sync0sync.c 39.6 KB
Newer Older
1 2 3 4 5 6 7 8 9 10 11 12 13 14 15 16 17 18 19 20 21 22 23 24 25 26 27 28 29 30 31 32 33 34 35 36 37 38 39 40 41 42 43 44 45 46 47 48 49 50 51 52 53 54 55 56 57 58 59 60 61 62 63 64 65 66 67 68 69 70 71 72 73 74 75 76 77 78 79 80 81 82 83 84 85 86 87 88 89 90 91 92 93 94 95 96 97 98 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360 361 362 363 364 365 366 367 368 369 370 371 372 373 374 375 376 377 378 379 380 381 382 383 384 385 386 387 388 389 390 391 392 393 394 395 396 397 398 399 400 401 402 403 404 405 406 407 408 409 410 411 412 413 414 415 416 417 418 419 420 421 422 423 424 425 426 427 428 429 430 431 432 433 434 435 436 437 438 439 440 441 442 443 444 445 446 447 448 449 450 451 452 453 454 455 456 457 458 459 460 461 462 463 464 465 466 467 468 469 470 471 472 473 474 475 476 477 478 479 480 481 482 483 484 485 486
/*****************************************************************************

Copyright (c) 1995, 2009, Innobase Oy. All Rights Reserved.
Copyright (c) 2008, Google Inc.

Portions of this file contain modifications contributed and copyrighted by
Google, Inc. Those modifications are gratefully acknowledged and are described
briefly in the InnoDB documentation. The contributions by Google are
incorporated with their permission, and subject to the conditions contained in
the file COPYING.Google.

This program is free software; you can redistribute it and/or modify it under
the terms of the GNU General Public License as published by the Free Software
Foundation; version 2 of the License.

This program is distributed in the hope that it will be useful, but WITHOUT
ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or FITNESS
FOR A PARTICULAR PURPOSE. See the GNU General Public License for more details.

You should have received a copy of the GNU General Public License along with
this program; if not, write to the Free Software Foundation, Inc., 59 Temple
Place, Suite 330, Boston, MA 02111-1307 USA

*****************************************************************************/

/**************************************************//**
@file sync/sync0sync.c
Mutex, the basic synchronization primitive

Created 9/5/1995 Heikki Tuuri
*******************************************************/

#include "sync0sync.h"
#ifdef UNIV_NONINL
#include "sync0sync.ic"
#endif

#include "sync0rw.h"
#include "buf0buf.h"
#include "srv0srv.h"
#include "buf0types.h"

/*
	REASONS FOR IMPLEMENTING THE SPIN LOCK MUTEX
	============================================

Semaphore operations in operating systems are slow: Solaris on a 1993 Sparc
takes 3 microseconds (us) for a lock-unlock pair and Windows NT on a 1995
Pentium takes 20 microseconds for a lock-unlock pair. Therefore, we have to
implement our own efficient spin lock mutex. Future operating systems may
provide efficient spin locks, but we cannot count on that.

Another reason for implementing a spin lock is that on multiprocessor systems
it can be more efficient for a processor to run a loop waiting for the
semaphore to be released than to switch to a different thread. A thread switch
takes 25 us on both platforms mentioned above. See Gray and Reuter's book
Transaction processing for background.

How long should the spin loop last before suspending the thread? On a
uniprocessor, spinning does not help at all, because if the thread owning the
mutex is not executing, it cannot be released. Spinning actually wastes
resources.

On a multiprocessor, we do not know if the thread owning the mutex is
executing or not. Thus it would make sense to spin as long as the operation
guarded by the mutex would typically last assuming that the thread is
executing. If the mutex is not released by that time, we may assume that the
thread owning the mutex is not executing and suspend the waiting thread.

A typical operation (where no i/o involved) guarded by a mutex or a read-write
lock may last 1 - 20 us on the current Pentium platform. The longest
operations are the binary searches on an index node.

We conclude that the best choice is to set the spin time at 20 us. Then the
system should work well on a multiprocessor. On a uniprocessor we have to
make sure that thread swithches due to mutex collisions are not frequent,
i.e., they do not happen every 100 us or so, because that wastes too much
resources. If the thread switches are not frequent, the 20 us wasted in spin
loop is not too much.

Empirical studies on the effect of spin time should be done for different
platforms.


	IMPLEMENTATION OF THE MUTEX
	===========================

For background, see Curt Schimmel's book on Unix implementation on modern
architectures. The key points in the implementation are atomicity and
serialization of memory accesses. The test-and-set instruction (XCHG in
Pentium) must be atomic. As new processors may have weak memory models, also
serialization of memory references may be necessary. The successor of Pentium,
P6, has at least one mode where the memory model is weak. As far as we know,
in Pentium all memory accesses are serialized in the program order and we do
not have to worry about the memory model. On other processors there are
special machine instructions called a fence, memory barrier, or storage
barrier (STBAR in Sparc), which can be used to serialize the memory accesses
to happen in program order relative to the fence instruction.

Leslie Lamport has devised a "bakery algorithm" to implement a mutex without
the atomic test-and-set, but his algorithm should be modified for weak memory
models. We do not use Lamport's algorithm, because we guess it is slower than
the atomic test-and-set.

Our mutex implementation works as follows: After that we perform the atomic
test-and-set instruction on the memory word. If the test returns zero, we
know we got the lock first. If the test returns not zero, some other thread
was quicker and got the lock: then we spin in a loop reading the memory word,
waiting it to become zero. It is wise to just read the word in the loop, not
perform numerous test-and-set instructions, because they generate memory
traffic between the cache and the main memory. The read loop can just access
the cache, saving bus bandwidth.

If we cannot acquire the mutex lock in the specified time, we reserve a cell
in the wait array, set the waiters byte in the mutex to 1. To avoid a race
condition, after setting the waiters byte and before suspending the waiting
thread, we still have to check that the mutex is reserved, because it may
have happened that the thread which was holding the mutex has just released
it and did not see the waiters byte set to 1, a case which would lead the
other thread to an infinite wait.

LEMMA 1: After a thread resets the event of a mutex (or rw_lock), some
=======
thread will eventually call os_event_set() on that particular event.
Thus no infinite wait is possible in this case.

Proof:	After making the reservation the thread sets the waiters field in the
mutex to 1. Then it checks that the mutex is still reserved by some thread,
or it reserves the mutex for itself. In any case, some thread (which may be
also some earlier thread, not necessarily the one currently holding the mutex)
will set the waiters field to 0 in mutex_exit, and then call
os_event_set() with the mutex as an argument.
Q.E.D.

LEMMA 2: If an os_event_set() call is made after some thread has called
=======
the os_event_reset() and before it starts wait on that event, the call
will not be lost to the second thread. This is true even if there is an
intervening call to os_event_reset() by another thread.
Thus no infinite wait is possible in this case.

Proof (non-windows platforms): os_event_reset() returns a monotonically
increasing value of signal_count. This value is increased at every
call of os_event_set() If thread A has called os_event_reset() followed
by thread B calling os_event_set() and then some other thread C calling
os_event_reset(), the is_set flag of the event will be set to FALSE;
but now if thread A calls os_event_wait_low() with the signal_count
value returned from the earlier call of os_event_reset(), it will
return immediately without waiting.
Q.E.D.

Proof (windows): If there is a writer thread which is forced to wait for
the lock, it may be able to set the state of rw_lock to RW_LOCK_WAIT_EX
The design of rw_lock ensures that there is one and only one thread
that is able to change the state to RW_LOCK_WAIT_EX and this thread is
guaranteed to acquire the lock after it is released by the current
holders and before any other waiter gets the lock.
On windows this thread waits on a separate event i.e.: wait_ex_event.
Since only one thread can wait on this event there is no chance
of this event getting reset before the writer starts wait on it.
Therefore, this thread is guaranteed to catch the os_set_event()
signalled unconditionally at the release of the lock.
Q.E.D. */

/* Number of spin waits on mutexes: for performance monitoring */

/** The number of iterations in the mutex_spin_wait() spin loop.
Intended for performance monitoring. */
static ib_int64_t	mutex_spin_round_count		= 0;
/** The number of mutex_spin_wait() calls.  Intended for
performance monitoring. */
static ib_int64_t	mutex_spin_wait_count		= 0;
/** The number of OS waits in mutex_spin_wait().  Intended for
performance monitoring. */
static ib_int64_t	mutex_os_wait_count		= 0;
/** The number of mutex_exit() calls. Intended for performance
monitoring. */
UNIV_INTERN ib_int64_t	mutex_exit_count		= 0;

/** The global array of wait cells for implementation of the database's own
mutexes and read-write locks */
UNIV_INTERN sync_array_t*	sync_primary_wait_array;

/** This variable is set to TRUE when sync_init is called */
UNIV_INTERN ibool	sync_initialized	= FALSE;

/** An acquired mutex or rw-lock and its level in the latching order */
typedef struct sync_level_struct	sync_level_t;
/** Mutexes or rw-locks held by a thread */
typedef struct sync_thread_struct	sync_thread_t;

#ifdef UNIV_SYNC_DEBUG
/** The latch levels currently owned by threads are stored in this data
structure; the size of this array is OS_THREAD_MAX_N */

UNIV_INTERN sync_thread_t*	sync_thread_level_arrays;

/** Mutex protecting sync_thread_level_arrays */
UNIV_INTERN mutex_t		sync_thread_mutex;
#endif /* UNIV_SYNC_DEBUG */

/** Global list of database mutexes (not OS mutexes) created. */
UNIV_INTERN ut_list_base_node_t  mutex_list;

/** Mutex protecting the mutex_list variable */
UNIV_INTERN mutex_t mutex_list_mutex;

#ifdef UNIV_SYNC_DEBUG
/** Latching order checks start when this is set TRUE */
UNIV_INTERN ibool	sync_order_checks_on	= FALSE;
#endif /* UNIV_SYNC_DEBUG */

/** Mutexes or rw-locks held by a thread */
struct sync_thread_struct{
	os_thread_id_t	id;	/*!< OS thread id */
	sync_level_t*	levels;	/*!< level array for this thread; if
				this is NULL this slot is unused */
};

/** Number of slots reserved for each OS thread in the sync level array */
#define SYNC_THREAD_N_LEVELS	10000

/** An acquired mutex or rw-lock and its level in the latching order */
struct sync_level_struct{
	void*	latch;	/*!< pointer to a mutex or an rw-lock; NULL means that
			the slot is empty */
	ulint	level;	/*!< level of the latch in the latching order */
};

/******************************************************************//**
Creates, or rather, initializes a mutex object in a specified memory
location (which must be appropriately aligned). The mutex is initialized
in the reset state. Explicit freeing of the mutex with mutex_free is
necessary only if the memory block containing it is freed. */
UNIV_INTERN
void
mutex_create_func(
/*==============*/
	mutex_t*	mutex,		/*!< in: pointer to memory */
#ifdef UNIV_DEBUG
	const char*	cmutex_name,	/*!< in: mutex name */
# ifdef UNIV_SYNC_DEBUG
	ulint		level,		/*!< in: level */
# endif /* UNIV_SYNC_DEBUG */
#endif /* UNIV_DEBUG */
	const char*	cfile_name,	/*!< in: file name where created */
	ulint		cline)		/*!< in: file line where created */
{
#if defined(HAVE_ATOMIC_BUILTINS)
	mutex_reset_lock_word(mutex);
#else
	os_fast_mutex_init(&(mutex->os_fast_mutex));
	mutex->lock_word = 0;
#endif
	mutex->event = os_event_create(NULL);
	mutex_set_waiters(mutex, 0);
#ifdef UNIV_DEBUG
	mutex->magic_n = MUTEX_MAGIC_N;
#endif /* UNIV_DEBUG */
#ifdef UNIV_SYNC_DEBUG
	mutex->line = 0;
	mutex->file_name = "not yet reserved";
	mutex->level = level;
#endif /* UNIV_SYNC_DEBUG */
	mutex->cfile_name = cfile_name;
	mutex->cline = cline;
	mutex->count_os_wait = 0;
#ifdef UNIV_DEBUG
	mutex->cmutex_name=	  cmutex_name;
	mutex->count_using=	  0;
	mutex->mutex_type=	  0;
	mutex->lspent_time=	  0;
	mutex->lmax_spent_time=     0;
	mutex->count_spin_loop= 0;
	mutex->count_spin_rounds=   0;
	mutex->count_os_yield=  0;
#endif /* UNIV_DEBUG */

	/* Check that lock_word is aligned; this is important on Intel */
	ut_ad(((ulint)(&(mutex->lock_word))) % 4 == 0);

	/* NOTE! The very first mutexes are not put to the mutex list */

	if ((mutex == &mutex_list_mutex)
#ifdef UNIV_SYNC_DEBUG
	    || (mutex == &sync_thread_mutex)
#endif /* UNIV_SYNC_DEBUG */
	    ) {

		return;
	}

	mutex_enter(&mutex_list_mutex);

	ut_ad(UT_LIST_GET_LEN(mutex_list) == 0
	      || UT_LIST_GET_FIRST(mutex_list)->magic_n == MUTEX_MAGIC_N);

	UT_LIST_ADD_FIRST(list, mutex_list, mutex);

	mutex_exit(&mutex_list_mutex);
}

/******************************************************************//**
Calling this function is obligatory only if the memory buffer containing
the mutex is freed. Removes a mutex object from the mutex list. The mutex
is checked to be in the reset state. */
UNIV_INTERN
void
mutex_free(
/*=======*/
	mutex_t*	mutex)	/*!< in: mutex */
{
	ut_ad(mutex_validate(mutex));
	ut_a(mutex_get_lock_word(mutex) == 0);
	ut_a(mutex_get_waiters(mutex) == 0);

	if (mutex != &mutex_list_mutex
#ifdef UNIV_SYNC_DEBUG
	    && mutex != &sync_thread_mutex
#endif /* UNIV_SYNC_DEBUG */
	    ) {

		mutex_enter(&mutex_list_mutex);

		ut_ad(!UT_LIST_GET_PREV(list, mutex)
		      || UT_LIST_GET_PREV(list, mutex)->magic_n
		      == MUTEX_MAGIC_N);
		ut_ad(!UT_LIST_GET_NEXT(list, mutex)
		      || UT_LIST_GET_NEXT(list, mutex)->magic_n
		      == MUTEX_MAGIC_N);

		UT_LIST_REMOVE(list, mutex_list, mutex);

		mutex_exit(&mutex_list_mutex);
	}

	os_event_free(mutex->event);

#if !defined(HAVE_ATOMIC_BUILTINS)
	os_fast_mutex_free(&(mutex->os_fast_mutex));
#endif
	/* If we free the mutex protecting the mutex list (freeing is
	not necessary), we have to reset the magic number AFTER removing
	it from the list. */
#ifdef UNIV_DEBUG
	mutex->magic_n = 0;
#endif /* UNIV_DEBUG */
}

/********************************************************************//**
NOTE! Use the corresponding macro in the header file, not this function
directly. Tries to lock the mutex for the current thread. If the lock is not
acquired immediately, returns with return value 1.
@return	0 if succeed, 1 if not */
UNIV_INTERN
ulint
mutex_enter_nowait_func(
/*====================*/
	mutex_t*	mutex,		/*!< in: pointer to mutex */
	const char*	file_name __attribute__((unused)),
					/*!< in: file name where mutex
					requested */
	ulint		line __attribute__((unused)))
					/*!< in: line where requested */
{
	ut_ad(mutex_validate(mutex));

	if (!mutex_test_and_set(mutex)) {

		ut_d(mutex->thread_id = os_thread_get_curr_id());
#ifdef UNIV_SYNC_DEBUG
		mutex_set_debug_info(mutex, file_name, line);
#endif

		return(0);	/* Succeeded! */
	}

	return(1);
}

#ifdef UNIV_DEBUG
/******************************************************************//**
Checks that the mutex has been initialized.
@return	TRUE */
UNIV_INTERN
ibool
mutex_validate(
/*===========*/
	const mutex_t*	mutex)	/*!< in: mutex */
{
	ut_a(mutex);
	ut_a(mutex->magic_n == MUTEX_MAGIC_N);

	return(TRUE);
}

/******************************************************************//**
Checks that the current thread owns the mutex. Works only in the debug
version.
@return	TRUE if owns */
UNIV_INTERN
ibool
mutex_own(
/*======*/
	const mutex_t*	mutex)	/*!< in: mutex */
{
	ut_ad(mutex_validate(mutex));

	return(mutex_get_lock_word(mutex) == 1
	       && os_thread_eq(mutex->thread_id, os_thread_get_curr_id()));
}
#endif /* UNIV_DEBUG */

/******************************************************************//**
Sets the waiters field in a mutex. */
UNIV_INTERN
void
mutex_set_waiters(
/*==============*/
	mutex_t*	mutex,	/*!< in: mutex */
	ulint		n)	/*!< in: value to set */
{
	volatile ulint*	ptr;		/* declared volatile to ensure that
					the value is stored to memory */
	ut_ad(mutex);

	ptr = &(mutex->waiters);

	*ptr = n;		/* Here we assume that the write of a single
				word in memory is atomic */
}

/******************************************************************//**
Reserves a mutex for the current thread. If the mutex is reserved, the
function spins a preset time (controlled by SYNC_SPIN_ROUNDS), waiting
for the mutex before suspending the thread. */
UNIV_INTERN
void
mutex_spin_wait(
/*============*/
	mutex_t*	mutex,		/*!< in: pointer to mutex */
	const char*	file_name,	/*!< in: file name where mutex
					requested */
	ulint		line)		/*!< in: line where requested */
{
	ulint	   index; /* index of the reserved wait cell */
	ulint	   i;	  /* spin round count */
#ifdef UNIV_DEBUG
	ib_int64_t lstart_time = 0, lfinish_time; /* for timing os_wait */
	ulint ltime_diff;
	ulint sec;
	ulint ms;
	uint timer_started = 0;
#endif /* UNIV_DEBUG */
	ut_ad(mutex);

	/* This update is not thread safe, but we don't mind if the count
	isn't exact. Moved out of ifdef that follows because we are willing
	to sacrifice the cost of counting this as the data is valuable.
	Count the number of calls to mutex_spin_wait. */
	mutex_spin_wait_count++;

mutex_loop:

	i = 0;

	/* Spin waiting for the lock word to become zero. Note that we do
	not have to assume that the read access to the lock word is atomic,
	as the actual locking is always committed with atomic test-and-set.
	In reality, however, all processors probably have an atomic read of
	a memory word. */

spin_loop:
	ut_d(mutex->count_spin_loop++);

	while (mutex_get_lock_word(mutex) != 0 && i < SYNC_SPIN_ROUNDS) {
		if (srv_spin_wait_delay) {
			ut_delay(ut_rnd_interval(0, srv_spin_wait_delay));
		}

		i++;
	}

	if (i == SYNC_SPIN_ROUNDS) {
#ifdef UNIV_DEBUG
		mutex->count_os_yield++;
Sergey Vojtovich's avatar
Sergey Vojtovich committed
487 488
#ifndef UNIV_HOTBACKUP
		if (timed_mutexes && timer_started == 0) {
489 490 491 492
			ut_usectime(&sec, &ms);
			lstart_time= (ib_int64_t)sec * 1000000 + ms;
			timer_started = 1;
		}
Sergey Vojtovich's avatar
Sergey Vojtovich committed
493
#endif /* UNIV_HOTBACKUP */
494 495 496 497 498 499 500 501 502 503 504 505 506 507 508 509 510 511 512 513 514 515 516 517 518 519 520 521 522 523 524 525 526 527 528 529 530 531 532 533 534 535 536 537 538 539 540 541 542 543 544 545 546 547 548 549 550 551 552 553 554 555 556 557 558 559 560 561 562 563 564 565 566 567 568 569 570 571 572 573 574 575 576 577 578 579 580 581 582 583 584 585 586 587
#endif /* UNIV_DEBUG */
		os_thread_yield();
	}

#ifdef UNIV_SRV_PRINT_LATCH_WAITS
	fprintf(stderr,
		"Thread %lu spin wait mutex at %p"
		" cfile %s cline %lu rnds %lu\n",
		(ulong) os_thread_pf(os_thread_get_curr_id()), (void*) mutex,
		mutex->cfile_name, (ulong) mutex->cline, (ulong) i);
#endif

	mutex_spin_round_count += i;

	ut_d(mutex->count_spin_rounds += i);

	if (mutex_test_and_set(mutex) == 0) {
		/* Succeeded! */

		ut_d(mutex->thread_id = os_thread_get_curr_id());
#ifdef UNIV_SYNC_DEBUG
		mutex_set_debug_info(mutex, file_name, line);
#endif

		goto finish_timing;
	}

	/* We may end up with a situation where lock_word is 0 but the OS
	fast mutex is still reserved. On FreeBSD the OS does not seem to
	schedule a thread which is constantly calling pthread_mutex_trylock
	(in mutex_test_and_set implementation). Then we could end up
	spinning here indefinitely. The following 'i++' stops this infinite
	spin. */

	i++;

	if (i < SYNC_SPIN_ROUNDS) {
		goto spin_loop;
	}

	sync_array_reserve_cell(sync_primary_wait_array, mutex,
				SYNC_MUTEX, file_name, line, &index);

	/* The memory order of the array reservation and the change in the
	waiters field is important: when we suspend a thread, we first
	reserve the cell and then set waiters field to 1. When threads are
	released in mutex_exit, the waiters field is first set to zero and
	then the event is set to the signaled state. */

	mutex_set_waiters(mutex, 1);

	/* Try to reserve still a few times */
	for (i = 0; i < 4; i++) {
		if (mutex_test_and_set(mutex) == 0) {
			/* Succeeded! Free the reserved wait cell */

			sync_array_free_cell(sync_primary_wait_array, index);

			ut_d(mutex->thread_id = os_thread_get_curr_id());
#ifdef UNIV_SYNC_DEBUG
			mutex_set_debug_info(mutex, file_name, line);
#endif

#ifdef UNIV_SRV_PRINT_LATCH_WAITS
			fprintf(stderr, "Thread %lu spin wait succeeds at 2:"
				" mutex at %p\n",
				(ulong) os_thread_pf(os_thread_get_curr_id()),
				(void*) mutex);
#endif

			goto finish_timing;

			/* Note that in this case we leave the waiters field
			set to 1. We cannot reset it to zero, as we do not
			know if there are other waiters. */
		}
	}

	/* Now we know that there has been some thread holding the mutex
	after the change in the wait array and the waiters field was made.
	Now there is no risk of infinite wait on the event. */

#ifdef UNIV_SRV_PRINT_LATCH_WAITS
	fprintf(stderr,
		"Thread %lu OS wait mutex at %p cfile %s cline %lu rnds %lu\n",
		(ulong) os_thread_pf(os_thread_get_curr_id()), (void*) mutex,
		mutex->cfile_name, (ulong) mutex->cline, (ulong) i);
#endif

	mutex_os_wait_count++;

	mutex->count_os_wait++;
#ifdef UNIV_DEBUG
	/* !!!!! Sometimes os_wait can be called without os_thread_yield */
Sergey Vojtovich's avatar
Sergey Vojtovich committed
588 589
#ifndef UNIV_HOTBACKUP
	if (timed_mutexes == 1 && timer_started == 0) {
590 591 592 593
		ut_usectime(&sec, &ms);
		lstart_time= (ib_int64_t)sec * 1000000 + ms;
		timer_started = 1;
	}
Sergey Vojtovich's avatar
Sergey Vojtovich committed
594
#endif /* UNIV_HOTBACKUP */
595 596 597 598 599 600 601 602 603 604 605 606 607 608 609 610 611 612 613 614 615 616 617 618 619 620 621 622 623 624 625 626 627 628 629 630 631 632 633 634 635 636 637 638 639 640 641 642 643 644 645 646 647 648 649 650 651 652 653 654 655 656 657 658 659 660 661 662 663 664 665 666 667 668 669 670 671 672 673 674 675 676 677 678 679 680 681 682 683 684 685 686 687 688 689 690 691 692 693 694 695 696 697 698 699 700 701 702 703 704 705 706 707 708 709 710 711 712 713 714 715 716 717 718 719 720 721 722 723 724 725 726 727 728 729 730 731 732 733 734 735 736 737 738 739 740 741 742 743 744 745 746 747 748 749 750 751 752 753 754 755 756 757 758 759 760 761 762 763 764 765 766 767 768 769 770 771 772 773 774 775 776 777 778 779 780 781 782 783 784 785 786 787 788 789 790 791 792 793 794 795 796 797 798 799 800 801 802 803 804 805 806 807 808 809 810 811 812 813 814 815 816 817 818 819 820 821 822 823 824 825 826 827 828 829 830 831 832 833 834 835 836 837 838 839 840 841 842 843 844 845 846 847 848 849 850 851 852 853 854 855 856 857 858 859 860 861 862 863 864 865 866 867 868 869 870 871 872 873 874 875 876 877 878 879 880 881 882 883 884 885 886 887 888 889 890 891 892 893 894 895 896 897 898 899 900 901 902 903 904 905 906 907 908 909 910 911 912 913 914 915 916 917 918 919 920 921 922 923 924 925 926 927 928 929 930 931 932 933 934 935 936 937 938 939 940 941 942 943 944 945 946 947 948 949 950 951 952 953 954 955 956 957 958 959 960 961 962 963 964 965 966 967 968 969 970 971 972 973 974 975 976 977 978 979 980 981 982 983 984 985 986 987 988 989 990 991 992 993 994 995 996 997 998 999 1000 1001 1002 1003 1004 1005 1006 1007 1008 1009 1010 1011 1012 1013 1014 1015 1016 1017 1018 1019 1020 1021 1022 1023 1024 1025 1026 1027 1028 1029 1030 1031 1032 1033 1034 1035 1036 1037 1038 1039 1040 1041 1042 1043 1044 1045 1046 1047 1048 1049 1050 1051 1052 1053 1054 1055 1056 1057 1058 1059 1060 1061 1062 1063 1064 1065 1066 1067 1068 1069 1070 1071 1072 1073 1074 1075 1076 1077 1078 1079 1080 1081 1082 1083 1084 1085 1086 1087 1088 1089 1090 1091 1092 1093 1094 1095 1096 1097 1098 1099 1100 1101 1102 1103 1104 1105 1106 1107 1108 1109 1110 1111 1112 1113 1114 1115 1116 1117 1118 1119 1120 1121 1122 1123 1124 1125 1126 1127 1128 1129 1130 1131 1132 1133 1134 1135 1136 1137 1138 1139 1140 1141 1142 1143 1144 1145 1146 1147 1148 1149 1150 1151 1152 1153 1154 1155 1156 1157 1158 1159 1160 1161 1162 1163 1164 1165 1166 1167 1168 1169 1170 1171 1172 1173 1174 1175 1176 1177 1178 1179 1180 1181 1182 1183 1184 1185 1186 1187 1188 1189 1190 1191 1192 1193 1194 1195 1196 1197 1198 1199 1200 1201 1202 1203 1204 1205 1206 1207 1208 1209 1210 1211 1212 1213 1214 1215 1216 1217 1218 1219 1220 1221 1222 1223 1224 1225 1226 1227 1228 1229 1230 1231 1232 1233 1234 1235 1236 1237 1238 1239 1240 1241 1242 1243 1244 1245 1246 1247 1248 1249 1250 1251 1252 1253 1254 1255 1256 1257 1258 1259 1260 1261 1262 1263 1264 1265 1266 1267 1268 1269 1270 1271 1272 1273 1274 1275 1276 1277 1278 1279 1280 1281 1282 1283 1284 1285 1286 1287 1288 1289 1290 1291 1292 1293 1294 1295 1296 1297 1298 1299 1300 1301 1302 1303 1304 1305 1306 1307 1308 1309 1310 1311 1312 1313 1314 1315 1316 1317 1318 1319 1320 1321 1322 1323 1324 1325 1326 1327 1328 1329 1330 1331 1332 1333 1334 1335 1336 1337 1338 1339 1340 1341 1342 1343 1344 1345 1346 1347 1348 1349 1350 1351 1352 1353 1354 1355 1356 1357 1358 1359 1360 1361 1362 1363 1364 1365 1366 1367 1368 1369 1370 1371 1372 1373 1374 1375 1376 1377 1378 1379 1380 1381 1382 1383 1384 1385 1386 1387 1388 1389 1390 1391 1392 1393 1394 1395 1396 1397 1398 1399 1400 1401 1402 1403 1404 1405 1406 1407 1408 1409 1410 1411 1412 1413 1414 1415 1416 1417
#endif /* UNIV_DEBUG */

	sync_array_wait_event(sync_primary_wait_array, index);
	goto mutex_loop;

finish_timing:
#ifdef UNIV_DEBUG
	if (timed_mutexes == 1 && timer_started==1) {
		ut_usectime(&sec, &ms);
		lfinish_time= (ib_int64_t)sec * 1000000 + ms;

		ltime_diff= (ulint) (lfinish_time - lstart_time);
		mutex->lspent_time += ltime_diff;

		if (mutex->lmax_spent_time < ltime_diff) {
			mutex->lmax_spent_time= ltime_diff;
		}
	}
#endif /* UNIV_DEBUG */
	return;
}

/******************************************************************//**
Releases the threads waiting in the primary wait array for this mutex. */
UNIV_INTERN
void
mutex_signal_object(
/*================*/
	mutex_t*	mutex)	/*!< in: mutex */
{
	mutex_set_waiters(mutex, 0);

	/* The memory order of resetting the waiters field and
	signaling the object is important. See LEMMA 1 above. */
	os_event_set(mutex->event);
	sync_array_object_signalled(sync_primary_wait_array);
}

#ifdef UNIV_SYNC_DEBUG
/******************************************************************//**
Sets the debug information for a reserved mutex. */
UNIV_INTERN
void
mutex_set_debug_info(
/*=================*/
	mutex_t*	mutex,		/*!< in: mutex */
	const char*	file_name,	/*!< in: file where requested */
	ulint		line)		/*!< in: line where requested */
{
	ut_ad(mutex);
	ut_ad(file_name);

	sync_thread_add_level(mutex, mutex->level);

	mutex->file_name = file_name;
	mutex->line	 = line;
}

/******************************************************************//**
Gets the debug information for a reserved mutex. */
UNIV_INTERN
void
mutex_get_debug_info(
/*=================*/
	mutex_t*	mutex,		/*!< in: mutex */
	const char**	file_name,	/*!< out: file where requested */
	ulint*		line,		/*!< out: line where requested */
	os_thread_id_t* thread_id)	/*!< out: id of the thread which owns
					the mutex */
{
	ut_ad(mutex);

	*file_name = mutex->file_name;
	*line	   = mutex->line;
	*thread_id = mutex->thread_id;
}

/******************************************************************//**
Prints debug info of currently reserved mutexes. */
static
void
mutex_list_print_info(
/*==================*/
	FILE*	file)		/*!< in: file where to print */
{
	mutex_t*	mutex;
	const char*	file_name;
	ulint		line;
	os_thread_id_t	thread_id;
	ulint		count		= 0;

	fputs("----------\n"
	      "MUTEX INFO\n"
	      "----------\n", file);

	mutex_enter(&mutex_list_mutex);

	mutex = UT_LIST_GET_FIRST(mutex_list);

	while (mutex != NULL) {
		count++;

		if (mutex_get_lock_word(mutex) != 0) {
			mutex_get_debug_info(mutex, &file_name, &line,
					     &thread_id);
			fprintf(file,
				"Locked mutex: addr %p thread %ld"
				" file %s line %ld\n",
				(void*) mutex, os_thread_pf(thread_id),
				file_name, line);
		}

		mutex = UT_LIST_GET_NEXT(list, mutex);
	}

	fprintf(file, "Total number of mutexes %ld\n", count);

	mutex_exit(&mutex_list_mutex);
}

/******************************************************************//**
Counts currently reserved mutexes. Works only in the debug version.
@return	number of reserved mutexes */
UNIV_INTERN
ulint
mutex_n_reserved(void)
/*==================*/
{
	mutex_t*	mutex;
	ulint		count		= 0;

	mutex_enter(&mutex_list_mutex);

	mutex = UT_LIST_GET_FIRST(mutex_list);

	while (mutex != NULL) {
		if (mutex_get_lock_word(mutex) != 0) {

			count++;
		}

		mutex = UT_LIST_GET_NEXT(list, mutex);
	}

	mutex_exit(&mutex_list_mutex);

	ut_a(count >= 1);

	return(count - 1); /* Subtract one, because this function itself
			   was holding one mutex (mutex_list_mutex) */
}

/******************************************************************//**
Returns TRUE if no mutex or rw-lock is currently locked. Works only in
the debug version.
@return	TRUE if no mutexes and rw-locks reserved */
UNIV_INTERN
ibool
sync_all_freed(void)
/*================*/
{
	return(mutex_n_reserved() + rw_lock_n_locked() == 0);
}

/******************************************************************//**
Gets the value in the nth slot in the thread level arrays.
@return	pointer to thread slot */
static
sync_thread_t*
sync_thread_level_arrays_get_nth(
/*=============================*/
	ulint	n)	/*!< in: slot number */
{
	ut_ad(n < OS_THREAD_MAX_N);

	return(sync_thread_level_arrays + n);
}

/******************************************************************//**
Looks for the thread slot for the calling thread.
@return	pointer to thread slot, NULL if not found */
static
sync_thread_t*
sync_thread_level_arrays_find_slot(void)
/*====================================*/

{
	sync_thread_t*	slot;
	os_thread_id_t	id;
	ulint		i;

	id = os_thread_get_curr_id();

	for (i = 0; i < OS_THREAD_MAX_N; i++) {

		slot = sync_thread_level_arrays_get_nth(i);

		if (slot->levels && os_thread_eq(slot->id, id)) {

			return(slot);
		}
	}

	return(NULL);
}

/******************************************************************//**
Looks for an unused thread slot.
@return	pointer to thread slot */
static
sync_thread_t*
sync_thread_level_arrays_find_free(void)
/*====================================*/

{
	sync_thread_t*	slot;
	ulint		i;

	for (i = 0; i < OS_THREAD_MAX_N; i++) {

		slot = sync_thread_level_arrays_get_nth(i);

		if (slot->levels == NULL) {

			return(slot);
		}
	}

	return(NULL);
}

/******************************************************************//**
Gets the value in the nth slot in the thread level array.
@return	pointer to level slot */
static
sync_level_t*
sync_thread_levels_get_nth(
/*=======================*/
	sync_level_t*	arr,	/*!< in: pointer to level array for an OS
				thread */
	ulint		n)	/*!< in: slot number */
{
	ut_ad(n < SYNC_THREAD_N_LEVELS);

	return(arr + n);
}

/******************************************************************//**
Checks if all the level values stored in the level array are greater than
the given limit.
@return	TRUE if all greater */
static
ibool
sync_thread_levels_g(
/*=================*/
	sync_level_t*	arr,	/*!< in: pointer to level array for an OS
				thread */
	ulint		limit)	/*!< in: level limit */
{
	sync_level_t*	slot;
	rw_lock_t*	lock;
	mutex_t*	mutex;
	ulint		i;

	for (i = 0; i < SYNC_THREAD_N_LEVELS; i++) {

		slot = sync_thread_levels_get_nth(arr, i);

		if (slot->latch != NULL) {
			if (slot->level <= limit) {

				lock = slot->latch;
				mutex = slot->latch;

				fprintf(stderr,
					"InnoDB: sync levels should be"
					" > %lu but a level is %lu\n",
					(ulong) limit, (ulong) slot->level);

				if (mutex->magic_n == MUTEX_MAGIC_N) {
					fprintf(stderr,
						"Mutex created at %s %lu\n",
						mutex->cfile_name,
						(ulong) mutex->cline);

					if (mutex_get_lock_word(mutex) != 0) {
						const char*	file_name;
						ulint		line;
						os_thread_id_t	thread_id;

						mutex_get_debug_info(
							mutex, &file_name,
							&line, &thread_id);

						fprintf(stderr,
							"InnoDB: Locked mutex:"
							" addr %p thread %ld"
							" file %s line %ld\n",
							(void*) mutex,
							os_thread_pf(
								thread_id),
							file_name,
							(ulong) line);
					} else {
						fputs("Not locked\n", stderr);
					}
				} else {
					rw_lock_print(lock);
				}

				return(FALSE);
			}
		}
	}

	return(TRUE);
}

/******************************************************************//**
Checks if the level value is stored in the level array.
@return	TRUE if stored */
static
ibool
sync_thread_levels_contain(
/*=======================*/
	sync_level_t*	arr,	/*!< in: pointer to level array for an OS
				thread */
	ulint		level)	/*!< in: level */
{
	sync_level_t*	slot;
	ulint		i;

	for (i = 0; i < SYNC_THREAD_N_LEVELS; i++) {

		slot = sync_thread_levels_get_nth(arr, i);

		if (slot->latch != NULL) {
			if (slot->level == level) {

				return(TRUE);
			}
		}
	}

	return(FALSE);
}

/******************************************************************//**
Checks that the level array for the current thread is empty.
@return	TRUE if empty except the exceptions specified below */
UNIV_INTERN
ibool
sync_thread_levels_empty_gen(
/*=========================*/
	ibool	dict_mutex_allowed)	/*!< in: TRUE if dictionary mutex is
					allowed to be owned by the thread,
					also purge_is_running mutex is
					allowed */
{
	sync_level_t*	arr;
	sync_thread_t*	thread_slot;
	sync_level_t*	slot;
	ulint		i;

	if (!sync_order_checks_on) {

		return(TRUE);
	}

	mutex_enter(&sync_thread_mutex);

	thread_slot = sync_thread_level_arrays_find_slot();

	if (thread_slot == NULL) {

		mutex_exit(&sync_thread_mutex);

		return(TRUE);
	}

	arr = thread_slot->levels;

	for (i = 0; i < SYNC_THREAD_N_LEVELS; i++) {

		slot = sync_thread_levels_get_nth(arr, i);

		if (slot->latch != NULL
		    && (!dict_mutex_allowed
			|| (slot->level != SYNC_DICT
			    && slot->level != SYNC_DICT_OPERATION))) {

			mutex_exit(&sync_thread_mutex);
			ut_error;

			return(FALSE);
		}
	}

	mutex_exit(&sync_thread_mutex);

	return(TRUE);
}

/******************************************************************//**
Checks that the level array for the current thread is empty.
@return	TRUE if empty */
UNIV_INTERN
ibool
sync_thread_levels_empty(void)
/*==========================*/
{
	return(sync_thread_levels_empty_gen(FALSE));
}

/******************************************************************//**
Adds a latch and its level in the thread level array. Allocates the memory
for the array if called first time for this OS thread. Makes the checks
against other latch levels stored in the array for this thread. */
UNIV_INTERN
void
sync_thread_add_level(
/*==================*/
	void*	latch,	/*!< in: pointer to a mutex or an rw-lock */
	ulint	level)	/*!< in: level in the latching order; if
			SYNC_LEVEL_VARYING, nothing is done */
{
	sync_level_t*	array;
	sync_level_t*	slot;
	sync_thread_t*	thread_slot;
	ulint		i;

	if (!sync_order_checks_on) {

		return;
	}

	if ((latch == (void*)&sync_thread_mutex)
	    || (latch == (void*)&mutex_list_mutex)
	    || (latch == (void*)&rw_lock_debug_mutex)
	    || (latch == (void*)&rw_lock_list_mutex)) {

		return;
	}

	if (level == SYNC_LEVEL_VARYING) {

		return;
	}

	mutex_enter(&sync_thread_mutex);

	thread_slot = sync_thread_level_arrays_find_slot();

	if (thread_slot == NULL) {
		/* We have to allocate the level array for a new thread */
		array = ut_malloc(sizeof(sync_level_t) * SYNC_THREAD_N_LEVELS);

		thread_slot = sync_thread_level_arrays_find_free();

		thread_slot->id = os_thread_get_curr_id();
		thread_slot->levels = array;

		for (i = 0; i < SYNC_THREAD_N_LEVELS; i++) {

			slot = sync_thread_levels_get_nth(array, i);

			slot->latch = NULL;
		}
	}

	array = thread_slot->levels;

	/* NOTE that there is a problem with _NODE and _LEAF levels: if the
	B-tree height changes, then a leaf can change to an internal node
	or the other way around. We do not know at present if this can cause
	unnecessary assertion failures below. */

	switch (level) {
	case SYNC_NO_ORDER_CHECK:
	case SYNC_EXTERN_STORAGE:
	case SYNC_TREE_NODE_FROM_HASH:
		/* Do no order checking */
		break;
	case SYNC_MEM_POOL:
	case SYNC_MEM_HASH:
	case SYNC_RECV:
	case SYNC_WORK_QUEUE:
	case SYNC_LOG:
	case SYNC_THR_LOCAL:
	case SYNC_ANY_LATCH:
	case SYNC_TRX_SYS_HEADER:
	case SYNC_FILE_FORMAT_TAG:
	case SYNC_DOUBLEWRITE:
	case SYNC_BUF_POOL:
	case SYNC_SEARCH_SYS:
	case SYNC_SEARCH_SYS_CONF:
	case SYNC_TRX_LOCK_HEAP:
	case SYNC_KERNEL:
	case SYNC_IBUF_BITMAP_MUTEX:
	case SYNC_RSEG:
	case SYNC_TRX_UNDO:
	case SYNC_PURGE_LATCH:
	case SYNC_PURGE_SYS:
	case SYNC_DICT_AUTOINC_MUTEX:
	case SYNC_DICT_OPERATION:
	case SYNC_DICT_HEADER:
	case SYNC_TRX_I_S_RWLOCK:
	case SYNC_TRX_I_S_LAST_READ:
		if (!sync_thread_levels_g(array, level)) {
			fprintf(stderr,
				"InnoDB: sync_thread_levels_g(array, %lu)"
				" does not hold!\n", level);
			ut_error;
		}
		break;
	case SYNC_BUF_BLOCK:
		/* Either the thread must own the buffer pool mutex
		(buf_pool_mutex), or it is allowed to latch only ONE
		buffer block (block->mutex or buf_pool_zip_mutex). */
		if (!sync_thread_levels_g(array, level)) {
			ut_a(sync_thread_levels_g(array, level - 1));
			ut_a(sync_thread_levels_contain(array, SYNC_BUF_POOL));
		}
		break;
	case SYNC_REC_LOCK:
		ut_a((sync_thread_levels_contain(array, SYNC_KERNEL)
		      && sync_thread_levels_g(array, SYNC_REC_LOCK - 1))
		     || sync_thread_levels_g(array, SYNC_REC_LOCK));
		break;
	case SYNC_IBUF_BITMAP:
		/* Either the thread must own the master mutex to all
		the bitmap pages, or it is allowed to latch only ONE
		bitmap page. */
		ut_a((sync_thread_levels_contain(array, SYNC_IBUF_BITMAP_MUTEX)
		      && sync_thread_levels_g(array, SYNC_IBUF_BITMAP - 1))
		     || sync_thread_levels_g(array, SYNC_IBUF_BITMAP));
		break;
	case SYNC_FSP_PAGE:
		ut_a(sync_thread_levels_contain(array, SYNC_FSP));
		break;
	case SYNC_FSP:
		ut_a(sync_thread_levels_contain(array, SYNC_FSP)
		     || sync_thread_levels_g(array, SYNC_FSP));
		break;
	case SYNC_TRX_UNDO_PAGE:
		ut_a(sync_thread_levels_contain(array, SYNC_TRX_UNDO)
		     || sync_thread_levels_contain(array, SYNC_RSEG)
		     || sync_thread_levels_contain(array, SYNC_PURGE_SYS)
		     || sync_thread_levels_g(array, SYNC_TRX_UNDO_PAGE));
		break;
	case SYNC_RSEG_HEADER:
		ut_a(sync_thread_levels_contain(array, SYNC_RSEG));
		break;
	case SYNC_RSEG_HEADER_NEW:
		ut_a(sync_thread_levels_contain(array, SYNC_KERNEL)
		     && sync_thread_levels_contain(array, SYNC_FSP_PAGE));
		break;
	case 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));
		break;
	case SYNC_TREE_NODE_NEW:
		ut_a(sync_thread_levels_contain(array, SYNC_FSP_PAGE)
		     || sync_thread_levels_contain(array, SYNC_IBUF_MUTEX));
		break;
	case SYNC_INDEX_TREE:
		ut_a((sync_thread_levels_contain(array, SYNC_IBUF_MUTEX)
		      && sync_thread_levels_contain(array, SYNC_FSP)
		      && sync_thread_levels_g(array, SYNC_FSP_PAGE - 1))
		     || sync_thread_levels_g(array, SYNC_TREE_NODE - 1));
		break;
	case SYNC_IBUF_MUTEX:
		ut_a(sync_thread_levels_g(array, SYNC_FSP_PAGE - 1));
		break;
	case SYNC_IBUF_PESS_INSERT_MUTEX:
		ut_a(sync_thread_levels_g(array, SYNC_FSP - 1)
		     && !sync_thread_levels_contain(array, SYNC_IBUF_MUTEX));
		break;
	case SYNC_IBUF_HEADER:
		ut_a(sync_thread_levels_g(array, SYNC_FSP - 1)
		     && !sync_thread_levels_contain(array, SYNC_IBUF_MUTEX)
		     && !sync_thread_levels_contain(
			     array, SYNC_IBUF_PESS_INSERT_MUTEX));
		break;
	case SYNC_DICT:
#ifdef UNIV_DEBUG
		ut_a(buf_debug_prints
		     || sync_thread_levels_g(array, SYNC_DICT));
#else /* UNIV_DEBUG */
		ut_a(sync_thread_levels_g(array, SYNC_DICT));
#endif /* UNIV_DEBUG */
		break;
	default:
		ut_error;
	}

	for (i = 0; i < SYNC_THREAD_N_LEVELS; i++) {

		slot = sync_thread_levels_get_nth(array, i);

		if (slot->latch == NULL) {
			slot->latch = latch;
			slot->level = level;

			break;
		}
	}

	ut_a(i < SYNC_THREAD_N_LEVELS);

	mutex_exit(&sync_thread_mutex);
}

/******************************************************************//**
Removes a latch from the thread level array if it is found there.
@return TRUE if found in the array; it is no error if the latch is
not found, as we presently are not able to determine the level for
every latch reservation the program does */
UNIV_INTERN
ibool
sync_thread_reset_level(
/*====================*/
	void*	latch)	/*!< in: pointer to a mutex or an rw-lock */
{
	sync_level_t*	array;
	sync_level_t*	slot;
	sync_thread_t*	thread_slot;
	ulint		i;

	if (!sync_order_checks_on) {

		return(FALSE);
	}

	if ((latch == (void*)&sync_thread_mutex)
	    || (latch == (void*)&mutex_list_mutex)
	    || (latch == (void*)&rw_lock_debug_mutex)
	    || (latch == (void*)&rw_lock_list_mutex)) {

		return(FALSE);
	}

	mutex_enter(&sync_thread_mutex);

	thread_slot = sync_thread_level_arrays_find_slot();

	if (thread_slot == NULL) {

		ut_error;

		mutex_exit(&sync_thread_mutex);
		return(FALSE);
	}

	array = thread_slot->levels;

	for (i = 0; i < SYNC_THREAD_N_LEVELS; i++) {

		slot = sync_thread_levels_get_nth(array, i);

		if (slot->latch == latch) {
			slot->latch = NULL;

			mutex_exit(&sync_thread_mutex);

			return(TRUE);
		}
	}

	if (((mutex_t*) latch)->magic_n != MUTEX_MAGIC_N) {
		rw_lock_t*	rw_lock;

		rw_lock = (rw_lock_t*) latch;

		if (rw_lock->level == SYNC_LEVEL_VARYING) {
			mutex_exit(&sync_thread_mutex);

			return(TRUE);
		}
	}

	ut_error;

	mutex_exit(&sync_thread_mutex);

	return(FALSE);
}
#endif /* UNIV_SYNC_DEBUG */

/******************************************************************//**
Initializes the synchronization data structures. */
UNIV_INTERN
void
sync_init(void)
/*===========*/
{
#ifdef UNIV_SYNC_DEBUG
	sync_thread_t*	thread_slot;
	ulint		i;
#endif /* UNIV_SYNC_DEBUG */

	ut_a(sync_initialized == FALSE);

	sync_initialized = TRUE;

	/* Create the primary system wait array which is protected by an OS
	mutex */

	sync_primary_wait_array = sync_array_create(OS_THREAD_MAX_N,
						    SYNC_ARRAY_OS_MUTEX);
#ifdef UNIV_SYNC_DEBUG
	/* Create the thread latch level array where the latch levels
	are stored for each OS thread */

	sync_thread_level_arrays = ut_malloc(OS_THREAD_MAX_N
					     * sizeof(sync_thread_t));
	for (i = 0; i < OS_THREAD_MAX_N; i++) {

		thread_slot = sync_thread_level_arrays_get_nth(i);
		thread_slot->levels = NULL;
	}
#endif /* UNIV_SYNC_DEBUG */
	/* Init the mutex list and create the mutex to protect it. */

	UT_LIST_INIT(mutex_list);
	mutex_create(&mutex_list_mutex, SYNC_NO_ORDER_CHECK);
#ifdef UNIV_SYNC_DEBUG
	mutex_create(&sync_thread_mutex, SYNC_NO_ORDER_CHECK);
#endif /* UNIV_SYNC_DEBUG */

	/* Init the rw-lock list and create the mutex to protect it. */

	UT_LIST_INIT(rw_lock_list);
	mutex_create(&rw_lock_list_mutex, SYNC_NO_ORDER_CHECK);

#ifdef UNIV_SYNC_DEBUG
	mutex_create(&rw_lock_debug_mutex, SYNC_NO_ORDER_CHECK);

	rw_lock_debug_event = os_event_create(NULL);
	rw_lock_debug_waiters = FALSE;
#endif /* UNIV_SYNC_DEBUG */
}

/******************************************************************//**
Frees the resources in InnoDB's own synchronization data structures. Use
os_sync_free() after calling this. */
UNIV_INTERN
void
sync_close(void)
/*===========*/
{
	mutex_t*	mutex;

	sync_array_free(sync_primary_wait_array);

	mutex = UT_LIST_GET_FIRST(mutex_list);

	while (mutex) {
		mutex_free(mutex);
		mutex = UT_LIST_GET_FIRST(mutex_list);
	}

	mutex_free(&mutex_list_mutex);
#ifdef UNIV_SYNC_DEBUG
	mutex_free(&sync_thread_mutex);
#endif /* UNIV_SYNC_DEBUG */
}

/*******************************************************************//**
Prints wait info of the sync system. */
UNIV_INTERN
void
sync_print_wait_info(
/*=================*/
	FILE*	file)		/*!< in: file where to print */
{
#ifdef UNIV_SYNC_DEBUG
	fprintf(file, "Mutex exits %llu, rws exits %llu, rwx exits %llu\n",
		mutex_exit_count, rw_s_exit_count, rw_x_exit_count);
#endif

	fprintf(file,
		"Mutex spin waits %llu, rounds %llu, OS waits %llu\n"
		"RW-shared spins %llu, OS waits %llu;"
		" RW-excl spins %llu, OS waits %llu\n",
		mutex_spin_wait_count,
		mutex_spin_round_count,
		mutex_os_wait_count,
		rw_s_spin_wait_count,
		rw_s_os_wait_count,
		rw_x_spin_wait_count,
		rw_x_os_wait_count);

	fprintf(file,
		"Spin rounds per wait: %.2f mutex, %.2f RW-shared, "
		"%.2f RW-excl\n",
		(double) mutex_spin_round_count /
		(mutex_spin_wait_count ? mutex_spin_wait_count : 1),
		(double) rw_s_spin_round_count /
		(rw_s_spin_wait_count ? rw_s_spin_wait_count : 1),
		(double) rw_x_spin_round_count /
		(rw_x_spin_wait_count ? rw_x_spin_wait_count : 1));
}

/*******************************************************************//**
Prints info of the sync system. */
UNIV_INTERN
void
sync_print(
/*=======*/
	FILE*	file)		/*!< in: file where to print */
{
#ifdef UNIV_SYNC_DEBUG
	mutex_list_print_info(file);

	rw_lock_list_print_info(file);
#endif /* UNIV_SYNC_DEBUG */

	sync_array_print_info(file, sync_primary_wait_array);

	sync_print_wait_info(file);
}