/storage/lwlock.h
+ *
+ *-------------------------------------------------------------------------
+ */
+#ifndef PROCARRAYLOCK_H
+#define PROCARRAYLOCK_H
+
+#include "storage/flexlock.h"
+
+typedef enum ProcArrayLockMode
+{
+	PAL_EXCLUSIVE,
+	PAL_SHARED
+} ProcArrayLockMode;
+
+extern void ProcArrayLockAcquire(ProcArrayLockMode mode);
+extern void ProcArrayLockClearTransaction(TransactionId latestXid);
+extern void ProcArrayLockRelease(void);
+
+#endif   /* PROCARRAYLOCK_H */

commit 24f8e349d085e646cb918c552cc8ead7d38f7013
Author: Pavan Deolasee <pavan@ubuntu.(none)>
Date:   Fri Nov 18 15:49:54 2011 +0530

    Implement a shared work Q mechanism. A process can queue its work for later
    execution if the protecting lock is currently not available. Backend which
    releases the last lock will finish the work and wake up the waiting process.

diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c
index 1a48485..59d2958 100644
--- a/src/backend/storage/ipc/procarray.c
+++ b/src/backend/storage/ipc/procarray.c
@@ -157,6 +157,7 @@ static int KnownAssignedXidsGetAndSetXmin(TransactionId *xarray,
 							   TransactionId xmax);
 static TransactionId KnownAssignedXidsGetOldestXmin(void);
 static void KnownAssignedXidsDisplay(int trace_level);
+static bool ProcArrayEndTransactionWQ(WorkQueueData *wqdata);
 
 /*
  * Report shared-memory space needed by CreateSharedProcArray.
@@ -331,8 +332,6 @@ ProcArrayRemove(PGPROC *proc, TransactionId latestXid)
 
 	elog(LOG, "failed to find proc %p in ProcArray", proc);
 }
-
-
 /*
  * ProcArrayEndTransaction -- mark a transaction as no longer running
  *
@@ -352,33 +351,24 @@ ProcArrayEndTransaction(PGPROC *proc, TransactionId latestXid)
 	if (TransactionIdIsValid(latestXid))
 	{
 		/*
-		 * We must lock ProcArrayLock while clearing proc->xid, so that we do
-		 * not exit the set of "running" transactions while someone else is
-		 * taking a snapshot.  See discussion in
-		 * src/backend/access/transam/README.
+		 * Use the shared work queue mechanism to get the work done. If the
+		 * ProcArrayLock is available, it will done immediately, otherwise it
+		 * will be queued up and some other backend (the one who releases the
+		 * lock last) will do it for us and wake us up
+		 *
+		 * Use the shared area in PGPROC to communicate with other backends. We
+		 * can write to the area without holding any lock becuase its
+		 * read/written by other backends only when we are sleeping in the
+		 * queue and only one backend can access it at any time
 		 */
-		Assert(TransactionIdIsValid(proc->xid));
-
-		LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
-
-		proc->xid = InvalidTransactionId;
-		proc->lxid = InvalidLocalTransactionId;
-		proc->xmin = InvalidTransactionId;
-		/* must be cleared with xid/xmin: */
-		proc->vacuumFlags &= ~PROC_VACUUM_STATE_MASK;
-		proc->inCommit = false; /* be sure this is cleared in abort */
-		proc->recoveryConflictPending = false;
-
-		/* Clear the subtransaction-XID cache too while holding the lock */
-		proc->subxids.nxids = 0;
-		proc->subxids.overflowed = false;
+		WorkQueueData *wqdata = &proc->wqdata;
 
-		/* Also advance global latestCompletedXid while holding the lock */
-		if (TransactionIdPrecedes(ShmemVariableCache->latestCompletedXid,
-								  latestXid))
-			ShmemVariableCache->latestCompletedXid = latestXid;
+		wqdata->wq_reqtype = WQ_END_TRANSACTION;
+		wqdata->wq_reqin.wqin_end_xact.proc = proc;
+		wqdata->wq_reqin.wqin_end_xact.latestXid = latestXid;
+		wqdata->wq_exec = ProcArrayEndTransactionWQ;
 
-		LWLockRelease(ProcArrayLock);
+		LWLockExecute(ProcArrayLock, LW_EXCLUSIVE);
 	}
 	else
 	{
@@ -401,6 +391,38 @@ ProcArrayEndTransaction(PGPROC *proc, TransactionId latestXid)
 	}
 }
 
+/*
+ * Do the real work for ProcArrayEndTransaction. We are called while holding
+ * the mutex, so be quick and fast
+ */
+static bool
+ProcArrayEndTransactionWQ(WorkQueueData *wqdata)
+{
+	volatile PGPROC *proc = (PGPROC *) wqdata->wq_reqin.wqin_end_xact.proc;
+	TransactionId latestXid = wqdata->wq_reqin.wqin_end_xact.latestXid;
+
+	Assert(TransactionIdIsValid(proc->xid));
+
+	proc->xid = InvalidTransactionId;
+	proc->lxid = InvalidLocalTransactionId;
+	proc->xmin = InvalidTransactionId;
+	/* must be cleared with xid/xmin: */
+	proc->vacuumFlags &= ~PROC_VACUUM_STATE_MASK;
+	proc->inCommit = false; /* be sure this is cleared in abort */
+	proc->recoveryConflictPending = false;
+
+	/* Clear the subtransaction-XID cache too while holding the lock */
+	proc->subxids.nxids = 0;
+	proc->subxids.overflowed = false;
+
+	/* Also advance global latestCompletedXid while holding the lock */
+	if (TransactionIdPrecedes(ShmemVariableCache->latestCompletedXid,
+				latestXid))
+		ShmemVariableCache->latestCompletedXid = latestXid;
+
+	return true;
+}
+
 
 /*
  * ProcArrayClearTransaction -- clear the transaction fields
diff --git a/src/backend/storage/lmgr/lwlock.c b/src/backend/storage/lmgr/lwlock.c
index 079eb29..ab2899e 100644
--- a/src/backend/storage/lmgr/lwlock.c
+++ b/src/backend/storage/lmgr/lwlock.c
@@ -43,6 +43,8 @@ typedef struct LWLock
 	bool		releaseOK;		/* T if ok to release waiters */
 	char		exclusive;		/* # of exclusive holders (0 or 1) */
 	int			shared;			/* # of shared holders (0..MaxBackends) */
+	PGPROC		*sq_head;		/* head of list of pending requests */
+	PGPROC		*sq_tail;		/* tail of list of pending request */
 	PGPROC	   *head;			/* head of list of waiting PGPROCs */
 	PGPROC	   *tail;			/* tail of list of waiting PGPROCs */
 	/* tail is undefined when head is NULL */
@@ -269,6 +271,8 @@ CreateLWLocks(void)
 		lock->lock.releaseOK = true;
 		lock->lock.exclusive = 0;
 		lock->lock.shared = 0;
+		lock->lock.sq_head = NULL;
+		lock->lock.sq_tail = NULL;
 		lock->lock.head = NULL;
 		lock->lock.tail = NULL;
 	}
@@ -571,7 +575,7 @@ void
 LWLockRelease(LWLockId lockid)
 {
 	volatile LWLock *lock = &(LWLockArray[lockid].lock);
-	PGPROC	   *head;
+	PGPROC	   *head, *wakeupq_head, *wakeupq_tail;
 	PGPROC	   *proc;
 	int			i;
 
@@ -610,21 +614,48 @@ LWLockRelease(LWLockId lockid)
 	 * 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)
 	{
-		if (lock->exclusive == 0 && lock->shared == 0 && lock->releaseOK)
+		/*
+		 * First process any pending requests and add those processes to the
+		 * list of to-be-awaken processes.
+		 */
+		head = lock->sq_head;
+		while (head != NULL)
+		{
+			proc = head;
+
+			Assert(proc->wqdata.wq_reqpending);
+			
+			/* Finish the work */
+			proc->wqdata.wq_exec(&proc->wqdata);
+			proc->wqdata.wq_reqpending = false;
+
+			head = proc->lwWaitLink;
+		}
+		
+		/*
+		 * We must wake up all those processes which are waiting for their
+		 * pending requests to be processed
+		 */
+		wakeupq_head = lock->sq_head;
+		wakeupq_tail = lock->sq_tail;
+
+		lock->sq_head = lock->sq_tail = NULL;
+
+		/*
+		 * 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.
+		 */
+		head = lock->head;
+		if (head != NULL)
 		{
-			/*
-			 * 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->lwWaitLink->lwExclusive)
 					proc = proc->lwWaitLink;
 			}
 			/* proc is now the last PGPROC to be released */
@@ -633,11 +664,19 @@ LWLockRelease(LWLockId lockid)
 			/* prevent additional wakeups until retryer gets to run */
 			lock->releaseOK = false;
 		}
+
+		/*
+		 * Add any other processes to be woken up to the list
+		 */
+		if (wakeupq_head)
+			wakeupq_tail->lwWaitLink = head;
 		else
-		{
-			/* lock is still held, can't awaken anything */
-			head = NULL;
-		}
+			wakeupq_head = head;
+	}
+	else
+	{
+		/* lock is still held, can't awaken anything */
+		wakeupq_head = NULL;
 	}
 
 	/* We are done updating shared state of the lock itself. */
@@ -648,6 +687,7 @@ LWLockRelease(LWLockId lockid)
 	/*
 	 * Awaken any waiters I removed from the queue.
 	 */
+	head = wakeupq_head;
 	while (head != NULL)
 	{
 		LOG_LWDEBUG("LWLockRelease", lockid, "release waiter");
@@ -685,6 +725,180 @@ LWLockReleaseAll(void)
 	}
 }
 
+/*
+ * Acquire the LWLock and execute the given task. If the lock is not available,
+ * we queue the task which will be executed when the lock becomes available.
+ * Even though, both exclusive and shared tasks may be performed, its useful to
+ * use this interface for tasks that need exclusive access to the lock. The
+ * task must be very short and must not perform any IO since it may be done
+ * while holding the spinlock. Also, any ereports must be avoided while
+ * executing the task.
+ *
+ * The caller can assume that the task is finished successfully when the
+ * function returns. If there is any out information, that will be returned
+ * in the MyProc->wqdata shared area.
+ */ 
+void
+LWLockExecute(LWLockId lockid, LWLockMode mode)
+{
+	volatile LWLock *lock = &(LWLockArray[lockid].lock);
+	PGPROC	   *proc = MyProc;
+	WorkQueueData	*wqdata = &proc->wqdata;
+	int			extraWaits = 0;
+	bool		mustwait;
+
+	PRINT_LWDEBUG("LWLockExecute", 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));
+
+	/*
+	 * 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, execute the task and we are done */
+	if (mode == LW_EXCLUSIVE)
+	{
+		if (lock->exclusive == 0 && lock->shared == 0)
+			mustwait = false;
+		else
+			mustwait = true;
+	}
+	else
+	{
+		if (lock->exclusive == 0)
+			mustwait = false;
+		else
+			mustwait = true;
+	}
+
+	if (!mustwait)
+	{
+		/*
+		 * No other process in critical section and none can get in until we
+		 * are holding the mutex
+		 *
+		 * Invoke callback for work execution. We are still holding the
+		 * mutex, so the callback must be short and quick. We rely that the
+		 * users of the function will be smart enough to know that
+		 *
+		 * XXX Alternatively, we can acquire the LWLock in the desired mode
+		 * (lock->shared/lock->exclusive must be incremented above), execute
+		 * the task and then release the lock. May be the caller can pass
+		 * additional information to choose the desired behavior. For now, just
+		 * do this while holding the mutex.
+		 */
+		wqdata->wq_exec(wqdata);
+		/* Can release the mutex now */
+		SpinLockRelease(&lock->mutex);
+	}
+	else
+	{
+		/*
+		 * 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;
+
+		/* Mark the request as pending. Its of not much use right now */
+		wqdata->wq_reqpending = true;
+
+		if (lock->sq_head == NULL)
+			lock->sq_head = proc;
+		else
+			lock->sq_tail->lwWaitLink = proc;
+		lock->sq_tail = proc;
+
+		/* XXX Should we set lock->releaseOK to true here ? */
+		/* 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("LWLockExecute", 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("LWLockExecute", lockid, "awakened");
+
+		TRACE_POSTGRESQL_LWLOCK_ACQUIRE(lockid, mode);
+		/*
+		 * 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();
+}
 
 /*
  * LWLockHeldByMe - test whether my process currently holds a lock
diff --git a/src/include/storage/lwlock.h b/src/include/storage/lwlock.h
index 438a48d..7c2d99a 100644
--- a/src/include/storage/lwlock.h
+++ b/src/include/storage/lwlock.h
@@ -14,6 +14,8 @@
 #ifndef LWLOCK_H
 #define LWLOCK_H
 
+#include "storage/wqueue.h"
+
 /*
  * It's a bit odd to declare NUM_BUFFER_PARTITIONS and NUM_LOCK_PARTITIONS
  * here, but we need them to set up enum LWLockId correctly, and having
@@ -114,5 +116,6 @@ extern Size LWLockShmemSize(void);
 extern void CreateLWLocks(void);
 
 extern void RequestAddinLWLocks(int n);
+extern void LWLockExecute(LWLockId lockid, LWLockMode mode);
 
 #endif   /* LWLOCK_H */
diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h
index 6e798b1..34874f2 100644
--- a/src/include/storage/proc.h
+++ b/src/include/storage/proc.h
@@ -154,6 +154,9 @@ struct PGPROC
 	Oid			fpRelId[FP_LOCK_SLOTS_PER_BACKEND]; /* slots for rel oids */
 	bool		fpVXIDLock;		/* are we holding a fast-path VXID lock? */
 	LocalTransactionId fpLocalTransactionId;	/* lxid for fast-path VXID lock */
+
+	/* Shared work queue */
+	WorkQueueData	wqdata;
 };
 
 /* NOTE: "typedef struct PGPROC PGPROC" appears in storage/lock.h. */
diff --git a/src/include/storage/wqueue.h b/src/include/storage/wqueue.h
new file mode 100644
index 0000000..7ee5cd3
--- /dev/null
+++ b/src/include/storage/wqueue.h
@@ -0,0 +1,44 @@
+/*
+ * wqueue.h
+ *
+ * 	Implement shared work queue for processing
+ *
+ * 	src/include/storage/wqueue.h
+ */
+
+#ifndef _WQUEUE_H
+#define _WQUEUE_H
+
+typedef enum WQRequestType
+{
+	WQ_NO_REQUEST = 0,
+	WQ_END_TRANSACTION
+} WQRequestType;
+
+typedef union WQRequestIn
+{
+	struct {
+		void			*proc;
+		TransactionId	latestXid;
+	} wqin_end_xact;
+} WQRequestIn;
+
+typedef union WQRequestOut
+{
+	bool	wqout_status;
+} WQRequestOut;
+
+struct WorkQueueData;
+
+typedef bool (*WQExecuteReq) (struct WorkQueueData *wqdata);
+
+typedef struct WorkQueueData
+{
+	WQRequestType	wq_reqtype;
+	bool			wq_reqpending;
+	WQRequestIn		wq_reqin;
+	WQRequestOut	wq_reqout;
+	WQExecuteReq	wq_exec;
+} WorkQueueData;
+
+#endif

diff --git a/contrib/pg_stat_statements/pg_stat_statements.c b/contrib/pg_stat_statements/pg_stat_statements.c
index 8dc3054..51b24d0 100644
--- a/contrib/pg_stat_statements/pg_stat_statements.c
+++ b/contrib/pg_stat_statements/pg_stat_statements.c
@@ -105,7 +105,7 @@ typedef struct pgssEntry
  */
 typedef struct pgssSharedState
 {
-	LWLockId	lock;			/* protects hashtable search/modification */
+	FlexLockId	lock;			/* protects hashtable search/modification */
 	int			query_size;		/* max query length in bytes */
 } pgssSharedState;
 
diff --git a/doc/src/sgml/config.sgml b/doc/src/sgml/config.sgml
index d1e628f..8517b36 100644
--- a/doc/src/sgml/config.sgml
+++ b/doc/src/sgml/config.sgml
@@ -6199,14 +6199,14 @@ LOG:  CleanUpLock: deleting: lock(0xb7acd844) id(24688,24696,0,0,0,1)
      </varlistentry>
 
      <varlistentry>
-      <term><varname>trace_lwlocks</varname> (<type>boolean</type>)</term>
+      <term><varname>trace_flexlocks</varname> (<type>boolean</type>)</term>
       <indexterm>
-       <primary><varname>trace_lwlocks</> configuration parameter</primary>
+       <primary><varname>trace_flexlocks</> configuration parameter</primary>
       </indexterm>
       <listitem>
        <para>
-        If on, emit information about lightweight lock usage.  Lightweight
-        locks are intended primarily to provide mutual exclusion of access
+        If on, emit information about FlexLock usage.  FlexLocks
+        are intended primarily to provide mutual exclusion of access
         to shared-memory data structures.
        </para>
        <para>
diff --git a/doc/src/sgml/monitoring.sgml b/doc/src/sgml/monitoring.sgml
index b9dc1d2..98ed0d3 100644
--- a/doc/src/sgml/monitoring.sgml
+++ b/doc/src/sgml/monitoring.sgml
@@ -1724,49 +1724,49 @@ SELECT pg_stat_get_backend_pid(s.backendid) AS procpid,
       or kilobytes of memory used for an internal sort.</entry>
     </row>
     <row>
-     <entry>lwlock-acquire</entry>
-     <entry>(LWLockId, LWLockMode)</entry>
-     <entry>Probe that fires when an LWLock has been acquired.
-      arg0 is the LWLock's ID.
-      arg1 is the requested lock mode, either exclusive or shared.</entry>
+     <entry>flexlock-acquire</entry>
+     <entry>(FlexLockId, FlexLockMode)</entry>
+     <entry>Probe that fires when an FlexLock has been acquired.
+      arg0 is the FlexLock's ID.
+      arg1 is the requested lock mode.</entry>
     </row>
     <row>
-     <entry>lwlock-release</entry>
-     <entry>(LWLockId)</entry>
-     <entry>Probe that fires when an LWLock has been released (but note
+     <entry>flexlock-release</entry>
+     <entry>(FlexLockId)</entry>
+     <entry>Probe that fires when a FlexLock has been released (but note
       that any released waiters have not yet been awakened).
-      arg0 is the LWLock's ID.</entry>
+      arg0 is the FlexLock's ID.</entry>
     </row>
     <row>
-     <entry>lwlock-wait-start</entry>
-     <entry>(LWLockId, LWLockMode)</entry>
-     <entry>Probe that fires when an LWLock was not immediately available and
+     <entry>flexlock-wait-start</entry>
+     <entry>(FlexLockId, FlexLockMode)</entry>
+     <entry>Probe that fires when an FlexLock was not immediately available and
       a server process has begun to wait for the lock to become available.
-      arg0 is the LWLock's ID.
+      arg0 is the FlexLock's ID.
       arg1 is the requested lock mode, either exclusive or shared.</entry>
     </row>
     <row>
-     <entry>lwlock-wait-done</entry>
-     <entry>(LWLockId, LWLockMode)</entry>
+     <entry>flexlock-wait-done</entry>
+     <entry>(FlexLockId, FlexLockMode)</entry>
      <entry>Probe that fires when a server process has been released from its
-      wait for an LWLock (it does not actually have the lock yet).
-      arg0 is the LWLock's ID.
+      wait for an FlexLock (it does not actually have the lock yet).
+      arg0 is the FlexLock's ID.
       arg1 is the requested lock mode, either exclusive or shared.</entry>
     </row>
     <row>
-     <entry>lwlock-condacquire</entry>
-     <entry>(LWLockId, LWLockMode)</entry>
-     <entry>Probe that fires when an LWLock was successfully acquired when the
-      caller specified no waiting.
-      arg0 is the LWLock's ID.
+     <entry>flexlock-condacquire</entry>
+     <entry>(FlexLockId, FlexLockMode)</entry>
+     <entry>Probe that fires when an FlexLock was successfully acquired when
+      the caller specified no waiting.
+      arg0 is the FlexLock's ID.
       arg1 is the requested lock mode, either exclusive or shared.</entry>
     </row>
     <row>
-     <entry>lwlock-condacquire-fail</entry>
-     <entry>(LWLockId, LWLockMode)</entry>
-     <entry>Probe that fires when an LWLock was not successfully acquired when
-      the caller specified no waiting.
-      arg0 is the LWLock's ID.
+     <entry>flexlock-condacquire-fail</entry>
+     <entry>(FlexLockId, FlexLockMode)</entry>
+     <entry>Probe that fires when an FlexLock was not successfully acquired
+      when the caller specified no waiting.
+      arg0 is the FlexLock's ID.
       arg1 is the requested lock mode, either exclusive or shared.</entry>
     </row>
     <row>
@@ -1813,11 +1813,11 @@ SELECT pg_stat_get_backend_pid(s.backendid) AS procpid,
      <entry>unsigned int</entry>
     </row>
     <row>
-     <entry>LWLockId</entry>
+     <entry>FlexLockId</entry>
      <entry>int</entry>
     </row>
     <row>
-     <entry>LWLockMode</entry>
+     <entry>FlexLockMode</entry>
      <entry>int</entry>
     </row>
     <row>
diff --git a/src/backend/access/transam/slru.c b/src/backend/access/transam/slru.c
index f7caa34..09d5862 100644
--- a/src/backend/access/transam/slru.c
+++ b/src/backend/access/transam/slru.c
@@ -151,7 +151,7 @@ SimpleLruShmemSize(int nslots, int nlsns)
 	sz += MAXALIGN(nslots * sizeof(bool));		/* page_dirty[] */
 	sz += MAXALIGN(nslots * sizeof(int));		/* page_number[] */
 	sz += MAXALIGN(nslots * sizeof(int));		/* page_lru_count[] */
-	sz += MAXALIGN(nslots * sizeof(LWLockId));	/* buffer_locks[] */
+	sz += MAXALIGN(nslots * sizeof(FlexLockId));		/* buffer_locks[] */
 
 	if (nlsns > 0)
 		sz += MAXALIGN(nslots * nlsns * sizeof(XLogRecPtr));	/* group_lsn[] */
@@ -161,7 +161,7 @@ SimpleLruShmemSize(int nslots, int nlsns)
 
 void
 SimpleLruInit(SlruCtl ctl, const char *name, int nslots, int nlsns,
-			  LWLockId ctllock, const char *subdir)
+			  FlexLockId ctllock, const char *subdir)
 {
 	SlruShared	shared;
 	bool		found;
@@ -202,8 +202,8 @@ SimpleLruInit(SlruCtl ctl, const char *name, int nslots, int nlsns,
 		offset += MAXALIGN(nslots * sizeof(int));
 		shared->page_lru_count = (int *) (ptr + offset);
 		offset += MAXALIGN(nslots * sizeof(int));
-		shared->buffer_locks = (LWLockId *) (ptr + offset);
-		offset += MAXALIGN(nslots * sizeof(LWLockId));
+		shared->buffer_locks = (FlexLockId *) (ptr + offset);
+		offset += MAXALIGN(nslots * sizeof(FlexLockId));
 
 		if (nlsns > 0)
 		{
diff --git a/src/backend/access/transam/twophase.c b/src/backend/access/transam/twophase.c
index d2fecb1..943929b 100644
--- a/src/backend/access/transam/twophase.c
+++ b/src/backend/access/transam/twophase.c
@@ -326,9 +326,9 @@ MarkAsPreparing(TransactionId xid, const char *gid,
 	proc->backendId = InvalidBackendId;
 	proc->databaseId = databaseid;
 	proc->roleId = owner;
-	proc->lwWaiting = false;
-	proc->lwExclusive = false;
-	proc->lwWaitLink = NULL;
+	proc->flWaitResult = 0;
+	proc->flWaitMode = 0;
+	proc->flWaitLink = NULL;
 	proc->waitLock = NULL;
 	proc->waitProcLock = NULL;
 	for (i = 0; i < NUM_LOCK_PARTITIONS; i++)
diff --git a/src/backend/access/transam/xact.c b/src/backend/access/transam/xact.c
index c383011..0da2ae5 100644
--- a/src/backend/access/transam/xact.c
+++ b/src/backend/access/transam/xact.c
@@ -2248,7 +2248,7 @@ AbortTransaction(void)
 	 * Releasing LW locks is critical since we might try to grab them again
 	 * while cleaning up!
 	 */
-	LWLockReleaseAll();
+	FlexLockReleaseAll();
 
 	/* Clean up buffer I/O and buffer context locks, too */
 	AbortBufferIO();
@@ -4138,7 +4138,7 @@ AbortSubTransaction(void)
 	 * FIXME This may be incorrect --- Are there some locks we should keep?
 	 * Buffer locks, for example?  I don't think so but I'm not sure.
 	 */
-	LWLockReleaseAll();
+	FlexLockReleaseAll();
 
 	AbortBufferIO();
 	UnlockBuffers();
diff --git a/src/backend/bootstrap/bootstrap.c b/src/backend/bootstrap/bootstrap.c
index 6bf2421..9ceee91 100644
--- a/src/backend/bootstrap/bootstrap.c
+++ b/src/backend/bootstrap/bootstrap.c
@@ -562,13 +562,13 @@ bootstrap_signals(void)
  * Begin shutdown of an auxiliary process.	This is approximately the equivalent
  * of ShutdownPostgres() in postinit.c.  We can't run transactions in an
  * auxiliary process, so most of the work of AbortTransaction() is not needed,
- * but we do need to make sure we've released any LWLocks we are holding.
+ * but we do need to make sure we've released any flex locks we are holding.
  * (This is only critical during an error exit.)
  */
 static void
 ShutdownAuxiliaryProcess(int code, Datum arg)
 {
-	LWLockReleaseAll();
+	FlexLockReleaseAll();
 }
 
 /* ----------------------------------------------------------------
diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c
index cacedab..f33f573 100644
--- a/src/backend/postmaster/bgwriter.c
+++ b/src/backend/postmaster/bgwriter.c
@@ -176,9 +176,10 @@ BackgroundWriterMain(void)
 		/*
 		 * These operations are really just a minimal subset of
 		 * AbortTransaction().	We don't have very many resources to worry
-		 * about in bgwriter, but we do have LWLocks, buffers, and temp files.
+		 * about in bgwriter, but we do have flex locks, buffers, and temp
+		 * files.
 		 */
-		LWLockReleaseAll();
+		FlexLockReleaseAll();
 		AbortBufferIO();
 		UnlockBuffers();
 		/* buffer pins are released here: */
diff --git a/src/backend/postmaster/checkpointer.c b/src/backend/postmaster/checkpointer.c
index e9ae1e8..49f07a7 100644
--- a/src/backend/postmaster/checkpointer.c
+++ b/src/backend/postmaster/checkpointer.c
@@ -281,9 +281,10 @@ CheckpointerMain(void)
 		/*
 		 * These operations are really just a minimal subset of
 		 * AbortTransaction().	We don't have very many resources to worry
-		 * about in checkpointer, but we do have LWLocks, buffers, and temp files.
+		 * about in checkpointer, but we do have flex locks, buffers, and temp
+		 * files.
 		 */
-		LWLockReleaseAll();
+		FlexLockReleaseAll();
 		AbortBufferIO();
 		UnlockBuffers();
 		/* buffer pins are released here: */
diff --git a/src/backend/postmaster/postmaster.c b/src/backend/postmaster/postmaster.c
index 963189d..59d18eb 100644
--- a/src/backend/postmaster/postmaster.c
+++ b/src/backend/postmaster/postmaster.c
@@ -109,6 +109,7 @@
 #include "postmaster/syslogger.h"
 #include "replication/walsender.h"
 #include "storage/fd.h"
+#include "storage/flexlock_internals.h"
 #include "storage/ipc.h"
 #include "storage/pg_shmem.h"
 #include "storage/pmsignal.h"
@@ -404,8 +405,6 @@ typedef struct
 typedef int InheritableSocket;
 #endif
 
-typedef struct LWLock LWLock;	/* ugly kluge */
-
 /*
  * Structure contains all variables passed to exec:ed backends
  */
@@ -426,7 +425,7 @@ typedef struct
 	slock_t    *ShmemLock;
 	VariableCache ShmemVariableCache;
 	Backend    *ShmemBackendArray;
-	LWLock	   *LWLockArray;
+	FlexLock   *FlexLockArray;
 	slock_t    *ProcStructLock;
 	PROC_HDR   *ProcGlobal;
 	PGPROC	   *AuxiliaryProcs;
@@ -4676,7 +4675,6 @@ MaxLivePostmasterChildren(void)
  * functions
  */
 extern slock_t *ShmemLock;
-extern LWLock *LWLockArray;
 extern slock_t *ProcStructLock;
 extern PGPROC *AuxiliaryProcs;
 extern PMSignalData *PMSignalState;
@@ -4721,7 +4719,7 @@ save_backend_variables(BackendParameters *param, Port *port,
 	param->ShmemVariableCache = ShmemVariableCache;
 	param->ShmemBackendArray = ShmemBackendArray;
 
-	param->LWLockArray = LWLockArray;
+	param->FlexLockArray = FlexLockArray;
 	param->ProcStructLock = ProcStructLock;
 	param->ProcGlobal = ProcGlobal;
 	param->AuxiliaryProcs = AuxiliaryProcs;
@@ -4945,7 +4943,7 @@ restore_backend_variables(BackendParameters *param, Port *port)
 	ShmemVariableCache = param->ShmemVariableCache;
 	ShmemBackendArray = param->ShmemBackendArray;
 
-	LWLockArray = param->LWLockArray;
+	FlexLockArray = param->FlexLockArray;
 	ProcStructLock = param->ProcStructLock;
 	ProcGlobal = param->ProcGlobal;
 	AuxiliaryProcs = param->AuxiliaryProcs;
diff --git a/src/backend/postmaster/walwriter.c b/src/backend/postmaster/walwriter.c
index 157728e..587443d 100644
--- a/src/backend/postmaster/walwriter.c
+++ b/src/backend/postmaster/walwriter.c
@@ -167,9 +167,9 @@ WalWriterMain(void)
 		/*
 		 * These operations are really just a minimal subset of
 		 * AbortTransaction().	We don't have very many resources to worry
-		 * about in walwriter, but we do have LWLocks, and perhaps buffers?
+		 * about in walwriter, but we do have flex locks, and perhaps buffers?
 		 */
-		LWLockReleaseAll();
+		FlexLockReleaseAll();
 		AbortBufferIO();
 		UnlockBuffers();
 		/* buffer pins are released here: */
diff --git a/src/backend/storage/buffer/bufmgr.c b/src/backend/storage/buffer/bufmgr.c
index 71fe8c6..4c4959c 100644
--- a/src/backend/storage/buffer/bufmgr.c
+++ b/src/backend/storage/buffer/bufmgr.c
@@ -141,7 +141,7 @@ PrefetchBuffer(Relation reln, ForkNumber forkNum, BlockNumber blockNum)
 	{
 		BufferTag	newTag;		/* identity of requested block */
 		uint32		newHash;	/* hash value for newTag */
-		LWLockId	newPartitionLock;	/* buffer partition lock for it */
+		FlexLockId	newPartitionLock;	/* buffer partition lock for it */
 		int			buf_id;
 
 		/* create a tag so we can lookup the buffer */
@@ -514,10 +514,10 @@ BufferAlloc(SMgrRelation smgr, char relpersistence, ForkNumber forkNum,
 {
 	BufferTag	newTag;			/* identity of requested block */
 	uint32		newHash;		/* hash value for newTag */
-	LWLockId	newPartitionLock;		/* buffer partition lock for it */
+	FlexLockId	newPartitionLock;		/* buffer partition lock for it */
 	BufferTag	oldTag;			/* previous identity of selected buffer */
 	uint32		oldHash;		/* hash value for oldTag */
-	LWLockId	oldPartitionLock;		/* buffer partition lock for it */
+	FlexLockId	oldPartitionLock;		/* buffer partition lock for it */
 	BufFlags	oldFlags;
 	int			buf_id;
 	volatile BufferDesc *buf;
@@ -857,7 +857,7 @@ InvalidateBuffer(volatile BufferDesc *buf)
 {
 	BufferTag	oldTag;
 	uint32		oldHash;		/* hash value for oldTag */
-	LWLockId	oldPartitionLock;		/* buffer partition lock for it */
+	FlexLockId	oldPartitionLock;		/* buffer partition lock for it */
 	BufFlags	oldFlags;
 
 	/* Save the original buffer tag before dropping the spinlock */
diff --git a/src/backend/storage/ipc/ipci.c b/src/backend/storage/ipc/ipci.c
index bb8b832..a2c570a 100644
--- a/src/backend/storage/ipc/ipci.c
+++ b/src/backend/storage/ipc/ipci.c
@@ -113,7 +113,7 @@ CreateSharedMemoryAndSemaphores(bool makePrivate, int port)
 		size = add_size(size, SUBTRANSShmemSize());
 		size = add_size(size, TwoPhaseShmemSize());
 		size = add_size(size, MultiXactShmemSize());
-		size = add_size(size, LWLockShmemSize());
+		size = add_size(size, FlexLockShmemSize());
 		size = add_size(size, ProcArrayShmemSize());
 		size = add_size(size, BackendStatusShmemSize());
 		size = add_size(size, SInvalShmemSize());
@@ -179,7 +179,7 @@ CreateSharedMemoryAndSemaphores(bool makePrivate, int port)
 	 * needed for InitShmemIndex.
 	 */
 	if (!IsUnderPostmaster)
-		CreateLWLocks();
+		CreateFlexLocks();
 
 	/*
 	 * Set up shmem.c index hashtable
diff --git a/src/backend/storage/lmgr/Makefile b/src/backend/storage/lmgr/Makefile
index e12a854..3730e51 100644
--- a/src/backend/storage/lmgr/Makefile
+++ b/src/backend/storage/lmgr/Makefile
@@ -12,7 +12,8 @@ subdir = src/backend/storage/lmgr
 top_builddir = ../../../..
 include $(top_builddir)/src/Makefile.global
 
-OBJS = lmgr.o lock.o proc.o deadlock.o lwlock.o spin.o s_lock.o predicate.o
+OBJS = flexlock.o lmgr.o lock.o proc.o deadlock.o lwlock.o spin.o s_lock.o \
+	predicate.o
 
 include $(top_srcdir)/src/backend/common.mk
 
diff --git a/src/backend/storage/lmgr/flexlock.c b/src/backend/storage/lmgr/flexlock.c
new file mode 100644
index 0000000..f517589
--- /dev/null
+++ b/src/backend/storage/lmgr/flexlock.c
@@ -0,0 +1,352 @@
+/*-------------------------------------------------------------------------
+ *
+ * flexlock.c
+ *	  Low-level routines for managing flex locks.
+ *
+ * Flex locks are intended primarily to provide mutual exclusion of access
+ * to shared-memory data structures.  Most, but not all, flex locks are
+ * lightweight locks (LWLocks).  This file contains support routines that
+ * are used for all types of flex locks, including lwlocks.  User-level
+ * locking should be done with the full lock manager --- which depends on
+ * LWLocks to protect its shared state.
+ *
+ * Portions Copyright (c) 1996-2011, PostgreSQL Global Development Group
+ * Portions Copyright (c) 1994, Regents of the University of California
+ *
+ * IDENTIFICATION
+ *	  src/backend/storage/lmgr/flexlock.c
+ *
+ *-------------------------------------------------------------------------
+ */
+
+#include "postgres.h"
+
+#include "miscadmin.h"
+#include "access/clog.h"
+#include "access/multixact.h"
+#include "access/subtrans.h"
+#include "commands/async.h"
+#include "storage/flexlock_internals.h"
+#include "storage/lwlock.h"
+#include "storage/predicate.h"
+#include "storage/proc.h"
+#include "storage/spin.h"
+#include "utils/elog.h"
+
+/*
+ * We use this structure to keep track of flex locks held, for release
+ * during error recovery.  The maximum size could be determined at runtime
+ * if necessary, but it seems unlikely that more than a few locks could
+ * ever be held simultaneously.
+ */
+#define MAX_SIMUL_FLEXLOCKS	100
+
+static int	num_held_flexlocks = 0;
+static FlexLockId held_flexlocks[MAX_SIMUL_FLEXLOCKS];
+
+static int	lock_addin_request = 0;
+static bool lock_addin_request_allowed = true;
+
+#ifdef LOCK_DEBUG
+bool		Trace_flexlocks = false;
+#endif
+
+/*
+ * This points to the array of FlexLocks in shared memory.  Backends inherit
+ * the pointer by fork from the postmaster (except in the EXEC_BACKEND case,
+ * where we have special measures to pass it down).
+ */
+FlexLockPadded *FlexLockArray = NULL;
+
+/* We use the ShmemLock spinlock to protect LWLockAssign */
+extern slock_t *ShmemLock;
+
+static void FlexLockInit(FlexLock *flex, char locktype);
+
+/*
+ * Compute number of FlexLocks to allocate.
+ */
+int
+NumFlexLocks(void)
+{
+	int			numLocks;
+
+	/*
+	 * Possibly this logic should be spread out among the affected modules,
+	 * the same way that shmem space estimation is done.  But for now, there
+	 * are few enough users of FlexLocks that we can get away with just keeping
+	 * the knowledge here.
+	 */
+
+	/* Predefined FlexLocks */
+	numLocks = (int) NumFixedFlexLocks;
+
+	/* bufmgr.c needs two for each shared buffer */
+	numLocks += 2 * NBuffers;
+
+	/* proc.c needs one for each backend or auxiliary process */
+	numLocks += MaxBackends + NUM_AUXILIARY_PROCS;
+
+	/* clog.c needs one per CLOG buffer */
+	numLocks += NUM_CLOG_BUFFERS;
+
+	/* subtrans.c needs one per SubTrans buffer */
+	numLocks += NUM_SUBTRANS_BUFFERS;
+
+	/* multixact.c needs two SLRU areas */
+	numLocks += NUM_MXACTOFFSET_BUFFERS + NUM_MXACTMEMBER_BUFFERS;
+
+	/* async.c needs one per Async buffer */
+	numLocks += NUM_ASYNC_BUFFERS;
+
+	/* predicate.c needs one per old serializable xid buffer */
+	numLocks += NUM_OLDSERXID_BUFFERS;
+
+	/*
+	 * Add any requested by loadable modules; for backwards-compatibility
+	 * reasons, allocate at least NUM_USER_DEFINED_FLEXLOCKS of them even if
+	 * there are no explicit requests.
+	 */
+	lock_addin_request_allowed = false;
+	numLocks += Max(lock_addin_request, NUM_USER_DEFINED_FLEXLOCKS);
+
+	return numLocks;
+}
+
+
+/*
+ * RequestAddinFlexLocks
+ *		Request that extra FlexLocks be allocated for use by
+ *		a loadable module.
+ *
+ * This is only useful if called from the _PG_init hook of a library that
+ * is loaded into the postmaster via shared_preload_libraries.	Once
+ * shared memory has been allocated, calls will be ignored.  (We could
+ * raise an error, but it seems better to make it a no-op, so that
+ * libraries containing such calls can be reloaded if needed.)
+ */
+void
+RequestAddinFlexLocks(int n)
+{
+	if (IsUnderPostmaster || !lock_addin_request_allowed)
+		return;					/* too late */
+	lock_addin_request += n;
+}
+
+
+/*
+ * Compute shmem space needed for FlexLocks.
+ */
+Size
+FlexLockShmemSize(void)
+{
+	Size		size;
+	int			numLocks = NumFlexLocks();
+
+	/* Space for the FlexLock array. */
+	size = mul_size(numLocks, FLEX_LOCK_BYTES);
+
+	/* Space for dynamic allocation counter, plus room for alignment. */
+	size = add_size(size, 2 * sizeof(int) + FLEX_LOCK_BYTES);
+
+	return size;
+}
+
+/*
+ * Allocate shmem space for FlexLocks and initialize the locks.
+ */
+void
+CreateFlexLocks(void)
+{
+	int			numLocks = NumFlexLocks();
+	Size		spaceLocks = FlexLockShmemSize();
+	FlexLockPadded *lock;
+	int		   *FlexLockCounter;
+	char	   *ptr;
+	int			id;
+
+	/* Allocate and zero space */
+	ptr = (char *) ShmemAlloc(spaceLocks);
+	memset(ptr, 0, spaceLocks);
+
+	/* Leave room for dynamic allocation counter */
+	ptr += 2 * sizeof(int);
+
+	/* Ensure desired alignment of FlexLock array */
+	ptr += FLEX_LOCK_BYTES - ((uintptr_t) ptr) % FLEX_LOCK_BYTES;
+
+	FlexLockArray = (FlexLockPadded *) ptr;
+
+	/* All of the "fixed" FlexLocks are LWLocks. */
+	for (id = 0, lock = FlexLockArray; id < NumFixedFlexLocks; id++, lock++)
+		FlexLockInit(&lock->flex, FLEXLOCK_TYPE_LWLOCK);
+
+	/*
+	 * Initialize the dynamic-allocation counter, which is stored just before
+	 * the first FlexLock.
+	 */
+	FlexLockCounter = (int *) ((char *) FlexLockArray - 2 * sizeof(int));
+	FlexLockCounter[0] = (int) NumFixedFlexLocks;
+	FlexLockCounter[1] = numLocks;
+}
+
+/*
+ * FlexLockAssign - assign a dynamically-allocated FlexLock number
+ *
+ * We interlock this using the same spinlock that is used to protect
+ * ShmemAlloc().  Interlocking is not really necessary during postmaster
+ * startup, but it is needed if any user-defined code tries to allocate
+ * LWLocks after startup.
+ */
+FlexLockId
+FlexLockAssign(char locktype)
+{
+	FlexLockId	result;
+
+	/* use volatile pointer to prevent code rearrangement */
+	volatile int *FlexLockCounter;
+
+	FlexLockCounter = (int *) ((char *) FlexLockArray - 2 * sizeof(int));
+	SpinLockAcquire(ShmemLock);
+	if (FlexLockCounter[0] >= FlexLockCounter[1])
+	{
+		SpinLockRelease(ShmemLock);
+		elog(ERROR, "no more FlexLockIds available");
+	}
+	result = (FlexLockId) (FlexLockCounter[0]++);
+	SpinLockRelease(ShmemLock);
+
+	FlexLockInit(&FlexLockArray[result].flex, locktype);
+
+	return result;
+}
+
+/*
+ * Initialize a FlexLock.
+ */
+static void
+FlexLockInit(FlexLock *flex, char locktype)
+{
+	SpinLockInit(&flex->mutex);
+	flex->releaseOK = true;
+	flex->locktype = locktype;
+	/*
+	 * We might need to think a little harder about what should happen here
+	 * if some future type of FlexLock requires more initialization than this.
+	 * For now, this will suffice.
+	 */
+}
+
+/*
+ * Add lock to list of locks held by this backend.
+ */
+void
+FlexLockRemember(FlexLockId id)
+{
+	if (num_held_flexlocks >= MAX_SIMUL_FLEXLOCKS)
+		elog(PANIC, "too many FlexLocks taken");
+	held_flexlocks[num_held_flexlocks++] = id;
+}
+
+/*
+ * Remove lock from list of locks held.  Usually, but not always, it will
+ * be the latest-acquired lock; so search array backwards.
+ */
+void
+FlexLockForget(FlexLockId id)
+{
+	int			i;
+
+	for (i = num_held_flexlocks; --i >= 0;)
+	{
+		if (id == held_flexlocks[i])
+			break;
+	}
+	if (i < 0)
+		elog(ERROR, "lock %d is not held", (int) id);
+	num_held_flexlocks--;
+	for (; i < num_held_flexlocks; i++)
+		held_flexlocks[i] = held_flexlocks[i + 1];
+}
+
+/*
+ * FlexLockWait - wait until awakened
+ *
+ * Since we share the process wait semaphore with the regular lock manager
+ * and ProcWaitForSignal, and we may need to acquire a FlexLock while one of
+ * those is pending, it is possible that we get awakened for a reason other
+ * than being signaled by a FlexLock release.  If so, loop back and wait again.
+ *
+ * Returns the number of "extra" waits absorbed so that, once we've gotten the
+ * FlexLock, we can 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.
+ */
+int
+FlexLockWait(FlexLockId id, int mode)
+{
+	int		extraWaits = 0;
+
+	FlexLockDebug("LWLockAcquire", id, "waiting");
+	TRACE_POSTGRESQL_FLEXLOCK_WAIT_START(id, mode);
+
+	for (;;)
+   	{
+		/* "false" means cannot accept cancel/die interrupt here. */
+		PGSemaphoreLock(&MyProc->sem, false);
+		/*
+		 * FLEXTODO: I think we should return this, instead of ignoring it.
+		 * Any non-zero value means "wake up".
+		 */
+		if (MyProc->flWaitResult)
+			break;
+		extraWaits++;
+   	}
+
+	TRACE_POSTGRESQL_FLEXLOCK_WAIT_DONE(id, mode);
+	FlexLockDebug("LWLockAcquire", id, "awakened");
+
+	return extraWaits;
+}
+
+/*
+ * FlexLockReleaseAll - release all currently-held locks
+ *
+ * Used to clean up after ereport(ERROR). An important difference between this
+ * function and retail LWLockRelease calls is that InterruptHoldoffCount is
+ * unchanged by this operation.  This is necessary since InterruptHoldoffCount
+ * has been set to an appropriate level earlier in error recovery. We could
+ * decrement it below zero if we allow it to drop for each released lock!
+ */
+void
+FlexLockReleaseAll(void)
+{
+	while (num_held_flexlocks > 0)
+	{
+		HOLD_INTERRUPTS();		/* match the upcoming RESUME_INTERRUPTS */
+
+		/*
+		 * FLEXTODO: When we have multiple types of flex locks, this will
+		 * need to call the appropriate release function for each lock type.
+		 */
+		LWLockRelease(held_flexlocks[num_held_flexlocks - 1]);
+	}
+}
+
+/*
+ * FlexLockHeldByMe - test whether my process currently holds a lock
+ *
+ * This is meant as debug support only.  We do not consider the lock mode.
+ */
+bool
+FlexLockHeldByMe(FlexLockId id)
+{
+	int			i;
+
+	for (i = 0; i < num_held_flexlocks; i++)
+	{
+		if (held_flexlocks[i] == id)
+			return true;
+	}
+	return false;
+}
diff --git a/src/backend/storage/lmgr/lock.c b/src/backend/storage/lmgr/lock.c
index 3ba4671..f594983 100644
--- a/src/backend/storage/lmgr/lock.c
+++ b/src/backend/storage/lmgr/lock.c
@@ -591,7 +591,7 @@ LockAcquireExtended(const LOCKTAG *locktag,
 	bool		found;
 	ResourceOwner owner;
 	uint32		hashcode;
-	LWLockId	partitionLock;
+	FlexLockId	partitionLock;
 	int			status;
 	bool		log_lock = false;
 
@@ -1546,7 +1546,7 @@ LockRelease(const LOCKTAG *locktag, LOCKMODE lockmode, bool sessionLock)
 	LOCALLOCK  *locallock;
 	LOCK	   *lock;
 	PROCLOCK   *proclock;
-	LWLockId	partitionLock;
+	FlexLockId	partitionLock;
 	bool		wakeupNeeded;
 
 	if (lockmethodid <= 0 || lockmethodid >= lengthof(LockMethods))
@@ -1912,7 +1912,7 @@ LockReleaseAll(LOCKMETHODID lockmethodid, bool allLocks)
 	 */
 	for (partition = 0; partition < NUM_LOCK_PARTITIONS; partition++)
 	{
-		LWLockId	partitionLock = FirstLockMgrLock + partition;
+		FlexLockId	partitionLock = FirstLockMgrLock + partition;
 		SHM_QUEUE  *procLocks = &(MyProc->myProcLocks[partition]);
 
 		proclock = (PROCLOCK *) SHMQueueNext(procLocks, procLocks,
@@ -2197,7 +2197,7 @@ static bool
 FastPathTransferRelationLocks(LockMethod lockMethodTable, const LOCKTAG *locktag,
 					  uint32 hashcode)
 {
-	LWLockId		partitionLock = LockHashPartitionLock(hashcode);
+	FlexLockId		partitionLock = LockHashPartitionLock(hashcode);
 	Oid				relid = locktag->locktag_field2;
 	uint32			i;
 
@@ -2281,7 +2281,7 @@ FastPathGetRelationLockEntry(LOCALLOCK *locallock)
 	LockMethod		lockMethodTable = LockMethods[DEFAULT_LOCKMETHOD];
 	LOCKTAG		   *locktag = &locallock->tag.lock;
 	PROCLOCK	   *proclock = NULL;
-	LWLockId		partitionLock = LockHashPartitionLock(locallock->hashcode);
+	FlexLockId		partitionLock = LockHashPartitionLock(locallock->hashcode);
 	Oid				relid = locktag->locktag_field2;
 	uint32			f;
 
@@ -2382,7 +2382,7 @@ GetLockConflicts(const LOCKTAG *locktag, LOCKMODE lockmode)
 	SHM_QUEUE  *procLocks;
 	PROCLOCK   *proclock;
 	uint32		hashcode;
-	LWLockId	partitionLock;
+	FlexLockId	partitionLock;
 	int			count = 0;
 	int			fast_count = 0;
 
@@ -2593,7 +2593,7 @@ LockRefindAndRelease(LockMethod lockMethodTable, PGPROC *proc,
 	PROCLOCKTAG proclocktag;
 	uint32		hashcode;
 	uint32		proclock_hashcode;
-	LWLockId	partitionLock;
+	FlexLockId	partitionLock;
 	bool		wakeupNeeded;
 
 	hashcode = LockTagHashCode(locktag);
@@ -2827,7 +2827,7 @@ PostPrepare_Locks(TransactionId xid)
 	 */
 	for (partition = 0; partition < NUM_LOCK_PARTITIONS; partition++)
 	{
-		LWLockId	partitionLock = FirstLockMgrLock + partition;
+		FlexLockId	partitionLock = FirstLockMgrLock + partition;
 		SHM_QUEUE  *procLocks = &(MyProc->myProcLocks[partition]);
 
 		proclock = (PROCLOCK *) SHMQueueNext(procLocks, procLocks,
@@ -3343,7 +3343,7 @@ lock_twophase_recover(TransactionId xid, uint16 info,
 	uint32		hashcode;
 	uint32		proclock_hashcode;
 	int			partition;
-	LWLockId	partitionLock;
+	FlexLockId	partitionLock;
 	LockMethod	lockMethodTable;
 
 	Assert(len == sizeof(TwoPhaseLockRecord));
diff --git a/src/backend/storage/lmgr/lwlock.c b/src/backend/storage/lmgr/lwlock.c
index 079eb29..ce6c931 100644
--- a/src/backend/storage/lmgr/lwlock.c
+++ b/src/backend/storage/lmgr/lwlock.c
@@ -21,74 +21,23 @@
  */
 #include "postgres.h"
 
-#include "access/clog.h"
-#include "access/multixact.h"
-#include "access/subtrans.h"
-#include "commands/async.h"
 #include "miscadmin.h"
 #include "pg_trace.h"
+#include "storage/flexlock_internals.h"
 #include "storage/ipc.h"
-#include "storage/predicate.h"
 #include "storage/proc.h"
 #include "storage/spin.h"
 
-
-/* We use the ShmemLock spinlock to protect LWLockAssign */
-extern slock_t *ShmemLock;
-
-
 typedef struct LWLock
 {
-	slock_t		mutex;			/* Protects LWLock and queue of PGPROCs */
-	bool		releaseOK;		/* T if ok to release waiters */
+	FlexLock	flex;			/* common FlexLock infrastructure */
 	char		exclusive;		/* # of exclusive holders (0 or 1) */
 	int			shared;			/* # of shared holders (0..MaxBackends) */
-	PGPROC	   *head;			/* head of list of waiting PGPROCs */
-	PGPROC	   *tail;			/* tail of list of waiting PGPROCs */
-	/* tail is undefined when head is NULL */
 } LWLock;
 
-/*
- * All the LWLock structs are allocated as an array in shared memory.
- * (LWLockIds are indexes into the array.)	We force the array stride to
- * be a power of 2, which saves a few cycles in indexing, but more
- * importantly also ensures that individual LWLocks don't cross cache line
- * boundaries.	This reduces cache contention problems, especially on AMD
- * Opterons.  (Of course, we have to also ensure that the array start
- * address is suitably aligned.)
- *
- * LWLock is between 16 and 32 bytes on all known platforms, so these two
- * cases are sufficient.
- */
-#define LWLOCK_PADDED_SIZE	(sizeof(LWLock) <= 16 ? 16 : 32)
-
-typedef union LWLockPadded
-{
-	LWLock		lock;
-	char		pad[LWLOCK_PADDED_SIZE];
-} LWLockPadded;
-
-/*
- * This points to the array of LWLocks in shared memory.  Backends inherit
- * the pointer by fork from the postmaster (except in the EXEC_BACKEND case,
- * where we have special measures to pass it down).
- */
-NON_EXEC_STATIC LWLockPadded *LWLockArray = NULL;
-
-
-/*
- * We use this structure to keep track of locked LWLocks for release
- * during error recovery.  The maximum size could be determined at runtime
- * if necessary, but it seems unlikely that more than a few locks could
- * ever be held simultaneously.
- */
-#define MAX_SIMUL_LWLOCKS	100
-
-static int	num_held_lwlocks = 0;
-static LWLockId held_lwlocks[MAX_SIMUL_LWLOCKS];
-
-static int	lock_addin_request = 0;
-static bool lock_addin_request_allowed = true;
+#define	LWLockPointer(lockid) \
+	(AssertMacro(FlexLockArray[lockid].flex.locktype == FLEXLOCK_TYPE_LWLOCK), \
+	 (volatile LWLock *) &FlexLockArray[lockid])
 
 #ifdef LWLOCK_STATS
 static int	counts_for_pid = 0;
@@ -98,27 +47,17 @@ static int *block_counts;
 #endif
 
 #ifdef LOCK_DEBUG
-bool		Trace_lwlocks = false;
-
 inline static void
-PRINT_LWDEBUG(const char *where, LWLockId lockid, const volatile LWLock *lock)
+PRINT_LWDEBUG(const char *where, FlexLockId lockid, const volatile LWLock *lock)
 {
-	if (Trace_lwlocks)
+	if (Trace_flexlocks)
 		elog(LOG, "%s(%d): excl %d shared %d head %p rOK %d",
 			 where, (int) lockid,
-			 (int) lock->exclusive, lock->shared, lock->head,
-			 (int) lock->releaseOK);
-}
-
-inline static void
-LOG_LWDEBUG(const char *where, LWLockId lockid, const char *msg)
-{
-	if (Trace_lwlocks)
-		elog(LOG, "%s(%d): %s", where, (int) lockid, msg);
+			 (int) lock->exclusive, lock->shared, lock->flex.head,
+			 (int) lock->flex.releaseOK);
 }
 #else							/* not LOCK_DEBUG */
 #define PRINT_LWDEBUG(a,b,c)
-#define LOG_LWDEBUG(a,b,c)
 #endif   /* LOCK_DEBUG */
 
 #ifdef LWLOCK_STATS
@@ -127,8 +66,8 @@ static void
 print_lwlock_stats(int code, Datum arg)
 {
 	int			i;
-	int		   *LWLockCounter = (int *) ((char *) LWLockArray - 2 * sizeof(int));
-	int			numLocks = LWLockCounter[1];
+	int		   *FlexLockCounter = (int *) ((char *) FlexLockArray - 2 * sizeof(int));
+	int			numLocks = FlexLockCounter[1];
 
 	/* Grab an LWLock to keep different backends from mixing reports */
 	LWLockAcquire(0, LW_EXCLUSIVE);
@@ -145,173 +84,15 @@ print_lwlock_stats(int code, Datum arg)
 }
 #endif   /* LWLOCK_STATS */
 
-
 /*
- * Compute number of LWLocks to allocate.
+ * LWLockAssign - initialize a new lwlock and return its ID
  */
-int
-NumLWLocks(void)
-{
-	int			numLocks;
-
-	/*
-	 * Possibly this logic should be spread out among the affected modules,
-	 * the same way that shmem space estimation is done.  But for now, there
-	 * are few enough users of LWLocks that we can get away with just keeping
-	 * the knowledge here.
-	 */
-
-	/* Predefined LWLocks */
-	numLocks = (int) NumFixedLWLocks;
-
-	/* bufmgr.c needs two for each shared buffer */
-	numLocks += 2 * NBuffers;
-
-	/* proc.c needs one for each backend or auxiliary process */
-	numLocks += MaxBackends + NUM_AUXILIARY_PROCS;
-
-	/* clog.c needs one per CLOG buffer */
-	numLocks += NUM_CLOG_BUFFERS;
-
-	/* subtrans.c needs one per SubTrans buffer */
-	numLocks += NUM_SUBTRANS_BUFFERS;
-
-	/* multixact.c needs two SLRU areas */
-	numLocks += NUM_MXACTOFFSET_BUFFERS + NUM_MXACTMEMBER_BUFFERS;
-
-	/* async.c needs one per Async buffer */
-	numLocks += NUM_ASYNC_BUFFERS;
-
-	/* predicate.c needs one per old serializable xid buffer */
-	numLocks += NUM_OLDSERXID_BUFFERS;
-
-	/*
-	 * Add any requested by loadable modules; for backwards-compatibility
-	 * reasons, allocate at least NUM_USER_DEFINED_LWLOCKS of them even if
-	 * there are no explicit requests.
-	 */
-	lock_addin_request_allowed = false;
-	numLocks += Max(lock_addin_request, NUM_USER_DEFINED_LWLOCKS);
-
-	return numLocks;
-}
-
-
-/*
- * RequestAddinLWLocks
- *		Request that extra LWLocks be allocated for use by
- *		a loadable module.
- *
- * This is only useful if called from the _PG_init hook of a library that
- * is loaded into the postmaster via shared_preload_libraries.	Once
- * shared memory has been allocated, calls will be ignored.  (We could
- * raise an error, but it seems better to make it a no-op, so that
- * libraries containing such calls can be reloaded if needed.)
- */
-void
-RequestAddinLWLocks(int n)
-{
-	if (IsUnderPostmaster || !lock_addin_request_allowed)
-		return;					/* too late */
-	lock_addin_request += n;
-}
-
-
-/*
- * Compute shmem space needed for LWLocks.
- */
-Size
-LWLockShmemSize(void)
-{
-	Size		size;
-	int			numLocks = NumLWLocks();
-
-	/* Space for the LWLock array. */
-	size = mul_size(numLocks, sizeof(LWLockPadded));
-
-	/* Space for dynamic allocation counter, plus room for alignment. */
-	size = add_size(size, 2 * sizeof(int) + LWLOCK_PADDED_SIZE);
-
-	return size;
-}
-
-
-/*
- * Allocate shmem space for LWLocks and initialize the locks.
- */
-void
-CreateLWLocks(void)
-{
-	int			numLocks = NumLWLocks();
-	Size		spaceLocks = LWLockShmemSize();
-	LWLockPadded *lock;
-	int		   *LWLockCounter;
-	char	   *ptr;
-	int			id;
-
-	/* Allocate space */
-	ptr = (char *) ShmemAlloc(spaceLocks);
-
-	/* Leave room for dynamic allocation counter */
-	ptr += 2 * sizeof(int);
-
-	/* Ensure desired alignment of LWLock array */
-	ptr += LWLOCK_PADDED_SIZE - ((uintptr_t) ptr) % LWLOCK_PADDED_SIZE;
-
-	LWLockArray = (LWLockPadded *) ptr;
-
-	/*
-	 * Initialize all LWLocks to "unlocked" state
-	 */
-	for (id = 0, lock = LWLockArray; id < numLocks; id++, lock++)
-	{
-		SpinLockInit(&lock->lock.mutex);
-		lock->lock.releaseOK = true;
-		lock->lock.exclusive = 0;
-		lock->lock.shared = 0;
-		lock->lock.head = NULL;
-		lock->lock.tail = NULL;
-	}
-
-	/*
-	 * Initialize the dynamic-allocation counter, which is stored just before
-	 * the first LWLock.
-	 */
-	LWLockCounter = (int *) ((char *) LWLockArray - 2 * sizeof(int));
-	LWLockCounter[0] = (int) NumFixedLWLocks;
-	LWLockCounter[1] = numLocks;
-}
-
-
-/*
- * LWLockAssign - assign a dynamically-allocated LWLock number
- *
- * We interlock this using the same spinlock that is used to protect
- * ShmemAlloc().  Interlocking is not really necessary during postmaster
- * startup, but it is needed if any user-defined code tries to allocate
- * LWLocks after startup.
- */
-LWLockId
+FlexLockId
 LWLockAssign(void)
 {
-	LWLockId	result;
-
-	/* use volatile pointer to prevent code rearrangement */
-	volatile int *LWLockCounter;
-
-	LWLockCounter = (int *) ((char *) LWLockArray - 2 * sizeof(int));
-	SpinLockAcquire(ShmemLock);
-	if (LWLockCounter[0] >= LWLockCounter[1])
-	{
-		SpinLockRelease(ShmemLock);
-		elog(ERROR, "no more LWLockIds available");
-	}
-	result = (LWLockId) (LWLockCounter[0]++);
-	SpinLockRelease(ShmemLock);
-	return result;
+	return FlexLockAssign(FLEXLOCK_TYPE_LWLOCK);
 }
 
-
 /*
  * LWLockAcquire - acquire a lightweight lock in the specified mode
  *
@@ -320,9 +101,9 @@ LWLockAssign(void)
  * Side effect: cancel/die interrupts are held off until lock release.
  */
 void
-LWLockAcquire(LWLockId lockid, LWLockMode mode)
+LWLockAcquire(FlexLockId lockid, LWLockMode mode)
 {
-	volatile LWLock *lock = &(LWLockArray[lockid].lock);
+	volatile LWLock *lock = LWLockPointer(lockid);
 	PGPROC	   *proc = MyProc;
 	bool		retry = false;
 	int			extraWaits = 0;
@@ -333,8 +114,8 @@ LWLockAcquire(LWLockId lockid, LWLockMode mode)
 	/* 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];
+		int		   *FlexLockCounter = (int *) ((char *) FlexLockArray - 2 * sizeof(int));
+		int			numLocks = FlexLockCounter[1];
 
 		sh_acquire_counts = calloc(numLocks, sizeof(int));
 		ex_acquire_counts = calloc(numLocks, sizeof(int));
@@ -356,10 +137,6 @@ LWLockAcquire(LWLockId lockid, LWLockMode mode)
 	 */
 	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
@@ -388,11 +165,11 @@ LWLockAcquire(LWLockId lockid, LWLockMode mode)
 		bool		mustwait;
 
 		/* Acquire mutex.  Time spent holding mutex should be short! */
-		SpinLockAcquire(&lock->mutex);
+		SpinLockAcquire(&lock->flex.mutex);
 
 		/* If retrying, allow LWLockRelease to release waiters again */
 		if (retry)
-			lock->releaseOK = true;
+			lock->flex.releaseOK = true;
 
 		/* If I can get the lock, do so quickly. */
 		if (mode == LW_EXCLUSIVE)
@@ -419,72 +196,30 @@ LWLockAcquire(LWLockId lockid, LWLockMode mode)
 		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;
+		/* Add myself to wait queue. */
+		FlexLockJoinWaitQueue(lock, (int) mode);
 
 		/* 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");
+		SpinLockRelease(&lock->flex.mutex);
+
+		/* Wait until awakened. */
+		extraWaits += FlexLockWait(lockid, mode);
 
 #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);
+	SpinLockRelease(&lock->flex.mutex);
 
-	TRACE_POSTGRESQL_LWLOCK_ACQUIRE(lockid, mode);
+	TRACE_POSTGRESQL_FLEXLOCK_ACQUIRE(lockid, mode);
 
 	/* Add lock to list of locks held by this backend */
-	held_lwlocks[num_held_lwlocks++] = lockid;
+	FlexLockRemember(lockid);
 
 	/*
 	 * Fix the process wait semaphore's count for any absorbed wakeups.
@@ -501,17 +236,13 @@ LWLockAcquire(LWLockId lockid, LWLockMode mode)
  * If successful, cancel/die interrupts are held off until lock release.
  */
 bool
-LWLockConditionalAcquire(LWLockId lockid, LWLockMode mode)
+LWLockConditionalAcquire(FlexLockId lockid, LWLockMode mode)
 {
-	volatile LWLock *lock = &(LWLockArray[lockid].lock);
+	volatile LWLock *lock = LWLockPointer(lockid);
 	bool		mustwait;
 
 	PRINT_LWDEBUG("LWLockConditionalAcquire", lockid, lock);
 
-	/* 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
@@ -520,7 +251,7 @@ LWLockConditionalAcquire(LWLockId lockid, LWLockMode mode)
 	HOLD_INTERRUPTS();
 
 	/* Acquire mutex.  Time spent holding mutex should be short! */
-	SpinLockAcquire(&lock->mutex);
+	SpinLockAcquire(&lock->flex.mutex);
 
 	/* If I can get the lock, do so quickly. */
 	if (mode == LW_EXCLUSIVE)
@@ -545,20 +276,20 @@ LWLockConditionalAcquire(LWLockId lockid, LWLockMode mode)
 	}
 
 	/* We are done updating shared state of the lock itself. */
-	SpinLockRelease(&lock->mutex);
+	SpinLockRelease(&lock->flex.mutex);
 
 	if (mustwait)
 	{
 		/* Failed to get lock, so release interrupt holdoff */
 		RESUME_INTERRUPTS();
-		LOG_LWDEBUG("LWLockConditionalAcquire", lockid, "failed");
-		TRACE_POSTGRESQL_LWLOCK_CONDACQUIRE_FAIL(lockid, mode);
+		FlexLockDebug("LWLockConditionalAcquire", lockid, "failed");
+		TRACE_POSTGRESQL_FLEXLOCK_CONDACQUIRE_FAIL(lockid, mode);
 	}
 	else
 	{
 		/* Add lock to list of locks held by this backend */
-		held_lwlocks[num_held_lwlocks++] = lockid;
-		TRACE_POSTGRESQL_LWLOCK_CONDACQUIRE(lockid, mode);
+		FlexLockRemember(lockid);
+		TRACE_POSTGRESQL_FLEXLOCK_CONDACQUIRE(lockid, mode);
 	}
 
 	return !mustwait;
@@ -568,32 +299,18 @@ LWLockConditionalAcquire(LWLockId lockid, LWLockMode mode)
  * LWLockRelease - release a previously acquired lock
  */
 void
-LWLockRelease(LWLockId lockid)
+LWLockRelease(FlexLockId lockid)
 {
-	volatile LWLock *lock = &(LWLockArray[lockid].lock);
+	volatile LWLock *lock = LWLockPointer(lockid);
 	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];
+	FlexLockForget(lockid);
 
 	/* Acquire mutex.  Time spent holding mutex should be short! */
-	SpinLockAcquire(&lock->mutex);
+	SpinLockAcquire(&lock->flex.mutex);
 
 	/* Release my hold on lock */
 	if (lock->exclusive > 0)
@@ -610,10 +327,10 @@ LWLockRelease(LWLockId lockid)
 	 * if someone has already awakened waiters that haven't yet acquired the
 	 * lock.
 	 */
-	head = lock->head;
+	head = lock->flex.head;
 	if (head != NULL)
 	{
-		if (lock->exclusive == 0 && lock->shared == 0 && lock->releaseOK)
+		if (lock->exclusive == 0 && lock->shared == 0 && lock->flex.releaseOK)
 		{
 			/*
 			 * Remove the to-be-awakened PGPROCs from the queue.  If the front
@@ -621,17 +338,17 @@ LWLockRelease(LWLockId lockid)
 			 * as many waiters as want shared access.
 			 */
 			proc = head;
-			if (!proc->lwExclusive)
+			if (proc->flWaitMode != LW_EXCLUSIVE)
 			{
-				while (proc->lwWaitLink != NULL &&
-					   !proc->lwWaitLink->lwExclusive)
-					proc = proc->lwWaitLink;
+				while (proc->flWaitLink != NULL &&
+					   proc->flWaitLink->flWaitMode != LW_EXCLUSIVE)
+					proc = proc->flWaitLink;
 			}
 			/* proc is now the last PGPROC to be released */
-			lock->head = proc->lwWaitLink;
-			proc->lwWaitLink = NULL;
+			lock->flex.head = proc->flWaitLink;
+			proc->flWaitLink = NULL;
 			/* prevent additional wakeups until retryer gets to run */
-			lock->releaseOK = false;
+			lock->flex.releaseOK = false;
 		}
 		else
 		{
@@ -641,20 +358,20 @@ LWLockRelease(LWLockId lockid)
 	}
 
 	/* We are done updating shared state of the lock itself. */
-	SpinLockRelease(&lock->mutex);
+	SpinLockRelease(&lock->flex.mutex);
 
-	TRACE_POSTGRESQL_LWLOCK_RELEASE(lockid);
+	TRACE_POSTGRESQL_FLEXLOCK_RELEASE(lockid);
 
 	/*
 	 * Awaken any waiters I removed from the queue.
 	 */
 	while (head != NULL)
 	{
-		LOG_LWDEBUG("LWLockRelease", lockid, "release waiter");
+		FlexLockDebug("LWLockRelease", lockid, "release waiter");
 		proc = head;
-		head = proc->lwWaitLink;
-		proc->lwWaitLink = NULL;
-		proc->lwWaiting = false;
+		head = proc->flWaitLink;
+		proc->flWaitLink = NULL;
+		proc->flWaitResult = 1;		/* any non-zero value will do */
 		PGSemaphoreUnlock(&proc->sem);
 	}
 
@@ -664,43 +381,17 @@ LWLockRelease(LWLockId lockid)
 	RESUME_INTERRUPTS();
 }
 
-
-/*
- * LWLockReleaseAll - release all currently-held locks
- *
- * Used to clean up after ereport(ERROR). An important difference between this
- * function and retail LWLockRelease calls is that InterruptHoldoffCount is
- * unchanged by this operation.  This is necessary since InterruptHoldoffCount
- * has been set to an appropriate level earlier in error recovery. We could
- * decrement it below zero if we allow it to drop for each released lock!
- */
-void
-LWLockReleaseAll(void)
-{
-	while (num_held_lwlocks > 0)
-	{
-		HOLD_INTERRUPTS();		/* match the upcoming RESUME_INTERRUPTS */
-
-		LWLockRelease(held_lwlocks[num_held_lwlocks - 1]);
-	}
-}
-
-
 /*
  * LWLockHeldByMe - test whether my process currently holds a lock
  *
- * This is meant as debug support only.  We do not distinguish whether the
- * lock is held shared or exclusive.
+ * The following convenience routine might not be worthwhile but for the fact
+ * that we've had a function by this name since long before FlexLocks existed.
+ * Callers who want to check whether an arbitrary FlexLock (that may or may not
+ * be an LWLock) is held can use FlexLockHeldByMe directly.
  */
 bool
-LWLockHeldByMe(LWLockId lockid)
+LWLockHeldByMe(FlexLockId lockid)
 {
-	int			i;
-
-	for (i = 0; i < num_held_lwlocks; i++)
-	{
-		if (held_lwlocks[i] == lockid)
-			return true;
-	}
-	return false;
+	AssertMacro(FlexLockArray[lockid].flex.locktype == FLEXLOCK_TYPE_LWLOCK);
+	return FlexLockHeldByMe(lockid);
 }
diff --git a/src/backend/storage/lmgr/predicate.c b/src/backend/storage/lmgr/predicate.c
index 345f6f5..15978a4 100644
--- a/src/backend/storage/lmgr/predicate.c
+++ b/src/backend/storage/lmgr/predicate.c
@@ -239,7 +239,7 @@
 #define PredicateLockHashPartition(hashcode) \
 	((hashcode) % NUM_PREDICATELOCK_PARTITIONS)
 #define PredicateLockHashPartitionLock(hashcode) \
-	((LWLockId) (FirstPredicateLockMgrLock + PredicateLockHashPartition(hashcode)))
+	((FlexLockId) (FirstPredicateLockMgrLock + PredicateLockHashPartition(hashcode)))
 
 #define NPREDICATELOCKTARGETENTS() \
 	mul_size(max_predicate_locks_per_xact, add_size(MaxBackends, max_prepared_xacts))
@@ -1840,7 +1840,7 @@ PageIsPredicateLocked(Relation relation, BlockNumber blkno)
 {
 	PREDICATELOCKTARGETTAG targettag;
 	uint32		targettaghash;
-	LWLockId	partitionLock;
+	FlexLockId	partitionLock;
 	PREDICATELOCKTARGET *target;
 
 	SET_PREDICATELOCKTARGETTAG_PAGE(targettag,
@@ -2073,7 +2073,7 @@ DeleteChildTargetLocks(const PREDICATELOCKTARGETTAG *newtargettag)
 		if (TargetTagIsCoveredBy(oldtargettag, *newtargettag))
 		{
 			uint32		oldtargettaghash;
-			LWLockId	partitionLock;
+			FlexLockId	partitionLock;
 			PREDICATELOCK *rmpredlock;
 
 			oldtargettaghash = PredicateLockTargetTagHashCode(&oldtargettag);
@@ -2285,7 +2285,7 @@ CreatePredicateLock(const PREDICATELOCKTARGETTAG *targettag,
 	PREDICATELOCKTARGET *target;
 	PREDICATELOCKTAG locktag;
 	PREDICATELOCK *lock;
-	LWLockId	partitionLock;
+	FlexLockId	partitionLock;
 	bool		found;
 
 	partitionLock = PredicateLockHashPartitionLock(targettaghash);
@@ -2586,10 +2586,10 @@ TransferPredicateLocksToNewTarget(PREDICATELOCKTARGETTAG oldtargettag,
 								  bool removeOld)
 {
 	uint32		oldtargettaghash;
-	LWLockId	oldpartitionLock;
+	FlexLockId	oldpartitionLock;
 	PREDICATELOCKTARGET *oldtarget;
 	uint32		newtargettaghash;
-	LWLockId	newpartitionLock;
+	FlexLockId	newpartitionLock;
 	bool		found;
 	bool		outOfShmem = false;
 
@@ -3578,7 +3578,7 @@ ClearOldPredicateLocks(void)
 			PREDICATELOCKTARGET *target;
 			PREDICATELOCKTARGETTAG targettag;
 			uint32		targettaghash;
-			LWLockId	partitionLock;
+			FlexLockId	partitionLock;
 
 			tag = predlock->tag;
 			target = tag.myTarget;
@@ -3656,7 +3656,7 @@ ReleaseOneSerializableXact(SERIALIZABLEXACT *sxact, bool partial,
 		PREDICATELOCKTARGET *target;
 		PREDICATELOCKTARGETTAG targettag;
 		uint32		targettaghash;
-		LWLockId	partitionLock;
+		FlexLockId	partitionLock;
 
 		nextpredlock = (PREDICATELOCK *)
 			SHMQueueNext(&(sxact->predicateLocks),
@@ -4034,7 +4034,7 @@ static void
 CheckTargetForConflictsIn(PREDICATELOCKTARGETTAG *targettag)
 {
 	uint32		targettaghash;
-	LWLockId	partitionLock;
+	FlexLockId	partitionLock;
 	PREDICATELOCKTARGET *target;
 	PREDICATELOCK *predlock;
 	PREDICATELOCK *mypredlock = NULL;
diff --git a/src/backend/storage/lmgr/proc.c b/src/backend/storage/lmgr/proc.c
index bcbc802..db01e9d 100644
--- a/src/backend/storage/lmgr/proc.c
+++ b/src/backend/storage/lmgr/proc.c
@@ -360,9 +360,9 @@ InitProcess(void)
 	/* NB -- autovac launcher intentionally does not set IS_AUTOVACUUM */
 	if (IsAutoVacuumWorkerProcess())
 		MyPgXact->vacuumFlags |= PROC_IS_AUTOVACUUM;
-	MyProc->lwWaiting = false;
-	MyProc->lwExclusive = false;
-	MyProc->lwWaitLink = NULL;
+	MyProc->flWaitResult = 0;
+	MyProc->flWaitMode = 0;
+	MyProc->flWaitLink = NULL;
 	MyProc->waitLock = NULL;
 	MyProc->waitProcLock = NULL;
 #ifdef USE_ASSERT_CHECKING
@@ -515,9 +515,9 @@ InitAuxiliaryProcess(void)
 	MyProc->roleId = InvalidOid;
 	MyPgXact->inCommit = false;
 	MyPgXact->vacuumFlags = 0;
-	MyProc->lwWaiting = false;
-	MyProc->lwExclusive = false;
-	MyProc->lwWaitLink = NULL;
+	MyProc->flWaitMode = 0;
+	MyProc->flWaitResult = 0;
+	MyProc->flWaitLink = NULL;
 	MyProc->waitLock = NULL;
 	MyProc->waitProcLock = NULL;
 #ifdef USE_ASSERT_CHECKING
@@ -643,7 +643,7 @@ IsWaitingForLock(void)
 void
 LockWaitCancel(void)
 {
-	LWLockId	partitionLock;
+	FlexLockId	partitionLock;
 
 	/* Nothing to do if we weren't waiting for a lock */
 	if (lockAwaited == NULL)
@@ -754,11 +754,11 @@ ProcKill(int code, Datum arg)
 #endif
 
 	/*
-	 * Release any LW locks I am holding.  There really shouldn't be any, but
-	 * it's cheap to check again before we cut the knees off the LWLock
+	 * Release any felx locks I am holding.  There really shouldn't be any, but
+	 * it's cheap to check again before we cut the knees off the flex lock
 	 * facility by releasing our PGPROC ...
 	 */
-	LWLockReleaseAll();
+	FlexLockReleaseAll();
 
 	/* Release ownership of the process's latch, too */
 	DisownLatch(&MyProc->procLatch);
@@ -815,8 +815,8 @@ AuxiliaryProcKill(int code, Datum arg)
 
 	Assert(MyProc == auxproc);
 
-	/* Release any LW locks I am holding (see notes above) */
-	LWLockReleaseAll();
+	/* Release any flex locks I am holding (see notes above) */
+	FlexLockReleaseAll();
 
 	/* Release ownership of the process's latch, too */
 	DisownLatch(&MyProc->procLatch);
@@ -901,7 +901,7 @@ ProcSleep(LOCALLOCK *locallock, LockMethod lockMethodTable)
 	LOCK	   *lock = locallock->lock;
 	PROCLOCK   *proclock = locallock->proclock;
 	uint32		hashcode = locallock->hashcode;
-	LWLockId	partitionLock = LockHashPartitionLock(hashcode);
+	FlexLockId	partitionLock = LockHashPartitionLock(hashcode);
 	PROC_QUEUE *waitQueue = &(lock->waitProcs);
 	LOCKMASK	myHeldLocks = MyProc->heldLocks;
 	bool		early_deadlock = false;
diff --git a/src/backend/utils/misc/check_guc b/src/backend/utils/misc/check_guc
index 293fb03..1a19e36 100755
--- a/src/backend/utils/misc/check_guc
+++ b/src/backend/utils/misc/check_guc
@@ -19,7 +19,7 @@
 INTENTIONALLY_NOT_INCLUDED="autocommit debug_deadlocks \
 is_superuser lc_collate lc_ctype lc_messages lc_monetary lc_numeric lc_time \
 pre_auth_delay role seed server_encoding server_version server_version_int \
-session_authorization trace_lock_oidmin trace_lock_table trace_locks trace_lwlocks \
+session_authorization trace_lock_oidmin trace_lock_table trace_locks trace_flexlocks \
 trace_notify trace_userlocks transaction_isolation transaction_read_only \
 zero_damaged_pages"
 
diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c
index da7b6d4..52de233 100644
--- a/src/backend/utils/misc/guc.c
+++ b/src/backend/utils/misc/guc.c
@@ -59,6 +59,7 @@
 #include "replication/walreceiver.h"
 #include "replication/walsender.h"
 #include "storage/bufmgr.h"
+#include "storage/flexlock_internals.h"
 #include "storage/standby.h"
 #include "storage/fd.h"
 #include "storage/predicate.h"
@@ -1071,12 +1072,12 @@ static struct config_bool ConfigureNamesBool[] =
 		NULL, NULL, NULL
 	},
 	{
-		{"trace_lwlocks", PGC_SUSET, DEVELOPER_OPTIONS,
+		{"trace_flexlocks", PGC_SUSET, DEVELOPER_OPTIONS,
 			gettext_noop("No description available."),
 			NULL,
 			GUC_NOT_IN_SAMPLE
 		},
-		&Trace_lwlocks,
+		&Trace_flexlocks,
 		false,
 		NULL, NULL, NULL
 	},
diff --git a/src/backend/utils/probes.d b/src/backend/utils/probes.d
index 71c5ab0..5b9cfe6 100644
--- a/src/backend/utils/probes.d
+++ b/src/backend/utils/probes.d
@@ -15,8 +15,8 @@
  * in probe definitions, as they cause compilation errors on Mac OS X 10.5.
  */
 #define LocalTransactionId unsigned int
-#define LWLockId int
-#define LWLockMode int
+#define FlexLockId int
+#define FlexLockMode int
 #define LOCKMODE int
 #define BlockNumber unsigned int
 #define Oid unsigned int
@@ -29,12 +29,12 @@ provider postgresql {
 	probe transaction__commit(LocalTransactionId);
 	probe transaction__abort(LocalTransactionId);
 
-	probe lwlock__acquire(LWLockId, LWLockMode);
-	probe lwlock__release(LWLockId);
-	probe lwlock__wait__start(LWLockId, LWLockMode);
-	probe lwlock__wait__done(LWLockId, LWLockMode);
-	probe lwlock__condacquire(LWLockId, LWLockMode);
-	probe lwlock__condacquire__fail(LWLockId, LWLockMode);
+	probe flexlock__acquire(FlexLockId, FlexLockMode);
+	probe flexlock__release(FlexLockId);
+	probe flexlock__wait__start(FlexLockId, FlexLockMode);
+	probe flexlock__wait__done(FlexLockId, FlexLockMode);
+	probe flexlock__condacquire(FlexLockId, FlexLockMode);
+	probe flexlock__condacquire__fail(FlexLockId, FlexLockMode);
 
 	probe lock__wait__start(unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, LOCKMODE);
 	probe lock__wait__done(unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, LOCKMODE);
diff --git a/src/include/access/slru.h b/src/include/access/slru.h
index e48743f..680a87f 100644
--- a/src/include/access/slru.h
+++ b/src/include/access/slru.h
@@ -55,7 +55,7 @@ typedef enum
  */
 typedef struct SlruSharedData
 {
-	LWLockId	ControlLock;
+	FlexLockId	ControlLock;
 
 	/* Number of buffers managed by this SLRU structure */
 	int			num_slots;
@@ -69,7 +69,7 @@ typedef struct SlruSharedData
 	bool	   *page_dirty;
 	int		   *page_number;
 	int		   *page_lru_count;
-	LWLockId   *buffer_locks;
+	FlexLockId *buffer_locks;
 
 	/*
 	 * Optional array of WAL flush LSNs associated with entries in the SLRU
@@ -136,7 +136,7 @@ typedef SlruCtlData *SlruCtl;
 
 extern Size SimpleLruShmemSize(int nslots, int nlsns);
 extern void SimpleLruInit(SlruCtl ctl, const char *name, int nslots, int nlsns,
-			  LWLockId ctllock, const char *subdir);
+			  FlexLockId ctllock, const char *subdir);
 extern int	SimpleLruZeroPage(SlruCtl ctl, int pageno);
 extern int SimpleLruReadPage(SlruCtl ctl, int pageno, bool write_ok,
 				  TransactionId xid);
diff --git a/src/include/pg_config_manual.h b/src/include/pg_config_manual.h
index 6c8e312..d3b74db 100644
--- a/src/include/pg_config_manual.h
+++ b/src/include/pg_config_manual.h
@@ -49,9 +49,9 @@
 #define SEQ_MINVALUE	(-SEQ_MAXVALUE)
 
 /*
- * Number of spare LWLocks to allocate for user-defined add-on code.
+ * Number of spare FlexLocks to allocate for user-defined add-on code.
  */
-#define NUM_USER_DEFINED_LWLOCKS	4
+#define NUM_USER_DEFINED_FLEXLOCKS	4
 
 /*
  * Define this if you want to allow the lo_import and lo_export SQL
diff --git a/src/include/storage/buf_internals.h b/src/include/storage/buf_internals.h
index b7d4ea5..ac7f665 100644
--- a/src/include/storage/buf_internals.h
+++ b/src/include/storage/buf_internals.h
@@ -103,7 +103,7 @@ typedef struct buftag
 #define BufTableHashPartition(hashcode) \
 	((hashcode) % NUM_BUFFER_PARTITIONS)
 #define BufMappingPartitionLock(hashcode) \
-	((LWLockId) (FirstBufMappingLock + BufTableHashPartition(hashcode)))
+	((FlexLockId) (FirstBufMappingLock + BufTableHashPartition(hashcode)))
 
 /*
  *	BufferDesc -- shared descriptor/state data for a single shared buffer.
@@ -143,8 +143,8 @@ typedef struct sbufdesc
 	int			buf_id;			/* buffer's index number (from 0) */
 	int			freeNext;		/* link in freelist chain */
 
-	LWLockId	io_in_progress_lock;	/* to wait for I/O to complete */
-	LWLockId	content_lock;	/* to lock access to buffer contents */
+	FlexLockId	io_in_progress_lock;	/* to wait for I/O to complete */
+	FlexLockId	content_lock;	/* to lock access to buffer contents */
 } BufferDesc;
 
 #define BufferDescriptorGetBuffer(bdesc) ((bdesc)->buf_id + 1)
diff --git a/src/include/storage/flexlock.h b/src/include/storage/flexlock.h
new file mode 100644
index 0000000..612c21a
--- /dev/null
+++ b/src/include/storage/flexlock.h
@@ -0,0 +1,102 @@
+/*-------------------------------------------------------------------------
+ *
+ * flexlock.h
+ *	  Flex lock manager
+ *
+ * Portions Copyright (c) 1996-2011, PostgreSQL Global Development Group
+ * Portions Copyright (c) 1994, Regents of the University of California
+ *
+ * src/include/storage/flexlock.h
+ *
+ *-------------------------------------------------------------------------
+ */
+#ifndef FLEXLOCK_H
+#define FLEXLOCK_H
+
+/*
+ * It's a bit odd to declare NUM_BUFFER_PARTITIONS and NUM_LOCK_PARTITIONS
+ * here, but we need them to set up enum FlexLockId correctly, and having
+ * this file include lock.h or bufmgr.h would be backwards.
+ */
+
+/* Number of partitions of the shared buffer mapping hashtable */
+#define NUM_BUFFER_PARTITIONS  16
+
+/* Number of partitions the shared lock tables are divided into */
+#define LOG2_NUM_LOCK_PARTITIONS  4
+#define NUM_LOCK_PARTITIONS  (1 << LOG2_NUM_LOCK_PARTITIONS)
+
+/* Number of partitions the shared predicate lock tables are divided into */
+#define LOG2_NUM_PREDICATELOCK_PARTITIONS  4
+#define NUM_PREDICATELOCK_PARTITIONS  (1 << LOG2_NUM_PREDICATELOCK_PARTITIONS)
+
+/*
+ * We have a number of predefined FlexLocks, plus a bunch of locks that are
+ * dynamically assigned (e.g., for shared buffers).  The FlexLock structures
+ * live in shared memory (since they contain shared data) and are identified
+ * by values of this enumerated type.  We abuse the notion of an enum somewhat
+ * by allowing values not listed in the enum declaration to be assigned.
+ * The extra value MaxDynamicFlexLock is there to keep the compiler from
+ * deciding that the enum can be represented as char or short ...
+ *
+ * If you remove a lock, please replace it with a placeholder. This retains
+ * the lock numbering, which is helpful for DTrace and other external
+ * debugging scripts.
+ */
+typedef enum FlexLockId
+{
+	BufFreelistLock,
+	ShmemIndexLock,
+	OidGenLock,
+	XidGenLock,
+	ProcArrayLock,
+	SInvalReadLock,
+	SInvalWriteLock,
+	WALInsertLock,
+	WALWriteLock,
+	ControlFileLock,
+	CheckpointLock,
+	CLogControlLock,
+	SubtransControlLock,
+	MultiXactGenLock,
+	MultiXactOffsetControlLock,
+	MultiXactMemberControlLock,
+	RelCacheInitLock,
+	BgWriterCommLock,
+	TwoPhaseStateLock,
+	TablespaceCreateLock,
+	BtreeVacuumLock,
+	AddinShmemInitLock,
+	AutovacuumLock,
+	AutovacuumScheduleLock,
+	SyncScanLock,
+	RelationMappingLock,
+	AsyncCtlLock,
+	AsyncQueueLock,
+	SerializableXactHashLock,
+	SerializableFinishedListLock,
+	SerializablePredicateLockListLock,
+	OldSerXidLock,
+	SyncRepLock,
+	/* Individual lock IDs end here */
+	FirstBufMappingLock,
+	FirstLockMgrLock = FirstBufMappingLock + NUM_BUFFER_PARTITIONS,
+	FirstPredicateLockMgrLock = FirstLockMgrLock + NUM_LOCK_PARTITIONS,
+
+	/* must be last except for MaxDynamicFlexLock: */
+	NumFixedFlexLocks = FirstPredicateLockMgrLock + NUM_PREDICATELOCK_PARTITIONS,
+
+	MaxDynamicFlexLock = 1000000000
+} FlexLockId;
+
+/* Shared memory setup. */
+extern int	NumFlexLocks(void);
+extern Size FlexLockShmemSize(void);
+extern void RequestAddinFlexLocks(int n);
+extern void CreateFlexLocks(void);
+
+/* Error recovery and debugging support functions. */
+extern void FlexLockReleaseAll(void);
+extern bool FlexLockHeldByMe(FlexLockId id);
+
+#endif   /* FLEXLOCK_H */
diff --git a/src/include/storage/flexlock_internals.h b/src/include/storage/flexlock_internals.h
new file mode 100644
index 0000000..5f78da7
--- /dev/null
+++ b/src/include/storage/flexlock_internals.h
@@ -0,0 +1,88 @@
+/*-------------------------------------------------------------------------
+ *
+ * flexlock_internals.h
+ *	  Flex lock internals.  Only files which implement a FlexLock
+ *    type should need to include this.  Merging this with flexlock.h
+ *    creates a circular header dependency, but even if it didn't, this
+ *    is cleaner.
+ *
+ * Portions Copyright (c) 1996-2011, PostgreSQL Global Development Group
+ * Portions Copyright (c) 1994, Regents of the University of California
+ *
+ * src/include/storage/flexlock_internals.h
+ *
+ *-------------------------------------------------------------------------
+ */
+#ifndef FLEXLOCK_INTERNALS_H
+#define FLEXLOCK_INTERNALS_H
+
+#include "pg_trace.h"
+#include "storage/flexlock.h"
+#include "storage/proc.h"
+#include "storage/s_lock.h"
+
+/*
+ * Individual FlexLock implementations each get this many bytes to store
+ * its state; of course, a given implementation could also allocate additional
+ * shmem elsewhere, but we provide this many bytes within the array.  The
+ * header fields common to all FlexLock types are included in this number.
+ * A power of two should probably be chosen, to avoid alignment issues and
+ * cache line splitting.  It might be useful to increase this on systems where
+ * a cache line is more than 64 bytes in size.
+ */
+#define FLEX_LOCK_BYTES		64
+
+typedef struct FlexLock
+{
+	char		locktype;		/* see FLEXLOCK_TYPE_* constants */
+	slock_t		mutex;			/* Protects FlexLock state and wait queues */
+	bool		releaseOK;		/* T if ok to release waiters */
+	PGPROC	   *head;			/* head of list of waiting PGPROCs */
+	PGPROC	   *tail;			/* tail of list of waiting PGPROCs */
+	/* tail is undefined when head is NULL */
+} FlexLock;
+
+#define FLEXLOCK_TYPE_LWLOCK			'l'
+
+typedef union FlexLockPadded
+{
+	FlexLock	flex;
+	char		pad[FLEX_LOCK_BYTES];
+} FlexLockPadded;
+
+extern FlexLockPadded *FlexLockArray;
+
+extern FlexLockId FlexLockAssign(char locktype);
+extern void FlexLockRemember(FlexLockId id);
+extern void FlexLockForget(FlexLockId id);
+extern int FlexLockWait(FlexLockId id, int mode);
+
+/*
+ * We must join the wait queue while holding the spinlock, so we define this
+ * as a macro, for speed.
+ */
+#define FlexLockJoinWaitQueue(lock, mode) \
+	do { \
+		Assert(MyProc != NULL); \
+		MyProc->flWaitResult = 0; \
+		MyProc->flWaitMode = mode; \
+		MyProc->flWaitLink = NULL; \
+		if (lock->flex.head == NULL) \
+			lock->flex.head = MyProc; \
+		else \
+			lock->flex.tail->flWaitLink = MyProc; \
+		lock->flex.tail = MyProc; \
+	} while (0)
+
+#ifdef LOCK_DEBUG
+extern bool	Trace_flexlocks;
+#define FlexLockDebug(where, id, msg) \
+	do { \
+		if (Trace_flexlocks) \
+			elog(LOG, "%s(%d): %s", where, (int) id, msg); \
+	} while (0)
+#else
+#define FlexLockDebug(where, id, msg)
+#endif
+
+#endif   /* FLEXLOCK_H */
diff --git a/src/include/storage/lock.h b/src/include/storage/lock.h
index e106ad5..ba87db2 100644
--- a/src/include/storage/lock.h
+++ b/src/include/storage/lock.h
@@ -471,7 +471,7 @@ typedef enum
 #define LockHashPartition(hashcode) \
 	((hashcode) % NUM_LOCK_PARTITIONS)
 #define LockHashPartitionLock(hashcode) \
-	((LWLockId) (FirstLockMgrLock + LockHashPartition(hashcode)))
+	((FlexLockId) (FirstLockMgrLock + LockHashPartition(hashcode)))
 
 
 /*
diff --git a/src/include/storage/lwlock.h b/src/include/storage/lwlock.h
index 438a48d..f68cddc 100644
--- a/src/include/storage/lwlock.h
+++ b/src/include/storage/lwlock.h
@@ -14,82 +14,7 @@
 #ifndef LWLOCK_H
 #define LWLOCK_H
 
-/*
- * It's a bit odd to declare NUM_BUFFER_PARTITIONS and NUM_LOCK_PARTITIONS
- * here, but we need them to set up enum LWLockId correctly, and having
- * this file include lock.h or bufmgr.h would be backwards.
- */
-
-/* Number of partitions of the shared buffer mapping hashtable */
-#define NUM_BUFFER_PARTITIONS  16
-
-/* Number of partitions the shared lock tables are divided into */
-#define LOG2_NUM_LOCK_PARTITIONS  4
-#define NUM_LOCK_PARTITIONS  (1 << LOG2_NUM_LOCK_PARTITIONS)
-
-/* Number of partitions the shared predicate lock tables are divided into */
-#define LOG2_NUM_PREDICATELOCK_PARTITIONS  4
-#define NUM_PREDICATELOCK_PARTITIONS  (1 << LOG2_NUM_PREDICATELOCK_PARTITIONS)
-
-/*
- * We have a number of predefined LWLocks, plus a bunch of LWLocks that are
- * dynamically assigned (e.g., for shared buffers).  The LWLock structures
- * live in shared memory (since they contain shared data) and are identified
- * by values of this enumerated type.  We abuse the notion of an enum somewhat
- * by allowing values not listed in the enum declaration to be assigned.
- * The extra value MaxDynamicLWLock is there to keep the compiler from
- * deciding that the enum can be represented as char or short ...
- *
- * If you remove a lock, please replace it with a placeholder. This retains
- * the lock numbering, which is helpful for DTrace and other external
- * debugging scripts.
- */
-typedef enum LWLockId
-{
-	BufFreelistLock,
-	ShmemIndexLock,
-	OidGenLock,
-	XidGenLock,
-	ProcArrayLock,
-	SInvalReadLock,
-	SInvalWriteLock,
-	WALInsertLock,
-	WALWriteLock,
-	ControlFileLock,
-	CheckpointLock,
-	CLogControlLock,
-	SubtransControlLock,
-	MultiXactGenLock,
-	MultiXactOffsetControlLock,
-	MultiXactMemberControlLock,
-	RelCacheInitLock,
-	BgWriterCommLock,
-	TwoPhaseStateLock,
-	TablespaceCreateLock,
-	BtreeVacuumLock,
-	AddinShmemInitLock,
-	AutovacuumLock,
-	AutovacuumScheduleLock,
-	SyncScanLock,
-	RelationMappingLock,
-	AsyncCtlLock,
-	AsyncQueueLock,
-	SerializableXactHashLock,
-	SerializableFinishedListLock,
-	SerializablePredicateLockListLock,
-	OldSerXidLock,
-	SyncRepLock,
-	/* Individual lock IDs end here */
-	FirstBufMappingLock,
-	FirstLockMgrLock = FirstBufMappingLock + NUM_BUFFER_PARTITIONS,
-	FirstPredicateLockMgrLock = FirstLockMgrLock + NUM_LOCK_PARTITIONS,
-
-	/* must be last except for MaxDynamicLWLock: */
-	NumFixedLWLocks = FirstPredicateLockMgrLock + NUM_PREDICATELOCK_PARTITIONS,
-
-	MaxDynamicLWLock = 1000000000
-} LWLockId;
-
+#include "storage/flexlock.h"
 
 typedef enum LWLockMode
 {
@@ -97,22 +22,10 @@ typedef enum LWLockMode
 	LW_SHARED
 } LWLockMode;
 
-
-#ifdef LOCK_DEBUG
-extern bool Trace_lwlocks;
-#endif
-
-extern LWLockId LWLockAssign(void);
-extern void LWLockAcquire(LWLockId lockid, LWLockMode mode);
-extern bool LWLockConditionalAcquire(LWLockId lockid, LWLockMode mode);
-extern void LWLockRelease(LWLockId lockid);
-extern void LWLockReleaseAll(void);
-extern bool LWLockHeldByMe(LWLockId lockid);
-
-extern int	NumLWLocks(void);
-extern Size LWLockShmemSize(void);
-extern void CreateLWLocks(void);
-
-extern void RequestAddinLWLocks(int n);
+extern FlexLockId LWLockAssign(void);
+extern void LWLockAcquire(FlexLockId lockid, LWLockMode mode);
+extern bool LWLockConditionalAcquire(FlexLockId lockid, LWLockMode mode);
+extern void LWLockRelease(FlexLockId lockid);
+extern bool LWLockHeldByMe(FlexLockId lockid);
 
 #endif   /* LWLOCK_H */
diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h
index c7cddc7..1f3a71d 100644
--- a/src/include/storage/proc.h
+++ b/src/include/storage/proc.h
@@ -99,10 +99,10 @@ struct PGPROC
 	 */
 	bool		recoveryConflictPending;
 
-	/* Info about LWLock the process is currently waiting for, if any. */
-	bool		lwWaiting;		/* true if waiting for an LW lock */
-	bool		lwExclusive;	/* true if waiting for exclusive access */
-	struct PGPROC *lwWaitLink;	/* next waiter for same LW lock */
+	/* Info about FlexLock the process is currently waiting for, if any. */
+	int			flWaitResult;	/* result of wait, or 0 if still waiting */
+	int			flWaitMode;		/* lock mode sought */
+	struct PGPROC *flWaitLink;	/* next waiter for same FlexLock */
 
 	/* Info about lock the process is currently waiting for, if any. */
 	/* waitLock and waitProcLock are NULL if not currently waiting. */
@@ -132,7 +132,7 @@ struct PGPROC
 	struct XidCache subxids;	/* cache for subtransaction XIDs */
 
 	/* Per-backend LWLock.  Protects fields below. */
-	LWLockId	backendLock;	/* protects the fields below */
+	FlexLockId	backendLock;	/* protects the fields below */
 
 	/* Lock manager data, recording fast-path locks taken by this backend. */
 	uint64		fpLockBits;		/* lock modes held for each fast-path slot */

diff --git a/src/backend/commands/analyze.c b/src/backend/commands/analyze.c
index 3143246..2e972ec 100644
--- a/src/backend/commands/analyze.c
+++ b/src/backend/commands/analyze.c
@@ -40,6 +40,7 @@
 #include "storage/lmgr.h"
 #include "storage/proc.h"
 #include "storage/procarray.h"
+#include "storage/procarraylock.h"
 #include "utils/acl.h"
 #include "utils/attoptcache.h"
 #include "utils/datum.h"
@@ -222,9 +223,9 @@ analyze_rel(Oid relid, VacuumStmt *vacstmt, BufferAccessStrategy bstrategy)
 	/*
 	 * OK, let's do it.  First let other backends know I'm in ANALYZE.
 	 */
-	LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
+	ProcArrayLockAcquire(PAL_EXCLUSIVE);
 	MyPgXact->vacuumFlags |= PROC_IN_ANALYZE;
-	LWLockRelease(ProcArrayLock);
+	ProcArrayLockRelease();
 
 	/*
 	 * Do the normal non-recursive ANALYZE.
@@ -249,9 +250,9 @@ analyze_rel(Oid relid, VacuumStmt *vacstmt, BufferAccessStrategy bstrategy)
 	 * Reset my PGPROC flag.  Note: we need this here, and not in vacuum_rel,
 	 * because the vacuum flag is cleared by the end-of-xact code.
 	 */
-	LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
+	ProcArrayLockAcquire(PAL_EXCLUSIVE);
 	MyPgXact->vacuumFlags &= ~PROC_IN_ANALYZE;
-	LWLockRelease(ProcArrayLock);
+	ProcArrayLockRelease();
 }
 
 /*
diff --git a/src/backend/commands/vacuum.c b/src/backend/commands/vacuum.c
index e70dbed..09aa32b 100644
--- a/src/backend/commands/vacuum.c
+++ b/src/backend/commands/vacuum.c
@@ -39,6 +39,7 @@
 #include "storage/lmgr.h"
 #include "storage/proc.h"
 #include "storage/procarray.h"
+#include "storage/procarraylock.h"
 #include "utils/acl.h"
 #include "utils/fmgroids.h"
 #include "utils/guc.h"
@@ -895,11 +896,11 @@ vacuum_rel(Oid relid, VacuumStmt *vacstmt, bool do_toast, bool for_wraparound)
 		 * MyProc->xid/xmin, else OldestXmin might appear to go backwards,
 		 * which is probably Not Good.
 		 */
-		LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
+		ProcArrayLockAcquire(PAL_EXCLUSIVE);
 		MyPgXact->vacuumFlags |= PROC_IN_VACUUM;
 		if (for_wraparound)
 			MyPgXact->vacuumFlags |= PROC_VACUUM_FOR_WRAPAROUND;
-		LWLockRelease(ProcArrayLock);
+		ProcArrayLockRelease();
 	}
 
 	/*
diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c
index 19ff524..d457e3f 100644
--- a/src/backend/storage/ipc/procarray.c
+++ b/src/backend/storage/ipc/procarray.c
@@ -52,6 +52,7 @@
 #include "access/twophase.h"
 #include "miscadmin.h"
 #include "storage/procarray.h"
+#include "storage/procarraylock.h"
 #include "storage/spin.h"
 #include "utils/builtins.h"
 #include "utils/snapmgr.h"
@@ -261,7 +262,7 @@ ProcArrayAdd(PGPROC *proc)
 	ProcArrayStruct *arrayP = procArray;
 	int			index;
 
-	LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
+	ProcArrayLockAcquire(PAL_EXCLUSIVE);
 
 	if (arrayP->numProcs >= arrayP->maxProcs)
 	{
@@ -270,7 +271,7 @@ ProcArrayAdd(PGPROC *proc)
 		 * fixed supply of PGPROC structs too, and so we should have failed
 		 * earlier.)
 		 */
-		LWLockRelease(ProcArrayLock);
+		ProcArrayLockRelease();
 		ereport(FATAL,
 				(errcode(ERRCODE_TOO_MANY_CONNECTIONS),
 				 errmsg("sorry, too many clients already")));
@@ -300,7 +301,7 @@ ProcArrayAdd(PGPROC *proc)
 	arrayP->pgprocnos[index] = proc->pgprocno;
 	arrayP->numProcs++;
 
-	LWLockRelease(ProcArrayLock);
+	ProcArrayLockRelease();
 }
 
 /*
@@ -325,7 +326,7 @@ ProcArrayRemove(PGPROC *proc, TransactionId latestXid)
 		DisplayXidCache();
 #endif
 
-	LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
+	ProcArrayLockAcquire(PAL_EXCLUSIVE);
 
 	if (TransactionIdIsValid(latestXid))
 	{
@@ -351,13 +352,13 @@ ProcArrayRemove(PGPROC *proc, TransactionId latestXid)
 					(arrayP->numProcs - index - 1) * sizeof (int));
 			arrayP->pgprocnos[arrayP->numProcs - 1] = -1; /* for debugging */
 			arrayP->numProcs--;
-			LWLockRelease(ProcArrayLock);
+			ProcArrayLockRelease();
 			return;
 		}
 	}
 
 	/* Ooops */
-	LWLockRelease(ProcArrayLock);
+	ProcArrayLockRelease();
 
 	elog(LOG, "failed to find proc %p in ProcArray", proc);
 }
@@ -383,54 +384,19 @@ ProcArrayEndTransaction(PGPROC *proc, TransactionId latestXid)
 
 	if (TransactionIdIsValid(latestXid))
 	{
-		/*
-		 * We must lock ProcArrayLock while clearing our advertised XID, so
-		 * that we do not exit the set of "running" transactions while someone
-		 * else is taking a snapshot.  See discussion in
-		 * src/backend/access/transam/README.
-		 */
-		Assert(TransactionIdIsValid(allPgXact[proc->pgprocno].xid));
-
-		LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
-
-		pgxact->xid = InvalidTransactionId;
-		proc->lxid = InvalidLocalTransactionId;
-		pgxact->xmin = InvalidTransactionId;
-		/* must be cleared with xid/xmin: */
-		pgxact->vacuumFlags &= ~PROC_VACUUM_STATE_MASK;
-		pgxact->inCommit = false; /* be sure this is cleared in abort */
-		proc->recoveryConflictPending = false;
-
-		/* Clear the subtransaction-XID cache too while holding the lock */
-		pgxact->nxids = 0;
-		pgxact->overflowed = false;
-
-		/* Also advance global latestCompletedXid while holding the lock */
-		if (TransactionIdPrecedes(ShmemVariableCache->latestCompletedXid,
-								  latestXid))
-			ShmemVariableCache->latestCompletedXid = latestXid;
-
-		LWLockRelease(ProcArrayLock);
+		Assert(proc == MyProc);
+		ProcArrayLockClearTransaction(latestXid);		
 	}
 	else
 	{
-		/*
-		 * If we have no XID, we don't need to lock, since we won't affect
-		 * anyone else's calculation of a snapshot.  We might change their
-		 * estimate of global xmin, but that's OK.
-		 */
-		Assert(!TransactionIdIsValid(allPgXact[proc->pgprocno].xid));
-
-		proc->lxid = InvalidLocalTransactionId;
 		pgxact->xmin = InvalidTransactionId;
 		/* must be cleared with xid/xmin: */
 		pgxact->vacuumFlags &= ~PROC_VACUUM_STATE_MASK;
-		pgxact->inCommit = false; /* be sure this is cleared in abort */
-		proc->recoveryConflictPending = false;
-
-		Assert(pgxact->nxids == 0);
-		Assert(pgxact->overflowed == false);
 	}
+
+	proc->lxid = InvalidLocalTransactionId;
+	pgxact->inCommit = false; /* be sure this is cleared in abort */
+	proc->recoveryConflictPending = false;
 }
 
 
@@ -562,7 +528,7 @@ ProcArrayApplyRecoveryInfo(RunningTransactions running)
 	/*
 	 * Nobody else is running yet, but take locks anyhow
 	 */
-	LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
+	ProcArrayLockAcquire(PAL_EXCLUSIVE);
 
 	/*
 	 * KnownAssignedXids is sorted so we cannot just add the xids, we have to
@@ -669,7 +635,7 @@ ProcArrayApplyRecoveryInfo(RunningTransactions running)
 	Assert(TransactionIdIsNormal(ShmemVariableCache->latestCompletedXid));
 	Assert(TransactionIdIsValid(ShmemVariableCache->nextXid));
 
-	LWLockRelease(ProcArrayLock);
+	ProcArrayLockRelease();
 
 	KnownAssignedXidsDisplay(trace_recovery(DEBUG3));
 	if (standbyState == STANDBY_SNAPSHOT_READY)
@@ -724,7 +690,7 @@ ProcArrayApplyXidAssignment(TransactionId topxid,
 	/*
 	 * Uses same locking as transaction commit
 	 */
-	LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
+	ProcArrayLockAcquire(PAL_EXCLUSIVE);
 
 	/*
 	 * Remove subxids from known-assigned-xacts.
@@ -737,7 +703,7 @@ ProcArrayApplyXidAssignment(TransactionId topxid,
 	if (TransactionIdPrecedes(procArray->lastOverflowedXid, max_xid))
 		procArray->lastOverflowedXid = max_xid;
 
-	LWLockRelease(ProcArrayLock);
+	ProcArrayLockRelease();
 }
 
 /*
@@ -829,7 +795,7 @@ TransactionIdIsInProgress(TransactionId xid)
 					 errmsg("out of memory")));
 	}
 
-	LWLockAcquire(ProcArrayLock, LW_SHARED);
+	ProcArrayLockAcquire(PAL_SHARED);
 
 	/*
 	 * Now that we have the lock, we can check latestCompletedXid; if the
@@ -837,7 +803,7 @@ TransactionIdIsInProgress(TransactionId xid)
 	 */
 	if (TransactionIdPrecedes(ShmemVariableCache->latestCompletedXid, xid))
 	{
-		LWLockRelease(ProcArrayLock);
+		ProcArrayLockRelease();
 		xc_by_latest_xid_inc();
 		return true;
 	}
@@ -865,7 +831,7 @@ TransactionIdIsInProgress(TransactionId xid)
 		 */
 		if (TransactionIdEquals(pxid, xid))
 		{
-			LWLockRelease(ProcArrayLock);
+			ProcArrayLockRelease();
 			xc_by_main_xid_inc();
 			return true;
 		}
@@ -887,7 +853,7 @@ TransactionIdIsInProgress(TransactionId xid)
 
 			if (TransactionIdEquals(cxid, xid))
 			{
-				LWLockRelease(ProcArrayLock);
+				ProcArrayLockRelease();
 				xc_by_child_xid_inc();
 				return true;
 			}
@@ -915,7 +881,7 @@ TransactionIdIsInProgress(TransactionId xid)
 
 		if (KnownAssignedXidExists(xid))
 		{
-			LWLockRelease(ProcArrayLock);
+			ProcArrayLockRelease();
 			xc_by_known_assigned_inc();
 			return true;
 		}
@@ -931,7 +897,7 @@ TransactionIdIsInProgress(TransactionId xid)
 			nxids = KnownAssignedXidsGet(xids, xid);
 	}
 
-	LWLockRelease(ProcArrayLock);
+	ProcArrayLockRelease();
 
 	/*
 	 * If none of the relevant caches overflowed, we know the Xid is not
@@ -997,7 +963,7 @@ TransactionIdIsActive(TransactionId xid)
 	if (TransactionIdPrecedes(xid, RecentXmin))
 		return false;
 
-	LWLockAcquire(ProcArrayLock, LW_SHARED);
+	ProcArrayLockAcquire(PAL_SHARED);
 
 	for (i = 0; i < arrayP->numProcs; i++)
 	{
@@ -1022,7 +988,7 @@ TransactionIdIsActive(TransactionId xid)
 		}
 	}
 
-	LWLockRelease(ProcArrayLock);
+	ProcArrayLockRelease();
 
 	return result;
 }
@@ -1085,7 +1051,7 @@ GetOldestXmin(bool allDbs, bool ignoreVacuum)
 	/* Cannot look for individual databases during recovery */
 	Assert(allDbs || !RecoveryInProgress());
 
-	LWLockAcquire(ProcArrayLock, LW_SHARED);
+	ProcArrayLockAcquire(PAL_SHARED);
 
 	/*
 	 * We initialize the MIN() calculation with latestCompletedXid + 1. This
@@ -1140,7 +1106,7 @@ GetOldestXmin(bool allDbs, bool ignoreVacuum)
 		 */
 		TransactionId kaxmin = KnownAssignedXidsGetOldestXmin();
 
-		LWLockRelease(ProcArrayLock);
+		ProcArrayLockRelease();
 
 		if (TransactionIdIsNormal(kaxmin) &&
 			TransactionIdPrecedes(kaxmin, result))
@@ -1151,7 +1117,7 @@ GetOldestXmin(bool allDbs, bool ignoreVacuum)
 		/*
 		 * No other information needed, so release the lock immediately.
 		 */
-		LWLockRelease(ProcArrayLock);
+		ProcArrayLockRelease();
 
 		/*
 		 * Compute the cutoff XID by subtracting vacuum_defer_cleanup_age,
@@ -1280,7 +1246,7 @@ GetSnapshotData(Snapshot snapshot)
 	 * It is sufficient to get shared lock on ProcArrayLock, even if we are
 	 * going to set MyProc->xmin.
 	 */
-	LWLockAcquire(ProcArrayLock, LW_SHARED);
+	ProcArrayLockAcquire(PAL_SHARED);
 
 	/* xmax is always latestCompletedXid + 1 */
 	xmax = ShmemVariableCache->latestCompletedXid;
@@ -1418,7 +1384,7 @@ GetSnapshotData(Snapshot snapshot)
 
 	if (!TransactionIdIsValid(MyPgXact->xmin))
 		MyPgXact->xmin = TransactionXmin = xmin;
-	LWLockRelease(ProcArrayLock);
+	ProcArrayLockRelease();
 
 	/*
 	 * Update globalxmin to include actual process xids.  This is a slightly
@@ -1475,7 +1441,7 @@ ProcArrayInstallImportedXmin(TransactionId xmin, TransactionId sourcexid)
 		return false;
 
 	/* Get lock so source xact can't end while we're doing this */
-	LWLockAcquire(ProcArrayLock, LW_SHARED);
+	ProcArrayLockAcquire(PAL_SHARED);
 
 	for (index = 0; index < arrayP->numProcs; index++)
 	{
@@ -1521,7 +1487,7 @@ ProcArrayInstallImportedXmin(TransactionId xmin, TransactionId sourcexid)
 		break;
 	}
 
-	LWLockRelease(ProcArrayLock);
+	ProcArrayLockRelease();
 
 	return result;
 }
@@ -1595,7 +1561,7 @@ GetRunningTransactionData(void)
 	 * Ensure that no xids enter or leave the procarray while we obtain
 	 * snapshot.
 	 */
-	LWLockAcquire(ProcArrayLock, LW_SHARED);
+	ProcArrayLockAcquire(PAL_SHARED);
 	LWLockAcquire(XidGenLock, LW_SHARED);
 
 	latestCompletedXid = ShmemVariableCache->latestCompletedXid;
@@ -1658,7 +1624,7 @@ GetRunningTransactionData(void)
 	CurrentRunningXacts->latestCompletedXid = latestCompletedXid;
 
 	/* We don't release XidGenLock here, the caller is responsible for that */
-	LWLockRelease(ProcArrayLock);
+	ProcArrayLockRelease();
 
 	Assert(TransactionIdIsValid(CurrentRunningXacts->nextXid));
 	Assert(TransactionIdIsValid(CurrentRunningXacts->oldestRunningXid));
@@ -1691,7 +1657,7 @@ GetOldestActiveTransactionId(void)
 
 	Assert(!RecoveryInProgress());
 
-	LWLockAcquire(ProcArrayLock, LW_SHARED);
+	ProcArrayLockAcquire(PAL_SHARED);
 
 	oldestRunningXid = ShmemVariableCache->nextXid;
 
@@ -1720,7 +1686,7 @@ GetOldestActiveTransactionId(void)
 		 */
 	}
 
-	LWLockRelease(ProcArrayLock);
+	ProcArrayLockRelease();
 
 	return oldestRunningXid;
 }
@@ -1753,7 +1719,7 @@ GetTransactionsInCommit(TransactionId **xids_p)
 	xids = (TransactionId *) palloc(arrayP->maxProcs * sizeof(TransactionId));
 	nxids = 0;
 
-	LWLockAcquire(ProcArrayLock, LW_SHARED);
+	ProcArrayLockAcquire(PAL_SHARED);
 
 	for (index = 0; index < arrayP->numProcs; index++)
 	{
@@ -1768,7 +1734,7 @@ GetTransactionsInCommit(TransactionId **xids_p)
 			xids[nxids++] = pxid;
 	}
 
-	LWLockRelease(ProcArrayLock);
+	ProcArrayLockRelease();
 
 	*xids_p = xids;
 	return nxids;
@@ -1790,7 +1756,7 @@ HaveTransactionsInCommit(TransactionId *xids, int nxids)
 	ProcArrayStruct *arrayP = procArray;
 	int			index;
 
-	LWLockAcquire(ProcArrayLock, LW_SHARED);
+	ProcArrayLockAcquire(PAL_SHARED);
 
 	for (index = 0; index < arrayP->numProcs; index++)
 	{
@@ -1818,7 +1784,7 @@ HaveTransactionsInCommit(TransactionId *xids, int nxids)
 		}
 	}
 
-	LWLockRelease(ProcArrayLock);
+	ProcArrayLockRelease();
 
 	return result;
 }
@@ -1840,7 +1806,7 @@ BackendPidGetProc(int pid)
 	if (pid == 0)				/* never match dummy PGPROCs */
 		return NULL;
 
-	LWLockAcquire(ProcArrayLock, LW_SHARED);
+	ProcArrayLockAcquire(PAL_SHARED);
 
 	for (index = 0; index < arrayP->numProcs; index++)
 	{
@@ -1853,7 +1819,7 @@ BackendPidGetProc(int pid)
 		}
 	}
 
-	LWLockRelease(ProcArrayLock);
+	ProcArrayLockRelease();
 
 	return result;
 }
@@ -1881,7 +1847,7 @@ BackendXidGetPid(TransactionId xid)
 	if (xid == InvalidTransactionId)	/* never match invalid xid */
 		return 0;
 
-	LWLockAcquire(ProcArrayLock, LW_SHARED);
+	ProcArrayLockAcquire(PAL_SHARED);
 
 	for (index = 0; index < arrayP->numProcs; index++)
 	{
@@ -1896,7 +1862,7 @@ BackendXidGetPid(TransactionId xid)
 		}
 	}
 
-	LWLockRelease(ProcArrayLock);
+	ProcArrayLockRelease();
 
 	return result;
 }
@@ -1951,7 +1917,7 @@ GetCurrentVirtualXIDs(TransactionId limitXmin, bool excludeXmin0,
 	vxids = (VirtualTransactionId *)
 		palloc(sizeof(VirtualTransactionId) * arrayP->maxProcs);
 
-	LWLockAcquire(ProcArrayLock, LW_SHARED);
+	ProcArrayLockAcquire(PAL_SHARED);
 
 	for (index = 0; index < arrayP->numProcs; index++)
 	{
@@ -1989,7 +1955,7 @@ GetCurrentVirtualXIDs(TransactionId limitXmin, bool excludeXmin0,
 		}
 	}
 
-	LWLockRelease(ProcArrayLock);
+	ProcArrayLockRelease();
 
 	*nvxids = count;
 	return vxids;
@@ -2048,7 +2014,7 @@ GetConflictingVirtualXIDs(TransactionId limitXmin, Oid dbOid)
 					 errmsg("out of memory")));
 	}
 
-	LWLockAcquire(ProcArrayLock, LW_SHARED);
+	ProcArrayLockAcquire(PAL_SHARED);
 
 	for (index = 0; index < arrayP->numProcs; index++)
 	{
@@ -2083,7 +2049,7 @@ GetConflictingVirtualXIDs(TransactionId limitXmin, Oid dbOid)
 		}
 	}
 
-	LWLockRelease(ProcArrayLock);
+	ProcArrayLockRelease();
 
 	/* add the terminator */
 	vxids[count].backendId = InvalidBackendId;
@@ -2104,7 +2070,7 @@ CancelVirtualTransaction(VirtualTransactionId vxid, ProcSignalReason sigmode)
 	int			index;
 	pid_t		pid = 0;
 
-	LWLockAcquire(ProcArrayLock, LW_SHARED);
+	ProcArrayLockAcquire(PAL_SHARED);
 
 	for (index = 0; index < arrayP->numProcs; index++)
 	{
@@ -2131,7 +2097,7 @@ CancelVirtualTransaction(VirtualTransactionId vxid, ProcSignalReason sigmode)
 		}
 	}
 
-	LWLockRelease(ProcArrayLock);
+	ProcArrayLockRelease();
 
 	return pid;
 }
@@ -2207,7 +2173,7 @@ CountDBBackends(Oid databaseid)
 	int			count = 0;
 	int			index;
 
-	LWLockAcquire(ProcArrayLock, LW_SHARED);
+	ProcArrayLockAcquire(PAL_SHARED);
 
 	for (index = 0; index < arrayP->numProcs; index++)
 	{
@@ -2221,7 +2187,7 @@ CountDBBackends(Oid databaseid)
 			count++;
 	}
 
-	LWLockRelease(ProcArrayLock);
+	ProcArrayLockRelease();
 
 	return count;
 }
@@ -2237,7 +2203,7 @@ CancelDBBackends(Oid databaseid, ProcSignalReason sigmode, bool conflictPending)
 	pid_t		pid = 0;
 
 	/* tell all backends to die */
-	LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
+	ProcArrayLockAcquire(PAL_EXCLUSIVE);
 
 	for (index = 0; index < arrayP->numProcs; index++)
 	{
@@ -2263,7 +2229,7 @@ CancelDBBackends(Oid databaseid, ProcSignalReason sigmode, bool conflictPending)
 		}
 	}
 
-	LWLockRelease(ProcArrayLock);
+	ProcArrayLockRelease();
 }
 
 /*
@@ -2276,7 +2242,7 @@ CountUserBackends(Oid roleid)
 	int			count = 0;
 	int			index;
 
-	LWLockAcquire(ProcArrayLock, LW_SHARED);
+	ProcArrayLockAcquire(PAL_SHARED);
 
 	for (index = 0; index < arrayP->numProcs; index++)
 	{
@@ -2289,7 +2255,7 @@ CountUserBackends(Oid roleid)
 			count++;
 	}
 
-	LWLockRelease(ProcArrayLock);
+	ProcArrayLockRelease();
 
 	return count;
 }
@@ -2337,7 +2303,7 @@ CountOtherDBBackends(Oid databaseId, int *nbackends, int *nprepared)
 
 		*nbackends = *nprepared = 0;
 
-		LWLockAcquire(ProcArrayLock, LW_SHARED);
+		ProcArrayLockAcquire(PAL_SHARED);
 
 		for (index = 0; index < arrayP->numProcs; index++)
 		{
@@ -2363,7 +2329,7 @@ CountOtherDBBackends(Oid databaseId, int *nbackends, int *nprepared)
 			}
 		}
 
-		LWLockRelease(ProcArrayLock);
+		ProcArrayLockRelease();
 
 		if (!found)
 			return false;		/* no conflicting backends, so done */
@@ -2416,7 +2382,7 @@ XidCacheRemoveRunningXids(TransactionId xid,
 	 * to abort subtransactions, but pending closer analysis we'd best be
 	 * conservative.
 	 */
-	LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
+	ProcArrayLockAcquire(PAL_EXCLUSIVE);
 
 	/*
 	 * Under normal circumstances xid and xids[] will be in increasing order,
@@ -2464,7 +2430,7 @@ XidCacheRemoveRunningXids(TransactionId xid,
 							  latestXid))
 		ShmemVariableCache->latestCompletedXid = latestXid;
 
-	LWLockRelease(ProcArrayLock);
+	ProcArrayLockRelease();
 }
 
 #ifdef XIDCACHE_DEBUG
@@ -2631,7 +2597,7 @@ ExpireTreeKnownAssignedTransactionIds(TransactionId xid, int nsubxids,
 	/*
 	 * Uses same locking as transaction commit
 	 */
-	LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
+	ProcArrayLockAcquire(PAL_EXCLUSIVE);
 
 	KnownAssignedXidsRemoveTree(xid, nsubxids, subxids);
 
@@ -2640,7 +2606,7 @@ ExpireTreeKnownAssignedTransactionIds(TransactionId xid, int nsubxids,
 							  max_xid))
 		ShmemVariableCache->latestCompletedXid = max_xid;
 
-	LWLockRelease(ProcArrayLock);
+	ProcArrayLockRelease();
 }
 
 /*
@@ -2650,9 +2616,9 @@ ExpireTreeKnownAssignedTransactionIds(TransactionId xid, int nsubxids,
 void
 ExpireAllKnownAssignedTransactionIds(void)
 {
-	LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
+	ProcArrayLockAcquire(PAL_EXCLUSIVE);
 	KnownAssignedXidsRemovePreceding(InvalidTransactionId);
-	LWLockRelease(ProcArrayLock);
+	ProcArrayLockRelease();
 }
 
 /*
@@ -2662,9 +2628,9 @@ ExpireAllKnownAssignedTransactionIds(void)
 void
 ExpireOldKnownAssignedTransactionIds(TransactionId xid)
 {
-	LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
+	ProcArrayLockAcquire(PAL_EXCLUSIVE);
 	KnownAssignedXidsRemovePreceding(xid);
-	LWLockRelease(ProcArrayLock);
+	ProcArrayLockRelease();
 }
 
 
@@ -2886,7 +2852,7 @@ KnownAssignedXidsAdd(TransactionId from_xid, TransactionId to_xid,
 	{
 		/* must hold lock to compress */
 		if (!exclusive_lock)
-			LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
+			ProcArrayLockAcquire(PAL_EXCLUSIVE);
 
 		KnownAssignedXidsCompress(true);
 
@@ -2894,7 +2860,7 @@ KnownAssignedXidsAdd(TransactionId from_xid, TransactionId to_xid,
 		/* note: we no longer care about the tail pointer */
 
 		if (!exclusive_lock)
-			LWLockRelease(ProcArrayLock);
+			ProcArrayLockRelease();
 
 		/*
 		 * If it still won't fit then we're out of memory
diff --git a/src/backend/storage/lmgr/Makefile b/src/backend/storage/lmgr/Makefile
index 3730e51..27eaa97 100644
--- a/src/backend/storage/lmgr/Makefile
+++ b/src/backend/storage/lmgr/Makefile
@@ -13,7 +13,7 @@ top_builddir = ../../../..
 include $(top_builddir)/src/Makefile.global
 
 OBJS = flexlock.o lmgr.o lock.o proc.o deadlock.o lwlock.o spin.o s_lock.o \
-	predicate.o
+	procarraylock.o predicate.o
 
 include $(top_srcdir)/src/backend/common.mk
 
diff --git a/src/backend/storage/lmgr/flexlock.c b/src/backend/storage/lmgr/flexlock.c
index f517589..f96437b 100644
--- a/src/backend/storage/lmgr/flexlock.c
+++ b/src/backend/storage/lmgr/flexlock.c
@@ -30,6 +30,7 @@
 #include "storage/lwlock.h"
 #include "storage/predicate.h"
 #include "storage/proc.h"
+#include "storage/procarraylock.h"
 #include "storage/spin.h"
 #include "utils/elog.h"
 
@@ -177,9 +178,14 @@ CreateFlexLocks(void)
 
 	FlexLockArray = (FlexLockPadded *) ptr;
 
-	/* All of the "fixed" FlexLocks are LWLocks. */
+	/* All of the "fixed" FlexLocks are LWLocks - except ProcArrayLock. */
 	for (id = 0, lock = FlexLockArray; id < NumFixedFlexLocks; id++, lock++)
-		FlexLockInit(&lock->flex, FLEXLOCK_TYPE_LWLOCK);
+	{
+		if (id == ProcArrayLock)
+			FlexLockInit(&lock->flex, FLEXLOCK_TYPE_PROCARRAYLOCK);
+		else
+			FlexLockInit(&lock->flex, FLEXLOCK_TYPE_LWLOCK);
+	}
 
 	/*
 	 * Initialize the dynamic-allocation counter, which is stored just before
@@ -323,13 +329,20 @@ FlexLockReleaseAll(void)
 {
 	while (num_held_flexlocks > 0)
 	{
+		FlexLockId	id;
+		FlexLock   *flex;
+
 		HOLD_INTERRUPTS();		/* match the upcoming RESUME_INTERRUPTS */
 
-		/*
-		 * FLEXTODO: When we have multiple types of flex locks, this will
-		 * need to call the appropriate release function for each lock type.
-		 */
-		LWLockRelease(held_flexlocks[num_held_flexlocks - 1]);
+		id = held_flexlocks[num_held_flexlocks - 1];
+		flex = &FlexLockArray[id].flex;
+		if (flex->locktype == FLEXLOCK_TYPE_LWLOCK)
+			LWLockRelease(id);
+		else
+		{
+			Assert(id == ProcArrayLock);
+			ProcArrayLockRelease();
+		}
 	}
 }
 
diff --git a/src/backend/storage/lmgr/proc.c b/src/backend/storage/lmgr/proc.c
index db01e9d..173b7cb 100644
--- a/src/backend/storage/lmgr/proc.c
+++ b/src/backend/storage/lmgr/proc.c
@@ -46,6 +46,7 @@
 #include "storage/pmsignal.h"
 #include "storage/proc.h"
 #include "storage/procarray.h"
+#include "storage/procarraylock.h"
 #include "storage/procsignal.h"
 #include "storage/spin.h"
 #include "utils/timestamp.h"
@@ -1083,7 +1084,7 @@ ProcSleep(LOCALLOCK *locallock, LockMethod lockMethodTable)
 			PGPROC	   *autovac = GetBlockingAutoVacuumPgproc();
 			PGXACT	   *autovac_pgxact = &ProcGlobal->allPgXact[autovac->pgprocno];
 
-			LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
+			ProcArrayLockAcquire(PAL_EXCLUSIVE);
 
 			/*
 			 * Only do it if the worker is not working to protect against Xid
@@ -1099,7 +1100,7 @@ ProcSleep(LOCALLOCK *locallock, LockMethod lockMethodTable)
 					 pid);
 
 				/* don't hold the lock across the kill() syscall */
-				LWLockRelease(ProcArrayLock);
+				ProcArrayLockRelease();
 
 				/* send the autovacuum worker Back to Old Kent Road */
 				if (kill(pid, SIGINT) < 0)
@@ -1111,7 +1112,7 @@ ProcSleep(LOCALLOCK *locallock, LockMethod lockMethodTable)
 				}
 			}
 			else
-				LWLockRelease(ProcArrayLock);
+				ProcArrayLockRelease();
 
 			/* prevent signal from being resent more than once */
 			allow_autovacuum_cancel = false;
diff --git a/src/backend/storage/lmgr/procarraylock.c b/src/backend/storage/lmgr/procarraylock.c
new file mode 100644
index 0000000..7cd4b6b
--- /dev/null
+++ b/src/backend/storage/lmgr/procarraylock.c
@@ -0,0 +1,344 @@
+/*-------------------------------------------------------------------------
+ *
+ * procarraylock.c
+ *	  Lock management for the ProcArray
+ *
+ * Because the ProcArray data structure is highly trafficked, it is
+ * critical that mutual exclusion for ProcArray options be as efficient
+ * as possible.  A particular problem is transaction end (commit or abort)
+ * which cannot be done in parallel with snapshot acquisition.  We
+ * therefore include some special hacks to deal with this case efficiently.
+ *
+ * Portions Copyright (c) 1996-2011, PostgreSQL Global Development Group
+ * Portions Copyright (c) 1994, Regents of the University of California
+ *
+ * IDENTIFICATION
+ *	  src/backend/storage/lmgr/procarraylock.c
+ *
+ *-------------------------------------------------------------------------
+ */
+#include "postgres.h"
+
+#include "miscadmin.h"
+#include "pg_trace.h"
+#include "access/transam.h"
+#include "storage/flexlock_internals.h"
+#include "storage/ipc.h"
+#include "storage/procarraylock.h"
+#include "storage/proc.h"
+#include "storage/spin.h"
+
+typedef struct ProcArrayLockStruct
+{
+	FlexLock	flex;			/* common FlexLock infrastructure */
+	char		exclusive;		/* # of exclusive holders (0 or 1) */
+	int			shared;			/* # of shared holders (0..MaxBackends) */
+	PGPROC	   *ending;			/* transactions wishing to clear state */
+	TransactionId	latest_ending_xid;	/* latest ending XID */
+} ProcArrayLockStruct;
+
+/* There is only one ProcArrayLock. */
+#define	ProcArrayLockPointer() \
+	(AssertMacro(FlexLockArray[ProcArrayLock].flex.locktype == \
+		FLEXLOCK_TYPE_PROCARRAYLOCK), \
+	 (volatile ProcArrayLockStruct *) &FlexLockArray[ProcArrayLock])
+
+/*
+ * ProcArrayLockAcquire - 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
+ProcArrayLockAcquire(ProcArrayLockMode mode)
+{
+	volatile ProcArrayLockStruct *lock = ProcArrayLockPointer();
+	PGPROC	   *proc = MyProc;
+	bool		retry = false;
+	int			extraWaits = 0;
+
+	/*
+	 * 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 ProcArrayLock.  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
+	 * ProcArrayLockRelease.  See comments in LWLockAcquire for an explanation
+	 * of why do we not attempt to hand off the lock directly.
+	 */
+	for (;;)
+	{
+		bool		mustwait;
+
+		/* Acquire mutex.  Time spent holding mutex should be short! */
+		SpinLockAcquire(&lock->flex.mutex);
+
+		/* If retrying, allow LWLockRelease to release waiters again */
+		if (retry)
+			lock->flex.releaseOK = true;
+
+		/* If I can get the lock, do so quickly. */
+		if (mode == PAL_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. */
+		FlexLockJoinWaitQueue(lock, (int) mode);
+
+		/* Can release the mutex now */
+		SpinLockRelease(&lock->flex.mutex);
+
+		/* Wait until awakened. */
+		extraWaits += FlexLockWait(ProcArrayLock, mode);
+
+		/* Now loop back and try to acquire lock again. */
+		retry = true;
+	}
+
+	/* We are done updating shared state of the lock itself. */
+	SpinLockRelease(&lock->flex.mutex);
+
+	TRACE_POSTGRESQL_FLEXLOCK_ACQUIRE(lockid, mode);
+
+	/* Add lock to list of locks held by this backend */
+	FlexLockRemember(ProcArrayLock);
+
+	/*
+	 * Fix the process wait semaphore's count for any absorbed wakeups.
+	 */
+	while (extraWaits-- > 0)
+		PGSemaphoreUnlock(&proc->sem);
+}
+
+/*
+ * ProcArrayLockClearTransaction - safely clear transaction details
+ *
+ * This can't be done while ProcArrayLock is held, but it's so fast that
+ * we can afford to do it while holding the spinlock, rather than acquiring
+ * and releasing the lock.
+ */
+void
+ProcArrayLockClearTransaction(TransactionId latestXid)
+{
+	volatile ProcArrayLockStruct *lock = ProcArrayLockPointer();
+	PGPROC	   *proc = MyProc;
+	int			extraWaits = 0;
+	bool		mustwait;
+
+	HOLD_INTERRUPTS();
+
+	/* Acquire mutex.  Time spent holding mutex should be short! */
+	SpinLockAcquire(&lock->flex.mutex);
+
+	if (lock->exclusive == 0 && lock->shared == 0)
+	{
+		{
+			volatile PGPROC *vproc = proc;
+			volatile PGXACT *pgxact = &ProcGlobal->allPgXact[vproc->pgprocno];
+			/* If there are no lockers, clear the critical PGPROC fields. */
+			pgxact->xid = InvalidTransactionId;
+	        pgxact->xmin = InvalidTransactionId;
+	        /* must be cleared with xid/xmin: */
+	        pgxact->vacuumFlags &= ~PROC_VACUUM_STATE_MASK;
+			pgxact->nxids = 0;
+			pgxact->overflowed = false;
+		}
+		mustwait = false;
+
+        /* Also advance global latestCompletedXid while holding the lock */
+        if (TransactionIdPrecedes(ShmemVariableCache->latestCompletedXid,
+                                  latestXid))
+            ShmemVariableCache->latestCompletedXid = latestXid;
+	}
+	else
+	{
+		/* Rats, must wait. */
+		proc->flWaitLink = lock->ending;
+		lock->ending = proc;
+		if (!TransactionIdIsValid(lock->latest_ending_xid) ||
+				TransactionIdPrecedes(lock->latest_ending_xid, latestXid)) 
+			lock->latest_ending_xid = latestXid;
+		mustwait = true;
+	}
+
+	/* Can release the mutex now */
+	SpinLockRelease(&lock->flex.mutex);
+
+	/*
+	 * If we were not able to perfom the operation immediately, we must wait.
+	 * But we need not retry after being awoken, because the last lock holder
+	 * to release the lock will do the work first, on our behalf.
+	 */
+	if (mustwait)
+	{
+		extraWaits += FlexLockWait(ProcArrayLock, 2);
+		while (extraWaits-- > 0)
+			PGSemaphoreUnlock(&proc->sem);
+	}
+
+	RESUME_INTERRUPTS();
+}
+
+/*
+ * ProcArrayLockRelease - release a previously acquired lock
+ */
+void
+ProcArrayLockRelease(void)
+{
+	volatile ProcArrayLockStruct *lock = ProcArrayLockPointer();
+	PGPROC	   *head;
+	PGPROC	   *ending = NULL;
+	PGPROC	   *proc;
+
+	FlexLockForget(ProcArrayLock);
+
+	/* Acquire mutex.  Time spent holding mutex should be short! */
+	SpinLockAcquire(&lock->flex.mutex);
+
+	/* Release my hold on lock */
+	if (lock->exclusive > 0)
+		lock->exclusive--;
+	else
+	{
+		Assert(lock->shared > 0);
+		lock->shared--;
+	}
+
+	/*
+	 * If the lock is now free, but there are some transactions trying to
+	 * end, we must clear the critical PGPROC fields for them, and save a
+	 * list of them so we can wake them up.
+	 */
+	if (lock->exclusive == 0 && lock->shared == 0 && lock->ending != NULL)
+	{
+		volatile PGPROC *vproc;
+
+		ending = lock->ending;
+		vproc = ending;
+
+		while (vproc != NULL)
+		{
+			volatile PGXACT *pgxact = &ProcGlobal->allPgXact[vproc->pgprocno];
+
+        	pgxact->xid = InvalidTransactionId;
+	        pgxact->xmin = InvalidTransactionId;
+	        /* must be cleared with xid/xmin: */
+	        pgxact->vacuumFlags &= ~PROC_VACUUM_STATE_MASK;
+			pgxact->nxids = 0;
+			pgxact->overflowed = false;
+			vproc = vproc->flWaitLink;
+		}
+
+		/* Also advance global latestCompletedXid */
+		if (TransactionIdPrecedes(ShmemVariableCache->latestCompletedXid,
+								  lock->latest_ending_xid))
+			ShmemVariableCache->latestCompletedXid = lock->latest_ending_xid;
+
+		/* Reset lock state. */
+		lock->ending = NULL;
+		lock->latest_ending_xid = InvalidTransactionId;
+	}
+
+	/*
+	 * 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->flex.head;
+	if (head != NULL)
+	{
+		if (lock->exclusive == 0 && lock->shared == 0 && lock->flex.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->flWaitMode != LW_EXCLUSIVE)
+			{
+				while (proc->flWaitLink != NULL &&
+					   proc->flWaitLink->flWaitMode != LW_EXCLUSIVE)
+					proc = proc->flWaitLink;
+			}
+			/* proc is now the last PGPROC to be released */
+			lock->flex.head = proc->flWaitLink;
+			proc->flWaitLink = NULL;
+			/* prevent additional wakeups until retryer gets to run */
+			lock->flex.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->flex.mutex);
+
+	TRACE_POSTGRESQL_FLEXLOCK_RELEASE(lockid);
+
+	/*
+	 * Awaken any waiters I removed from the queue.
+	 */
+	while (head != NULL)
+	{
+		FlexLockDebug("LWLockRelease", lockid, "release waiter");
+		proc = head;
+		head = proc->flWaitLink;
+		proc->flWaitLink = NULL;
+		proc->flWaitResult = 1;		/* any non-zero value will do */
+		PGSemaphoreUnlock(&proc->sem);
+	}
+
+	/*
+	 * Also awaken any processes whose critical PGPROC fields I cleared
+	 */
+	while (ending != NULL)
+	{
+		FlexLockDebug("LWLockRelease", lockid, "release ending");
+		proc = ending;
+		ending = proc->flWaitLink;
+		proc->flWaitLink = NULL;
+		proc->flWaitResult = 1;		/* any non-zero value will do */
+		PGSemaphoreUnlock(&proc->sem);
+	}
+
+	/*
+	 * Now okay to allow cancel/die interrupts.
+	 */
+	RESUME_INTERRUPTS();
+}
diff --git a/src/include/storage/flexlock_internals.h b/src/include/storage/flexlock_internals.h
index 5f78da7..d1bca45 100644
--- a/src/include/storage/flexlock_internals.h
+++ b/src/include/storage/flexlock_internals.h
@@ -43,6 +43,7 @@ typedef struct FlexLock
 } FlexLock;
 
 #define FLEXLOCK_TYPE_LWLOCK			'l'
+#define FLEXLOCK_TYPE_PROCARRAYLOCK		'p'
 
 typedef union FlexLockPadded
 {
diff --git a/src/include/storage/procarraylock.h b/src/include/storage/procarraylock.h
new file mode 100644
index 0000000..678ca6f
--- /dev/null
+++ b/src/include/storage/procarraylock.h
@@ -0,0 +1,28 @@
+/*-------------------------------------------------------------------------
+ *
+ * procarraylock.h
+ *	  Lock management for the ProcArray
+ *
+ * Portions Copyright (c) 1996-2011, PostgreSQL Global Development Group
+ * Portions Copyright (c) 1994, Regents of the University of California
+ *
+ * src/include/storage/lwlock.h
+ *
+ *-------------------------------------------------------------------------
+ */
+#ifndef PROCARRAYLOCK_H
+#define PROCARRAYLOCK_H
+
+#include "storage/flexlock.h"
+
+typedef enum ProcArrayLockMode
+{
+	PAL_EXCLUSIVE,
+	PAL_SHARED
+} ProcArrayLockMode;
+
+extern void ProcArrayLockAcquire(ProcArrayLockMode mode);
+extern void ProcArrayLockClearTransaction(TransactionId latestXid);
+extern void ProcArrayLockRelease(void);
+
+#endif   /* PROCARRAYLOCK_H */

diff --git a/contrib/pg_stat_statements/pg_stat_statements.c b/contrib/pg_stat_statements/pg_stat_statements.c
index 51b24d0..6167e36 100644
--- a/contrib/pg_stat_statements/pg_stat_statements.c
+++ b/contrib/pg_stat_statements/pg_stat_statements.c
@@ -260,7 +260,7 @@ _PG_init(void)
 	 * resources in pgss_shmem_startup().
 	 */
 	RequestAddinShmemSpace(pgss_memsize());
-	RequestAddinLWLocks(1);
+	RequestAddinFlexLocks(1);
 
 	/*
 	 * Install hooks.
diff --git a/src/backend/postmaster/postmaster.c b/src/backend/postmaster/postmaster.c
index 59d18eb..a07a4c9 100644
--- a/src/backend/postmaster/postmaster.c
+++ b/src/backend/postmaster/postmaster.c
@@ -109,7 +109,6 @@
 #include "postmaster/syslogger.h"
 #include "replication/walsender.h"
 #include "storage/fd.h"
-#include "storage/flexlock_internals.h"
 #include "storage/ipc.h"
 #include "storage/pg_shmem.h"
 #include "storage/pmsignal.h"
diff --git a/src/backend/storage/lmgr/flexlock.c b/src/backend/storage/lmgr/flexlock.c
index f96437b..6145951 100644
--- a/src/backend/storage/lmgr/flexlock.c
+++ b/src/backend/storage/lmgr/flexlock.c
@@ -22,17 +22,16 @@
 #include "postgres.h"
 
 #include "miscadmin.h"
+#include "pg_trace.h"
 #include "access/clog.h"
 #include "access/multixact.h"
 #include "access/subtrans.h"
 #include "commands/async.h"
+#include "storage/flexlock.h"
 #include "storage/flexlock_internals.h"
-#include "storage/lwlock.h"
 #include "storage/predicate.h"
-#include "storage/proc.h"
 #include "storage/procarraylock.h"
 #include "storage/spin.h"
-#include "utils/elog.h"
 
 /*
  * We use this structure to keep track of flex locks held, for release
diff --git a/src/backend/storage/lmgr/proc.c b/src/backend/storage/lmgr/proc.c
index 173b7cb..10ec83b 100644
--- a/src/backend/storage/lmgr/proc.c
+++ b/src/backend/storage/lmgr/proc.c
@@ -755,7 +755,7 @@ ProcKill(int code, Datum arg)
 #endif
 
 	/*
-	 * Release any felx locks I am holding.  There really shouldn't be any, but
+	 * Release any flex locks I am holding.  There really shouldn't be any, but
 	 * it's cheap to check again before we cut the knees off the flex lock
 	 * facility by releasing our PGPROC ...
 	 */
diff --git a/src/include/storage/flexlock_internals.h b/src/include/storage/flexlock_internals.h
index d1bca45..a5c5711 100644
--- a/src/include/storage/flexlock_internals.h
+++ b/src/include/storage/flexlock_internals.h
@@ -16,8 +16,6 @@
 #ifndef FLEXLOCK_INTERNALS_H
 #define FLEXLOCK_INTERNALS_H
 
-#include "pg_trace.h"
-#include "storage/flexlock.h"
 #include "storage/proc.h"
 #include "storage/s_lock.h"
 

diff --git a/src/backend/storage/lmgr/procarraylock.c b/src/backend/storage/lmgr/procarraylock.c
index 7cd4b6b..13b51cb 100644
--- a/src/backend/storage/lmgr/procarraylock.c
+++ b/src/backend/storage/lmgr/procarraylock.c
@@ -153,9 +153,10 @@ ProcArrayLockClearTransaction(TransactionId latestXid)
 {
 	volatile ProcArrayLockStruct *lock = ProcArrayLockPointer();
 	PGPROC	   *proc = MyProc;
-	int			extraWaits = 0;
 	bool		mustwait;
 
+	Assert(TransactionIdIsValid(latestXid));
+
 	HOLD_INTERRUPTS();
 
 	/* Acquire mutex.  Time spent holding mutex should be short! */
@@ -186,8 +187,11 @@ ProcArrayLockClearTransaction(TransactionId latestXid)
 		/* Rats, must wait. */
 		proc->flWaitLink = lock->ending;
 		lock->ending = proc;
-		if (!TransactionIdIsValid(lock->latest_ending_xid) ||
-				TransactionIdPrecedes(lock->latest_ending_xid, latestXid)) 
+		/*
+		 * lock->latest_ending_xid may be invalid, but invalid transaction
+		 * IDs always precede valid ones.
+		 */
+		if (TransactionIdPrecedes(lock->latest_ending_xid, latestXid)) 
 			lock->latest_ending_xid = latestXid;
 		mustwait = true;
 	}
@@ -202,7 +206,9 @@ ProcArrayLockClearTransaction(TransactionId latestXid)
 	 */
 	if (mustwait)
 	{
-		extraWaits += FlexLockWait(ProcArrayLock, 2);
+		int			extraWaits;
+
+		extraWaits = FlexLockWait(ProcArrayLock, 2);
 		while (extraWaits-- > 0)
 			PGSemaphoreUnlock(&proc->sem);
 	}
@@ -247,7 +253,7 @@ ProcArrayLockRelease(void)
 		ending = lock->ending;
 		vproc = ending;
 
-		while (vproc != NULL)
+		do
 		{
 			volatile PGXACT *pgxact = &ProcGlobal->allPgXact[vproc->pgprocno];
 
@@ -258,7 +264,7 @@ ProcArrayLockRelease(void)
 			pgxact->nxids = 0;
 			pgxact->overflowed = false;
 			vproc = vproc->flWaitLink;
-		}
+		} while (vproc != NULL);
 
 		/* Also advance global latestCompletedXid */
 		if (TransactionIdPrecedes(ShmemVariableCache->latestCompletedXid,

diff --git a/contrib/pg_stat_statements/pg_stat_statements.c b/contrib/pg_stat_statements/pg_stat_statements.c
index 8dc3054..6167e36 100644
--- a/contrib/pg_stat_statements/pg_stat_statements.c
+++ b/contrib/pg_stat_statements/pg_stat_statements.c
@@ -105,7 +105,7 @@ typedef struct pgssEntry
  */
 typedef struct pgssSharedState
 {
-	LWLockId	lock;			/* protects hashtable search/modification */
+	FlexLockId	lock;			/* protects hashtable search/modification */
 	int			query_size;		/* max query length in bytes */
 } pgssSharedState;
 
@@ -260,7 +260,7 @@ _PG_init(void)
 	 * resources in pgss_shmem_startup().
 	 */
 	RequestAddinShmemSpace(pgss_memsize());
-	RequestAddinLWLocks(1);
+	RequestAddinFlexLocks(1);
 
 	/*
 	 * Install hooks.
diff --git a/doc/src/sgml/config.sgml b/doc/src/sgml/config.sgml
index d1e628f..8517b36 100644
--- a/doc/src/sgml/config.sgml
+++ b/doc/src/sgml/config.sgml
@@ -6199,14 +6199,14 @@ LOG:  CleanUpLock: deleting: lock(0xb7acd844) id(24688,24696,0,0,0,1)
      </varlistentry>
 
      <varlistentry>
-      <term><varname>trace_lwlocks</varname> (<type>boolean</type>)</term>
+      <term><varname>trace_flexlocks</varname> (<type>boolean</type>)</term>
       <indexterm>
-       <primary><varname>trace_lwlocks</> configuration parameter</primary>
+       <primary><varname>trace_flexlocks</> configuration parameter</primary>
       </indexterm>
       <listitem>
        <para>
-        If on, emit information about lightweight lock usage.  Lightweight
-        locks are intended primarily to provide mutual exclusion of access
+        If on, emit information about FlexLock usage.  FlexLocks
+        are intended primarily to provide mutual exclusion of access
         to shared-memory data structures.
        </para>
        <para>
diff --git a/doc/src/sgml/monitoring.sgml b/doc/src/sgml/monitoring.sgml
index d6056a2..f110253 100644
--- a/doc/src/sgml/monitoring.sgml
+++ b/doc/src/sgml/monitoring.sgml
@@ -1733,49 +1733,49 @@ SELECT pg_stat_get_backend_pid(s.backendid) AS procpid,
       or kilobytes of memory used for an internal sort.</entry>
     </row>
     <row>
-     <entry>lwlock-acquire</entry>
-     <entry>(LWLockId, LWLockMode)</entry>
-     <entry>Probe that fires when an LWLock has been acquired.
-      arg0 is the LWLock's ID.
-      arg1 is the requested lock mode, either exclusive or shared.</entry>
+     <entry>flexlock-acquire</entry>
+     <entry>(FlexLockId, FlexLockMode)</entry>
+     <entry>Probe that fires when an FlexLock has been acquired.
+      arg0 is the FlexLock's ID.
+      arg1 is the requested lock mode.</entry>
     </row>
     <row>
-     <entry>lwlock-release</entry>
-     <entry>(LWLockId)</entry>
-     <entry>Probe that fires when an LWLock has been released (but note
+     <entry>flexlock-release</entry>
+     <entry>(FlexLockId)</entry>
+     <entry>Probe that fires when a FlexLock has been released (but note
       that any released waiters have not yet been awakened).
-      arg0 is the LWLock's ID.</entry>
+      arg0 is the FlexLock's ID.</entry>
     </row>
     <row>
-     <entry>lwlock-wait-start</entry>
-     <entry>(LWLockId, LWLockMode)</entry>
-     <entry>Probe that fires when an LWLock was not immediately available and
+     <entry>flexlock-wait-start</entry>
+     <entry>(FlexLockId, FlexLockMode)</entry>
+     <entry>Probe that fires when an FlexLock was not immediately available and
       a server process has begun to wait for the lock to become available.
-      arg0 is the LWLock's ID.
+      arg0 is the FlexLock's ID.
       arg1 is the requested lock mode, either exclusive or shared.</entry>
     </row>
     <row>
-     <entry>lwlock-wait-done</entry>
-     <entry>(LWLockId, LWLockMode)</entry>
+     <entry>flexlock-wait-done</entry>
+     <entry>(FlexLockId, FlexLockMode)</entry>
      <entry>Probe that fires when a server process has been released from its
-      wait for an LWLock (it does not actually have the lock yet).
-      arg0 is the LWLock's ID.
+      wait for an FlexLock (it does not actually have the lock yet).
+      arg0 is the FlexLock's ID.
       arg1 is the requested lock mode, either exclusive or shared.</entry>
     </row>
     <row>
-     <entry>lwlock-condacquire</entry>
-     <entry>(LWLockId, LWLockMode)</entry>
-     <entry>Probe that fires when an LWLock was successfully acquired when the
-      caller specified no waiting.
-      arg0 is the LWLock's ID.
+     <entry>flexlock-condacquire</entry>
+     <entry>(FlexLockId, FlexLockMode)</entry>
+     <entry>Probe that fires when an FlexLock was successfully acquired when
+      the caller specified no waiting.
+      arg0 is the FlexLock's ID.
       arg1 is the requested lock mode, either exclusive or shared.</entry>
     </row>
     <row>
-     <entry>lwlock-condacquire-fail</entry>
-     <entry>(LWLockId, LWLockMode)</entry>
-     <entry>Probe that fires when an LWLock was not successfully acquired when
-      the caller specified no waiting.
-      arg0 is the LWLock's ID.
+     <entry>flexlock-condacquire-fail</entry>
+     <entry>(FlexLockId, FlexLockMode)</entry>
+     <entry>Probe that fires when an FlexLock was not successfully acquired
+      when the caller specified no waiting.
+      arg0 is the FlexLock's ID.
       arg1 is the requested lock mode, either exclusive or shared.</entry>
     </row>
     <row>
@@ -1822,11 +1822,11 @@ SELECT pg_stat_get_backend_pid(s.backendid) AS procpid,
      <entry>unsigned int</entry>
     </row>
     <row>
-     <entry>LWLockId</entry>
+     <entry>FlexLockId</entry>
      <entry>int</entry>
     </row>
     <row>
-     <entry>LWLockMode</entry>
+     <entry>FlexLockMode</entry>
      <entry>int</entry>
     </row>
     <row>
diff --git a/src/backend/access/transam/slru.c b/src/backend/access/transam/slru.c
index f7caa34..09d5862 100644
--- a/src/backend/access/transam/slru.c
+++ b/src/backend/access/transam/slru.c
@@ -151,7 +151,7 @@ SimpleLruShmemSize(int nslots, int nlsns)
 	sz += MAXALIGN(nslots * sizeof(bool));		/* page_dirty[] */
 	sz += MAXALIGN(nslots * sizeof(int));		/* page_number[] */
 	sz += MAXALIGN(nslots * sizeof(int));		/* page_lru_count[] */
-	sz += MAXALIGN(nslots * sizeof(LWLockId));	/* buffer_locks[] */
+	sz += MAXALIGN(nslots * sizeof(FlexLockId));		/* buffer_locks[] */
 
 	if (nlsns > 0)
 		sz += MAXALIGN(nslots * nlsns * sizeof(XLogRecPtr));	/* group_lsn[] */
@@ -161,7 +161,7 @@ SimpleLruShmemSize(int nslots, int nlsns)
 
 void
 SimpleLruInit(SlruCtl ctl, const char *name, int nslots, int nlsns,
-			  LWLockId ctllock, const char *subdir)
+			  FlexLockId ctllock, const char *subdir)
 {
 	SlruShared	shared;
 	bool		found;
@@ -202,8 +202,8 @@ SimpleLruInit(SlruCtl ctl, const char *name, int nslots, int nlsns,
 		offset += MAXALIGN(nslots * sizeof(int));
 		shared->page_lru_count = (int *) (ptr + offset);
 		offset += MAXALIGN(nslots * sizeof(int));
-		shared->buffer_locks = (LWLockId *) (ptr + offset);
-		offset += MAXALIGN(nslots * sizeof(LWLockId));
+		shared->buffer_locks = (FlexLockId *) (ptr + offset);
+		offset += MAXALIGN(nslots * sizeof(FlexLockId));
 
 		if (nlsns > 0)
 		{
diff --git a/src/backend/access/transam/twophase.c b/src/backend/access/transam/twophase.c
index d2fecb1..943929b 100644
--- a/src/backend/access/transam/twophase.c
+++ b/src/backend/access/transam/twophase.c
@@ -326,9 +326,9 @@ MarkAsPreparing(TransactionId xid, const char *gid,
 	proc->backendId = InvalidBackendId;
 	proc->databaseId = databaseid;
 	proc->roleId = owner;
-	proc->lwWaiting = false;
-	proc->lwExclusive = false;
-	proc->lwWaitLink = NULL;
+	proc->flWaitResult = 0;
+	proc->flWaitMode = 0;
+	proc->flWaitLink = NULL;
 	proc->waitLock = NULL;
 	proc->waitProcLock = NULL;
 	for (i = 0; i < NUM_LOCK_PARTITIONS; i++)
diff --git a/src/backend/access/transam/xact.c b/src/backend/access/transam/xact.c
index c383011..0da2ae5 100644
--- a/src/backend/access/transam/xact.c
+++ b/src/backend/access/transam/xact.c
@@ -2248,7 +2248,7 @@ AbortTransaction(void)
 	 * Releasing LW locks is critical since we might try to grab them again
 	 * while cleaning up!
 	 */
-	LWLockReleaseAll();
+	FlexLockReleaseAll();
 
 	/* Clean up buffer I/O and buffer context locks, too */
 	AbortBufferIO();
@@ -4138,7 +4138,7 @@ AbortSubTransaction(void)
 	 * FIXME This may be incorrect --- Are there some locks we should keep?
 	 * Buffer locks, for example?  I don't think so but I'm not sure.
 	 */
-	LWLockReleaseAll();
+	FlexLockReleaseAll();
 
 	AbortBufferIO();
 	UnlockBuffers();
diff --git a/src/backend/bootstrap/bootstrap.c b/src/backend/bootstrap/bootstrap.c
index 6bf2421..9ceee91 100644
--- a/src/backend/bootstrap/bootstrap.c
+++ b/src/backend/bootstrap/bootstrap.c
@@ -562,13 +562,13 @@ bootstrap_signals(void)
  * Begin shutdown of an auxiliary process.	This is approximately the equivalent
  * of ShutdownPostgres() in postinit.c.  We can't run transactions in an
  * auxiliary process, so most of the work of AbortTransaction() is not needed,
- * but we do need to make sure we've released any LWLocks we are holding.
+ * but we do need to make sure we've released any flex locks we are holding.
  * (This is only critical during an error exit.)
  */
 static void
 ShutdownAuxiliaryProcess(int code, Datum arg)
 {
-	LWLockReleaseAll();
+	FlexLockReleaseAll();
 }
 
 /* ----------------------------------------------------------------
diff --git a/src/backend/postmaster/bgwriter.c b/src/backend/postmaster/bgwriter.c
index cacedab..f33f573 100644
--- a/src/backend/postmaster/bgwriter.c
+++ b/src/backend/postmaster/bgwriter.c
@@ -176,9 +176,10 @@ BackgroundWriterMain(void)
 		/*
 		 * These operations are really just a minimal subset of
 		 * AbortTransaction().	We don't have very many resources to worry
-		 * about in bgwriter, but we do have LWLocks, buffers, and temp files.
+		 * about in bgwriter, but we do have flex locks, buffers, and temp
+		 * files.
 		 */
-		LWLockReleaseAll();
+		FlexLockReleaseAll();
 		AbortBufferIO();
 		UnlockBuffers();
 		/* buffer pins are released here: */
diff --git a/src/backend/postmaster/checkpointer.c b/src/backend/postmaster/checkpointer.c
index e9ae1e8..49f07a7 100644
--- a/src/backend/postmaster/checkpointer.c
+++ b/src/backend/postmaster/checkpointer.c
@@ -281,9 +281,10 @@ CheckpointerMain(void)
 		/*
 		 * These operations are really just a minimal subset of
 		 * AbortTransaction().	We don't have very many resources to worry
-		 * about in checkpointer, but we do have LWLocks, buffers, and temp files.
+		 * about in checkpointer, but we do have flex locks, buffers, and temp
+		 * files.
 		 */
-		LWLockReleaseAll();
+		FlexLockReleaseAll();
 		AbortBufferIO();
 		UnlockBuffers();
 		/* buffer pins are released here: */
diff --git a/src/backend/postmaster/postmaster.c b/src/backend/postmaster/postmaster.c
index 963189d..a07a4c9 100644
--- a/src/backend/postmaster/postmaster.c
+++ b/src/backend/postmaster/postmaster.c
@@ -404,8 +404,6 @@ typedef struct
 typedef int InheritableSocket;
 #endif
 
-typedef struct LWLock LWLock;	/* ugly kluge */
-
 /*
  * Structure contains all variables passed to exec:ed backends
  */
@@ -426,7 +424,7 @@ typedef struct
 	slock_t    *ShmemLock;
 	VariableCache ShmemVariableCache;
 	Backend    *ShmemBackendArray;
-	LWLock	   *LWLockArray;
+	FlexLock   *FlexLockArray;
 	slock_t    *ProcStructLock;
 	PROC_HDR   *ProcGlobal;
 	PGPROC	   *AuxiliaryProcs;
@@ -4676,7 +4674,6 @@ MaxLivePostmasterChildren(void)
  * functions
  */
 extern slock_t *ShmemLock;
-extern LWLock *LWLockArray;
 extern slock_t *ProcStructLock;
 extern PGPROC *AuxiliaryProcs;
 extern PMSignalData *PMSignalState;
@@ -4721,7 +4718,7 @@ save_backend_variables(BackendParameters *param, Port *port,
 	param->ShmemVariableCache = ShmemVariableCache;
 	param->ShmemBackendArray = ShmemBackendArray;
 
-	param->LWLockArray = LWLockArray;
+	param->FlexLockArray = FlexLockArray;
 	param->ProcStructLock = ProcStructLock;
 	param->ProcGlobal = ProcGlobal;
 	param->AuxiliaryProcs = AuxiliaryProcs;
@@ -4945,7 +4942,7 @@ restore_backend_variables(BackendParameters *param, Port *port)
 	ShmemVariableCache = param->ShmemVariableCache;
 	ShmemBackendArray = param->ShmemBackendArray;
 
-	LWLockArray = param->LWLockArray;
+	FlexLockArray = param->FlexLockArray;
 	ProcStructLock = param->ProcStructLock;
 	ProcGlobal = param->ProcGlobal;
 	AuxiliaryProcs = param->AuxiliaryProcs;
diff --git a/src/backend/postmaster/walwriter.c b/src/backend/postmaster/walwriter.c
index 157728e..587443d 100644
--- a/src/backend/postmaster/walwriter.c
+++ b/src/backend/postmaster/walwriter.c
@@ -167,9 +167,9 @@ WalWriterMain(void)
 		/*
 		 * These operations are really just a minimal subset of
 		 * AbortTransaction().	We don't have very many resources to worry
-		 * about in walwriter, but we do have LWLocks, and perhaps buffers?
+		 * about in walwriter, but we do have flex locks, and perhaps buffers?
 		 */
-		LWLockReleaseAll();
+		FlexLockReleaseAll();
 		AbortBufferIO();
 		UnlockBuffers();
 		/* buffer pins are released here: */
diff --git a/src/backend/storage/buffer/bufmgr.c b/src/backend/storage/buffer/bufmgr.c
index 71fe8c6..4c4959c 100644
--- a/src/backend/storage/buffer/bufmgr.c
+++ b/src/backend/storage/buffer/bufmgr.c
@@ -141,7 +141,7 @@ PrefetchBuffer(Relation reln, ForkNumber forkNum, BlockNumber blockNum)
 	{
 		BufferTag	newTag;		/* identity of requested block */
 		uint32		newHash;	/* hash value for newTag */
-		LWLockId	newPartitionLock;	/* buffer partition lock for it */
+		FlexLockId	newPartitionLock;	/* buffer partition lock for it */
 		int			buf_id;
 
 		/* create a tag so we can lookup the buffer */
@@ -514,10 +514,10 @@ BufferAlloc(SMgrRelation smgr, char relpersistence, ForkNumber forkNum,
 {
 	BufferTag	newTag;			/* identity of requested block */
 	uint32		newHash;		/* hash value for newTag */
-	LWLockId	newPartitionLock;		/* buffer partition lock for it */
+	FlexLockId	newPartitionLock;		/* buffer partition lock for it */
 	BufferTag	oldTag;			/* previous identity of selected buffer */
 	uint32		oldHash;		/* hash value for oldTag */
-	LWLockId	oldPartitionLock;		/* buffer partition lock for it */
+	FlexLockId	oldPartitionLock;		/* buffer partition lock for it */
 	BufFlags	oldFlags;
 	int			buf_id;
 	volatile BufferDesc *buf;
@@ -857,7 +857,7 @@ InvalidateBuffer(volatile BufferDesc *buf)
 {
 	BufferTag	oldTag;
 	uint32		oldHash;		/* hash value for oldTag */
-	LWLockId	oldPartitionLock;		/* buffer partition lock for it */
+	FlexLockId	oldPartitionLock;		/* buffer partition lock for it */
 	BufFlags	oldFlags;
 
 	/* Save the original buffer tag before dropping the spinlock */
diff --git a/src/backend/storage/ipc/ipci.c b/src/backend/storage/ipc/ipci.c
index bb8b832..a2c570a 100644
--- a/src/backend/storage/ipc/ipci.c
+++ b/src/backend/storage/ipc/ipci.c
@@ -113,7 +113,7 @@ CreateSharedMemoryAndSemaphores(bool makePrivate, int port)
 		size = add_size(size, SUBTRANSShmemSize());
 		size = add_size(size, TwoPhaseShmemSize());
 		size = add_size(size, MultiXactShmemSize());
-		size = add_size(size, LWLockShmemSize());
+		size = add_size(size, FlexLockShmemSize());
 		size = add_size(size, ProcArrayShmemSize());
 		size = add_size(size, BackendStatusShmemSize());
 		size = add_size(size, SInvalShmemSize());
@@ -179,7 +179,7 @@ CreateSharedMemoryAndSemaphores(bool makePrivate, int port)
 	 * needed for InitShmemIndex.
 	 */
 	if (!IsUnderPostmaster)
-		CreateLWLocks();
+		CreateFlexLocks();
 
 	/*
 	 * Set up shmem.c index hashtable
diff --git a/src/backend/storage/lmgr/Makefile b/src/backend/storage/lmgr/Makefile
index e12a854..3730e51 100644
--- a/src/backend/storage/lmgr/Makefile
+++ b/src/backend/storage/lmgr/Makefile
@@ -12,7 +12,8 @@ subdir = src/backend/storage/lmgr
 top_builddir = ../../../..
 include $(top_builddir)/src/Makefile.global
 
-OBJS = lmgr.o lock.o proc.o deadlock.o lwlock.o spin.o s_lock.o predicate.o
+OBJS = flexlock.o lmgr.o lock.o proc.o deadlock.o lwlock.o spin.o s_lock.o \
+	predicate.o
 
 include $(top_srcdir)/src/backend/common.mk
 
diff --git a/src/backend/storage/lmgr/flexlock.c b/src/backend/storage/lmgr/flexlock.c
new file mode 100644
index 0000000..cf0004b
--- /dev/null
+++ b/src/backend/storage/lmgr/flexlock.c
@@ -0,0 +1,271 @@
+/*-------------------------------------------------------------------------
+ *
+ * flexlock.c
+ *	  Low-level routines for managing flex locks.
+ *
+ * Flex locks are intended primarily to provide mutual exclusion of access
+ * to shared-memory data structures.  Most, but not all, flex locks are
+ * lightweight locks (LWLocks).  This file contains support routines that
+ * are used for all types of flex locks, including lwlocks.  User-level
+ * locking should be done with the full lock manager --- which depends on
+ * LWLocks to protect its shared state.
+ *
+ * Portions Copyright (c) 1996-2011, PostgreSQL Global Development Group
+ * Portions Copyright (c) 1994, Regents of the University of California
+ *
+ * IDENTIFICATION
+ *	  src/backend/storage/lmgr/flexlock.c
+ *
+ *-------------------------------------------------------------------------
+ */
+
+#include "postgres.h"
+
+#include "miscadmin.h"
+#include "pg_trace.h"
+#include "access/clog.h"
+#include "access/multixact.h"
+#include "access/subtrans.h"
+#include "commands/async.h"
+#include "storage/flexlock.h"
+#include "storage/flexlock_internals.h"
+#include "storage/predicate.h"
+#include "storage/spin.h"
+
+int	num_held_flexlocks = 0;
+FlexLockId held_flexlocks[MAX_SIMUL_FLEXLOCKS];
+
+static int	lock_addin_request = 0;
+static bool lock_addin_request_allowed = true;
+
+#ifdef LOCK_DEBUG
+bool		Trace_flexlocks = false;
+#endif
+
+/*
+ * This points to the array of FlexLocks in shared memory.  Backends inherit
+ * the pointer by fork from the postmaster (except in the EXEC_BACKEND case,
+ * where we have special measures to pass it down).
+ */
+FlexLockPadded *FlexLockArray = NULL;
+
+/* We use the ShmemLock spinlock to protect LWLockAssign */
+extern slock_t *ShmemLock;
+
+static void FlexLockInit(FlexLock *flex, char locktype);
+
+/*
+ * Compute number of FlexLocks to allocate.
+ */
+int
+NumFlexLocks(void)
+{
+	int			numLocks;
+
+	/*
+	 * Possibly this logic should be spread out among the affected modules,
+	 * the same way that shmem space estimation is done.  But for now, there
+	 * are few enough users of FlexLocks that we can get away with just keeping
+	 * the knowledge here.
+	 */
+
+	/* Predefined FlexLocks */
+	numLocks = (int) NumFixedFlexLocks;
+
+	/* bufmgr.c needs two for each shared buffer */
+	numLocks += 2 * NBuffers;
+
+	/* proc.c needs one for each backend or auxiliary process */
+	numLocks += MaxBackends + NUM_AUXILIARY_PROCS;
+
+	/* clog.c needs one per CLOG buffer */
+	numLocks += NUM_CLOG_BUFFERS;
+
+	/* subtrans.c needs one per SubTrans buffer */
+	numLocks += NUM_SUBTRANS_BUFFERS;
+
+	/* multixact.c needs two SLRU areas */
+	numLocks += NUM_MXACTOFFSET_BUFFERS + NUM_MXACTMEMBER_BUFFERS;
+
+	/* async.c needs one per Async buffer */
+	numLocks += NUM_ASYNC_BUFFERS;
+
+	/* predicate.c needs one per old serializable xid buffer */
+	numLocks += NUM_OLDSERXID_BUFFERS;
+
+	/*
+	 * Add any requested by loadable modules; for backwards-compatibility
+	 * reasons, allocate at least NUM_USER_DEFINED_FLEXLOCKS of them even if
+	 * there are no explicit requests.
+	 */
+	lock_addin_request_allowed = false;
+	numLocks += Max(lock_addin_request, NUM_USER_DEFINED_FLEXLOCKS);
+
+	return numLocks;
+}
+
+
+/*
+ * RequestAddinFlexLocks
+ *		Request that extra FlexLocks be allocated for use by
+ *		a loadable module.
+ *
+ * This is only useful if called from the _PG_init hook of a library that
+ * is loaded into the postmaster via shared_preload_libraries.	Once
+ * shared memory has been allocated, calls will be ignored.  (We could
+ * raise an error, but it seems better to make it a no-op, so that
+ * libraries containing such calls can be reloaded if needed.)
+ */
+void
+RequestAddinFlexLocks(int n)
+{
+	if (IsUnderPostmaster || !lock_addin_request_allowed)
+		return;					/* too late */
+	lock_addin_request += n;
+}
+
+
+/*
+ * Compute shmem space needed for FlexLocks.
+ */
+Size
+FlexLockShmemSize(void)
+{
+	Size		size;
+	int			numLocks = NumFlexLocks();
+
+	/* Space for the FlexLock array. */
+	size = mul_size(numLocks, FLEX_LOCK_BYTES);
+
+	/* Space for dynamic allocation counter, plus room for alignment. */
+	size = add_size(size, 2 * sizeof(int) + FLEX_LOCK_BYTES);
+
+	return size;
+}
+
+/*
+ * Allocate shmem space for FlexLocks and initialize the locks.
+ */
+void
+CreateFlexLocks(void)
+{
+	int			numLocks = NumFlexLocks();
+	Size		spaceLocks = FlexLockShmemSize();
+	FlexLockPadded *lock;
+	int		   *FlexLockCounter;
+	char	   *ptr;
+	int			id;
+
+	/* Allocate and zero space */
+	ptr = (char *) ShmemAlloc(spaceLocks);
+	memset(ptr, 0, spaceLocks);
+
+	/* Leave room for dynamic allocation counter */
+	ptr += 2 * sizeof(int);
+
+	/* Ensure desired alignment of FlexLock array */
+	ptr += FLEX_LOCK_BYTES - ((uintptr_t) ptr) % FLEX_LOCK_BYTES;
+
+	FlexLockArray = (FlexLockPadded *) ptr;
+
+	/* All of the "fixed" FlexLocks are LWLocks. */
+	for (id = 0, lock = FlexLockArray; id < NumFixedFlexLocks; id++, lock++)
+		FlexLockInit(&lock->flex, FLEXLOCK_TYPE_LWLOCK);
+
+	/*
+	 * Initialize the dynamic-allocation counter, which is stored just before
+	 * the first FlexLock.
+	 */
+	FlexLockCounter = (int *) ((char *) FlexLockArray - 2 * sizeof(int));
+	FlexLockCounter[0] = (int) NumFixedFlexLocks;
+	FlexLockCounter[1] = numLocks;
+}
+
+/*
+ * FlexLockAssign - assign a dynamically-allocated FlexLock number
+ *
+ * We interlock this using the same spinlock that is used to protect
+ * ShmemAlloc().  Interlocking is not really necessary during postmaster
+ * startup, but it is needed if any user-defined code tries to allocate
+ * LWLocks after startup.
+ */
+FlexLockId
+FlexLockAssign(char locktype)
+{
+	FlexLockId	result;
+
+	/* use volatile pointer to prevent code rearrangement */
+	volatile int *FlexLockCounter;
+
+	FlexLockCounter = (int *) ((char *) FlexLockArray - 2 * sizeof(int));
+	SpinLockAcquire(ShmemLock);
+	if (FlexLockCounter[0] >= FlexLockCounter[1])
+	{
+		SpinLockRelease(ShmemLock);
+		elog(ERROR, "no more FlexLockIds available");
+	}
+	result = (FlexLockId) (FlexLockCounter[0]++);
+	SpinLockRelease(ShmemLock);
+
+	FlexLockInit(&FlexLockArray[result].flex, locktype);
+
+	return result;
+}
+
+/*
+ * Initialize a FlexLock.
+ */
+static void
+FlexLockInit(FlexLock *flex, char locktype)
+{
+	SpinLockInit(&flex->mutex);
+	flex->releaseOK = true;
+	flex->locktype = locktype;
+	/*
+	 * We might need to think a little harder about what should happen here
+	 * if some future type of FlexLock requires more initialization than this.
+	 * For now, this will suffice.
+	 */
+}
+
+/*
+ * FlexLockReleaseAll - release all currently-held locks
+ *
+ * Used to clean up after ereport(ERROR). An important difference between this
+ * function and retail LWLockRelease calls is that InterruptHoldoffCount is
+ * unchanged by this operation.  This is necessary since InterruptHoldoffCount
+ * has been set to an appropriate level earlier in error recovery. We could
+ * decrement it below zero if we allow it to drop for each released lock!
+ */
+void
+FlexLockReleaseAll(void)
+{
+	while (num_held_flexlocks > 0)
+	{
+		HOLD_INTERRUPTS();		/* match the upcoming RESUME_INTERRUPTS */
+
+		/*
+		 * FLEXTODO: When we have multiple types of flex locks, this will
+		 * need to call the appropriate release function for each lock type.
+		 */
+		LWLockRelease(held_flexlocks[num_held_flexlocks - 1]);
+	}
+}
+
+/*
+ * FlexLockHeldByMe - test whether my process currently holds a lock
+ *
+ * This is meant as debug support only.  We do not consider the lock mode.
+ */
+bool
+FlexLockHeldByMe(FlexLockId id)
+{
+	int			i;
+
+	for (i = 0; i < num_held_flexlocks; i++)
+	{
+		if (held_flexlocks[i] == id)
+			return true;
+	}
+	return false;
+}
diff --git a/src/backend/storage/lmgr/lock.c b/src/backend/storage/lmgr/lock.c
index 3ba4671..f594983 100644
--- a/src/backend/storage/lmgr/lock.c
+++ b/src/backend/storage/lmgr/lock.c
@@ -591,7 +591,7 @@ LockAcquireExtended(const LOCKTAG *locktag,
 	bool		found;
 	ResourceOwner owner;
 	uint32		hashcode;
-	LWLockId	partitionLock;
+	FlexLockId	partitionLock;
 	int			status;
 	bool		log_lock = false;
 
@@ -1546,7 +1546,7 @@ LockRelease(const LOCKTAG *locktag, LOCKMODE lockmode, bool sessionLock)
 	LOCALLOCK  *locallock;
 	LOCK	   *lock;
 	PROCLOCK   *proclock;
-	LWLockId	partitionLock;
+	FlexLockId	partitionLock;
 	bool		wakeupNeeded;
 
 	if (lockmethodid <= 0 || lockmethodid >= lengthof(LockMethods))
@@ -1912,7 +1912,7 @@ LockReleaseAll(LOCKMETHODID lockmethodid, bool allLocks)
 	 */
 	for (partition = 0; partition < NUM_LOCK_PARTITIONS; partition++)
 	{
-		LWLockId	partitionLock = FirstLockMgrLock + partition;
+		FlexLockId	partitionLock = FirstLockMgrLock + partition;
 		SHM_QUEUE  *procLocks = &(MyProc->myProcLocks[partition]);
 
 		proclock = (PROCLOCK *) SHMQueueNext(procLocks, procLocks,
@@ -2197,7 +2197,7 @@ static bool
 FastPathTransferRelationLocks(LockMethod lockMethodTable, const LOCKTAG *locktag,
 					  uint32 hashcode)
 {
-	LWLockId		partitionLock = LockHashPartitionLock(hashcode);
+	FlexLockId		partitionLock = LockHashPartitionLock(hashcode);
 	Oid				relid = locktag->locktag_field2;
 	uint32			i;
 
@@ -2281,7 +2281,7 @@ FastPathGetRelationLockEntry(LOCALLOCK *locallock)
 	LockMethod		lockMethodTable = LockMethods[DEFAULT_LOCKMETHOD];
 	LOCKTAG		   *locktag = &locallock->tag.lock;
 	PROCLOCK	   *proclock = NULL;
-	LWLockId		partitionLock = LockHashPartitionLock(locallock->hashcode);
+	FlexLockId		partitionLock = LockHashPartitionLock(locallock->hashcode);
 	Oid				relid = locktag->locktag_field2;
 	uint32			f;
 
@@ -2382,7 +2382,7 @@ GetLockConflicts(const LOCKTAG *locktag, LOCKMODE lockmode)
 	SHM_QUEUE  *procLocks;
 	PROCLOCK   *proclock;
 	uint32		hashcode;
-	LWLockId	partitionLock;
+	FlexLockId	partitionLock;
 	int			count = 0;
 	int			fast_count = 0;
 
@@ -2593,7 +2593,7 @@ LockRefindAndRelease(LockMethod lockMethodTable, PGPROC *proc,
 	PROCLOCKTAG proclocktag;
 	uint32		hashcode;
 	uint32		proclock_hashcode;
-	LWLockId	partitionLock;
+	FlexLockId	partitionLock;
 	bool		wakeupNeeded;
 
 	hashcode = LockTagHashCode(locktag);
@@ -2827,7 +2827,7 @@ PostPrepare_Locks(TransactionId xid)
 	 */
 	for (partition = 0; partition < NUM_LOCK_PARTITIONS; partition++)
 	{
-		LWLockId	partitionLock = FirstLockMgrLock + partition;
+		FlexLockId	partitionLock = FirstLockMgrLock + partition;
 		SHM_QUEUE  *procLocks = &(MyProc->myProcLocks[partition]);
 
 		proclock = (PROCLOCK *) SHMQueueNext(procLocks, procLocks,
@@ -3343,7 +3343,7 @@ lock_twophase_recover(TransactionId xid, uint16 info,
 	uint32		hashcode;
 	uint32		proclock_hashcode;
 	int			partition;
-	LWLockId	partitionLock;
+	FlexLockId	partitionLock;
 	LockMethod	lockMethodTable;
 
 	Assert(len == sizeof(TwoPhaseLockRecord));
diff --git a/src/backend/storage/lmgr/lwlock.c b/src/backend/storage/lmgr/lwlock.c
index 079eb29..bdb5f6e 100644
--- a/src/backend/storage/lmgr/lwlock.c
+++ b/src/backend/storage/lmgr/lwlock.c
@@ -21,74 +21,23 @@
  */
 #include "postgres.h"
 
-#include "access/clog.h"
-#include "access/multixact.h"
-#include "access/subtrans.h"
-#include "commands/async.h"
 #include "miscadmin.h"
 #include "pg_trace.h"
+#include "storage/flexlock_internals.h"
 #include "storage/ipc.h"
-#include "storage/predicate.h"
 #include "storage/proc.h"
 #include "storage/spin.h"
 
-
-/* We use the ShmemLock spinlock to protect LWLockAssign */
-extern slock_t *ShmemLock;
-
-
 typedef struct LWLock
 {
-	slock_t		mutex;			/* Protects LWLock and queue of PGPROCs */
-	bool		releaseOK;		/* T if ok to release waiters */
+	FlexLock	flex;			/* common FlexLock infrastructure */
 	char		exclusive;		/* # of exclusive holders (0 or 1) */
 	int			shared;			/* # of shared holders (0..MaxBackends) */
-	PGPROC	   *head;			/* head of list of waiting PGPROCs */
-	PGPROC	   *tail;			/* tail of list of waiting PGPROCs */
-	/* tail is undefined when head is NULL */
 } LWLock;
 
-/*
- * All the LWLock structs are allocated as an array in shared memory.
- * (LWLockIds are indexes into the array.)	We force the array stride to
- * be a power of 2, which saves a few cycles in indexing, but more
- * importantly also ensures that individual LWLocks don't cross cache line
- * boundaries.	This reduces cache contention problems, especially on AMD
- * Opterons.  (Of course, we have to also ensure that the array start
- * address is suitably aligned.)
- *
- * LWLock is between 16 and 32 bytes on all known platforms, so these two
- * cases are sufficient.
- */
-#define LWLOCK_PADDED_SIZE	(sizeof(LWLock) <= 16 ? 16 : 32)
-
-typedef union LWLockPadded
-{
-	LWLock		lock;
-	char		pad[LWLOCK_PADDED_SIZE];
-} LWLockPadded;
-
-/*
- * This points to the array of LWLocks in shared memory.  Backends inherit
- * the pointer by fork from the postmaster (except in the EXEC_BACKEND case,
- * where we have special measures to pass it down).
- */
-NON_EXEC_STATIC LWLockPadded *LWLockArray = NULL;
-
-
-/*
- * We use this structure to keep track of locked LWLocks for release
- * during error recovery.  The maximum size could be determined at runtime
- * if necessary, but it seems unlikely that more than a few locks could
- * ever be held simultaneously.
- */
-#define MAX_SIMUL_LWLOCKS	100
-
-static int	num_held_lwlocks = 0;
-static LWLockId held_lwlocks[MAX_SIMUL_LWLOCKS];
-
-static int	lock_addin_request = 0;
-static bool lock_addin_request_allowed = true;
+#define	LWLockPointer(lockid) \
+	(AssertMacro(FlexLockArray[lockid].flex.locktype == FLEXLOCK_TYPE_LWLOCK), \
+	 (volatile LWLock *) &FlexLockArray[lockid])
 
 #ifdef LWLOCK_STATS
 static int	counts_for_pid = 0;
@@ -98,27 +47,17 @@ static int *block_counts;
 #endif
 
 #ifdef LOCK_DEBUG
-bool		Trace_lwlocks = false;
-
 inline static void
-PRINT_LWDEBUG(const char *where, LWLockId lockid, const volatile LWLock *lock)
+PRINT_LWDEBUG(const char *where, FlexLockId lockid, const volatile LWLock *lock)
 {
-	if (Trace_lwlocks)
+	if (Trace_flexlocks)
 		elog(LOG, "%s(%d): excl %d shared %d head %p rOK %d",
 			 where, (int) lockid,
-			 (int) lock->exclusive, lock->shared, lock->head,
-			 (int) lock->releaseOK);
-}
-
-inline static void
-LOG_LWDEBUG(const char *where, LWLockId lockid, const char *msg)
-{
-	if (Trace_lwlocks)
-		elog(LOG, "%s(%d): %s", where, (int) lockid, msg);
+			 (int) lock->exclusive, lock->shared, lock->flex.head,
+			 (int) lock->flex.releaseOK);
 }
 #else							/* not LOCK_DEBUG */
 #define PRINT_LWDEBUG(a,b,c)
-#define LOG_LWDEBUG(a,b,c)
 #endif   /* LOCK_DEBUG */
 
 #ifdef LWLOCK_STATS
@@ -127,8 +66,8 @@ static void
 print_lwlock_stats(int code, Datum arg)
 {
 	int			i;
-	int		   *LWLockCounter = (int *) ((char *) LWLockArray - 2 * sizeof(int));
-	int			numLocks = LWLockCounter[1];
+	int		   *FlexLockCounter = (int *) ((char *) FlexLockArray - 2 * sizeof(int));
+	int			numLocks = FlexLockCounter[1];
 
 	/* Grab an LWLock to keep different backends from mixing reports */
 	LWLockAcquire(0, LW_EXCLUSIVE);
@@ -145,173 +84,15 @@ print_lwlock_stats(int code, Datum arg)
 }
 #endif   /* LWLOCK_STATS */
 
-
 /*
- * Compute number of LWLocks to allocate.
+ * LWLockAssign - initialize a new lwlock and return its ID
  */
-int
-NumLWLocks(void)
-{
-	int			numLocks;
-
-	/*
-	 * Possibly this logic should be spread out among the affected modules,
-	 * the same way that shmem space estimation is done.  But for now, there
-	 * are few enough users of LWLocks that we can get away with just keeping
-	 * the knowledge here.
-	 */
-
-	/* Predefined LWLocks */
-	numLocks = (int) NumFixedLWLocks;
-
-	/* bufmgr.c needs two for each shared buffer */
-	numLocks += 2 * NBuffers;
-
-	/* proc.c needs one for each backend or auxiliary process */
-	numLocks += MaxBackends + NUM_AUXILIARY_PROCS;
-
-	/* clog.c needs one per CLOG buffer */
-	numLocks += NUM_CLOG_BUFFERS;
-
-	/* subtrans.c needs one per SubTrans buffer */
-	numLocks += NUM_SUBTRANS_BUFFERS;
-
-	/* multixact.c needs two SLRU areas */
-	numLocks += NUM_MXACTOFFSET_BUFFERS + NUM_MXACTMEMBER_BUFFERS;
-
-	/* async.c needs one per Async buffer */
-	numLocks += NUM_ASYNC_BUFFERS;
-
-	/* predicate.c needs one per old serializable xid buffer */
-	numLocks += NUM_OLDSERXID_BUFFERS;
-
-	/*
-	 * Add any requested by loadable modules; for backwards-compatibility
-	 * reasons, allocate at least NUM_USER_DEFINED_LWLOCKS of them even if
-	 * there are no explicit requests.
-	 */
-	lock_addin_request_allowed = false;
-	numLocks += Max(lock_addin_request, NUM_USER_DEFINED_LWLOCKS);
-
-	return numLocks;
-}
-
-
-/*
- * RequestAddinLWLocks
- *		Request that extra LWLocks be allocated for use by
- *		a loadable module.
- *
- * This is only useful if called from the _PG_init hook of a library that
- * is loaded into the postmaster via shared_preload_libraries.	Once
- * shared memory has been allocated, calls will be ignored.  (We could
- * raise an error, but it seems better to make it a no-op, so that
- * libraries containing such calls can be reloaded if needed.)
- */
-void
-RequestAddinLWLocks(int n)
-{
-	if (IsUnderPostmaster || !lock_addin_request_allowed)
-		return;					/* too late */
-	lock_addin_request += n;
-}
-
-
-/*
- * Compute shmem space needed for LWLocks.
- */
-Size
-LWLockShmemSize(void)
-{
-	Size		size;
-	int			numLocks = NumLWLocks();
-
-	/* Space for the LWLock array. */
-	size = mul_size(numLocks, sizeof(LWLockPadded));
-
-	/* Space for dynamic allocation counter, plus room for alignment. */
-	size = add_size(size, 2 * sizeof(int) + LWLOCK_PADDED_SIZE);
-
-	return size;
-}
-
-
-/*
- * Allocate shmem space for LWLocks and initialize the locks.
- */
-void
-CreateLWLocks(void)
-{
-	int			numLocks = NumLWLocks();
-	Size		spaceLocks = LWLockShmemSize();
-	LWLockPadded *lock;
-	int		   *LWLockCounter;
-	char	   *ptr;
-	int			id;
-
-	/* Allocate space */
-	ptr = (char *) ShmemAlloc(spaceLocks);
-
-	/* Leave room for dynamic allocation counter */
-	ptr += 2 * sizeof(int);
-
-	/* Ensure desired alignment of LWLock array */
-	ptr += LWLOCK_PADDED_SIZE - ((uintptr_t) ptr) % LWLOCK_PADDED_SIZE;
-
-	LWLockArray = (LWLockPadded *) ptr;
-
-	/*
-	 * Initialize all LWLocks to "unlocked" state
-	 */
-	for (id = 0, lock = LWLockArray; id < numLocks; id++, lock++)
-	{
-		SpinLockInit(&lock->lock.mutex);
-		lock->lock.releaseOK = true;
-		lock->lock.exclusive = 0;
-		lock->lock.shared = 0;
-		lock->lock.head = NULL;
-		lock->lock.tail = NULL;
-	}
-
-	/*
-	 * Initialize the dynamic-allocation counter, which is stored just before
-	 * the first LWLock.
-	 */
-	LWLockCounter = (int *) ((char *) LWLockArray - 2 * sizeof(int));
-	LWLockCounter[0] = (int) NumFixedLWLocks;
-	LWLockCounter[1] = numLocks;
-}
-
-
-/*
- * LWLockAssign - assign a dynamically-allocated LWLock number
- *
- * We interlock this using the same spinlock that is used to protect
- * ShmemAlloc().  Interlocking is not really necessary during postmaster
- * startup, but it is needed if any user-defined code tries to allocate
- * LWLocks after startup.
- */
-LWLockId
+FlexLockId
 LWLockAssign(void)
 {
-	LWLockId	result;
-
-	/* use volatile pointer to prevent code rearrangement */
-	volatile int *LWLockCounter;
-
-	LWLockCounter = (int *) ((char *) LWLockArray - 2 * sizeof(int));
-	SpinLockAcquire(ShmemLock);
-	if (LWLockCounter[0] >= LWLockCounter[1])
-	{
-		SpinLockRelease(ShmemLock);
-		elog(ERROR, "no more LWLockIds available");
-	}
-	result = (LWLockId) (LWLockCounter[0]++);
-	SpinLockRelease(ShmemLock);
-	return result;
+	return FlexLockAssign(FLEXLOCK_TYPE_LWLOCK);
 }
 
-
 /*
  * LWLockAcquire - acquire a lightweight lock in the specified mode
  *
@@ -320,9 +101,9 @@ LWLockAssign(void)
  * Side effect: cancel/die interrupts are held off until lock release.
  */
 void
-LWLockAcquire(LWLockId lockid, LWLockMode mode)
+LWLockAcquire(FlexLockId lockid, LWLockMode mode)
 {
-	volatile LWLock *lock = &(LWLockArray[lockid].lock);
+	volatile LWLock *lock = LWLockPointer(lockid);
 	PGPROC	   *proc = MyProc;
 	bool		retry = false;
 	int			extraWaits = 0;
@@ -333,8 +114,8 @@ LWLockAcquire(LWLockId lockid, LWLockMode mode)
 	/* 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];
+		int		   *FlexLockCounter = (int *) ((char *) FlexLockArray - 2 * sizeof(int));
+		int			numLocks = FlexLockCounter[1];
 
 		sh_acquire_counts = calloc(numLocks, sizeof(int));
 		ex_acquire_counts = calloc(numLocks, sizeof(int));
@@ -356,10 +137,6 @@ LWLockAcquire(LWLockId lockid, LWLockMode mode)
 	 */
 	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
@@ -388,11 +165,11 @@ LWLockAcquire(LWLockId lockid, LWLockMode mode)
 		bool		mustwait;
 
 		/* Acquire mutex.  Time spent holding mutex should be short! */
-		SpinLockAcquire(&lock->mutex);
+		SpinLockAcquire(&lock->flex.mutex);
 
 		/* If retrying, allow LWLockRelease to release waiters again */
 		if (retry)
-			lock->releaseOK = true;
+			lock->flex.releaseOK = true;
 
 		/* If I can get the lock, do so quickly. */
 		if (mode == LW_EXCLUSIVE)
@@ -419,72 +196,30 @@ LWLockAcquire(LWLockId lockid, LWLockMode mode)
 		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;
+		/* Add myself to wait queue. */
+		FlexLockJoinWaitQueue(lock, (int) mode);
 
 		/* 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");
+		SpinLockRelease(&lock->flex.mutex);
+
+		/* Wait until awakened. */
+		FlexLockWait(lockid, mode, extraWaits);
 
 #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);
+	SpinLockRelease(&lock->flex.mutex);
 
-	TRACE_POSTGRESQL_LWLOCK_ACQUIRE(lockid, mode);
+	TRACE_POSTGRESQL_FLEXLOCK_ACQUIRE(lockid, mode);
 
 	/* Add lock to list of locks held by this backend */
-	held_lwlocks[num_held_lwlocks++] = lockid;
+	FlexLockRemember(lockid);
 
 	/*
 	 * Fix the process wait semaphore's count for any absorbed wakeups.
@@ -501,17 +236,13 @@ LWLockAcquire(LWLockId lockid, LWLockMode mode)
  * If successful, cancel/die interrupts are held off until lock release.
  */
 bool
-LWLockConditionalAcquire(LWLockId lockid, LWLockMode mode)
+LWLockConditionalAcquire(FlexLockId lockid, LWLockMode mode)
 {
-	volatile LWLock *lock = &(LWLockArray[lockid].lock);
+	volatile LWLock *lock = LWLockPointer(lockid);
 	bool		mustwait;
 
 	PRINT_LWDEBUG("LWLockConditionalAcquire", lockid, lock);
 
-	/* 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
@@ -520,7 +251,7 @@ LWLockConditionalAcquire(LWLockId lockid, LWLockMode mode)
 	HOLD_INTERRUPTS();
 
 	/* Acquire mutex.  Time spent holding mutex should be short! */
-	SpinLockAcquire(&lock->mutex);
+	SpinLockAcquire(&lock->flex.mutex);
 
 	/* If I can get the lock, do so quickly. */
 	if (mode == LW_EXCLUSIVE)
@@ -545,20 +276,20 @@ LWLockConditionalAcquire(LWLockId lockid, LWLockMode mode)
 	}
 
 	/* We are done updating shared state of the lock itself. */
-	SpinLockRelease(&lock->mutex);
+	SpinLockRelease(&lock->flex.mutex);
 
 	if (mustwait)
 	{
 		/* Failed to get lock, so release interrupt holdoff */
 		RESUME_INTERRUPTS();
-		LOG_LWDEBUG("LWLockConditionalAcquire", lockid, "failed");
-		TRACE_POSTGRESQL_LWLOCK_CONDACQUIRE_FAIL(lockid, mode);
+		FlexLockDebug("LWLockConditionalAcquire", lockid, "failed");
+		TRACE_POSTGRESQL_FLEXLOCK_CONDACQUIRE_FAIL(lockid, mode);
 	}
 	else
 	{
 		/* Add lock to list of locks held by this backend */
-		held_lwlocks[num_held_lwlocks++] = lockid;
-		TRACE_POSTGRESQL_LWLOCK_CONDACQUIRE(lockid, mode);
+		FlexLockRemember(lockid);
+		TRACE_POSTGRESQL_FLEXLOCK_CONDACQUIRE(lockid, mode);
 	}
 
 	return !mustwait;
@@ -568,32 +299,18 @@ LWLockConditionalAcquire(LWLockId lockid, LWLockMode mode)
  * LWLockRelease - release a previously acquired lock
  */
 void
-LWLockRelease(LWLockId lockid)
+LWLockRelease(FlexLockId lockid)
 {
-	volatile LWLock *lock = &(LWLockArray[lockid].lock);
+	volatile LWLock *lock = LWLockPointer(lockid);
 	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];
+	FlexLockForget(lockid);
 
 	/* Acquire mutex.  Time spent holding mutex should be short! */
-	SpinLockAcquire(&lock->mutex);
+	SpinLockAcquire(&lock->flex.mutex);
 
 	/* Release my hold on lock */
 	if (lock->exclusive > 0)
@@ -610,10 +327,10 @@ LWLockRelease(LWLockId lockid)
 	 * if someone has already awakened waiters that haven't yet acquired the
 	 * lock.
 	 */
-	head = lock->head;
+	head = lock->flex.head;
 	if (head != NULL)
 	{
-		if (lock->exclusive == 0 && lock->shared == 0 && lock->releaseOK)
+		if (lock->exclusive == 0 && lock->shared == 0 && lock->flex.releaseOK)
 		{
 			/*
 			 * Remove the to-be-awakened PGPROCs from the queue.  If the front
@@ -621,17 +338,17 @@ LWLockRelease(LWLockId lockid)
 			 * as many waiters as want shared access.
 			 */
 			proc = head;
-			if (!proc->lwExclusive)
+			if (proc->flWaitMode != LW_EXCLUSIVE)
 			{
-				while (proc->lwWaitLink != NULL &&
-					   !proc->lwWaitLink->lwExclusive)
-					proc = proc->lwWaitLink;
+				while (proc->flWaitLink != NULL &&
+					   proc->flWaitLink->flWaitMode != LW_EXCLUSIVE)
+					proc = proc->flWaitLink;
 			}
 			/* proc is now the last PGPROC to be released */
-			lock->head = proc->lwWaitLink;
-			proc->lwWaitLink = NULL;
+			lock->flex.head = proc->flWaitLink;
+			proc->flWaitLink = NULL;
 			/* prevent additional wakeups until retryer gets to run */
-			lock->releaseOK = false;
+			lock->flex.releaseOK = false;
 		}
 		else
 		{
@@ -641,20 +358,20 @@ LWLockRelease(LWLockId lockid)
 	}
 
 	/* We are done updating shared state of the lock itself. */
-	SpinLockRelease(&lock->mutex);
+	SpinLockRelease(&lock->flex.mutex);
 
-	TRACE_POSTGRESQL_LWLOCK_RELEASE(lockid);
+	TRACE_POSTGRESQL_FLEXLOCK_RELEASE(lockid);
 
 	/*
 	 * Awaken any waiters I removed from the queue.
 	 */
 	while (head != NULL)
 	{
-		LOG_LWDEBUG("LWLockRelease", lockid, "release waiter");
+		FlexLockDebug("LWLockRelease", lockid, "release waiter");
 		proc = head;
-		head = proc->lwWaitLink;
-		proc->lwWaitLink = NULL;
-		proc->lwWaiting = false;
+		head = proc->flWaitLink;
+		proc->flWaitLink = NULL;
+		proc->flWaitResult = 1;		/* any non-zero value will do */
 		PGSemaphoreUnlock(&proc->sem);
 	}
 
@@ -664,43 +381,17 @@ LWLockRelease(LWLockId lockid)
 	RESUME_INTERRUPTS();
 }
 
-
-/*
- * LWLockReleaseAll - release all currently-held locks
- *
- * Used to clean up after ereport(ERROR). An important difference between this
- * function and retail LWLockRelease calls is that InterruptHoldoffCount is
- * unchanged by this operation.  This is necessary since InterruptHoldoffCount
- * has been set to an appropriate level earlier in error recovery. We could
- * decrement it below zero if we allow it to drop for each released lock!
- */
-void
-LWLockReleaseAll(void)
-{
-	while (num_held_lwlocks > 0)
-	{
-		HOLD_INTERRUPTS();		/* match the upcoming RESUME_INTERRUPTS */
-
-		LWLockRelease(held_lwlocks[num_held_lwlocks - 1]);
-	}
-}
-
-
 /*
  * LWLockHeldByMe - test whether my process currently holds a lock
  *
- * This is meant as debug support only.  We do not distinguish whether the
- * lock is held shared or exclusive.
+ * The following convenience routine might not be worthwhile but for the fact
+ * that we've had a function by this name since long before FlexLocks existed.
+ * Callers who want to check whether an arbitrary FlexLock (that may or may not
+ * be an LWLock) is held can use FlexLockHeldByMe directly.
  */
 bool
-LWLockHeldByMe(LWLockId lockid)
+LWLockHeldByMe(FlexLockId lockid)
 {
-	int			i;
-
-	for (i = 0; i < num_held_lwlocks; i++)
-	{
-		if (held_lwlocks[i] == lockid)
-			return true;
-	}
-	return false;
+	AssertMacro(FlexLockArray[lockid].flex.locktype == FLEXLOCK_TYPE_LWLOCK);
+	return FlexLockHeldByMe(lockid);
 }
diff --git a/src/backend/storage/lmgr/predicate.c b/src/backend/storage/lmgr/predicate.c
index 345f6f5..15978a4 100644
--- a/src/backend/storage/lmgr/predicate.c
+++ b/src/backend/storage/lmgr/predicate.c
@@ -239,7 +239,7 @@
 #define PredicateLockHashPartition(hashcode) \
 	((hashcode) % NUM_PREDICATELOCK_PARTITIONS)
 #define PredicateLockHashPartitionLock(hashcode) \
-	((LWLockId) (FirstPredicateLockMgrLock + PredicateLockHashPartition(hashcode)))
+	((FlexLockId) (FirstPredicateLockMgrLock + PredicateLockHashPartition(hashcode)))
 
 #define NPREDICATELOCKTARGETENTS() \
 	mul_size(max_predicate_locks_per_xact, add_size(MaxBackends, max_prepared_xacts))
@@ -1840,7 +1840,7 @@ PageIsPredicateLocked(Relation relation, BlockNumber blkno)
 {
 	PREDICATELOCKTARGETTAG targettag;
 	uint32		targettaghash;
-	LWLockId	partitionLock;
+	FlexLockId	partitionLock;
 	PREDICATELOCKTARGET *target;
 
 	SET_PREDICATELOCKTARGETTAG_PAGE(targettag,
@@ -2073,7 +2073,7 @@ DeleteChildTargetLocks(const PREDICATELOCKTARGETTAG *newtargettag)
 		if (TargetTagIsCoveredBy(oldtargettag, *newtargettag))
 		{
 			uint32		oldtargettaghash;
-			LWLockId	partitionLock;
+			FlexLockId	partitionLock;
 			PREDICATELOCK *rmpredlock;
 
 			oldtargettaghash = PredicateLockTargetTagHashCode(&oldtargettag);
@@ -2285,7 +2285,7 @@ CreatePredicateLock(const PREDICATELOCKTARGETTAG *targettag,
 	PREDICATELOCKTARGET *target;
 	PREDICATELOCKTAG locktag;
 	PREDICATELOCK *lock;
-	LWLockId	partitionLock;
+	FlexLockId	partitionLock;
 	bool		found;
 
 	partitionLock = PredicateLockHashPartitionLock(targettaghash);
@@ -2586,10 +2586,10 @@ TransferPredicateLocksToNewTarget(PREDICATELOCKTARGETTAG oldtargettag,
 								  bool removeOld)
 {
 	uint32		oldtargettaghash;
-	LWLockId	oldpartitionLock;
+	FlexLockId	oldpartitionLock;
 	PREDICATELOCKTARGET *oldtarget;
 	uint32		newtargettaghash;
-	LWLockId	newpartitionLock;
+	FlexLockId	newpartitionLock;
 	bool		found;
 	bool		outOfShmem = false;
 
@@ -3578,7 +3578,7 @@ ClearOldPredicateLocks(void)
 			PREDICATELOCKTARGET *target;
 			PREDICATELOCKTARGETTAG targettag;
 			uint32		targettaghash;
-			LWLockId	partitionLock;
+			FlexLockId	partitionLock;
 
 			tag = predlock->tag;
 			target = tag.myTarget;
@@ -3656,7 +3656,7 @@ ReleaseOneSerializableXact(SERIALIZABLEXACT *sxact, bool partial,
 		PREDICATELOCKTARGET *target;
 		PREDICATELOCKTARGETTAG targettag;
 		uint32		targettaghash;
-		LWLockId	partitionLock;
+		FlexLockId	partitionLock;
 
 		nextpredlock = (PREDICATELOCK *)
 			SHMQueueNext(&(sxact->predicateLocks),
@@ -4034,7 +4034,7 @@ static void
 CheckTargetForConflictsIn(PREDICATELOCKTARGETTAG *targettag)
 {
 	uint32		targettaghash;
-	LWLockId	partitionLock;
+	FlexLockId	partitionLock;
 	PREDICATELOCKTARGET *target;
 	PREDICATELOCK *predlock;
 	PREDICATELOCK *mypredlock = NULL;
diff --git a/src/backend/storage/lmgr/proc.c b/src/backend/storage/lmgr/proc.c
index bcbc802..b402999 100644
--- a/src/backend/storage/lmgr/proc.c
+++ b/src/backend/storage/lmgr/proc.c
@@ -360,9 +360,9 @@ InitProcess(void)
 	/* NB -- autovac launcher intentionally does not set IS_AUTOVACUUM */
 	if (IsAutoVacuumWorkerProcess())
 		MyPgXact->vacuumFlags |= PROC_IS_AUTOVACUUM;
-	MyProc->lwWaiting = false;
-	MyProc->lwExclusive = false;
-	MyProc->lwWaitLink = NULL;
+	MyProc->flWaitResult = 0;
+	MyProc->flWaitMode = 0;
+	MyProc->flWaitLink = NULL;
 	MyProc->waitLock = NULL;
 	MyProc->waitProcLock = NULL;
 #ifdef USE_ASSERT_CHECKING
@@ -515,9 +515,9 @@ InitAuxiliaryProcess(void)
 	MyProc->roleId = InvalidOid;
 	MyPgXact->inCommit = false;
 	MyPgXact->vacuumFlags = 0;
-	MyProc->lwWaiting = false;
-	MyProc->lwExclusive = false;
-	MyProc->lwWaitLink = NULL;
+	MyProc->flWaitMode = 0;
+	MyProc->flWaitResult = 0;
+	MyProc->flWaitLink = NULL;
 	MyProc->waitLock = NULL;
 	MyProc->waitProcLock = NULL;
 #ifdef USE_ASSERT_CHECKING
@@ -643,7 +643,7 @@ IsWaitingForLock(void)
 void
 LockWaitCancel(void)
 {
-	LWLockId	partitionLock;
+	FlexLockId	partitionLock;
 
 	/* Nothing to do if we weren't waiting for a lock */
 	if (lockAwaited == NULL)
@@ -754,11 +754,11 @@ ProcKill(int code, Datum arg)
 #endif
 
 	/*
-	 * Release any LW locks I am holding.  There really shouldn't be any, but
-	 * it's cheap to check again before we cut the knees off the LWLock
+	 * Release any flex locks I am holding.  There really shouldn't be any, but
+	 * it's cheap to check again before we cut the knees off the flex lock
 	 * facility by releasing our PGPROC ...
 	 */
-	LWLockReleaseAll();
+	FlexLockReleaseAll();
 
 	/* Release ownership of the process's latch, too */
 	DisownLatch(&MyProc->procLatch);
@@ -815,8 +815,8 @@ AuxiliaryProcKill(int code, Datum arg)
 
 	Assert(MyProc == auxproc);
 
-	/* Release any LW locks I am holding (see notes above) */
-	LWLockReleaseAll();
+	/* Release any flex locks I am holding (see notes above) */
+	FlexLockReleaseAll();
 
 	/* Release ownership of the process's latch, too */
 	DisownLatch(&MyProc->procLatch);
@@ -901,7 +901,7 @@ ProcSleep(LOCALLOCK *locallock, LockMethod lockMethodTable)
 	LOCK	   *lock = locallock->lock;
 	PROCLOCK   *proclock = locallock->proclock;
 	uint32		hashcode = locallock->hashcode;
-	LWLockId	partitionLock = LockHashPartitionLock(hashcode);
+	FlexLockId	partitionLock = LockHashPartitionLock(hashcode);
 	PROC_QUEUE *waitQueue = &(lock->waitProcs);
 	LOCKMASK	myHeldLocks = MyProc->heldLocks;
 	bool		early_deadlock = false;
diff --git a/src/backend/utils/misc/check_guc b/src/backend/utils/misc/check_guc
index 293fb03..1a19e36 100755
--- a/src/backend/utils/misc/check_guc
+++ b/src/backend/utils/misc/check_guc
@@ -19,7 +19,7 @@
 INTENTIONALLY_NOT_INCLUDED="autocommit debug_deadlocks \
 is_superuser lc_collate lc_ctype lc_messages lc_monetary lc_numeric lc_time \
 pre_auth_delay role seed server_encoding server_version server_version_int \
-session_authorization trace_lock_oidmin trace_lock_table trace_locks trace_lwlocks \
+session_authorization trace_lock_oidmin trace_lock_table trace_locks trace_flexlocks \
 trace_notify trace_userlocks transaction_isolation transaction_read_only \
 zero_damaged_pages"
 
diff --git a/src/backend/utils/misc/guc.c b/src/backend/utils/misc/guc.c
index da7b6d4..52de233 100644
--- a/src/backend/utils/misc/guc.c
+++ b/src/backend/utils/misc/guc.c
@@ -59,6 +59,7 @@
 #include "replication/walreceiver.h"
 #include "replication/walsender.h"
 #include "storage/bufmgr.h"
+#include "storage/flexlock_internals.h"
 #include "storage/standby.h"
 #include "storage/fd.h"
 #include "storage/predicate.h"
@@ -1071,12 +1072,12 @@ static struct config_bool ConfigureNamesBool[] =
 		NULL, NULL, NULL
 	},
 	{
-		{"trace_lwlocks", PGC_SUSET, DEVELOPER_OPTIONS,
+		{"trace_flexlocks", PGC_SUSET, DEVELOPER_OPTIONS,
 			gettext_noop("No description available."),
 			NULL,
 			GUC_NOT_IN_SAMPLE
 		},
-		&Trace_lwlocks,
+		&Trace_flexlocks,
 		false,
 		NULL, NULL, NULL
 	},
diff --git a/src/backend/utils/probes.d b/src/backend/utils/probes.d
index 71c5ab0..5b9cfe6 100644
--- a/src/backend/utils/probes.d
+++ b/src/backend/utils/probes.d
@@ -15,8 +15,8 @@
  * in probe definitions, as they cause compilation errors on Mac OS X 10.5.
  */
 #define LocalTransactionId unsigned int
-#define LWLockId int
-#define LWLockMode int
+#define FlexLockId int
+#define FlexLockMode int
 #define LOCKMODE int
 #define BlockNumber unsigned int
 #define Oid unsigned int
@@ -29,12 +29,12 @@ provider postgresql {
 	probe transaction__commit(LocalTransactionId);
 	probe transaction__abort(LocalTransactionId);
 
-	probe lwlock__acquire(LWLockId, LWLockMode);
-	probe lwlock__release(LWLockId);
-	probe lwlock__wait__start(LWLockId, LWLockMode);
-	probe lwlock__wait__done(LWLockId, LWLockMode);
-	probe lwlock__condacquire(LWLockId, LWLockMode);
-	probe lwlock__condacquire__fail(LWLockId, LWLockMode);
+	probe flexlock__acquire(FlexLockId, FlexLockMode);
+	probe flexlock__release(FlexLockId);
+	probe flexlock__wait__start(FlexLockId, FlexLockMode);
+	probe flexlock__wait__done(FlexLockId, FlexLockMode);
+	probe flexlock__condacquire(FlexLockId, FlexLockMode);
+	probe flexlock__condacquire__fail(FlexLockId, FlexLockMode);
 
 	probe lock__wait__start(unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, LOCKMODE);
 	probe lock__wait__done(unsigned int, unsigned int, unsigned int, unsigned int, unsigned int, LOCKMODE);
diff --git a/src/include/access/slru.h b/src/include/access/slru.h
index e48743f..680a87f 100644
--- a/src/include/access/slru.h
+++ b/src/include/access/slru.h
@@ -55,7 +55,7 @@ typedef enum
  */
 typedef struct SlruSharedData
 {
-	LWLockId	ControlLock;
+	FlexLockId	ControlLock;
 
 	/* Number of buffers managed by this SLRU structure */
 	int			num_slots;
@@ -69,7 +69,7 @@ typedef struct SlruSharedData
 	bool	   *page_dirty;
 	int		   *page_number;
 	int		   *page_lru_count;
-	LWLockId   *buffer_locks;
+	FlexLockId *buffer_locks;
 
 	/*
 	 * Optional array of WAL flush LSNs associated with entries in the SLRU
@@ -136,7 +136,7 @@ typedef SlruCtlData *SlruCtl;
 
 extern Size SimpleLruShmemSize(int nslots, int nlsns);
 extern void SimpleLruInit(SlruCtl ctl, const char *name, int nslots, int nlsns,
-			  LWLockId ctllock, const char *subdir);
+			  FlexLockId ctllock, const char *subdir);
 extern int	SimpleLruZeroPage(SlruCtl ctl, int pageno);
 extern int SimpleLruReadPage(SlruCtl ctl, int pageno, bool write_ok,
 				  TransactionId xid);
diff --git a/src/include/pg_config_manual.h b/src/include/pg_config_manual.h
index 6c8e312..d3b74db 100644
--- a/src/include/pg_config_manual.h
+++ b/src/include/pg_config_manual.h
@@ -49,9 +49,9 @@
 #define SEQ_MINVALUE	(-SEQ_MAXVALUE)
 
 /*
- * Number of spare LWLocks to allocate for user-defined add-on code.
+ * Number of spare FlexLocks to allocate for user-defined add-on code.
  */
-#define NUM_USER_DEFINED_LWLOCKS	4
+#define NUM_USER_DEFINED_FLEXLOCKS	4
 
 /*
  * Define this if you want to allow the lo_import and lo_export SQL
diff --git a/src/include/storage/buf_internals.h b/src/include/storage/buf_internals.h
index b7d4ea5..ac7f665 100644
--- a/src/include/storage/buf_internals.h
+++ b/src/include/storage/buf_internals.h
@@ -103,7 +103,7 @@ typedef struct buftag
 #define BufTableHashPartition(hashcode) \
 	((hashcode) % NUM_BUFFER_PARTITIONS)
 #define BufMappingPartitionLock(hashcode) \
-	((LWLockId) (FirstBufMappingLock + BufTableHashPartition(hashcode)))
+	((FlexLockId) (FirstBufMappingLock + BufTableHashPartition(hashcode)))
 
 /*
  *	BufferDesc -- shared descriptor/state data for a single shared buffer.
@@ -143,8 +143,8 @@ typedef struct sbufdesc
 	int			buf_id;			/* buffer's index number (from 0) */
 	int			freeNext;		/* link in freelist chain */
 
-	LWLockId	io_in_progress_lock;	/* to wait for I/O to complete */
-	LWLockId	content_lock;	/* to lock access to buffer contents */
+	FlexLockId	io_in_progress_lock;	/* to wait for I/O to complete */
+	FlexLockId	content_lock;	/* to lock access to buffer contents */
 } BufferDesc;
 
 #define BufferDescriptorGetBuffer(bdesc) ((bdesc)->buf_id + 1)
diff --git a/src/include/storage/flexlock.h b/src/include/storage/flexlock.h
new file mode 100644
index 0000000..612c21a
--- /dev/null
+++ b/src/include/storage/flexlock.h
@@ -0,0 +1,102 @@
+/*-------------------------------------------------------------------------
+ *
+ * flexlock.h
+ *	  Flex lock manager
+ *
+ * Portions Copyright (c) 1996-2011, PostgreSQL Global Development Group
+ * Portions Copyright (c) 1994, Regents of the University of California
+ *
+ * src/include/storage/flexlock.h
+ *
+ *-------------------------------------------------------------------------
+ */
+#ifndef FLEXLOCK_H
+#define FLEXLOCK_H
+
+/*
+ * It's a bit odd to declare NUM_BUFFER_PARTITIONS and NUM_LOCK_PARTITIONS
+ * here, but we need them to set up enum FlexLockId correctly, and having
+ * this file include lock.h or bufmgr.h would be backwards.
+ */
+
+/* Number of partitions of the shared buffer mapping hashtable */
+#define NUM_BUFFER_PARTITIONS  16
+
+/* Number of partitions the shared lock tables are divided into */
+#define LOG2_NUM_LOCK_PARTITIONS  4
+#define NUM_LOCK_PARTITIONS  (1 << LOG2_NUM_LOCK_PARTITIONS)
+
+/* Number of partitions the shared predicate lock tables are divided into */
+#define LOG2_NUM_PREDICATELOCK_PARTITIONS  4
+#define NUM_PREDICATELOCK_PARTITIONS  (1 << LOG2_NUM_PREDICATELOCK_PARTITIONS)
+
+/*
+ * We have a number of predefined FlexLocks, plus a bunch of locks that are
+ * dynamically assigned (e.g., for shared buffers).  The FlexLock structures
+ * live in shared memory (since they contain shared data) and are identified
+ * by values of this enumerated type.  We abuse the notion of an enum somewhat
+ * by allowing values not listed in the enum declaration to be assigned.
+ * The extra value MaxDynamicFlexLock is there to keep the compiler from
+ * deciding that the enum can be represented as char or short ...
+ *
+ * If you remove a lock, please replace it with a placeholder. This retains
+ * the lock numbering, which is helpful for DTrace and other external
+ * debugging scripts.
+ */
+typedef enum FlexLockId
+{
+	BufFreelistLock,
+	ShmemIndexLock,
+	OidGenLock,
+	XidGenLock,
+	ProcArrayLock,
+	SInvalReadLock,
+	SInvalWriteLock,
+	WALInsertLock,
+	WALWriteLock,
+	ControlFileLock,
+	CheckpointLock,
+	CLogControlLock,
+	SubtransControlLock,
+	MultiXactGenLock,
+	MultiXactOffsetControlLock,
+	MultiXactMemberControlLock,
+	RelCacheInitLock,
+	BgWriterCommLock,
+	TwoPhaseStateLock,
+	TablespaceCreateLock,
+	BtreeVacuumLock,
+	AddinShmemInitLock,
+	AutovacuumLock,
+	AutovacuumScheduleLock,
+	SyncScanLock,
+	RelationMappingLock,
+	AsyncCtlLock,
+	AsyncQueueLock,
+	SerializableXactHashLock,
+	SerializableFinishedListLock,
+	SerializablePredicateLockListLock,
+	OldSerXidLock,
+	SyncRepLock,
+	/* Individual lock IDs end here */
+	FirstBufMappingLock,
+	FirstLockMgrLock = FirstBufMappingLock + NUM_BUFFER_PARTITIONS,
+	FirstPredicateLockMgrLock = FirstLockMgrLock + NUM_LOCK_PARTITIONS,
+
+	/* must be last except for MaxDynamicFlexLock: */
+	NumFixedFlexLocks = FirstPredicateLockMgrLock + NUM_PREDICATELOCK_PARTITIONS,
+
+	MaxDynamicFlexLock = 1000000000
+} FlexLockId;
+
+/* Shared memory setup. */
+extern int	NumFlexLocks(void);
+extern Size FlexLockShmemSize(void);
+extern void RequestAddinFlexLocks(int n);
+extern void CreateFlexLocks(void);
+
+/* Error recovery and debugging support functions. */
+extern void FlexLockReleaseAll(void);
+extern bool FlexLockHeldByMe(FlexLockId id);
+
+#endif   /* FLEXLOCK_H */
diff --git a/src/include/storage/flexlock_internals.h b/src/include/storage/flexlock_internals.h
new file mode 100644
index 0000000..a5a6bde
--- /dev/null
+++ b/src/include/storage/flexlock_internals.h
@@ -0,0 +1,146 @@
+/*-------------------------------------------------------------------------
+ *
+ * flexlock_internals.h
+ *	  Flex lock internals.  Only files which implement a FlexLock
+ *    type should need to include this.  Merging this with flexlock.h
+ *    creates a circular header dependency, but even if it didn't, this
+ *    is cleaner.
+ *
+ * Portions Copyright (c) 1996-2011, PostgreSQL Global Development Group
+ * Portions Copyright (c) 1994, Regents of the University of California
+ *
+ * src/include/storage/flexlock_internals.h
+ *
+ *-------------------------------------------------------------------------
+ */
+#ifndef FLEXLOCK_INTERNALS_H
+#define FLEXLOCK_INTERNALS_H
+
+#include "storage/proc.h"
+#include "storage/s_lock.h"
+
+/*
+ * Individual FlexLock implementations each get this many bytes to store
+ * its state; of course, a given implementation could also allocate additional
+ * shmem elsewhere, but we provide this many bytes within the array.  The
+ * header fields common to all FlexLock types are included in this number.
+ * A power of two should probably be chosen, to avoid alignment issues and
+ * cache line splitting.  It might be useful to increase this on systems where
+ * a cache line is more than 64 bytes in size.
+ */
+#define FLEX_LOCK_BYTES		64
+
+typedef struct FlexLock
+{
+	char		locktype;		/* see FLEXLOCK_TYPE_* constants */
+	slock_t		mutex;			/* Protects FlexLock state and wait queues */
+	bool		releaseOK;		/* T if ok to release waiters */
+	PGPROC	   *head;			/* head of list of waiting PGPROCs */
+	PGPROC	   *tail;			/* tail of list of waiting PGPROCs */
+	/* tail is undefined when head is NULL */
+} FlexLock;
+
+#define FLEXLOCK_TYPE_LWLOCK			'l'
+
+typedef union FlexLockPadded
+{
+	FlexLock	flex;
+	char		pad[FLEX_LOCK_BYTES];
+} FlexLockPadded;
+
+extern FlexLockPadded *FlexLockArray;
+
+extern FlexLockId FlexLockAssign(char locktype);
+
+/*
+ * We use this structure to keep track of flex locks held, for release
+ * during error recovery.  The maximum size could be determined at runtime
+ * if necessary, but it seems unlikely that more than a few locks could
+ * ever be held simultaneously.
+ */
+#define MAX_SIMUL_FLEXLOCKS 100
+extern int num_held_flexlocks;
+extern FlexLockId held_flexlocks[MAX_SIMUL_FLEXLOCKS];
+
+/* We define the following operations as macros, for speed. */
+
+/* Remember that we've acquired a FlexLock. */
+#define FlexLockRemember(lock) \
+	do { \
+		if (num_held_flexlocks >= MAX_SIMUL_FLEXLOCKS) \
+			elog(PANIC, "too many FlexLocks taken"); \
+		held_flexlocks[num_held_flexlocks++] = lock; \
+	} while (0)
+
+/*
+ * Remove lock from list of locks held.  Usually, but not always, it will
+ * be the latest-acquired lock; so search array backwards.
+ */
+#define FlexLockForget(lock) \
+	do { \
+		int	i; \
+		for (i = num_held_flexlocks; --i >= 0;) \
+			if (lock == held_flexlocks[i]) \
+				break; \
+		if (i < 0) \
+			elog(ERROR, "lock %d is not held", (int) lock); \
+		num_held_flexlocks--; \
+		for (; i < num_held_flexlocks; i++) \
+			held_flexlocks[i] = held_flexlocks[i + 1]; \
+	} while (0)
+
+/*
+ * FlexLockWait - wait until awakened
+ *
+ * Since we share the process wait semaphore with the regular lock manager
+ * and ProcWaitForSignal, and we may need to acquire a FlexLock while one of
+ * those is pending, it is possible that we get awakened for a reason other
+ * than being signaled by a FlexLock release.  If so, loop back and wait again.
+ *
+ * Returns the number of "extra" waits absorbed so that, once we've gotten the
+ * FlexLock, we can 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.
+ */
+#define FlexLockWait(lock, mode, extraWaits) \
+	do { \
+		FlexLockDebug("FlexLockWait", lock, "waiting"); \
+		TRACE_POSTGRESQL_FLEXLOCK_WAIT_START(lock, mode); \
+		for (;;) \
+	   	{ \
+			/* "false" means cannot accept cancel/die interrupt here. */ \
+			PGSemaphoreLock(&MyProc->sem, false); \
+			/* any non-zero value means "wake up */ \
+			if (MyProc->flWaitResult) \
+				break; \
+			extraWaits++; \
+   		} \
+		TRACE_POSTGRESQL_FLEXLOCK_WAIT_DONE(lock, mode); \
+		FlexLockDebug("FlexLockWait", lock, "awakened"); \
+	} while (0)
+
+#define FlexLockJoinWaitQueue(lock, mode) \
+	do { \
+		Assert(MyProc != NULL); \
+		MyProc->flWaitResult = 0; \
+		MyProc->flWaitMode = mode; \
+		MyProc->flWaitLink = NULL; \
+		if (lock->flex.head == NULL) \
+			lock->flex.head = MyProc; \
+		else \
+			lock->flex.tail->flWaitLink = MyProc; \
+		lock->flex.tail = MyProc; \
+	} while (0)
+
+#ifdef LOCK_DEBUG
+extern bool	Trace_flexlocks;
+#define FlexLockDebug(where, lock, msg) \
+	do { \
+		if (Trace_flexlocks) \
+			elog(LOG, "%s(%d): %s", where, (int) lock, msg); \
+	} while (0)
+#else
+#define FlexLockDebug(where, lock, msg)
+#endif
+
+#endif   /* FLEXLOCK_H */
diff --git a/src/include/storage/lock.h b/src/include/storage/lock.h
index e106ad5..ba87db2 100644
--- a/src/include/storage/lock.h
+++ b/src/include/storage/lock.h
@@ -471,7 +471,7 @@ typedef enum
 #define LockHashPartition(hashcode) \
 	((hashcode) % NUM_LOCK_PARTITIONS)
 #define LockHashPartitionLock(hashcode) \
-	((LWLockId) (FirstLockMgrLock + LockHashPartition(hashcode)))
+	((FlexLockId) (FirstLockMgrLock + LockHashPartition(hashcode)))
 
 
 /*
diff --git a/src/include/storage/lwlock.h b/src/include/storage/lwlock.h
index 438a48d..f68cddc 100644
--- a/src/include/storage/lwlock.h
+++ b/src/include/storage/lwlock.h
@@ -14,82 +14,7 @@
 #ifndef LWLOCK_H
 #define LWLOCK_H
 
-/*
- * It's a bit odd to declare NUM_BUFFER_PARTITIONS and NUM_LOCK_PARTITIONS
- * here, but we need them to set up enum LWLockId correctly, and having
- * this file include lock.h or bufmgr.h would be backwards.
- */
-
-/* Number of partitions of the shared buffer mapping hashtable */
-#define NUM_BUFFER_PARTITIONS  16
-
-/* Number of partitions the shared lock tables are divided into */
-#define LOG2_NUM_LOCK_PARTITIONS  4
-#define NUM_LOCK_PARTITIONS  (1 << LOG2_NUM_LOCK_PARTITIONS)
-
-/* Number of partitions the shared predicate lock tables are divided into */
-#define LOG2_NUM_PREDICATELOCK_PARTITIONS  4
-#define NUM_PREDICATELOCK_PARTITIONS  (1 << LOG2_NUM_PREDICATELOCK_PARTITIONS)
-
-/*
- * We have a number of predefined LWLocks, plus a bunch of LWLocks that are
- * dynamically assigned (e.g., for shared buffers).  The LWLock structures
- * live in shared memory (since they contain shared data) and are identified
- * by values of this enumerated type.  We abuse the notion of an enum somewhat
- * by allowing values not listed in the enum declaration to be assigned.
- * The extra value MaxDynamicLWLock is there to keep the compiler from
- * deciding that the enum can be represented as char or short ...
- *
- * If you remove a lock, please replace it with a placeholder. This retains
- * the lock numbering, which is helpful for DTrace and other external
- * debugging scripts.
- */
-typedef enum LWLockId
-{
-	BufFreelistLock,
-	ShmemIndexLock,
-	OidGenLock,
-	XidGenLock,
-	ProcArrayLock,
-	SInvalReadLock,
-	SInvalWriteLock,
-	WALInsertLock,
-	WALWriteLock,
-	ControlFileLock,
-	CheckpointLock,
-	CLogControlLock,
-	SubtransControlLock,
-	MultiXactGenLock,
-	MultiXactOffsetControlLock,
-	MultiXactMemberControlLock,
-	RelCacheInitLock,
-	BgWriterCommLock,
-	TwoPhaseStateLock,
-	TablespaceCreateLock,
-	BtreeVacuumLock,
-	AddinShmemInitLock,
-	AutovacuumLock,
-	AutovacuumScheduleLock,
-	SyncScanLock,
-	RelationMappingLock,
-	AsyncCtlLock,
-	AsyncQueueLock,
-	SerializableXactHashLock,
-	SerializableFinishedListLock,
-	SerializablePredicateLockListLock,
-	OldSerXidLock,
-	SyncRepLock,
-	/* Individual lock IDs end here */
-	FirstBufMappingLock,
-	FirstLockMgrLock = FirstBufMappingLock + NUM_BUFFER_PARTITIONS,
-	FirstPredicateLockMgrLock = FirstLockMgrLock + NUM_LOCK_PARTITIONS,
-
-	/* must be last except for MaxDynamicLWLock: */
-	NumFixedLWLocks = FirstPredicateLockMgrLock + NUM_PREDICATELOCK_PARTITIONS,
-
-	MaxDynamicLWLock = 1000000000
-} LWLockId;
-
+#include "storage/flexlock.h"
 
 typedef enum LWLockMode
 {
@@ -97,22 +22,10 @@ typedef enum LWLockMode
 	LW_SHARED
 } LWLockMode;
 
-
-#ifdef LOCK_DEBUG
-extern bool Trace_lwlocks;
-#endif
-
-extern LWLockId LWLockAssign(void);
-extern void LWLockAcquire(LWLockId lockid, LWLockMode mode);
-extern bool LWLockConditionalAcquire(LWLockId lockid, LWLockMode mode);
-extern void LWLockRelease(LWLockId lockid);
-extern void LWLockReleaseAll(void);
-extern bool LWLockHeldByMe(LWLockId lockid);
-
-extern int	NumLWLocks(void);
-extern Size LWLockShmemSize(void);
-extern void CreateLWLocks(void);
-
-extern void RequestAddinLWLocks(int n);
+extern FlexLockId LWLockAssign(void);
+extern void LWLockAcquire(FlexLockId lockid, LWLockMode mode);
+extern bool LWLockConditionalAcquire(FlexLockId lockid, LWLockMode mode);
+extern void LWLockRelease(FlexLockId lockid);
+extern bool LWLockHeldByMe(FlexLockId lockid);
 
 #endif   /* LWLOCK_H */
diff --git a/src/include/storage/proc.h b/src/include/storage/proc.h
index c7cddc7..1f3a71d 100644
--- a/src/include/storage/proc.h
+++ b/src/include/storage/proc.h
@@ -99,10 +99,10 @@ struct PGPROC
 	 */
 	bool		recoveryConflictPending;
 
-	/* Info about LWLock the process is currently waiting for, if any. */
-	bool		lwWaiting;		/* true if waiting for an LW lock */
-	bool		lwExclusive;	/* true if waiting for exclusive access */
-	struct PGPROC *lwWaitLink;	/* next waiter for same LW lock */
+	/* Info about FlexLock the process is currently waiting for, if any. */
+	int			flWaitResult;	/* result of wait, or 0 if still waiting */
+	int			flWaitMode;		/* lock mode sought */
+	struct PGPROC *flWaitLink;	/* next waiter for same FlexLock */
 
 	/* Info about lock the process is currently waiting for, if any. */
 	/* waitLock and waitProcLock are NULL if not currently waiting. */
@@ -132,7 +132,7 @@ struct PGPROC
 	struct XidCache subxids;	/* cache for subtransaction XIDs */
 
 	/* Per-backend LWLock.  Protects fields below. */
-	LWLockId	backendLock;	/* protects the fields below */
+	FlexLockId	backendLock;	/* protects the fields below */
 
 	/* Lock manager data, recording fast-path locks taken by this backend. */
 	uint64		fpLockBits;		/* lock modes held for each fast-path slot */

diff --git a/src/backend/commands/analyze.c b/src/backend/commands/analyze.c
index c3d3958..4d27c53 100644
--- a/src/backend/commands/analyze.c
+++ b/src/backend/commands/analyze.c
@@ -40,6 +40,7 @@
 #include "storage/lmgr.h"
 #include "storage/proc.h"
 #include "storage/procarray.h"
+#include "storage/procarraylock.h"
 #include "utils/acl.h"
 #include "utils/attoptcache.h"
 #include "utils/datum.h"
@@ -222,9 +223,9 @@ analyze_rel(Oid relid, VacuumStmt *vacstmt, BufferAccessStrategy bstrategy)
 	/*
 	 * OK, let's do it.  First let other backends know I'm in ANALYZE.
 	 */
-	LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
+	ProcArrayLockAcquire(PAL_EXCLUSIVE);
 	MyPgXact->vacuumFlags |= PROC_IN_ANALYZE;
-	LWLockRelease(ProcArrayLock);
+	ProcArrayLockRelease();
 
 	/*
 	 * Do the normal non-recursive ANALYZE.
@@ -249,9 +250,9 @@ analyze_rel(Oid relid, VacuumStmt *vacstmt, BufferAccessStrategy bstrategy)
 	 * Reset my PGPROC flag.  Note: we need this here, and not in vacuum_rel,
 	 * because the vacuum flag is cleared by the end-of-xact code.
 	 */
-	LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
+	ProcArrayLockAcquire(PAL_EXCLUSIVE);
 	MyPgXact->vacuumFlags &= ~PROC_IN_ANALYZE;
-	LWLockRelease(ProcArrayLock);
+	ProcArrayLockRelease();
 }
 
 /*
diff --git a/src/backend/commands/vacuum.c b/src/backend/commands/vacuum.c
index e70dbed..09aa32b 100644
--- a/src/backend/commands/vacuum.c
+++ b/src/backend/commands/vacuum.c
@@ -39,6 +39,7 @@
 #include "storage/lmgr.h"
 #include "storage/proc.h"
 #include "storage/procarray.h"
+#include "storage/procarraylock.h"
 #include "utils/acl.h"
 #include "utils/fmgroids.h"
 #include "utils/guc.h"
@@ -895,11 +896,11 @@ vacuum_rel(Oid relid, VacuumStmt *vacstmt, bool do_toast, bool for_wraparound)
 		 * MyProc->xid/xmin, else OldestXmin might appear to go backwards,
 		 * which is probably Not Good.
 		 */
-		LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
+		ProcArrayLockAcquire(PAL_EXCLUSIVE);
 		MyPgXact->vacuumFlags |= PROC_IN_VACUUM;
 		if (for_wraparound)
 			MyPgXact->vacuumFlags |= PROC_VACUUM_FOR_WRAPAROUND;
-		LWLockRelease(ProcArrayLock);
+		ProcArrayLockRelease();
 	}
 
 	/*
diff --git a/src/backend/storage/ipc/procarray.c b/src/backend/storage/ipc/procarray.c
index 19ff524..d457e3f 100644
--- a/src/backend/storage/ipc/procarray.c
+++ b/src/backend/storage/ipc/procarray.c
@@ -52,6 +52,7 @@
 #include "access/twophase.h"
 #include "miscadmin.h"
 #include "storage/procarray.h"
+#include "storage/procarraylock.h"
 #include "storage/spin.h"
 #include "utils/builtins.h"
 #include "utils/snapmgr.h"
@@ -261,7 +262,7 @@ ProcArrayAdd(PGPROC *proc)
 	ProcArrayStruct *arrayP = procArray;
 	int			index;
 
-	LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
+	ProcArrayLockAcquire(PAL_EXCLUSIVE);
 
 	if (arrayP->numProcs >= arrayP->maxProcs)
 	{
@@ -270,7 +271,7 @@ ProcArrayAdd(PGPROC *proc)
 		 * fixed supply of PGPROC structs too, and so we should have failed
 		 * earlier.)
 		 */
-		LWLockRelease(ProcArrayLock);
+		ProcArrayLockRelease();
 		ereport(FATAL,
 				(errcode(ERRCODE_TOO_MANY_CONNECTIONS),
 				 errmsg("sorry, too many clients already")));
@@ -300,7 +301,7 @@ ProcArrayAdd(PGPROC *proc)
 	arrayP->pgprocnos[index] = proc->pgprocno;
 	arrayP->numProcs++;
 
-	LWLockRelease(ProcArrayLock);
+	ProcArrayLockRelease();
 }
 
 /*
@@ -325,7 +326,7 @@ ProcArrayRemove(PGPROC *proc, TransactionId latestXid)
 		DisplayXidCache();
 #endif
 
-	LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
+	ProcArrayLockAcquire(PAL_EXCLUSIVE);
 
 	if (TransactionIdIsValid(latestXid))
 	{
@@ -351,13 +352,13 @@ ProcArrayRemove(PGPROC *proc, TransactionId latestXid)
 					(arrayP->numProcs - index - 1) * sizeof (int));
 			arrayP->pgprocnos[arrayP->numProcs - 1] = -1; /* for debugging */
 			arrayP->numProcs--;
-			LWLockRelease(ProcArrayLock);
+			ProcArrayLockRelease();
 			return;
 		}
 	}
 
 	/* Ooops */
-	LWLockRelease(ProcArrayLock);
+	ProcArrayLockRelease();
 
 	elog(LOG, "failed to find proc %p in ProcArray", proc);
 }
@@ -383,54 +384,19 @@ ProcArrayEndTransaction(PGPROC *proc, TransactionId latestXid)
 
 	if (TransactionIdIsValid(latestXid))
 	{
-		/*
-		 * We must lock ProcArrayLock while clearing our advertised XID, so
-		 * that we do not exit the set of "running" transactions while someone
-		 * else is taking a snapshot.  See discussion in
-		 * src/backend/access/transam/README.
-		 */
-		Assert(TransactionIdIsValid(allPgXact[proc->pgprocno].xid));
-
-		LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
-
-		pgxact->xid = InvalidTransactionId;
-		proc->lxid = InvalidLocalTransactionId;
-		pgxact->xmin = InvalidTransactionId;
-		/* must be cleared with xid/xmin: */
-		pgxact->vacuumFlags &= ~PROC_VACUUM_STATE_MASK;
-		pgxact->inCommit = false; /* be sure this is cleared in abort */
-		proc->recoveryConflictPending = false;
-
-		/* Clear the subtransaction-XID cache too while holding the lock */
-		pgxact->nxids = 0;
-		pgxact->overflowed = false;
-
-		/* Also advance global latestCompletedXid while holding the lock */
-		if (TransactionIdPrecedes(ShmemVariableCache->latestCompletedXid,
-								  latestXid))
-			ShmemVariableCache->latestCompletedXid = latestXid;
-
-		LWLockRelease(ProcArrayLock);
+		Assert(proc == MyProc);
+		ProcArrayLockClearTransaction(latestXid);		
 	}
 	else
 	{
-		/*
-		 * If we have no XID, we don't need to lock, since we won't affect
-		 * anyone else's calculation of a snapshot.  We might change their
-		 * estimate of global xmin, but that's OK.
-		 */
-		Assert(!TransactionIdIsValid(allPgXact[proc->pgprocno].xid));
-
-		proc->lxid = InvalidLocalTransactionId;
 		pgxact->xmin = InvalidTransactionId;
 		/* must be cleared with xid/xmin: */
 		pgxact->vacuumFlags &= ~PROC_VACUUM_STATE_MASK;
-		pgxact->inCommit = false; /* be sure this is cleared in abort */
-		proc->recoveryConflictPending = false;
-
-		Assert(pgxact->nxids == 0);
-		Assert(pgxact->overflowed == false);
 	}
+
+	proc->lxid = InvalidLocalTransactionId;
+	pgxact->inCommit = false; /* be sure this is cleared in abort */
+	proc->recoveryConflictPending = false;
 }
 
 
@@ -562,7 +528,7 @@ ProcArrayApplyRecoveryInfo(RunningTransactions running)
 	/*
 	 * Nobody else is running yet, but take locks anyhow
 	 */
-	LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
+	ProcArrayLockAcquire(PAL_EXCLUSIVE);
 
 	/*
 	 * KnownAssignedXids is sorted so we cannot just add the xids, we have to
@@ -669,7 +635,7 @@ ProcArrayApplyRecoveryInfo(RunningTransactions running)
 	Assert(TransactionIdIsNormal(ShmemVariableCache->latestCompletedXid));
 	Assert(TransactionIdIsValid(ShmemVariableCache->nextXid));
 
-	LWLockRelease(ProcArrayLock);
+	ProcArrayLockRelease();
 
 	KnownAssignedXidsDisplay(trace_recovery(DEBUG3));
 	if (standbyState == STANDBY_SNAPSHOT_READY)
@@ -724,7 +690,7 @@ ProcArrayApplyXidAssignment(TransactionId topxid,
 	/*
 	 * Uses same locking as transaction commit
 	 */
-	LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
+	ProcArrayLockAcquire(PAL_EXCLUSIVE);
 
 	/*
 	 * Remove subxids from known-assigned-xacts.
@@ -737,7 +703,7 @@ ProcArrayApplyXidAssignment(TransactionId topxid,
 	if (TransactionIdPrecedes(procArray->lastOverflowedXid, max_xid))
 		procArray->lastOverflowedXid = max_xid;
 
-	LWLockRelease(ProcArrayLock);
+	ProcArrayLockRelease();
 }
 
 /*
@@ -829,7 +795,7 @@ TransactionIdIsInProgress(TransactionId xid)
 					 errmsg("out of memory")));
 	}
 
-	LWLockAcquire(ProcArrayLock, LW_SHARED);
+	ProcArrayLockAcquire(PAL_SHARED);
 
 	/*
 	 * Now that we have the lock, we can check latestCompletedXid; if the
@@ -837,7 +803,7 @@ TransactionIdIsInProgress(TransactionId xid)
 	 */
 	if (TransactionIdPrecedes(ShmemVariableCache->latestCompletedXid, xid))
 	{
-		LWLockRelease(ProcArrayLock);
+		ProcArrayLockRelease();
 		xc_by_latest_xid_inc();
 		return true;
 	}
@@ -865,7 +831,7 @@ TransactionIdIsInProgress(TransactionId xid)
 		 */
 		if (TransactionIdEquals(pxid, xid))
 		{
-			LWLockRelease(ProcArrayLock);
+			ProcArrayLockRelease();
 			xc_by_main_xid_inc();
 			return true;
 		}
@@ -887,7 +853,7 @@ TransactionIdIsInProgress(TransactionId xid)
 
 			if (TransactionIdEquals(cxid, xid))
 			{
-				LWLockRelease(ProcArrayLock);
+				ProcArrayLockRelease();
 				xc_by_child_xid_inc();
 				return true;
 			}
@@ -915,7 +881,7 @@ TransactionIdIsInProgress(TransactionId xid)
 
 		if (KnownAssignedXidExists(xid))
 		{
-			LWLockRelease(ProcArrayLock);
+			ProcArrayLockRelease();
 			xc_by_known_assigned_inc();
 			return true;
 		}
@@ -931,7 +897,7 @@ TransactionIdIsInProgress(TransactionId xid)
 			nxids = KnownAssignedXidsGet(xids, xid);
 	}
 
-	LWLockRelease(ProcArrayLock);
+	ProcArrayLockRelease();
 
 	/*
 	 * If none of the relevant caches overflowed, we know the Xid is not
@@ -997,7 +963,7 @@ TransactionIdIsActive(TransactionId xid)
 	if (TransactionIdPrecedes(xid, RecentXmin))
 		return false;
 
-	LWLockAcquire(ProcArrayLock, LW_SHARED);
+	ProcArrayLockAcquire(PAL_SHARED);
 
 	for (i = 0; i < arrayP->numProcs; i++)
 	{
@@ -1022,7 +988,7 @@ TransactionIdIsActive(TransactionId xid)
 		}
 	}
 
-	LWLockRelease(ProcArrayLock);
+	ProcArrayLockRelease();
 
 	return result;
 }
@@ -1085,7 +1051,7 @@ GetOldestXmin(bool allDbs, bool ignoreVacuum)
 	/* Cannot look for individual databases during recovery */
 	Assert(allDbs || !RecoveryInProgress());
 
-	LWLockAcquire(ProcArrayLock, LW_SHARED);
+	ProcArrayLockAcquire(PAL_SHARED);
 
 	/*
 	 * We initialize the MIN() calculation with latestCompletedXid + 1. This
@@ -1140,7 +1106,7 @@ GetOldestXmin(bool allDbs, bool ignoreVacuum)
 		 */
 		TransactionId kaxmin = KnownAssignedXidsGetOldestXmin();
 
-		LWLockRelease(ProcArrayLock);
+		ProcArrayLockRelease();
 
 		if (TransactionIdIsNormal(kaxmin) &&
 			TransactionIdPrecedes(kaxmin, result))
@@ -1151,7 +1117,7 @@ GetOldestXmin(bool allDbs, bool ignoreVacuum)
 		/*
 		 * No other information needed, so release the lock immediately.
 		 */
-		LWLockRelease(ProcArrayLock);
+		ProcArrayLockRelease();
 
 		/*
 		 * Compute the cutoff XID by subtracting vacuum_defer_cleanup_age,
@@ -1280,7 +1246,7 @@ GetSnapshotData(Snapshot snapshot)
 	 * It is sufficient to get shared lock on ProcArrayLock, even if we are
 	 * going to set MyProc->xmin.
 	 */
-	LWLockAcquire(ProcArrayLock, LW_SHARED);
+	ProcArrayLockAcquire(PAL_SHARED);
 
 	/* xmax is always latestCompletedXid + 1 */
 	xmax = ShmemVariableCache->latestCompletedXid;
@@ -1418,7 +1384,7 @@ GetSnapshotData(Snapshot snapshot)
 
 	if (!TransactionIdIsValid(MyPgXact->xmin))
 		MyPgXact->xmin = TransactionXmin = xmin;
-	LWLockRelease(ProcArrayLock);
+	ProcArrayLockRelease();
 
 	/*
 	 * Update globalxmin to include actual process xids.  This is a slightly
@@ -1475,7 +1441,7 @@ ProcArrayInstallImportedXmin(TransactionId xmin, TransactionId sourcexid)
 		return false;
 
 	/* Get lock so source xact can't end while we're doing this */
-	LWLockAcquire(ProcArrayLock, LW_SHARED);
+	ProcArrayLockAcquire(PAL_SHARED);
 
 	for (index = 0; index < arrayP->numProcs; index++)
 	{
@@ -1521,7 +1487,7 @@ ProcArrayInstallImportedXmin(TransactionId xmin, TransactionId sourcexid)
 		break;
 	}
 
-	LWLockRelease(ProcArrayLock);
+	ProcArrayLockRelease();
 
 	return result;
 }
@@ -1595,7 +1561,7 @@ GetRunningTransactionData(void)
 	 * Ensure that no xids enter or leave the procarray while we obtain
 	 * snapshot.
 	 */
-	LWLockAcquire(ProcArrayLock, LW_SHARED);
+	ProcArrayLockAcquire(PAL_SHARED);
 	LWLockAcquire(XidGenLock, LW_SHARED);
 
 	latestCompletedXid = ShmemVariableCache->latestCompletedXid;
@@ -1658,7 +1624,7 @@ GetRunningTransactionData(void)
 	CurrentRunningXacts->latestCompletedXid = latestCompletedXid;
 
 	/* We don't release XidGenLock here, the caller is responsible for that */
-	LWLockRelease(ProcArrayLock);
+	ProcArrayLockRelease();
 
 	Assert(TransactionIdIsValid(CurrentRunningXacts->nextXid));
 	Assert(TransactionIdIsValid(CurrentRunningXacts->oldestRunningXid));
@@ -1691,7 +1657,7 @@ GetOldestActiveTransactionId(void)
 
 	Assert(!RecoveryInProgress());
 
-	LWLockAcquire(ProcArrayLock, LW_SHARED);
+	ProcArrayLockAcquire(PAL_SHARED);
 
 	oldestRunningXid = ShmemVariableCache->nextXid;
 
@@ -1720,7 +1686,7 @@ GetOldestActiveTransactionId(void)
 		 */
 	}
 
-	LWLockRelease(ProcArrayLock);
+	ProcArrayLockRelease();
 
 	return oldestRunningXid;
 }
@@ -1753,7 +1719,7 @@ GetTransactionsInCommit(TransactionId **xids_p)
 	xids = (TransactionId *) palloc(arrayP->maxProcs * sizeof(TransactionId));
 	nxids = 0;
 
-	LWLockAcquire(ProcArrayLock, LW_SHARED);
+	ProcArrayLockAcquire(PAL_SHARED);
 
 	for (index = 0; index < arrayP->numProcs; index++)
 	{
@@ -1768,7 +1734,7 @@ GetTransactionsInCommit(TransactionId **xids_p)
 			xids[nxids++] = pxid;
 	}
 
-	LWLockRelease(ProcArrayLock);
+	ProcArrayLockRelease();
 
 	*xids_p = xids;
 	return nxids;
@@ -1790,7 +1756,7 @@ HaveTransactionsInCommit(TransactionId *xids, int nxids)
 	ProcArrayStruct *arrayP = procArray;
 	int			index;
 
-	LWLockAcquire(ProcArrayLock, LW_SHARED);
+	ProcArrayLockAcquire(PAL_SHARED);
 
 	for (index = 0; index < arrayP->numProcs; index++)
 	{
@@ -1818,7 +1784,7 @@ HaveTransactionsInCommit(TransactionId *xids, int nxids)
 		}
 	}
 
-	LWLockRelease(ProcArrayLock);
+	ProcArrayLockRelease();
 
 	return result;
 }
@@ -1840,7 +1806,7 @@ BackendPidGetProc(int pid)
 	if (pid == 0)				/* never match dummy PGPROCs */
 		return NULL;
 
-	LWLockAcquire(ProcArrayLock, LW_SHARED);
+	ProcArrayLockAcquire(PAL_SHARED);
 
 	for (index = 0; index < arrayP->numProcs; index++)
 	{
@@ -1853,7 +1819,7 @@ BackendPidGetProc(int pid)
 		}
 	}
 
-	LWLockRelease(ProcArrayLock);
+	ProcArrayLockRelease();
 
 	return result;
 }
@@ -1881,7 +1847,7 @@ BackendXidGetPid(TransactionId xid)
 	if (xid == InvalidTransactionId)	/* never match invalid xid */
 		return 0;
 
-	LWLockAcquire(ProcArrayLock, LW_SHARED);
+	ProcArrayLockAcquire(PAL_SHARED);
 
 	for (index = 0; index < arrayP->numProcs; index++)
 	{
@@ -1896,7 +1862,7 @@ BackendXidGetPid(TransactionId xid)
 		}
 	}
 
-	LWLockRelease(ProcArrayLock);
+	ProcArrayLockRelease();
 
 	return result;
 }
@@ -1951,7 +1917,7 @@ GetCurrentVirtualXIDs(TransactionId limitXmin, bool excludeXmin0,
 	vxids = (VirtualTransactionId *)
 		palloc(sizeof(VirtualTransactionId) * arrayP->maxProcs);
 
-	LWLockAcquire(ProcArrayLock, LW_SHARED);
+	ProcArrayLockAcquire(PAL_SHARED);
 
 	for (index = 0; index < arrayP->numProcs; index++)
 	{
@@ -1989,7 +1955,7 @@ GetCurrentVirtualXIDs(TransactionId limitXmin, bool excludeXmin0,
 		}
 	}
 
-	LWLockRelease(ProcArrayLock);
+	ProcArrayLockRelease();
 
 	*nvxids = count;
 	return vxids;
@@ -2048,7 +2014,7 @@ GetConflictingVirtualXIDs(TransactionId limitXmin, Oid dbOid)
 					 errmsg("out of memory")));
 	}
 
-	LWLockAcquire(ProcArrayLock, LW_SHARED);
+	ProcArrayLockAcquire(PAL_SHARED);
 
 	for (index = 0; index < arrayP->numProcs; index++)
 	{
@@ -2083,7 +2049,7 @@ GetConflictingVirtualXIDs(TransactionId limitXmin, Oid dbOid)
 		}
 	}
 
-	LWLockRelease(ProcArrayLock);
+	ProcArrayLockRelease();
 
 	/* add the terminator */
 	vxids[count].backendId = InvalidBackendId;
@@ -2104,7 +2070,7 @@ CancelVirtualTransaction(VirtualTransactionId vxid, ProcSignalReason sigmode)
 	int			index;
 	pid_t		pid = 0;
 
-	LWLockAcquire(ProcArrayLock, LW_SHARED);
+	ProcArrayLockAcquire(PAL_SHARED);
 
 	for (index = 0; index < arrayP->numProcs; index++)
 	{
@@ -2131,7 +2097,7 @@ CancelVirtualTransaction(VirtualTransactionId vxid, ProcSignalReason sigmode)
 		}
 	}
 
-	LWLockRelease(ProcArrayLock);
+	ProcArrayLockRelease();
 
 	return pid;
 }
@@ -2207,7 +2173,7 @@ CountDBBackends(Oid databaseid)
 	int			count = 0;
 	int			index;
 
-	LWLockAcquire(ProcArrayLock, LW_SHARED);
+	ProcArrayLockAcquire(PAL_SHARED);
 
 	for (index = 0; index < arrayP->numProcs; index++)
 	{
@@ -2221,7 +2187,7 @@ CountDBBackends(Oid databaseid)
 			count++;
 	}
 
-	LWLockRelease(ProcArrayLock);
+	ProcArrayLockRelease();
 
 	return count;
 }
@@ -2237,7 +2203,7 @@ CancelDBBackends(Oid databaseid, ProcSignalReason sigmode, bool conflictPending)
 	pid_t		pid = 0;
 
 	/* tell all backends to die */
-	LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
+	ProcArrayLockAcquire(PAL_EXCLUSIVE);
 
 	for (index = 0; index < arrayP->numProcs; index++)
 	{
@@ -2263,7 +2229,7 @@ CancelDBBackends(Oid databaseid, ProcSignalReason sigmode, bool conflictPending)
 		}
 	}
 
-	LWLockRelease(ProcArrayLock);
+	ProcArrayLockRelease();
 }
 
 /*
@@ -2276,7 +2242,7 @@ CountUserBackends(Oid roleid)
 	int			count = 0;
 	int			index;
 
-	LWLockAcquire(ProcArrayLock, LW_SHARED);
+	ProcArrayLockAcquire(PAL_SHARED);
 
 	for (index = 0; index < arrayP->numProcs; index++)
 	{
@@ -2289,7 +2255,7 @@ CountUserBackends(Oid roleid)
 			count++;
 	}
 
-	LWLockRelease(ProcArrayLock);
+	ProcArrayLockRelease();
 
 	return count;
 }
@@ -2337,7 +2303,7 @@ CountOtherDBBackends(Oid databaseId, int *nbackends, int *nprepared)
 
 		*nbackends = *nprepared = 0;
 
-		LWLockAcquire(ProcArrayLock, LW_SHARED);
+		ProcArrayLockAcquire(PAL_SHARED);
 
 		for (index = 0; index < arrayP->numProcs; index++)
 		{
@@ -2363,7 +2329,7 @@ CountOtherDBBackends(Oid databaseId, int *nbackends, int *nprepared)
 			}
 		}
 
-		LWLockRelease(ProcArrayLock);
+		ProcArrayLockRelease();
 
 		if (!found)
 			return false;		/* no conflicting backends, so done */
@@ -2416,7 +2382,7 @@ XidCacheRemoveRunningXids(TransactionId xid,
 	 * to abort subtransactions, but pending closer analysis we'd best be
 	 * conservative.
 	 */
-	LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
+	ProcArrayLockAcquire(PAL_EXCLUSIVE);
 
 	/*
 	 * Under normal circumstances xid and xids[] will be in increasing order,
@@ -2464,7 +2430,7 @@ XidCacheRemoveRunningXids(TransactionId xid,
 							  latestXid))
 		ShmemVariableCache->latestCompletedXid = latestXid;
 
-	LWLockRelease(ProcArrayLock);
+	ProcArrayLockRelease();
 }
 
 #ifdef XIDCACHE_DEBUG
@@ -2631,7 +2597,7 @@ ExpireTreeKnownAssignedTransactionIds(TransactionId xid, int nsubxids,
 	/*
 	 * Uses same locking as transaction commit
 	 */
-	LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
+	ProcArrayLockAcquire(PAL_EXCLUSIVE);
 
 	KnownAssignedXidsRemoveTree(xid, nsubxids, subxids);
 
@@ -2640,7 +2606,7 @@ ExpireTreeKnownAssignedTransactionIds(TransactionId xid, int nsubxids,
 							  max_xid))
 		ShmemVariableCache->latestCompletedXid = max_xid;
 
-	LWLockRelease(ProcArrayLock);
+	ProcArrayLockRelease();
 }
 
 /*
@@ -2650,9 +2616,9 @@ ExpireTreeKnownAssignedTransactionIds(TransactionId xid, int nsubxids,
 void
 ExpireAllKnownAssignedTransactionIds(void)
 {
-	LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
+	ProcArrayLockAcquire(PAL_EXCLUSIVE);
 	KnownAssignedXidsRemovePreceding(InvalidTransactionId);
-	LWLockRelease(ProcArrayLock);
+	ProcArrayLockRelease();
 }
 
 /*
@@ -2662,9 +2628,9 @@ ExpireAllKnownAssignedTransactionIds(void)
 void
 ExpireOldKnownAssignedTransactionIds(TransactionId xid)
 {
-	LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
+	ProcArrayLockAcquire(PAL_EXCLUSIVE);
 	KnownAssignedXidsRemovePreceding(xid);
-	LWLockRelease(ProcArrayLock);
+	ProcArrayLockRelease();
 }
 
 
@@ -2886,7 +2852,7 @@ KnownAssignedXidsAdd(TransactionId from_xid, TransactionId to_xid,
 	{
 		/* must hold lock to compress */
 		if (!exclusive_lock)
-			LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
+			ProcArrayLockAcquire(PAL_EXCLUSIVE);
 
 		KnownAssignedXidsCompress(true);
 
@@ -2894,7 +2860,7 @@ KnownAssignedXidsAdd(TransactionId from_xid, TransactionId to_xid,
 		/* note: we no longer care about the tail pointer */
 
 		if (!exclusive_lock)
-			LWLockRelease(ProcArrayLock);
+			ProcArrayLockRelease();
 
 		/*
 		 * If it still won't fit then we're out of memory
diff --git a/src/backend/storage/lmgr/Makefile b/src/backend/storage/lmgr/Makefile
index 3730e51..27eaa97 100644
--- a/src/backend/storage/lmgr/Makefile
+++ b/src/backend/storage/lmgr/Makefile
@@ -13,7 +13,7 @@ top_builddir = ../../../..
 include $(top_builddir)/src/Makefile.global
 
 OBJS = flexlock.o lmgr.o lock.o proc.o deadlock.o lwlock.o spin.o s_lock.o \
-	predicate.o
+	procarraylock.o predicate.o
 
 include $(top_srcdir)/src/backend/common.mk
 
diff --git a/src/backend/storage/lmgr/flexlock.c b/src/backend/storage/lmgr/flexlock.c
index cf0004b..434e9c7 100644
--- a/src/backend/storage/lmgr/flexlock.c
+++ b/src/backend/storage/lmgr/flexlock.c
@@ -30,6 +30,7 @@
 #include "storage/flexlock.h"
 #include "storage/flexlock_internals.h"
 #include "storage/predicate.h"
+#include "storage/procarraylock.h"
 #include "storage/spin.h"
 
 int	num_held_flexlocks = 0;
@@ -168,9 +169,14 @@ CreateFlexLocks(void)
 
 	FlexLockArray = (FlexLockPadded *) ptr;
 
-	/* All of the "fixed" FlexLocks are LWLocks. */
+	/* All of the "fixed" FlexLocks are LWLocks - except ProcArrayLock. */
 	for (id = 0, lock = FlexLockArray; id < NumFixedFlexLocks; id++, lock++)
-		FlexLockInit(&lock->flex, FLEXLOCK_TYPE_LWLOCK);
+	{
+		if (id == ProcArrayLock)
+			FlexLockInit(&lock->flex, FLEXLOCK_TYPE_PROCARRAYLOCK);
+		else
+			FlexLockInit(&lock->flex, FLEXLOCK_TYPE_LWLOCK);
+	}
 
 	/*
 	 * Initialize the dynamic-allocation counter, which is stored just before
@@ -242,13 +248,20 @@ FlexLockReleaseAll(void)
 {
 	while (num_held_flexlocks > 0)
 	{
+		FlexLockId	id;
+		FlexLock   *flex;
+
 		HOLD_INTERRUPTS();		/* match the upcoming RESUME_INTERRUPTS */
 
-		/*
-		 * FLEXTODO: When we have multiple types of flex locks, this will
-		 * need to call the appropriate release function for each lock type.
-		 */
-		LWLockRelease(held_flexlocks[num_held_flexlocks - 1]);
+		id = held_flexlocks[num_held_flexlocks - 1];
+		flex = &FlexLockArray[id].flex;
+		if (flex->locktype == FLEXLOCK_TYPE_LWLOCK)
+			LWLockRelease(id);
+		else
+		{
+			Assert(id == ProcArrayLock);
+			ProcArrayLockRelease();
+		}
 	}
 }
 
diff --git a/src/backend/storage/lmgr/proc.c b/src/backend/storage/lmgr/proc.c
index b402999..10ec83b 100644
--- a/src/backend/storage/lmgr/proc.c
+++ b/src/backend/storage/lmgr/proc.c
@@ -46,6 +46,7 @@
 #include "storage/pmsignal.h"
 #include "storage/proc.h"
 #include "storage/procarray.h"
+#include "storage/procarraylock.h"
 #include "storage/procsignal.h"
 #include "storage/spin.h"
 #include "utils/timestamp.h"
@@ -1083,7 +1084,7 @@ ProcSleep(LOCALLOCK *locallock, LockMethod lockMethodTable)
 			PGPROC	   *autovac = GetBlockingAutoVacuumPgproc();
 			PGXACT	   *autovac_pgxact = &ProcGlobal->allPgXact[autovac->pgprocno];
 
-			LWLockAcquire(ProcArrayLock, LW_EXCLUSIVE);
+			ProcArrayLockAcquire(PAL_EXCLUSIVE);
 
 			/*
 			 * Only do it if the worker is not working to protect against Xid
@@ -1099,7 +1100,7 @@ ProcSleep(LOCALLOCK *locallock, LockMethod lockMethodTable)
 					 pid);
 
 				/* don't hold the lock across the kill() syscall */
-				LWLockRelease(ProcArrayLock);
+				ProcArrayLockRelease();
 
 				/* send the autovacuum worker Back to Old Kent Road */
 				if (kill(pid, SIGINT) < 0)
@@ -1111,7 +1112,7 @@ ProcSleep(LOCALLOCK *locallock, LockMethod lockMethodTable)
 				}
 			}
 			else
-				LWLockRelease(ProcArrayLock);
+				ProcArrayLockRelease();
 
 			/* prevent signal from being resent more than once */
 			allow_autovacuum_cancel = false;
diff --git a/src/backend/storage/lmgr/procarraylock.c b/src/backend/storage/lmgr/procarraylock.c
new file mode 100644
index 0000000..e4fdd2d
--- /dev/null
+++ b/src/backend/storage/lmgr/procarraylock.c
@@ -0,0 +1,344 @@
+/*-------------------------------------------------------------------------
+ *
+ * procarraylock.c
+ *	  Lock management for the ProcArray
+ *
+ * Because the ProcArray data structure is highly trafficked, it is
+ * critical that mutual exclusion for ProcArray options be as efficient
+ * as possible.  A particular problem is transaction end (commit or abort)
+ * which cannot be done in parallel with snapshot acquisition.  We
+ * therefore include some special hacks to deal with this case efficiently.
+ *
+ * Portions Copyright (c) 1996-2011, PostgreSQL Global Development Group
+ * Portions Copyright (c) 1994, Regents of the University of California
+ *
+ * IDENTIFICATION
+ *	  src/backend/storage/lmgr/procarraylock.c
+ *
+ *-------------------------------------------------------------------------
+ */
+#include "postgres.h"
+
+#include "miscadmin.h"
+#include "pg_trace.h"
+#include "access/transam.h"
+#include "storage/flexlock_internals.h"
+#include "storage/ipc.h"
+#include "storage/procarraylock.h"
+#include "storage/proc.h"
+#include "storage/spin.h"
+
+typedef struct ProcArrayLockStruct
+{
+	FlexLock	flex;			/* common FlexLock infrastructure */
+	char		exclusive;		/* # of exclusive holders (0 or 1) */
+	int			shared;			/* # of shared holders (0..MaxBackends) */
+	PGPROC	   *ending;			/* transactions wishing to clear state */
+	TransactionId	latest_ending_xid;	/* latest ending XID */
+} ProcArrayLockStruct;
+
+/* There is only one ProcArrayLock. */
+#define	ProcArrayLockPointer() \
+	(AssertMacro(FlexLockArray[ProcArrayLock].flex.locktype == \
+		FLEXLOCK_TYPE_PROCARRAYLOCK), \
+	 (volatile ProcArrayLockStruct *) &FlexLockArray[ProcArrayLock])
+
+/*
+ * ProcArrayLockAcquire - 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
+ProcArrayLockAcquire(ProcArrayLockMode mode)
+{
+	volatile ProcArrayLockStruct *lock = ProcArrayLockPointer();
+	PGPROC	   *proc = MyProc;
+	bool		retry = false;
+	int			extraWaits = 0;
+
+	/*
+	 * 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 ProcArrayLock.  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
+	 * ProcArrayLockRelease.  See comments in LWLockAcquire for an explanation
+	 * of why do we not attempt to hand off the lock directly.
+	 */
+	for (;;)
+	{
+		bool		mustwait;
+
+		/* Acquire mutex.  Time spent holding mutex should be short! */
+		SpinLockAcquire(&lock->flex.mutex);
+
+		/* If retrying, allow LWLockRelease to release waiters again */
+		if (retry)
+			lock->flex.releaseOK = true;
+
+		/* If I can get the lock, do so quickly. */
+		if (mode == PAL_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. */
+		FlexLockJoinWaitQueue(lock, (int) mode);
+
+		/* Can release the mutex now */
+		SpinLockRelease(&lock->flex.mutex);
+
+		/* Wait until awakened. */
+		FlexLockWait(ProcArrayLock, mode, extraWaits);
+
+		/* Now loop back and try to acquire lock again. */
+		retry = true;
+	}
+
+	/* We are done updating shared state of the lock itself. */
+	SpinLockRelease(&lock->flex.mutex);
+
+	TRACE_POSTGRESQL_FLEXLOCK_ACQUIRE(lockid, mode);
+
+	/* Add lock to list of locks held by this backend */
+	FlexLockRemember(ProcArrayLock);
+
+	/*
+	 * Fix the process wait semaphore's count for any absorbed wakeups.
+	 */
+	while (extraWaits-- > 0)
+		PGSemaphoreUnlock(&proc->sem);
+}
+
+/*
+ * ProcArrayLockClearTransaction - safely clear transaction details
+ *
+ * This can't be done while ProcArrayLock is held, but it's so fast that
+ * we can afford to do it while holding the spinlock, rather than acquiring
+ * and releasing the lock.
+ */
+void
+ProcArrayLockClearTransaction(TransactionId latestXid)
+{
+	volatile ProcArrayLockStruct *lock = ProcArrayLockPointer();
+	PGPROC	   *proc = MyProc;
+	int			extraWaits = 0;
+	bool		mustwait;
+
+	HOLD_INTERRUPTS();
+
+	/* Acquire mutex.  Time spent holding mutex should be short! */
+	SpinLockAcquire(&lock->flex.mutex);
+
+	if (lock->exclusive == 0 && lock->shared == 0)
+	{
+		{
+			volatile PGPROC *vproc = proc;
+			volatile PGXACT *pgxact = &ProcGlobal->allPgXact[vproc->pgprocno];
+			/* If there are no lockers, clear the critical PGPROC fields. */
+			pgxact->xid = InvalidTransactionId;
+	        pgxact->xmin = InvalidTransactionId;
+	        /* must be cleared with xid/xmin: */
+	        pgxact->vacuumFlags &= ~PROC_VACUUM_STATE_MASK;
+			pgxact->nxids = 0;
+			pgxact->overflowed = false;
+		}
+		mustwait = false;
+
+        /* Also advance global latestCompletedXid while holding the lock */
+        if (TransactionIdPrecedes(ShmemVariableCache->latestCompletedXid,
+                                  latestXid))
+            ShmemVariableCache->latestCompletedXid = latestXid;
+	}
+	else
+	{
+		/* Rats, must wait. */
+		proc->flWaitLink = lock->ending;
+		lock->ending = proc;
+		if (!TransactionIdIsValid(lock->latest_ending_xid) ||
+				TransactionIdPrecedes(lock->latest_ending_xid, latestXid)) 
+			lock->latest_ending_xid = latestXid;
+		mustwait = true;
+	}
+
+	/* Can release the mutex now */
+	SpinLockRelease(&lock->flex.mutex);
+
+	/*
+	 * If we were not able to perfom the operation immediately, we must wait.
+	 * But we need not retry after being awoken, because the last lock holder
+	 * to release the lock will do the work first, on our behalf.
+	 */
+	if (mustwait)
+	{
+		FlexLockWait(ProcArrayLock, 2, extraWaits);
+		while (extraWaits-- > 0)
+			PGSemaphoreUnlock(&proc->sem);
+	}
+
+	RESUME_INTERRUPTS();
+}
+
+/*
+ * ProcArrayLockRelease - release a previously acquired lock
+ */
+void
+ProcArrayLockRelease(void)
+{
+	volatile ProcArrayLockStruct *lock = ProcArrayLockPointer();
+	PGPROC	   *head;
+	PGPROC	   *ending = NULL;
+	PGPROC	   *proc;
+
+	FlexLockForget(ProcArrayLock);
+
+	/* Acquire mutex.  Time spent holding mutex should be short! */
+	SpinLockAcquire(&lock->flex.mutex);
+
+	/* Release my hold on lock */
+	if (lock->exclusive > 0)
+		lock->exclusive--;
+	else
+	{
+		Assert(lock->shared > 0);
+		lock->shared--;
+	}
+
+	/*
+	 * If the lock is now free, but there are some transactions trying to
+	 * end, we must clear the critical PGPROC fields for them, and save a
+	 * list of them so we can wake them up.
+	 */
+	if (lock->exclusive == 0 && lock->shared == 0 && lock->ending != NULL)
+	{
+		volatile PGPROC *vproc;
+
+		ending = lock->ending;
+		vproc = ending;
+
+		while (vproc != NULL)
+		{
+			volatile PGXACT *pgxact = &ProcGlobal->allPgXact[vproc->pgprocno];
+
+        	pgxact->xid = InvalidTransactionId;
+	        pgxact->xmin = InvalidTransactionId;
+	        /* must be cleared with xid/xmin: */
+	        pgxact->vacuumFlags &= ~PROC_VACUUM_STATE_MASK;
+			pgxact->nxids = 0;
+			pgxact->overflowed = false;
+			vproc = vproc->flWaitLink;
+		}
+
+		/* Also advance global latestCompletedXid */
+		if (TransactionIdPrecedes(ShmemVariableCache->latestCompletedXid,
+								  lock->latest_ending_xid))
+			ShmemVariableCache->latestCompletedXid = lock->latest_ending_xid;
+
+		/* Reset lock state. */
+		lock->ending = NULL;
+		lock->latest_ending_xid = InvalidTransactionId;
+	}
+
+	/*
+	 * 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->flex.head;
+	if (head != NULL)
+	{
+		if (lock->exclusive == 0 && lock->shared == 0 && lock->flex.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->flWaitMode != LW_EXCLUSIVE)
+			{
+				while (proc->flWaitLink != NULL &&
+					   proc->flWaitLink->flWaitMode != LW_EXCLUSIVE)
+					proc = proc->flWaitLink;
+			}
+			/* proc is now the last PGPROC to be released */
+			lock->flex.head = proc->flWaitLink;
+			proc->flWaitLink = NULL;
+			/* prevent additional wakeups until retryer gets to run */
+			lock->flex.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->flex.mutex);
+
+	TRACE_POSTGRESQL_FLEXLOCK_RELEASE(lockid);
+
+	/*
+	 * Awaken any waiters I removed from the queue.
+	 */
+	while (head != NULL)
+	{
+		FlexLockDebug("LWLockRelease", lockid, "release waiter");
+		proc = head;
+		head = proc->flWaitLink;
+		proc->flWaitLink = NULL;
+		proc->flWaitResult = 1;		/* any non-zero value will do */
+		PGSemaphoreUnlock(&proc->sem);
+	}
+
+	/*
+	 * Also awaken any processes whose critical PGPROC fields I cleared
+	 */
+	while (ending != NULL)
+	{
+		FlexLockDebug("LWLockRelease", lockid, "release ending");
+		proc = ending;
+		ending = proc->flWaitLink;
+		proc->flWaitLink = NULL;
+		proc->flWaitResult = 1;		/* any non-zero value will do */
+		PGSemaphoreUnlock(&proc->sem);
+	}
+
+	/*
+	 * Now okay to allow cancel/die interrupts.
+	 */
+	RESUME_INTERRUPTS();
+}
diff --git a/src/include/storage/flexlock_internals.h b/src/include/storage/flexlock_internals.h
index a5a6bde..5749f8b 100644
--- a/src/include/storage/flexlock_internals.h
+++ b/src/include/storage/flexlock_internals.h
@@ -41,6 +41,7 @@ typedef struct FlexLock
 } FlexLock;
 
 #define FLEXLOCK_TYPE_LWLOCK			'l'
+#define FLEXLOCK_TYPE_PROCARRAYLOCK		'p'
 
 typedef union FlexLockPadded
 {
diff --git a/src/include/storage/procarraylock.h b/src/include/storage/procarraylock.h
new file mode 100644
index 0000000..678ca6f
--- /dev/null
+++ b/src/include/storage/procarraylock.h
@@ -0,0 +1,28 @@
+/*-------------------------------------------------------------------------
+ *
+ * procarraylock.h
+ *	  Lock management for the ProcArray
+ *
+ * Portions Copyright (c) 1996-2011, PostgreSQL Global Development Group
+ * Portions Copyright (c) 1994, Regents of the University of California
+ *
+ * src/include/storage/lwlock.h
+ *
+ *-------------------------------------------------------------------------
+ */
+#ifndef PROCARRAYLOCK_H
+#define PROCARRAYLOCK_H
+
+#include "storage/flexlock.h"
+
+typedef enum ProcArrayLockMode
+{
+	PAL_EXCLUSIVE,
+	PAL_SHARED
+} ProcArrayLockMode;
+
+extern void ProcArrayLockAcquire(ProcArrayLockMode mode);
+extern void ProcArrayLockClearTransaction(TransactionId latestXid);
+extern void ProcArrayLockRelease(void);
+
+#endif   /* PROCARRAYLOCK_H */