This ChangeSet must be null-merged to 5.1. Applied innodb-5.0-ss982, -ss998, -ss1003

Fixes:
- Bug #15815: Very poor performance with multiple queries running concurrently
- Bug #22868: 'Thread thrashing' with > 50 concurrent conns under an upd-intensive workloadw
- Bug #23769: Debug assertion failure with innodb_locks_unsafe_for_binlog
- Bug #24089: Race condition in fil_flush_file_spaces()

innobase/buf/buf0flu.c

@@ -114,6 +114,7 @@ buf_flush_ready_for_replace(
 {
 #ifdef UNIV_SYNC_DEBUG
 	ut_ad(mutex_own(&(buf_pool->mutex)));
+	ut_ad(mutex_own(&block->mutex));
 #endif /* UNIV_SYNC_DEBUG */
 	if (block->state != BUF_BLOCK_FILE_PAGE) {
 		ut_print_timestamp(stderr);
@@ -148,6 +149,7 @@ buf_flush_ready_for_flush(
 {
 #ifdef UNIV_SYNC_DEBUG
 	ut_ad(mutex_own(&(buf_pool->mutex)));
+	ut_ad(mutex_own(&(block->mutex)));
 #endif /* UNIV_SYNC_DEBUG */
 	ut_a(block->state == BUF_BLOCK_FILE_PAGE);
 
@@ -539,8 +541,15 @@ buf_flush_try_page(
 
 	ut_a(!block || block->state == BUF_BLOCK_FILE_PAGE);
 
+	if (!block) {
+		mutex_exit(&(buf_pool->mutex));
+		return(0);
+	}
+
+	mutex_enter(&block->mutex);
+
 	if (flush_type == BUF_FLUSH_LIST
-	    && block && buf_flush_ready_for_flush(block, flush_type)) {
+	    && buf_flush_ready_for_flush(block, flush_type)) {
 	
 		block->io_fix = BUF_IO_WRITE;
 
@@ -578,6 +587,7 @@ buf_flush_try_page(
 			locked = TRUE;
 		}
 
+		mutex_exit(&block->mutex);
 		mutex_exit(&(buf_pool->mutex));
 
 		if (!locked) {
@@ -598,8 +608,8 @@ buf_flush_try_page(
 		
 		return(1);
 
-	} else if (flush_type == BUF_FLUSH_LRU && block
-			&& buf_flush_ready_for_flush(block, flush_type)) {
+	} else if (flush_type == BUF_FLUSH_LRU
+		   && buf_flush_ready_for_flush(block, flush_type)) {
 
 		/* VERY IMPORTANT:
 		Because any thread may call the LRU flush, even when owning
@@ -639,14 +649,15 @@ buf_flush_try_page(
 		buf_pool mutex: this ensures that the latch is acquired
 		immediately. */
 		
+		mutex_exit(&block->mutex);
 		mutex_exit(&(buf_pool->mutex));
 
 		buf_flush_write_block_low(block);
 
 		return(1);
 
-	} else if (flush_type == BUF_FLUSH_SINGLE_PAGE && block
-			&& buf_flush_ready_for_flush(block, flush_type)) {
+	} else if (flush_type == BUF_FLUSH_SINGLE_PAGE
+		   && buf_flush_ready_for_flush(block, flush_type)) {
 	
 		block->io_fix = BUF_IO_WRITE;
 
@@ -672,6 +683,7 @@ buf_flush_try_page(
 
 		(buf_pool->n_flush[flush_type])++;
 
+		mutex_exit(&block->mutex);
 		mutex_exit(&(buf_pool->mutex));
 
 		rw_lock_s_lock_gen(&(block->lock), BUF_IO_WRITE);
@@ -688,11 +700,12 @@ buf_flush_try_page(
 		buf_flush_write_block_low(block);
 		
 		return(1);
-	} else {
-		mutex_exit(&(buf_pool->mutex));
+	}
 
-		return(0);
-	}		
+	mutex_exit(&block->mutex);
+	mutex_exit(&(buf_pool->mutex));
+
+	return(0);
 }
 
 /***************************************************************
@@ -737,34 +750,48 @@ buf_flush_try_neighbors(
 		block = buf_page_hash_get(space, i);
 		ut_a(!block || block->state == BUF_BLOCK_FILE_PAGE);
 
-		if (block && flush_type == BUF_FLUSH_LRU && i != offset
-		    && !block->old) {
+		if (!block) {
+
+			continue;
+
+		} else if (flush_type == BUF_FLUSH_LRU && i != offset
+			   && !block->old) {
 
 		        /* We avoid flushing 'non-old' blocks in an LRU flush,
 		        because the flushed blocks are soon freed */
 
 		        continue;
-		}
+		} else {
+
+			mutex_enter(&block->mutex);
+
+			if (buf_flush_ready_for_flush(block, flush_type)
+			    && (i == offset || block->buf_fix_count == 0)) {
+				/* We only try to flush those
+				neighbors != offset where the buf fix count is
+				zero, as we then know that we probably can
+				latch the page without a semaphore wait.
+				Semaphore waits are expensive because we must
+				flush the doublewrite buffer before we start
+				waiting. */
 
-		if (block && buf_flush_ready_for_flush(block, flush_type)
-		   && (i == offset || block->buf_fix_count == 0)) {
-			/* We only try to flush those neighbors != offset
-			where the buf fix count is zero, as we then know that
-			we probably can latch the page without a semaphore
-			wait. Semaphore waits are expensive because we must
-			flush the doublewrite buffer before we start
-			waiting. */
+				mutex_exit(&block->mutex);
 
-			mutex_exit(&(buf_pool->mutex));
+				mutex_exit(&(buf_pool->mutex));
 
-			/* Note: as we release the buf_pool mutex above, in
-			buf_flush_try_page we cannot be sure the page is still
-			in a flushable state: therefore we check it again
-			inside that function. */
+				/* Note: as we release the buf_pool mutex
+				above, in buf_flush_try_page we cannot be sure
+				the page is still in a flushable state:
+				therefore we check it again inside that
+				function. */
 
-			count += buf_flush_try_page(space, i, flush_type);
+				count += buf_flush_try_page(space, i,
+							    flush_type);
 
-			mutex_enter(&(buf_pool->mutex));
+				mutex_enter(&(buf_pool->mutex));
+			} else {
+				mutex_exit(&block->mutex);
+			}
 		}
 	}
 				
@@ -858,12 +885,15 @@ buf_flush_batch(
 	    	while ((block != NULL) && !found) {
 			ut_a(block->state == BUF_BLOCK_FILE_PAGE);
 
+			mutex_enter(&block->mutex);
+
 			if (buf_flush_ready_for_flush(block, flush_type)) {
 
 				found = TRUE;
 				space = block->space;
 				offset = block->offset;
 	    
+				mutex_exit(&block->mutex);
 				mutex_exit(&(buf_pool->mutex));
 
 				old_page_count = page_count;
@@ -881,10 +911,14 @@ buf_flush_batch(
 
 			} else if (flush_type == BUF_FLUSH_LRU) {
 
+				mutex_exit(&block->mutex);
+
 				block = UT_LIST_GET_PREV(LRU, block);
 			} else {
 				ut_ad(flush_type == BUF_FLUSH_LIST);
 
+				mutex_exit(&block->mutex);
+
 				block = UT_LIST_GET_PREV(flush_list, block);
 			}
 	    	}
@@ -966,10 +1000,14 @@ buf_flush_LRU_recommendation(void)
 	       				+ BUF_FLUSH_EXTRA_MARGIN)
 	       && (distance < BUF_LRU_FREE_SEARCH_LEN)) {
 
+		mutex_enter(&block->mutex);
+
 		if (buf_flush_ready_for_replace(block)) {
 			n_replaceable++;
 		}
 
+		mutex_exit(&block->mutex);
+
 		distance++;
 			
 		block = UT_LIST_GET_PREV(LRU, block);

innobase/buf/buf0lru.c

@@ -86,6 +86,9 @@ scan_again:
 	block = UT_LIST_GET_LAST(buf_pool->LRU);
 
 	while (block != NULL) {
+
+		mutex_enter(&block->mutex);
+
 	        ut_a(block->state == BUF_BLOCK_FILE_PAGE);
 
 		if (block->space == id
@@ -112,6 +115,8 @@ scan_again:
 			if (block->is_hashed) {
 				page_no = block->offset;
 			
+				mutex_exit(&block->mutex);
+
 				mutex_exit(&(buf_pool->mutex));
 
 				/* Note that the following call will acquire
@@ -138,6 +143,7 @@ scan_again:
 			buf_LRU_block_free_hashed_page(block);
 		}
 next_page:
+		mutex_exit(&block->mutex);
 		block = UT_LIST_GET_PREV(LRU, block);
 	}
 
@@ -211,6 +217,9 @@ buf_LRU_search_and_free_block(
 
 	while (block != NULL) {
 	        ut_a(block->in_LRU_list);
+
+		mutex_enter(&block->mutex);
+
 		if (buf_flush_ready_for_replace(block)) {
 
 #ifdef UNIV_DEBUG
@@ -225,6 +234,7 @@ buf_LRU_search_and_free_block(
 			buf_LRU_block_remove_hashed_page(block);
 
 			mutex_exit(&(buf_pool->mutex));
+			mutex_exit(&block->mutex);
 
 			/* Remove possible adaptive hash index built on the
 			page; in the case of AWE the block may not have a
@@ -233,15 +243,21 @@ buf_LRU_search_and_free_block(
 			if (block->frame) {
 				btr_search_drop_page_hash_index(block->frame);
 			}
-			mutex_enter(&(buf_pool->mutex));
 
 			ut_a(block->buf_fix_count == 0);
 
+			mutex_enter(&(buf_pool->mutex));
+			mutex_enter(&block->mutex);
+
 			buf_LRU_block_free_hashed_page(block);
 			freed = TRUE;
+			mutex_exit(&block->mutex);
 
 			break;
 		}
+
+		mutex_exit(&block->mutex);
+
 		block = UT_LIST_GET_PREV(LRU, block);
 		distance++;
 
@@ -415,8 +431,12 @@ loop:
 			}
 		}
 		
+		mutex_enter(&block->mutex);
+
 		block->state = BUF_BLOCK_READY_FOR_USE;
 
+		mutex_exit(&block->mutex);
+
 		mutex_exit(&(buf_pool->mutex));
 
 		if (started_monitor) {
@@ -818,6 +838,7 @@ buf_LRU_block_free_non_file_page(
 {
 #ifdef UNIV_SYNC_DEBUG
 	ut_ad(mutex_own(&(buf_pool->mutex)));
+	ut_ad(mutex_own(&block->mutex));
 #endif /* UNIV_SYNC_DEBUG */
 	ut_ad(block);
 	
@@ -857,6 +878,7 @@ buf_LRU_block_remove_hashed_page(
 {
 #ifdef UNIV_SYNC_DEBUG
 	ut_ad(mutex_own(&(buf_pool->mutex)));
+	ut_ad(mutex_own(&block->mutex));
 #endif /* UNIV_SYNC_DEBUG */
 	ut_ad(block);
 	
@@ -914,6 +936,7 @@ buf_LRU_block_free_hashed_page(
 {
 #ifdef UNIV_SYNC_DEBUG
 	ut_ad(mutex_own(&(buf_pool->mutex)));
+	ut_ad(mutex_own(&block->mutex));
 #endif /* UNIV_SYNC_DEBUG */
 	ut_a(block->state == BUF_BLOCK_REMOVE_HASH);
 

innobase/dict/dict0crea.c

@@ -724,8 +724,10 @@ dict_truncate_index_tree(
 				/* out: new root page number, or
 				FIL_NULL on failure */
 	dict_table_t*	table,	/* in: the table the index belongs to */
-	rec_t*		rec,	/* in: record in the clustered index of
-				SYS_INDEXES table */
+	btr_pcur_t*	pcur,	/* in/out: persistent cursor pointing to
+				record in the clustered index of
+				SYS_INDEXES table. The cursor may be
+				repositioned in this call. */
 	mtr_t*		mtr)	/* in: mtr having the latch
 				on the record page. The mtr may be
 				committed and restarted in this call. */
@@ -734,6 +736,7 @@ dict_truncate_index_tree(
 	ulint		space;
 	ulint		type;
 	dulint		index_id;
+	rec_t*		rec;
 	byte*		ptr;
 	ulint		len;
 	ulint		comp;
@@ -744,6 +747,7 @@ dict_truncate_index_tree(
 #endif /* UNIV_SYNC_DEBUG */
 
 	ut_a(!dict_sys->sys_indexes->comp);
+	rec = btr_pcur_get_rec(pcur);
 	ptr = rec_get_nth_field_old(rec, DICT_SYS_INDEXES_PAGE_NO_FIELD, &len);
 
 	ut_ad(len == 4);
@@ -809,10 +813,11 @@ dict_truncate_index_tree(
 	/* We will need to commit the mini-transaction in order to avoid
 	deadlocks in the btr_create() call, because otherwise we would
 	be freeing and allocating pages in the same mini-transaction. */
+	btr_pcur_store_position(pcur, mtr);
 	mtr_commit(mtr);
-	/* mtr_commit() will invalidate rec. */
-	rec = NULL;
+
 	mtr_start(mtr);
+	btr_pcur_restore_position(BTR_MODIFY_LEAF, pcur, mtr);
 
 	/* Find the index corresponding to this SYS_INDEXES record. */
 	for (index = UT_LIST_GET_FIRST(table->indexes);

innobase/fil/fil0fil.c

@@ -4285,29 +4285,47 @@ fil_flush_file_spaces(
 {
 	fil_system_t*	system	= fil_system;
 	fil_space_t*	space;
+	ulint*		space_ids;
+	ulint		n_space_ids;
+	ulint		i;
 
 	mutex_enter(&(system->mutex));
 
-	space = UT_LIST_GET_FIRST(system->unflushed_spaces);
+	n_space_ids = UT_LIST_GET_LEN(system->unflushed_spaces);
+	if (n_space_ids == 0) {
 
-	while (space) {
-		if (space->purpose == purpose && !space->is_being_deleted) {
+		mutex_exit(&system->mutex);
+		return;
+	}
 
-			space->n_pending_flushes++; /* prevent dropping of the
-						    space while we are
-						    flushing */
-			mutex_exit(&(system->mutex));
+	/* Assemble a list of space ids to flush.  Previously, we
+	traversed system->unflushed_spaces and called UT_LIST_GET_NEXT()
+	on a space that was just removed from the list by fil_flush().
+	Thus, the space could be dropped and the memory overwritten. */
+	space_ids = mem_alloc(n_space_ids * sizeof *space_ids);
 
-			fil_flush(space->id);
+	n_space_ids = 0;
 
-			mutex_enter(&(system->mutex));
+	for (space = UT_LIST_GET_FIRST(system->unflushed_spaces);
+	     space;
+	     space = UT_LIST_GET_NEXT(unflushed_spaces, space)) {
+
+		if (space->purpose == purpose && !space->is_being_deleted) {
 
-			space->n_pending_flushes--;
+			space_ids[n_space_ids++] = space->id;
 		}
-		space = UT_LIST_GET_NEXT(unflushed_spaces, space);
 	}
-	
-	mutex_exit(&(system->mutex));
+
+	mutex_exit(&system->mutex);
+
+	/* Flush the spaces.  It will not hurt to call fil_flush() on
+	a non-existing space id. */
+	for (i = 0; i < n_space_ids; i++) {
+
+		fil_flush(space_ids[i]);
+	}
+
+	mem_free(space_ids);
 }
 
 /**********************************************************************

innobase/include/buf0buf.h

@@ -461,8 +461,8 @@ Gets the mutex number protecting the page record lock hash chain in the lock
 table. */
 UNIV_INLINE
 mutex_t*
-buf_frame_get_lock_mutex(
-/*=====================*/
+buf_frame_get_mutex(
+/*================*/
 			/* out: mutex */
 	byte*	ptr);	/* in: pointer to within a buffer frame */
 /***********************************************************************
@@ -713,7 +713,10 @@ struct buf_block_struct{
 
 	ulint		magic_n;	/* magic number to check */
 	ulint		state;		/* state of the control block:
-					BUF_BLOCK_NOT_USED, ... */
+					BUF_BLOCK_NOT_USED, ...; changing
+					this is only allowed when a thread
+					has BOTH the buffer pool mutex AND
+					block->mutex locked */
 	byte*		frame;		/* pointer to buffer frame which
 					is of size UNIV_PAGE_SIZE, and
 					aligned to an address divisible by
@@ -731,8 +734,12 @@ struct buf_block_struct{
 	ulint		offset;		/* page number within the space */
 	ulint		lock_hash_val;	/* hashed value of the page address
 					in the record lock hash table */
-	mutex_t*	lock_mutex;	/* mutex protecting the chain in the
-					record lock hash table */
+	mutex_t		mutex;		/* mutex protecting this block:
+					state (also protected by the buffer
+					pool mutex), io_fix, buf_fix_count,
+					and accessed; we introduce this new
+					mutex in InnoDB-5.1 to relieve
+					contention on the buffer pool mutex */
 	rw_lock_t	lock;		/* read-write lock of the buffer
 					frame */
 	buf_block_t*	hash;		/* node used in chaining to the page
@@ -788,20 +795,27 @@ struct buf_block_struct{
 					in heuristic algorithms, because of
 					the possibility of a wrap-around! */
 	ulint		freed_page_clock;/* the value of freed_page_clock
-					buffer pool when this block was
-					last time put to the head of the
-					LRU list */
+					of the buffer pool when this block was
+					the last time put to the head of the
+					LRU list; a thread is allowed to
+					read this for heuristic purposes
+					without holding any mutex or latch */
 	ibool		old;		/* TRUE if the block is in the old
 					blocks in the LRU list */
 	ibool		accessed;	/* TRUE if the page has been accessed
 					while in the buffer pool: read-ahead
 					may read in pages which have not been
-					accessed yet */
+					accessed yet; this is protected by
+					block->mutex; a thread is allowed to
+					read this for heuristic purposes
+					without holding any mutex or latch */
 	ulint		buf_fix_count;	/* count of how manyfold this block
-					is currently bufferfixed */
+					is currently bufferfixed; this is
+					protected by block->mutex */
 	ulint		io_fix;		/* if a read is pending to the frame,
 					io_fix is BUF_IO_READ, in the case
-					of a write BUF_IO_WRITE, otherwise 0 */
+					of a write BUF_IO_WRITE, otherwise 0;
+					this is protected by block->mutex */
 	/* 4. Optimistic search field */
 
 	dulint		modify_clock;	/* this clock is incremented every
@@ -962,7 +976,9 @@ struct buf_pool_struct{
 					number of buffer blocks removed from
 					the end of the LRU list; NOTE that
 					this counter may wrap around at 4
-					billion! */
+					billion! A thread is allowed to
+					read this for heuristic purposes
+					without holding any mutex or latch */
 	ulint		LRU_flush_ended;/* when an LRU flush ends for a page,
 					this is incremented by one; this is
 					set to zero when a buffer block is

innobase/include/buf0buf.ic

@@ -330,8 +330,8 @@ Gets the mutex number protecting the page record lock hash chain in the lock
 table. */
 UNIV_INLINE
 mutex_t*
-buf_frame_get_lock_mutex(
-/*=====================*/
+buf_frame_get_mutex(
+/*================*/
 			/* out: mutex */
 	byte*	ptr)	/* in: pointer to within a buffer frame */
 {
@@ -339,7 +339,7 @@ buf_frame_get_lock_mutex(
 
 	block = buf_block_align(ptr);
 
-	return(block->lock_mutex);
+	return(&block->mutex);
 }
 
 /*************************************************************************
@@ -512,6 +512,7 @@ buf_block_buf_fix_inc_debug(
 	ret = rw_lock_s_lock_func_nowait(&(block->debug_latch), file, line);
 
 	ut_ad(ret == TRUE);
+	ut_ad(mutex_own(&block->mutex));
 #endif
 	block->buf_fix_count++;
 }
@@ -524,6 +525,9 @@ buf_block_buf_fix_inc(
 /*==================*/
 	buf_block_t*	block)	/* in: block to bufferfix */
 {
+#ifdef UNIV_SYNC_DEBUG
+	ut_ad(mutex_own(&block->mutex));
+#endif
 	block->buf_fix_count++;
 }
 #endif /* UNIV_SYNC_DEBUG */
@@ -618,23 +622,24 @@ buf_page_release(
 	
 	ut_ad(block);
 
-	mutex_enter_fast(&(buf_pool->mutex));
-
 	ut_a(block->state == BUF_BLOCK_FILE_PAGE);
 	ut_a(block->buf_fix_count > 0);
 
 	if (rw_latch == RW_X_LATCH && mtr->modifications) {
-
+		mutex_enter(&buf_pool->mutex);
 		buf_flush_note_modification(block, mtr);
+		mutex_exit(&buf_pool->mutex);
 	}
 
+	mutex_enter(&block->mutex);
+
 #ifdef UNIV_SYNC_DEBUG
 	rw_lock_s_unlock(&(block->debug_latch));
 #endif
 	buf_fix_count = block->buf_fix_count;
 	block->buf_fix_count = buf_fix_count - 1;
 
-	mutex_exit(&(buf_pool->mutex));
+	mutex_exit(&block->mutex);
 
 	if (rw_latch == RW_S_LATCH) {
 		rw_lock_s_unlock(&(block->lock));

innobase/include/dict0crea.h

@@ -62,8 +62,10 @@ dict_truncate_index_tree(
 				/* out: new root page number, or
 				FIL_NULL on failure */
 	dict_table_t*	table,	/* in: the table the index belongs to */
-	rec_t*		rec,	/* in: record in the clustered index of
-				SYS_INDEXES table */
+	btr_pcur_t*	pcur,	/* in/out: persistent cursor pointing to
+				record in the clustered index of
+				SYS_INDEXES table. The cursor may be
+				repositioned in this call. */
 	mtr_t*		mtr);	/* in: mtr having the latch
 				on the record page. The mtr may be
 				committed and restarted in this call. */

innobase/include/sync0arr.h

@@ -75,17 +75,12 @@ sync_array_free_cell(
 	sync_array_t*	arr,	/* in: wait array */
         ulint    	index);  /* in: index of the cell in array */
 /**************************************************************************
-Looks for the cells in the wait array which refer
-to the wait object specified,
-and sets their corresponding events to the signaled state. In this
-way releases the threads waiting for the object to contend for the object.
-It is possible that no such cell is found, in which case does nothing. */
+Note that one of the wait objects was signalled. */
 
 void
-sync_array_signal_object(
-/*=====================*/
-	sync_array_t*	arr,	/* in: wait array */
-	void*		object);/* in: wait object */
+sync_array_object_signalled(
+/*========================*/
+	sync_array_t*	arr);	/* in: wait array */
 /**************************************************************************
 If the wakeup algorithm does not work perfectly at semaphore relases,
 this function will do the waking (see the comment in mutex_exit). This

innobase/include/sync0rw.h

@@ -411,6 +411,7 @@ blocked by readers, a writer may queue for the lock by setting the writer
 field. Then no new readers are allowed in. */
 
 struct rw_lock_struct {
+	os_event_t	event;	/* Used by sync0arr.c for thread queueing */
 	ulint	reader_count;	/* Number of readers who have locked this
 				lock in the shared mode */
 	ulint	writer; 	/* This field is set to RW_LOCK_EX if there

innobase/include/sync0rw.ic

@@ -382,7 +382,8 @@ rw_lock_s_unlock_func(
 	mutex_exit(mutex);
 
 	if (UNIV_UNLIKELY(sg)) {
-		sync_array_signal_object(sync_primary_wait_array, lock);
+		os_event_set(lock->event);
+		sync_array_object_signalled(sync_primary_wait_array);
 	}
 
         ut_ad(rw_lock_validate(lock));
@@ -462,7 +463,8 @@ rw_lock_x_unlock_func(
 	mutex_exit(&(lock->mutex));
 
 	if (UNIV_UNLIKELY(sg)) {
-		sync_array_signal_object(sync_primary_wait_array, lock);
+		os_event_set(lock->event);
+		sync_array_object_signalled(sync_primary_wait_array);
 	}
 
         ut_ad(rw_lock_validate(lock));

innobase/include/sync0sync.h

@@ -453,6 +453,7 @@ Do not use its fields directly! The structure used in the spin lock
 implementation of a mutual exclusion semaphore. */
 
 struct mutex_struct {
+	os_event_t	event;	/* Used by sync0arr.c for the wait queue */
 	ulint	lock_word;	/* This ulint is the target of the atomic
 				test-and-set instruction in Win32 */
 #if !defined(_WIN32) || !defined(UNIV_CAN_USE_X86_ASSEMBLER)

innobase/os/os0sync.c

@@ -21,6 +21,7 @@ Created 9/6/1995 Heikki Tuuri
 
 /* Type definition for an operating system mutex struct */
 struct os_mutex_struct{ 
+	os_event_t	event;	/* Used by sync0arr.c for queing threads */
 	void*		handle;	/* OS handle to mutex */
 	ulint		count;	/* we use this counter to check
 				that the same thread does not
@@ -35,6 +36,7 @@ struct os_mutex_struct{
 /* Mutex protecting counts and the lists of OS mutexes and events */
 os_mutex_t	os_sync_mutex;
 ibool		os_sync_mutex_inited	= FALSE;
+ibool		os_sync_free_called	= FALSE;
 
 /* This is incremented by 1 in os_thread_create and decremented by 1 in
 os_thread_exit */
@@ -50,6 +52,10 @@ ulint	os_event_count		= 0;
 ulint	os_mutex_count		= 0;
 ulint	os_fast_mutex_count	= 0;
 
+/* Because a mutex is embedded inside an event and there is an
+event embedded inside a mutex, on free, this generates a recursive call.
+This version of the free event function doesn't acquire the global lock */
+static void os_event_free_internal(os_event_t	event);
 
 /*************************************************************
 Initializes global event and OS 'slow' mutex lists. */
@@ -76,6 +82,7 @@ os_sync_free(void)
 	os_event_t	event;
 	os_mutex_t	mutex;
 
+	os_sync_free_called = TRUE;
 	event = UT_LIST_GET_FIRST(os_event_list);
 
 	while (event) {
@@ -99,6 +106,7 @@ os_sync_free(void)
 
 	      mutex = UT_LIST_GET_FIRST(os_mutex_list);
 	}
+	os_sync_free_called = FALSE;
 }
 
 /*************************************************************
@@ -146,14 +154,21 @@ os_event_create(
 	event->signal_count = 0;
 #endif /* __WIN__ */
 
-        /* Put to the list of events */
-	os_mutex_enter(os_sync_mutex);
+	/* The os_sync_mutex can be NULL because during startup an event
+	can be created [ because it's embedded in the mutex/rwlock ] before
+	this module has been initialized */
+	if (os_sync_mutex != NULL) {
+		os_mutex_enter(os_sync_mutex);
+	}
 
+	/* Put to the list of events */
 	UT_LIST_ADD_FIRST(os_event_list, os_event_list, event);
 
 	os_event_count++;
 
-	os_mutex_exit(os_sync_mutex);
+	if (os_sync_mutex != NULL) {
+		os_mutex_exit(os_sync_mutex);
+	}
 
 	return(event);
 }
@@ -255,6 +270,35 @@ os_event_reset(
 #endif
 }
 
+/**************************************************************
+Frees an event object, without acquiring the global lock. */
+static
+void
+os_event_free_internal(
+/*===================*/
+	os_event_t	event)	/* in: event to free */
+{
+#ifdef __WIN__
+	ut_a(event);
+
+	ut_a(CloseHandle(event->handle));
+#else
+	ut_a(event);
+
+	/* This is to avoid freeing the mutex twice */
+	os_fast_mutex_free(&(event->os_mutex));
+
+	ut_a(0 == pthread_cond_destroy(&(event->cond_var)));
+#endif
+	/* Remove from the list of events */
+
+	UT_LIST_REMOVE(os_event_list, os_event_list, event);
+
+	os_event_count--;
+
+	ut_free(event);
+}
+
 /**************************************************************
 Frees an event object. */
 
@@ -456,6 +500,7 @@ os_mutex_create(
 
 	mutex_str->handle = mutex;
 	mutex_str->count = 0;
+	mutex_str->event = os_event_create(NULL);
 
 	if (os_sync_mutex_inited) {
 		/* When creating os_sync_mutex itself we cannot reserve it */
@@ -532,6 +577,10 @@ os_mutex_free(
 {
 	ut_a(mutex);
 
+	if (!os_sync_free_called) {
+		os_event_free_internal(mutex->event);
+	}
+
 	if (os_sync_mutex_inited) {
 		os_mutex_enter(os_sync_mutex);
 	}

innobase/row/row0mysql.c

@@ -2920,12 +2920,10 @@ do not allow the TRUNCATE. We also reserve the data dictionary latch. */
 			goto next_rec;
 		}
 
-		btr_pcur_store_position(&pcur, &mtr);
+		/* This call may commit and restart mtr
+		and reposition pcur. */
+		root_page_no = dict_truncate_index_tree(table, &pcur, &mtr);
 
-		/* This call may commit and restart mtr. */
-		root_page_no = dict_truncate_index_tree(table, rec, &mtr);
-
-		btr_pcur_restore_position(BTR_MODIFY_LEAF, &pcur, &mtr);
 		rec = btr_pcur_get_rec(&pcur);
 
 		if (root_page_no != FIL_NULL) {

innobase/row/row0sel.c

@@ -1323,6 +1323,12 @@ rec_loop:
 						ULINT_UNDEFINED, &heap);
 
 		if (srv_locks_unsafe_for_binlog) {
+
+			if (page_rec_is_supremum(rec)) {
+
+				goto next_rec;
+			}
+
 			lock_type = LOCK_REC_NOT_GAP;
 		} else {
 			lock_type = LOCK_ORDINARY;

innobase/srv/srv0start.c

@@ -1142,10 +1142,7 @@ innobase_start_or_create_for_mysql(void)
 
 #if defined(__NETWARE__)
 
-/* Create less event semaphores because Win 98/ME had difficulty creating
-40000 event semaphores.
-Comment from Novell, Inc.: also, these just take a lot of memory on
-NetWare. */
+	/* Comment from Novell, Inc.: These take a lot of memory on NetWare.*/
         srv_max_n_threads = 1000;
 #else
         if (srv_pool_size >= 1000 * 1024) {

innobase/sync/sync0arr.c

@@ -62,9 +62,6 @@ struct sync_cell_struct {
 	ibool		waiting;	/* TRUE if the thread has already
 					called sync_array_event_wait
 					on this cell */
-	ibool		event_set;	/* TRUE if the event is set */
-        os_event_t 	event;   	/* operating system event
-                                        semaphore handle */
 	time_t		reservation_time;/* time when the thread reserved
 					the wait cell */
 };
@@ -218,10 +215,7 @@ sync_array_create(
         for (i = 0; i < n_cells; i++) {
 		cell = sync_array_get_nth_cell(arr, i);        	
                 cell->wait_object = NULL;
-
-                /* Create an operating system event semaphore with no name */
-                cell->event = os_event_create(NULL);
-		cell->event_set = FALSE; /* it is created in reset state */
+		cell->waiting = FALSE;
 	}
 
 	return(arr);
@@ -235,19 +229,12 @@ sync_array_free(
 /*============*/
 	sync_array_t*	arr)	/* in, own: sync wait array */
 {
-        ulint           i;
-        sync_cell_t*   	cell;
 	ulint		protection;
 
         ut_a(arr->n_reserved == 0);
         
 	sync_array_validate(arr);
         
-        for (i = 0; i < arr->n_cells; i++) {
-		cell = sync_array_get_nth_cell(arr, i);        	
-		os_event_free(cell->event);
-        }
-
 	protection = arr->protection;
 
         /* Release the mutex protecting the wait array complex */
@@ -292,28 +279,20 @@ sync_array_validate(
         sync_array_exit(arr);
 }
 
-/***********************************************************************
-Puts the cell event in set state. */
-static
-void
-sync_cell_event_set(
-/*================*/
-	sync_cell_t*	cell)	/* in: array cell */
-{
-	os_event_set(cell->event);
-	cell->event_set = TRUE;
-}		
-
 /***********************************************************************
 Puts the cell event in reset state. */
 static
 void
 sync_cell_event_reset(
 /*==================*/
-	sync_cell_t*	cell)	/* in: array cell */
+	ulint		type,	/* in: lock type mutex/rw_lock */
+	void*		object) /* in: the rw_lock/mutex object */
 {
-	os_event_reset(cell->event);
-	cell->event_set = FALSE;
+	if (type == SYNC_MUTEX) {
+		os_event_reset(((mutex_t *) object)->event);
+	} else {
+		os_event_reset(((rw_lock_t *) object)->event);
+	}
 }		
 
 /**********************************************************************
@@ -346,14 +325,7 @@ sync_array_reserve_cell(
 
                 if (cell->wait_object == NULL) {
 
-                        /* Make sure the event is reset */
-			if (cell->event_set) {	
-                        	sync_cell_event_reset(cell);
-			}
-
-			cell->reservation_time = time(NULL);
-			cell->thread = os_thread_get_curr_id();
-
+			cell->waiting = FALSE;
 			cell->wait_object = object;
 
 			if (type == SYNC_MUTEX) {
@@ -363,7 +335,6 @@ sync_array_reserve_cell(
 			}
 				
 			cell->request_type = type;
-			cell->waiting = FALSE;
 			
 			cell->file = file;
 			cell->line = line;
@@ -373,6 +344,13 @@ sync_array_reserve_cell(
 			*index = i;
 
 			sync_array_exit(arr);
+
+			/* Make sure the event is reset */
+			sync_cell_event_reset(type, object);
+
+			cell->reservation_time = time(NULL);
+
+			cell->thread = os_thread_get_curr_id();
                         
                         return;
                 }
@@ -408,7 +386,12 @@ sync_array_wait_event(
 	ut_a(!cell->waiting);
 	ut_ad(os_thread_get_curr_id() == cell->thread);
 
-       	event = cell->event;
+	if (cell->request_type == SYNC_MUTEX) {
+		event = ((mutex_t*) cell->wait_object)->event;
+	} else {
+		event = ((rw_lock_t*) cell->wait_object)->event;
+	}
+
        	cell->waiting = TRUE;
 
 #ifdef UNIV_SYNC_DEBUG
@@ -510,10 +493,6 @@ sync_array_cell_print(
         if (!cell->waiting) {
 		fputs("wait has ended\n", file);
 	}
-
-        if (cell->event_set) {
-		fputs("wait is ending\n", file);
-	}
 }
 
 #ifdef UNIV_SYNC_DEBUG
@@ -623,7 +602,7 @@ sync_array_detect_deadlock(
 	
 	depth++;
 	
-	if (cell->event_set || !cell->waiting) {
+	if (!cell->waiting) {
 
 		return(FALSE); /* No deadlock here */
 	}
@@ -802,6 +781,7 @@ sync_array_free_cell(
 
         ut_a(cell->wait_object != NULL);
 
+	cell->waiting = FALSE;
 	cell->wait_object =  NULL;
 
 	ut_a(arr->n_reserved > 0);
@@ -811,44 +791,17 @@ sync_array_free_cell(
 }
 
 /**************************************************************************
-Looks for the cells in the wait array which refer to the wait object
-specified, and sets their corresponding events to the signaled state. In this
-way releases the threads waiting for the object to contend for the object.
-It is possible that no such cell is found, in which case does nothing. */
+Increments the signalled count. */
 
 void
-sync_array_signal_object(
-/*=====================*/
-	sync_array_t*	arr,	/* in: wait array */
-	void*		object)	/* in: wait object */
+sync_array_object_signalled(
+/*========================*/
+	sync_array_t*	arr)	/* in: wait array */
 {
-        sync_cell_t*   	cell;
-        ulint           count;
-        ulint           i;
-
         sync_array_enter(arr);
 
 	arr->sg_count++;
 
-	i = 0;
-	count = 0;
-
-        while (count < arr->n_reserved) {
-
-        	cell = sync_array_get_nth_cell(arr, i);
-
-                if (cell->wait_object != NULL) {
-
-                        count++;
-                        if (cell->wait_object == object) {
-
-                        	sync_cell_event_set(cell);
-                        }
-                }
-
-                i++;
-        }
-
         sync_array_exit(arr);
 }
 
@@ -881,7 +834,17 @@ sync_arr_wake_threads_if_sema_free(void)
 
                         if (sync_arr_cell_can_wake_up(cell)) {
 
-                        	sync_cell_event_set(cell);
+				if (cell->request_type == SYNC_MUTEX) {
+					mutex_t*	mutex;
+
+					mutex = cell->wait_object;
+					os_event_set(mutex->event);
+				} else {
+					rw_lock_t*	lock;
+
+					lock = cell->wait_object;
+					os_event_set(lock->event);
+				}
                         }
                 }
 
@@ -911,7 +874,7 @@ sync_array_print_long_waits(void)
 
         	cell = sync_array_get_nth_cell(sync_primary_wait_array, i);
 
-                if (cell->wait_object != NULL
+		if (cell->wait_object != NULL && cell->waiting
 		    && difftime(time(NULL), cell->reservation_time) > 240) {
 			fputs("InnoDB: Warning: a long semaphore wait:\n",
 				stderr);
@@ -919,7 +882,7 @@ sync_array_print_long_waits(void)
 			noticed = TRUE;
                 }
 
-                if (cell->wait_object != NULL
+		if (cell->wait_object != NULL && cell->waiting
 		    && difftime(time(NULL), cell->reservation_time)
 		    > fatal_timeout) {
 			fatal = TRUE;

innobase/sync/sync0rw.c

@@ -128,6 +128,7 @@ rw_lock_create_func(
 	lock->last_x_file_name = "not yet reserved";
 	lock->last_s_line = 0;
 	lock->last_x_line = 0;
+	lock->event = os_event_create(NULL);
 
 	mutex_enter(&rw_lock_list_mutex);
 	
@@ -163,6 +164,7 @@ rw_lock_free(
 	mutex_free(rw_lock_get_mutex(lock));
 
 	mutex_enter(&rw_lock_list_mutex);
+	os_event_free(lock->event);
 
 	if (UT_LIST_GET_PREV(list, lock)) {
 		ut_a(UT_LIST_GET_PREV(list, lock)->magic_n == RW_LOCK_MAGIC_N);

innobase/sync/sync0sync.c

@@ -212,6 +212,7 @@ mutex_create_func(
 	os_fast_mutex_init(&(mutex->os_fast_mutex));
 	mutex->lock_word = 0;
 #endif
+	mutex->event = os_event_create(NULL);
 	mutex_set_waiters(mutex, 0);
 	mutex->magic_n = MUTEX_MAGIC_N;
 #ifdef UNIV_SYNC_DEBUG
@@ -288,6 +289,8 @@ mutex_free(
 		mutex_exit(&mutex_list_mutex);
 	}
 
+	os_event_free(mutex->event);
+
 #if !defined(_WIN32) || !defined(UNIV_CAN_USE_X86_ASSEMBLER) 
 	os_fast_mutex_free(&(mutex->os_fast_mutex));
 #endif
@@ -564,8 +567,8 @@ mutex_signal_object(
 
 	/* The memory order of resetting the waiters field and
 	signaling the object is important. See LEMMA 1 above. */
-
-	sync_array_signal_object(sync_primary_wait_array, mutex);
+	os_event_set(mutex->event);
+	sync_array_object_signalled(sync_primary_wait_array);
 }
 
 #ifdef UNIV_SYNC_DEBUG
@@ -1114,6 +1117,7 @@ sync_thread_add_level(
 		ut_a(sync_thread_levels_g(array, SYNC_PURGE_SYS));
 	} else if (level == SYNC_TREE_NODE) {
 		ut_a(sync_thread_levels_contain(array, SYNC_INDEX_TREE)
+		     || sync_thread_levels_contain(array, SYNC_DICT_OPERATION)
 		     || sync_thread_levels_g(array, SYNC_TREE_NODE - 1));
 	} else if (level == SYNC_TREE_NODE_FROM_HASH) {
 		ut_a(1);