Merge lgrimmer@build.mysql.com:/home/bk/mysql-4.0

into mysql.com:/space/my/mysql-4.0

innobase/com/com0shm.c

@@ -103,7 +103,8 @@ struct com_shm_endpoint_struct{
 				the area currently may contain a datagram;
 				NOTE: automatic event */
 	os_event_t	empty; 	/* this is in the signaled state if the area
-				currently may be empty; NOTE: automatic event */
+				currently may be empty; NOTE: automatic
+				event */
 	ip_mutex_hdl_t*	ip_mutex; /* handle to the interprocess mutex
 				protecting the shared memory */
 	UT_LIST_NODE_T(com_shm_endpoint_t) list; /* If the endpoint struct
@@ -793,16 +794,18 @@ com_shm_create_or_open(
 
 	ut_strcpy(buf + len, (char*)"_IBSHM_EV_NE"),
 
-	event_ne = os_event_create_auto(buf);
+	event_ne = os_event_create(buf);
 
 	ut_ad(event_ne);
 
 	ut_strcpy(buf + len, (char*)"_IBSHM_EV_EM"),
 
-	event_em = os_event_create_auto(buf);
+	event_em = os_event_create(buf);
 
 	ut_ad(event_em);
 
+	ut_a(0); /* event_ne and event_em should be auto events! */
+
 	com_shm_endpoint_set_shm(ep, shm);
 	com_shm_endpoint_set_map(ep, map);
 

innobase/include/os0sync.h

@@ -13,13 +13,26 @@ Created 9/6/1995 Heikki Tuuri
 #include "ut0lst.h"
 
 #ifdef __WIN__
+
 #define os_fast_mutex_t CRITICAL_SECTION
-typedef HANDLE          os_event_t;
+
+typedef HANDLE          os_native_event_t;
+
+typedef struct os_event_struct os_event_struct_t;
+typedef os_event_struct_t*     os_event_t;
+
+struct os_event_struct {
+	os_native_event_t		  handle;
+					/* Windows event */
+	UT_LIST_NODE_T(os_event_struct_t) os_event_list;
+					/* list of all created events */
+};
 #else
 typedef pthread_mutex_t	os_fast_mutex_t;
 
 typedef struct os_event_struct os_event_struct_t;
 typedef os_event_struct_t*     os_event_t;
+
 struct os_event_struct {
 	os_fast_mutex_t	os_mutex;	/* this mutex protects the next
 					fields */
@@ -39,16 +52,16 @@ typedef os_mutex_str_t*		os_mutex_t;
 
 #define OS_SYNC_TIME_EXCEEDED	1
 
-/* Mutex protecting the thread count and event and OS 'slow' mutex lists */
+/* Mutex protecting counts and the event and OS 'slow' mutex lists */
 extern os_mutex_t	os_sync_mutex;
 
 /* This is incremented by 1 in os_thread_create and decremented by 1 in
 os_thread_exit */
 extern ulint		os_thread_count;
 
-/* The following are approximate counters for debugging in Unix */
 extern ulint		os_event_count;
 extern ulint		os_mutex_count;
+extern ulint		os_fast_mutex_count;
 
 /*************************************************************
 Initializes global event and OS 'slow' mutex lists. */
@@ -57,15 +70,14 @@ void
 os_sync_init(void);
 /*==============*/
 /*************************************************************
-Frees created events (not in Windows) and OS 'slow' mutexes. */
+Frees created events and OS 'slow' mutexes. */
 
 void
 os_sync_free(void);
 /*==============*/
-/*************************************************************
-Creates an event semaphore, i.e., a semaphore which may
-just have two states: signaled and nonsignaled.
-The created event is manual reset: it must be reset
+/************************************************************* 
+Creates an event semaphore, i.e., a semaphore which may just have two states:
+signaled and nonsignaled. The created event is manual reset: it must be reset
 explicitly by calling sync_os_reset_event. */
 
 os_event_t
@@ -74,10 +86,10 @@ os_event_create(
 			/* out: the event handle */
 	char*	name);	/* in: the name of the event, if NULL
 			the event is created without a name */
+#ifdef __WIN__
 /*************************************************************
-Creates an auto-reset event semaphore, i.e., an event
-which is automatically reset when a single thread is
-released. */
+Creates an auto-reset event semaphore, i.e., an event which is automatically
+reset when a single thread is released. Works only in Windows. */
 
 os_event_t
 os_event_create_auto(
@@ -85,6 +97,7 @@ os_event_create_auto(
 			/* out: the event handle */
 	char*	name);	/* in: the name of the event, if NULL
 			the event is created without a name */
+#endif
 /**************************************************************
 Sets an event semaphore to the signaled state: lets waiting threads
 proceed. */
@@ -120,7 +133,7 @@ os_event_wait(
 	os_event_t	event);	/* in: event to wait */
 /**************************************************************
 Waits for an event object until it is in the signaled state or
-a timeout is exceeded. */
+a timeout is exceeded. In Unix the timeout is always infinite. */
 
 ulint
 os_event_wait_time(
@@ -131,8 +144,9 @@ os_event_wait_time(
 	os_event_t	event,	/* in: event to wait */
 	ulint		time);	/* in: timeout in microseconds, or
 				OS_SYNC_INFINITE_TIME */
+#ifdef __WIN__
 /**************************************************************
-Waits for any event in an event array. Returns if even a single
+Waits for any event in an OS native event array. Returns if even a single
 one is signaled or becomes signaled. */
 
 ulint
@@ -140,14 +154,15 @@ os_event_wait_multiple(
 /*===================*/
 					/* out: index of the event
 					which was signaled */
-	ulint		n,		/* in: number of events in the
+	ulint		        n,	/* in: number of events in the
 					array */
-	os_event_t* 	event_array);	/* in: pointer to an array of event
+	os_native_event_t* 	native_event_array);
+					/* in: pointer to an array of event
 					handles */
+#endif
 /*************************************************************
-Creates an operating system mutex semaphore.
-Because these are slow, the mutex semaphore of the database
-itself (sync_mutex_t) should be used where possible. */
+Creates an operating system mutex semaphore. Because these are slow, the
+mutex semaphore of InnoDB itself (mutex_t) should be used where possible. */
 
 os_mutex_t
 os_mutex_create(

innobase/include/os0sync.ic

@@ -44,4 +44,3 @@ os_fast_mutex_trylock(
 #endif
 #endif
 }
-

innobase/include/os0thread.h

@@ -65,7 +65,9 @@ os_thread_pf(
 /********************************************************************
 Creates a new thread of execution. The execution starts from
 the function given. The start function takes a void* parameter
-and returns a ulint. */
+and returns a ulint.
+NOTE: We count the number of threads in os_thread_exit(). A created
+thread should always use that to exit and not use return() to exit. */
 
 os_thread_t
 os_thread_create(

innobase/os/os0file.c

@@ -80,6 +80,8 @@ struct os_aio_slot_struct{
 					which pending aio operation was
 					completed */
 #ifdef WIN_ASYNC_IO
+        os_event_t	event;		/* event object we need in the
+					OVERLAPPED struct */
 	OVERLAPPED	control;	/* Windows control block for the
 					aio request */
 #elif defined(POSIX_ASYNC_IO)
@@ -107,11 +109,14 @@ struct os_aio_array_struct{
 	ulint		n_reserved;/* Number of reserved slots in the
 				  aio array outside the ibuf segment */
 	os_aio_slot_t* 	slots;	  /* Pointer to the slots in the array */
-	os_event_t*	events;	  /* Pointer to an array of event handles
-				  where we copied the handles from slots,
-				  in the same order. This can be used in
-				  WaitForMultipleObjects; used only in
+#ifdef __WIN__
+	os_native_event_t* native_events;	 
+				  /* Pointer to an array of OS native event
+				  handles where we copied the handles from
+				  slots, in the same order. This can be used
+				  in WaitForMultipleObjects; used only in
 				  Windows */
+#endif
 };
 
 /* Array of events used in simulated aio */
@@ -296,7 +301,7 @@ os_file_handle_error(
 				operation */
 	os_file_t	file,	/* in: file pointer */
 	char*		name,	/* in: name of a file or NULL */
-	const char*	operation) /* in: type of operation */
+	const char*	operation)/* in: operation */
 {
 	ulint	err;
 
@@ -338,8 +343,8 @@ os_file_handle_error(
 	        if (name) {
 	                fprintf(stderr, "InnoDB: File name %s\n", name);
 	        }
-		fprintf(stderr, "InnoDB: system call %s\n", operation);
-
+	  
+		fprintf(stderr, "InnoDB: System call %s.\n", operation);
 		fprintf(stderr, "InnoDB: Cannot continue operation.\n");
 
 		fflush(stderr);
@@ -422,9 +427,8 @@ try_again:
 		*success = FALSE;
 
 		retry = os_file_handle_error(file, name,
-					     create_mode == OS_FILE_OPEN ?
-					     "open" : "create");
-
+				create_mode == OS_FILE_OPEN ?
+				"open" : "create");
 		if (retry) {
 			goto try_again;
 		}
@@ -465,10 +469,8 @@ try_again:
 		*success = FALSE;
 
 		retry = os_file_handle_error(file, name,
-					     create_mode == OS_FILE_OPEN ?
-					     "open" : "create");
-
-
+				create_mode == OS_FILE_OPEN ?
+				"open" : "create");
 		if (retry) {
 			goto try_again;
 		}
@@ -576,9 +578,8 @@ try_again:
 		*success = FALSE;
 
 		retry = os_file_handle_error(file, name,
-					     create_mode == OS_FILE_OPEN ?
-					     "open" : "create");
-
+				create_mode == OS_FILE_OPEN ?
+				"open" : "create");
 		if (retry) {
 			goto try_again;
 		}
@@ -625,9 +626,8 @@ try_again:
 		*success = FALSE;
 
 		retry = os_file_handle_error(file, name,
-					     create_mode == OS_FILE_OPEN ?
-					     "open" : "create");
-
+				create_mode == OS_FILE_OPEN ?
+				"open" : "create");
 		if (retry) {
 			goto try_again;
 		}
@@ -1319,19 +1319,22 @@ os_aio_array_create(
 	array->n_segments	= n_segments;
 	array->n_reserved	= 0;
 	array->slots		= ut_malloc(n * sizeof(os_aio_slot_t));
-	array->events		= ut_malloc(n * sizeof(os_event_t));
-	
+#ifdef __WIN__
+	array->native_events	= ut_malloc(n * sizeof(os_native_event_t));
+#endif	
 	for (i = 0; i < n; i++) {
 		slot = os_aio_array_get_nth_slot(array, i);
 
 		slot->pos = i;
 		slot->reserved = FALSE;
 #ifdef WIN_ASYNC_IO
+		slot->event = os_event_create(NULL);
+
 		over = &(slot->control);
 
-		over->hEvent = os_event_create(NULL);
+		over->hEvent = slot->event->handle;
 
-		*((array->events) + i) = over->hEvent;
+		*((array->native_events) + i) = over->hEvent;
 #endif
 	}
 	
@@ -1429,7 +1432,7 @@ os_aio_array_wake_win_aio_at_shutdown(
 
 	for (i = 0; i < array->n_slots; i++) {
 
-	        os_event_set(*(array->events + i));
+	        os_event_set((array->slots + i)->event);
 	}
 }
 #endif
@@ -1689,7 +1692,7 @@ loop:
 	control = &(slot->control);
 	control->Offset = (DWORD)offset;
 	control->OffsetHigh = (DWORD)offset_high;
-	os_event_reset(control->hEvent);
+	os_event_reset(slot->event);
 
 #elif defined(POSIX_ASYNC_IO)
 
@@ -1747,7 +1750,7 @@ os_aio_array_free_slot(
 	}
 
 #ifdef WIN_ASYNC_IO		
-	os_event_reset(slot->control.hEvent);
+	os_event_reset(slot->event);
 #endif
 	os_mutex_exit(array->mutex);
 }
@@ -1872,7 +1875,7 @@ os_aio(
 				offset where to read or write */
 	ulint		offset_high, /* in: most significant 32 bits of
 				offset */
-	ulint		n,	/* in: number of bytes to read or write */	
+	ulint		n,	/* in: number of bytes to read or write */
 	void*		message1,/* in: messages for the aio handler (these
 				can be used to identify a completed aio
 				operation); if mode is OS_AIO_SYNC, these
@@ -1916,7 +1919,8 @@ os_aio(
 		wait in the Windows case. */
 
 		if (type == OS_FILE_READ) {
-			return(os_file_read(file, buf, offset, offset_high, n));
+			return(os_file_read(file, buf, offset,
+							offset_high, n));
 		}
 
 		ut_a(type == OS_FILE_WRITE);
@@ -1994,8 +1998,7 @@ try_again:
 #ifdef WIN_ASYNC_IO
 	if (os_aio_use_native_aio) {
 		if ((ret && len == n)
-			|| (!ret && GetLastError() == ERROR_IO_PENDING)) {	
-
+		    || (!ret && GetLastError() == ERROR_IO_PENDING)) {
 			/* aio was queued successfully! */
 		
 	    		if (mode == OS_AIO_SYNC) {
@@ -2025,8 +2028,8 @@ try_again:
 
 	os_aio_array_free_slot(array, slot);
 
-	retry = os_file_handle_error(file, name, "aio");
-
+	retry = os_file_handle_error(file, name,
+			type == OS_FILE_READ ? "aio read" : "aio write");
 	if (retry) {
 
 		goto try_again;
@@ -2091,15 +2094,15 @@ os_aio_windows_handle(
 	n = array->n_slots / array->n_segments;
 
 	if (array == os_aio_sync_array) {
-		srv_io_thread_op_info[orig_seg] = "wait Windows aio for 1 page";
-
-		ut_ad(pos < array->n_slots); 
-		os_event_wait(array->events[pos]);
+		srv_io_thread_op_info[orig_seg] =
+						"wait Windows aio for 1 page";
+		os_event_wait(os_aio_array_get_nth_slot(array, pos)->event);
 		i = pos;
 	} else {
 		srv_io_thread_op_info[orig_seg] =
 						"wait Windows aio";
-		i = os_event_wait_multiple(n, (array->events) + segment * n);
+		i = os_event_wait_multiple(n,
+				(array->native_events) + segment * n);
 	}
 
 	os_mutex_enter(array->mutex);
@@ -2124,7 +2127,7 @@ os_aio_windows_handle(
 		         ut_a(TRUE == os_file_flush(slot->file));
 		}
 	} else {
-		os_file_handle_error(slot->file, slot->name, "aio");
+		os_file_handle_error(slot->file, slot->name, "Windows aio");
 		
 		ret_val = FALSE;
 	}		  

innobase/os/os0sync.c

@@ -32,24 +32,23 @@ struct os_mutex_struct{
 				/* list of all 'slow' OS mutexes created */
 };
 
-/* Mutex protecting the thread count and the lists of OS mutexes
-and events */
+/* Mutex protecting counts and the lists of OS mutexes and events */
 os_mutex_t	os_sync_mutex;
 ibool		os_sync_mutex_inited	= FALSE;
 
 /* This is incremented by 1 in os_thread_create and decremented by 1 in
 os_thread_exit */
-ulint		os_thread_count		= 0;
+ulint	os_thread_count		= 0;
 
-/* The list of all events created (not in Windows) */
+/* The list of all events created */
 UT_LIST_BASE_NODE_T(os_event_struct_t)	os_event_list;
 
 /* The list of all OS 'slow' mutexes */
 UT_LIST_BASE_NODE_T(os_mutex_str_t)	os_mutex_list;
 
-/* The following are approximate counters for debugging in Unix */
-ulint		os_event_count		= 0;
-ulint		os_mutex_count		= 0;
+ulint	os_event_count		= 0;
+ulint	os_mutex_count		= 0;
+ulint	os_fast_mutex_count	= 0;
 
 
 /*************************************************************
@@ -68,7 +67,7 @@ os_sync_init(void)
 }
 
 /*************************************************************
-Frees created events (not in Windows) and OS 'slow' mutexes. */
+Frees created events and OS 'slow' mutexes. */
 
 void
 os_sync_free(void)
@@ -89,6 +88,12 @@ os_sync_free(void)
 	mutex = UT_LIST_GET_FIRST(os_mutex_list);
 
 	while (mutex) {
+	      if (mutex == os_sync_mutex) {
+		      /* Set the flag to FALSE so that we do not try to
+		      reserve os_sync_mutex any more in remaining freeing
+		      operations in shutdown */
+		      os_sync_mutex_inited = FALSE;
+	      }
 
 	      os_mutex_free(mutex);
 
@@ -97,10 +102,9 @@ os_sync_free(void)
 }
 
 /*************************************************************
-Creates an event semaphore, i.e., a semaphore which may
-just have two states: signaled and nonsignaled.
-The created event is manual reset: it must be reset
-explicitly by calling sync_os_reset_event. */
+Creates an event semaphore, i.e., a semaphore which may just have two
+states: signaled and nonsignaled. The created event is manual reset: it
+must be reset explicitly by calling sync_os_reset_event. */
 
 os_event_t
 os_event_create(
@@ -112,20 +116,18 @@ os_event_create(
 #ifdef __WIN__
         os_event_t event;
 
-	event = CreateEvent(NULL,	/* No security attributes */
+	event = ut_malloc(sizeof(struct os_event_struct));
+
+	event->handle = CreateEvent(NULL,/* No security attributes */
 			TRUE,		/* Manual reset */
 			FALSE,		/* Initial state nonsignaled */
 			name);
-	if (!event) {
+	if (!event->handle) {
 	        fprintf(stderr,
 "InnoDB: Could not create a Windows event semaphore; Windows error %lu\n",
 		  (ulint)GetLastError());
 	}
-
-	ut_a(event);
-
-	return(event);
-#else
+#else /* Unix */
 	os_event_t	event;
 
 	UT_NOT_USED(name);
@@ -141,7 +143,9 @@ os_event_create(
 	ut_a(0 == pthread_cond_init(&(event->cond_var), NULL));
 #endif
 	event->is_set = FALSE;
+#endif /* __WIN__ */
 
+        /* Put to the list of events */
 	os_mutex_enter(os_sync_mutex);
 
 	UT_LIST_ADD_FIRST(os_event_list, os_event_list, event);
@@ -151,13 +155,12 @@ os_event_create(
 	os_mutex_exit(os_sync_mutex);
 
 	return(event);
-#endif
 }
 
+#ifdef __WIN__
 /*************************************************************
-Creates an auto-reset event semaphore, i.e., an event
-which is automatically reset when a single thread is
-released. */
+Creates an auto-reset event semaphore, i.e., an event which is automatically
+reset when a single thread is released. Works only in Windows. */
 
 os_event_t
 os_event_create_auto(
@@ -166,26 +169,33 @@ os_event_create_auto(
 	char*	name)	/* in: the name of the event, if NULL
 			the event is created without a name */
 {
-#ifdef __WIN__
         os_event_t event;
 
-	event = CreateEvent(NULL,	/* No security attributes */
+	event = ut_malloc(sizeof(struct os_event_struct));
+
+	event->handle = CreateEvent(NULL,/* No security attributes */
 			FALSE,		/* Auto-reset */
 			FALSE,		/* Initial state nonsignaled */
 			name);
-	ut_a(event);
 
-	return(event);
-#else
-	/* Does nothing in Posix because we do not need this with MySQL  */
+	if (!event->handle) {
+	        fprintf(stderr,
+"InnoDB: Could not create a Windows auto event semaphore; Windows error %lu\n",
+		  (ulint)GetLastError());
+	}
 
-	UT_NOT_USED(name);
+        /* Put to the list of events */
+	os_mutex_enter(os_sync_mutex);
 
-	ut_a(0);
+	UT_LIST_ADD_FIRST(os_event_list, os_event_list, event);
 
-	return(NULL);
-#endif
+	os_event_count++;
+
+	os_mutex_exit(os_sync_mutex);
+
+	return(event);
 }
+#endif
 
 /**************************************************************
 Sets an event semaphore to the signaled state: lets waiting threads
@@ -198,7 +208,7 @@ os_event_set(
 {
 #ifdef __WIN__	
 	ut_a(event);
-	ut_a(SetEvent(event));
+	ut_a(SetEvent(event->handle));
 #else
 	ut_a(event);
 
@@ -227,7 +237,7 @@ os_event_reset(
 #ifdef __WIN__
 	ut_a(event);
 
-	ut_a(ResetEvent(event));
+	ut_a(ResetEvent(event->handle));
 #else
 	ut_a(event);
 
@@ -255,12 +265,14 @@ os_event_free(
 #ifdef __WIN__
 	ut_a(event);
 
-	ut_a(CloseHandle(event));
+	ut_a(CloseHandle(event->handle));
 #else
 	ut_a(event);
 
 	os_fast_mutex_free(&(event->os_mutex));
 	ut_a(0 == pthread_cond_destroy(&(event->cond_var)));
+#endif
+        /* Remove from the list of events */
 
 	os_mutex_enter(os_sync_mutex);
 
@@ -271,7 +283,6 @@ os_event_free(
 	os_mutex_exit(os_sync_mutex);
 
 	ut_free(event);
-#endif
 }
 
 /**************************************************************
@@ -291,7 +302,7 @@ os_event_wait(
 	ut_a(event);
 
 	/* Specify an infinite time limit for waiting */
-	err = WaitForSingleObject(event, INFINITE);
+	err = WaitForSingleObject(event->handle, INFINITE);
 
 	ut_a(err == WAIT_OBJECT_0);
 
@@ -324,7 +335,7 @@ loop:
 
 /**************************************************************
 Waits for an event object until it is in the signaled state or
-a timeout is exceeded. */
+a timeout is exceeded. In Unix the timeout is always infinite. */
 
 ulint
 os_event_wait_time(
@@ -341,9 +352,9 @@ os_event_wait_time(
 	ut_a(event);
 
 	if (time != OS_SYNC_INFINITE_TIME) {
-		err = WaitForSingleObject(event, time / 1000);
+		err = WaitForSingleObject(event->handle, time / 1000);
 	} else {
-		err = WaitForSingleObject(event, INFINITE);
+		err = WaitForSingleObject(event->handle, INFINITE);
 	}
 	
 	if (err == WAIT_OBJECT_0) {
@@ -367,8 +378,9 @@ os_event_wait_time(
 #endif
 }
 
+#ifdef __WIN__
 /**************************************************************
-Waits for any event in an event array. Returns if even a single
+Waits for any event in an OS native event array. Returns if even a single
 one is signaled or becomes signaled. */
 
 ulint
@@ -376,18 +388,18 @@ os_event_wait_multiple(
 /*===================*/
 					/* out: index of the event
 					which was signaled */
-	ulint		n,		/* in: number of events in the
+	ulint			n,	/* in: number of events in the
 					array */
-	os_event_t*	event_array)	/* in: pointer to an array of event
+	os_native_event_t*	native_event_array)
+					/* in: pointer to an array of event
 					handles */
 {
-#ifdef __WIN__
 	DWORD	index;
 
-	ut_a(event_array);
+	ut_a(native_event_array);
 	ut_a(n > 0);
 
-	index = WaitForMultipleObjects(n, event_array,
+	index = WaitForMultipleObjects(n, native_event_array,
 					FALSE,	   /* Wait for any 1 event */
 					INFINITE); /* Infinite wait time
 						   limit */
@@ -399,21 +411,12 @@ os_event_wait_multiple(
 	}
 
 	return(index - WAIT_OBJECT_0);
-#else
-	ut_a(n == 0);
-	
-	/* In Posix we can only wait for a single event */
-
-	os_event_wait(*event_array);
-
-	return(0);
-#endif
 }
+#endif
 
 /*************************************************************
-Creates an operating system mutex semaphore.
-Because these are slow, the mutex semaphore of the database
-itself (sync_mutex_t) should be used where possible. */
+Creates an operating system mutex semaphore. Because these are slow, the
+mutex semaphore of InnoDB itself (mutex_t) should be used where possible. */
 
 os_mutex_t
 os_mutex_create(
@@ -430,50 +433,35 @@ os_mutex_create(
 			FALSE,		/* Initial state: no owner */
 			name);
 	ut_a(mutex);
-
-	mutex_str = ut_malloc(sizeof(os_mutex_str_t));
-
-	mutex_str->handle = mutex;
-	mutex_str->count = 0;
-
-	if (os_sync_mutex_inited) {
-	        os_mutex_enter(os_sync_mutex);
-	}
-
-	UT_LIST_ADD_FIRST(os_mutex_list, os_mutex_list, mutex_str);
-
-	if (os_sync_mutex_inited) {
-		os_mutex_exit(os_sync_mutex);
-	}
-
-	return(mutex_str);
 #else
-	os_fast_mutex_t*	os_mutex;
+	os_fast_mutex_t*	mutex;
 	os_mutex_t		mutex_str;
 
 	UT_NOT_USED(name);
 	
-	os_mutex = ut_malloc(sizeof(os_fast_mutex_t));
-
-	os_fast_mutex_init(os_mutex);
+	mutex = ut_malloc(sizeof(os_fast_mutex_t));
 
+	os_fast_mutex_init(mutex);
+#endif	
 	mutex_str = ut_malloc(sizeof(os_mutex_str_t));
 
-	mutex_str->handle = os_mutex;
+	mutex_str->handle = mutex;
 	mutex_str->count = 0;
 
 	if (os_sync_mutex_inited) {
+		/* When creating os_sync_mutex itself we cannot reserve it */
 	        os_mutex_enter(os_sync_mutex);
 	}
 
 	UT_LIST_ADD_FIRST(os_mutex_list, os_mutex_list, mutex_str);
 
+	os_mutex_count++;
+
 	if (os_sync_mutex_inited) {
 		os_mutex_exit(os_sync_mutex);
 	}
 
 	return(mutex_str);
-#endif	
 }
 
 /**************************************************************
@@ -513,21 +501,14 @@ os_mutex_exit(
 /*==========*/
 	os_mutex_t	mutex)	/* in: mutex to release */
 {
-#ifdef __WIN__
 	ut_a(mutex);
 
 	ut_a(mutex->count == 1);
 
 	(mutex->count)--;
-
+#ifdef __WIN__
 	ut_a(ReleaseMutex(mutex->handle));
 #else
-	ut_a(mutex);
-
-	ut_a(mutex->count == 1);
-
-	(mutex->count)--;
-
 	os_fast_mutex_unlock(mutex->handle);
 #endif	
 }
@@ -540,25 +521,25 @@ os_mutex_free(
 /*==========*/
 	os_mutex_t	mutex)	/* in: mutex to free */
 {
-#ifdef __WIN__
 	ut_a(mutex);
 
-	os_mutex_enter(os_sync_mutex);
+	if (os_sync_mutex_inited) {
+		os_mutex_enter(os_sync_mutex);
+	}
 
 	UT_LIST_REMOVE(os_mutex_list, os_mutex_list, mutex);
+	
+	os_mutex_count--;
 
-	os_mutex_exit(os_sync_mutex);
+	if (os_sync_mutex_inited) {
+		os_mutex_exit(os_sync_mutex);
+	}
 
+#ifdef __WIN__
 	ut_a(CloseHandle(mutex->handle));
 
 	ut_free(mutex);
 #else
-	os_mutex_enter(os_sync_mutex);
-
-	UT_LIST_REMOVE(os_mutex_list, os_mutex_list, mutex);
-
-	os_mutex_exit(os_sync_mutex);
-
 	os_fast_mutex_free(mutex->handle);
 	ut_free(mutex->handle);
 	ut_free(mutex);
@@ -583,8 +564,19 @@ os_fast_mutex_init(
 #else
 	ut_a(0 == pthread_mutex_init(fast_mutex, MY_MUTEX_INIT_FAST));
 #endif
-	os_mutex_count++;
 #endif
+	if (os_sync_mutex_inited) {
+		/* When creating os_sync_mutex itself (in Unix) we cannot
+		reserve it */
+
+		os_mutex_enter(os_sync_mutex);
+	}
+
+	os_fast_mutex_count++;
+
+	if (os_sync_mutex_inited) {
+	        os_mutex_exit(os_sync_mutex);
+	}
 }
 
 /**************************************************************
@@ -631,6 +623,17 @@ os_fast_mutex_free(
 	DeleteCriticalSection((LPCRITICAL_SECTION) fast_mutex);
 #else
 	ut_a(0 == pthread_mutex_destroy(fast_mutex));
-	os_mutex_count--;
 #endif
+	if (os_sync_mutex_inited) {
+		/* When freeing the last mutexes, we have
+		already freed os_sync_mutex */
+
+		os_mutex_enter(os_sync_mutex);
+	}
+
+	os_fast_mutex_count--;
+
+	if (os_sync_mutex_inited) {
+		os_mutex_exit(os_sync_mutex);
+	}
 }

innobase/os/os0thread.c

@@ -186,6 +186,10 @@ os_thread_exit(
 	void*	exit_value)	/* in: exit value; in Windows this void*
 				is cast as a DWORD */
 {
+#ifdef UNIV_DEBUG_THREAD_CREATION
+        printf("A thread exits.\n");
+        printf("Thread id %lu\n", os_thread_pf(os_thread_get_curr_id()));
+#endif
 	os_mutex_enter(os_sync_mutex);
 	os_thread_count--;
 	os_mutex_exit(os_sync_mutex);

innobase/srv/srv0srv.c

@@ -856,6 +856,7 @@ srv_release_max_if_no_queries(void)
 	mutex_exit(&kernel_mutex);
 }
 
+#ifdef notdefined
 /***********************************************************************
 Releases one utility thread if no queries are active and
 the high-water mark 2 for the utility is exceeded. */
@@ -890,7 +891,6 @@ srv_release_one_if_no_queries(void)
 	mutex_exit(&kernel_mutex);
 }
 
-#ifdef notdefined
 /***********************************************************************
 Decrements the utility meter by the value given and suspends the calling
 thread, which must be an utility thread of the type given, if necessary. */
@@ -1000,6 +1000,8 @@ srv_communication_init(
 	
 	ut_a(ret == 0);
 }
+
+#ifdef notdefined
 	
 /*************************************************************************
 Implements the recovery utility. */
@@ -1060,6 +1062,7 @@ srv_purge_thread(
 
 	return(0);
 }
+#endif /* notdefined */
 
 /*************************************************************************
 Creates the utility threads. */
@@ -1090,6 +1093,7 @@ srv_create_utility_threads(void)
 	ut_a(thread); */
 }
 
+#ifdef notdefined
 /*************************************************************************
 Implements the communication threads. */
 static
@@ -1139,6 +1143,7 @@ srv_com_thread(
 
 	return(0);
 }
+#endif
 
 /*************************************************************************
 Creates the communication threads. */
@@ -1159,6 +1164,7 @@ srv_create_com_threads(void)
 	}
 }
 
+#ifdef notdefined
 /*************************************************************************
 Implements the worker threads. */
 static
@@ -1203,6 +1209,7 @@ srv_worker_thread(
 
 	return(0);
 }
+#endif
 
 /*************************************************************************
 Creates the worker threads. */
@@ -2456,6 +2463,10 @@ srv_lock_timeout_and_monitor_thread(
 	char*		buf;
 	ulint		i;
 
+#ifdef UNIV_DEBUG_THREAD_CREATION
+	printf("Lock timeout thread starts\n");
+	printf("Thread id %lu\n", os_thread_pf(os_thread_get_curr_id()));
+#endif
 	UT_NOT_USED(arg);
 	srv_last_monitor_time = time(NULL);
 	last_table_monitor_time = time(NULL);
@@ -2596,6 +2607,10 @@ loop:
 exit_func:
 	srv_lock_timeout_and_monitor_active = FALSE;
 
+	/* We count the number of threads in os_thread_exit(). A created
+	thread should always use that to exit and not use return() to exit. */
+
+	os_thread_exit(NULL);
 #ifndef __WIN__
         return(NULL);
 #else
@@ -2621,6 +2636,10 @@ srv_error_monitor_thread(
 	ulint	cnt	= 0;
 
 	UT_NOT_USED(arg);
+#ifdef UNIV_DEBUG_THREAD_CREATION
+	printf("Error monitor thread starts\n");
+	printf("Thread id %lu\n", os_thread_pf(os_thread_get_curr_id()));
+#endif
 loop:
 	srv_error_monitor_active = TRUE;
 
@@ -2657,6 +2676,9 @@ loop:
 
 	srv_error_monitor_active = FALSE;
 
+	/* We count the number of threads in os_thread_exit(). A created
+	thread should always use that to exit and not use return() to exit. */
+
 	os_thread_exit(NULL);
 
 #ifndef __WIN__
@@ -2737,6 +2759,10 @@ srv_master_thread(
 	
 	UT_NOT_USED(arg);
 
+#ifdef UNIV_DEBUG_THREAD_CREATION
+	printf("Master thread starts\n");
+	printf("Thread id %lu\n", os_thread_pf(os_thread_get_curr_id()));
+#endif
 	srv_main_thread_process_no = os_proc_get_number();
 	srv_main_thread_id = os_thread_pf(os_thread_get_curr_id());
 	
@@ -2972,6 +2998,15 @@ background_loop:
 
 	n_tables_to_drop = row_drop_tables_for_mysql_in_background();
 
+	if (n_tables_to_drop > 0) {
+	        /* Do not monopolize the CPU even if there are tables waiting
+		in the background drop queue. (It is essentially a bug if
+		MySQL tries to drop a table while there are still open handles
+		to it and we had to put it to the background drop queue.) */
+
+		os_thread_sleep(100000);
+	}
+ 
 	srv_main_thread_op_info = (char*)"purging";
 
 	if (srv_fast_shutdown && srv_shutdown_state > 0) {
@@ -3110,6 +3145,13 @@ suspend_thread:
 
 	goto loop;
 
+	/* We count the number of threads in os_thread_exit(). A created
+	thread should always use that to exit and not use return() to exit.
+	The thread actually never comes here because it is exited in an
+	os_event_wait(). */
+	
+	os_thread_exit(NULL);
+
 #ifndef __WIN__
         return(NULL);
 #else

innobase/srv/srv0start.c

@@ -414,8 +414,10 @@ io_handler_thread(
 	
 	segment = *((ulint*)arg);
 
-/*	printf("Io handler thread %lu starts\n", segment); */
-
+#ifdef UNIV_DEBUG_THREAD_CREATION
+	printf("Io handler thread %lu starts\n", segment);
+	printf("Thread id %lu\n", os_thread_pf(os_thread_get_curr_id()));
+#endif
 	for (i = 0;; i++) {
 		fil_aio_wait(segment);
 
@@ -424,6 +426,13 @@ io_handler_thread(
 		mutex_exit(&ios_mutex);
 	}
 
+	/* We count the number of threads in os_thread_exit(). A created
+	thread should always use that to exit and not use return() to exit.
+	The thread actually never comes here because it is exited in an
+	os_event_wait(). */
+
+	os_thread_exit(NULL);
+
 #ifndef __WIN__
 	return(NULL);
 #else
@@ -1546,21 +1555,32 @@ innobase_shutdown_for_mysql(void)
 		      os_thread_count);
 	}
 
-	/* 3. Free all InnoDB's own mutexes */
+	/* 3. Free all InnoDB's own mutexes and the os_fast_mutexes inside
+	them */
 
 	sync_close();
 
-	/* 4. Free all OS synchronization primitives (in Windows currently
-	events are not freed) */
+	/* 4. Free the os_conc_mutex and all os_events and os_mutexes */
 
 	srv_free();
 	os_sync_free();
 
-	/* 5. Free all allocated memory (and the os_fast_mutex created in
+	/* 5. Free all allocated memory and the os_fast_mutex created in
 	ut0mem.c */
 
         ut_free_all_mem();
 
+	if (os_thread_count != 0
+	    || os_event_count != 0
+	    || os_mutex_count != 0
+	    || os_fast_mutex_count != 0) {
+	        fprintf(stderr,
+"InnoDB: Warning: some resources were not cleaned up in shutdown:\n"
+"InnoDB: threads %lu, events %lu, os_mutexes %lu, os_fast_mutexes %lu\n",
+		      os_thread_count, os_event_count, os_mutex_count,
+		      os_fast_mutex_count);
+	}
+
 	if (srv_print_verbose_log) {
 	        ut_print_timestamp(stderr);
 	        fprintf(stderr, "  InnoDB: Shutdown completed\n");

sql/slave.cc

@@ -2081,8 +2081,13 @@ static int exec_relay_log_event(THD* thd, RELAY_LOG_INFO* rli)
   else
   {
     sql_print_error("\
-Could not parse log event entry, check the master for binlog corruption\n\
-This may also be a network problem, or just a bug in the master or slave code.\
+Could not parse relay log event entry. The possible reasons are: the master's \
+binary log is corrupted (you can check this by running 'mysqlbinlog' on the \
+binary log), the slave's relay log is corrupted (you can check this by running \
+'mysqlbinlog' on the relay log), a network problem, or a bug in the master's \
+or slave's MySQL code. If you want to check the master's binary log or slave's \
+relay log, you will be able to know their names by issuing 'SHOW SLAVE STATUS' \
+on this slave.\
 ");
     return 1;
   }