summaryrefslogtreecommitdiff
path: root/lib/libpthread/pthread_mutex.c
blob: cf05088d22d3fe6fb4d91a55bf1c364c7a696880 (plain)
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
487
488
489
490
491
492
493
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
588
589
590
591
592
593
594
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
/*	$NetBSD: pthread_mutex.c,v 1.51 2008/08/02 19:46:30 matt Exp $	*/

/*-
 * Copyright (c) 2001, 2003, 2006, 2007, 2008 The NetBSD Foundation, Inc.
 * All rights reserved.
 *
 * This code is derived from software contributed to The NetBSD Foundation
 * by Nathan J. Williams, by Jason R. Thorpe, and by Andrew Doran.
 *
 * Redistribution and use in source and binary forms, with or without
 * modification, are permitted provided that the following conditions
 * are met:
 * 1. Redistributions of source code must retain the above copyright
 *    notice, this list of conditions and the following disclaimer.
 * 2. Redistributions in binary form must reproduce the above copyright
 *    notice, this list of conditions and the following disclaimer in the
 *    documentation and/or other materials provided with the distribution.
 *
 * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
 * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
 * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
 * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
 * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
 * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
 * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
 * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
 * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
 * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
 * POSSIBILITY OF SUCH DAMAGE.
 */

/*
 * To track threads waiting for mutexes to be released, we use lockless
 * lists built on atomic operations and memory barriers.
 *
 * A simple spinlock would be faster and make the code easier to
 * follow, but spinlocks are problematic in userspace.  If a thread is
 * preempted by the kernel while holding a spinlock, any other thread
 * attempting to acquire that spinlock will needlessly busy wait.
 *
 * There is no good way to know that the holding thread is no longer
 * running, nor to request a wake-up once it has begun running again. 
 * Of more concern, threads in the SCHED_FIFO class do not have a
 * limited time quantum and so could spin forever, preventing the
 * thread holding the spinlock from getting CPU time: it would never
 * be released.
 */

#include <sys/cdefs.h>
__RCSID("$NetBSD: pthread_mutex.c,v 1.51 2008/08/02 19:46:30 matt Exp $");

#include <sys/types.h>
#include <sys/lwpctl.h>
#include <sys/lock.h>

#include <errno.h>
#include <limits.h>
#include <stdlib.h>
#include <string.h>
#include <stdio.h>

#include "pthread.h"
#include "pthread_int.h"

#define	MUTEX_WAITERS_BIT		((uintptr_t)0x01)
#define	MUTEX_RECURSIVE_BIT		((uintptr_t)0x02)
#define	MUTEX_DEFERRED_BIT		((uintptr_t)0x04)
#define	MUTEX_THREAD			((uintptr_t)-16L)

#define	MUTEX_HAS_WAITERS(x)		((uintptr_t)(x) & MUTEX_WAITERS_BIT)
#define	MUTEX_RECURSIVE(x)		((uintptr_t)(x) & MUTEX_RECURSIVE_BIT)
#define	MUTEX_OWNER(x)			((uintptr_t)(x) & MUTEX_THREAD)

#if __GNUC_PREREQ__(3, 0)
#define	NOINLINE		__attribute ((noinline))
#else
#define	NOINLINE		/* nothing */
#endif

static void	pthread__mutex_wakeup(pthread_t, pthread_mutex_t *);
static int	pthread__mutex_lock_slow(pthread_mutex_t *);
static int	pthread__mutex_unlock_slow(pthread_mutex_t *);
static void	pthread__mutex_pause(void);

int		_pthread_mutex_held_np(pthread_mutex_t *);
pthread_t	_pthread_mutex_owner_np(pthread_mutex_t *);

__weak_alias(pthread_mutex_held_np,_pthread_mutex_held_np)
__weak_alias(pthread_mutex_owner_np,_pthread_mutex_owner_np)

__strong_alias(__libc_mutex_init,pthread_mutex_init)
__strong_alias(__libc_mutex_lock,pthread_mutex_lock)
__strong_alias(__libc_mutex_trylock,pthread_mutex_trylock)
__strong_alias(__libc_mutex_unlock,pthread_mutex_unlock)
__strong_alias(__libc_mutex_destroy,pthread_mutex_destroy)

__strong_alias(__libc_mutexattr_init,pthread_mutexattr_init)
__strong_alias(__libc_mutexattr_destroy,pthread_mutexattr_destroy)
__strong_alias(__libc_mutexattr_settype,pthread_mutexattr_settype)

__strong_alias(__libc_thr_once,pthread_once)

int
pthread_mutex_init(pthread_mutex_t *ptm, const pthread_mutexattr_t *attr)
{
	intptr_t type;

	if (attr == NULL)
		type = PTHREAD_MUTEX_NORMAL;
	else
		type = (intptr_t)attr->ptma_private;

	switch (type) {
	case PTHREAD_MUTEX_ERRORCHECK:
		__cpu_simple_lock_set(&ptm->ptm_errorcheck);
		ptm->ptm_owner = NULL;
		break;
	case PTHREAD_MUTEX_RECURSIVE:
		__cpu_simple_lock_clear(&ptm->ptm_errorcheck);
		ptm->ptm_owner = (void *)MUTEX_RECURSIVE_BIT;
		break;
	default:
		__cpu_simple_lock_clear(&ptm->ptm_errorcheck);
		ptm->ptm_owner = NULL;
		break;
	}

	ptm->ptm_magic = _PT_MUTEX_MAGIC;
	ptm->ptm_waiters = NULL;
	ptm->ptm_recursed = 0;

	return 0;
}


int
pthread_mutex_destroy(pthread_mutex_t *ptm)
{

	pthread__error(EINVAL, "Invalid mutex",
	    ptm->ptm_magic == _PT_MUTEX_MAGIC);
	pthread__error(EBUSY, "Destroying locked mutex",
	    MUTEX_OWNER(ptm->ptm_owner) == 0);

	ptm->ptm_magic = _PT_MUTEX_DEAD;
	return 0;
}

int
pthread_mutex_lock(pthread_mutex_t *ptm)
{
	pthread_t self;
	void *val;

	self = pthread__self();
	val = atomic_cas_ptr(&ptm->ptm_owner, NULL, self);
	if (__predict_true(val == NULL)) {
#ifndef PTHREAD__ATOMIC_IS_MEMBAR
		membar_enter();
#endif
		return 0;
	}
	return pthread__mutex_lock_slow(ptm);
}

/* We want function call overhead. */
NOINLINE static void
pthread__mutex_pause(void)
{

	pthread__smt_pause();
}

/*
 * Spin while the holder is running.  'lwpctl' gives us the true
 * status of the thread.  pt_blocking is set by libpthread in order
 * to cut out system call and kernel spinlock overhead on remote CPUs
 * (could represent many thousands of clock cycles).  pt_blocking also
 * makes this thread yield if the target is calling sched_yield().
 */
NOINLINE static void *
pthread__mutex_spin(pthread_mutex_t *ptm, pthread_t owner)
{
	pthread_t thread;
	unsigned int count, i;

	for (count = 2;; owner = ptm->ptm_owner) {
		thread = (pthread_t)MUTEX_OWNER(owner);
		if (thread == NULL)
			break;
		if (thread->pt_lwpctl->lc_curcpu == LWPCTL_CPU_NONE ||
		    thread->pt_blocking)
			break;
		if (count < 128) 
			count += count;
		for (i = count; i != 0; i--)
			pthread__mutex_pause();
	}

	return owner;
}

NOINLINE static int
pthread__mutex_lock_slow(pthread_mutex_t *ptm)
{
	void *waiters, *new, *owner, *next;
	pthread_t self;

	pthread__error(EINVAL, "Invalid mutex",
	    ptm->ptm_magic == _PT_MUTEX_MAGIC);

	owner = ptm->ptm_owner;
	self = pthread__self();

	/* Recursive or errorcheck? */
	if (MUTEX_OWNER(owner) == (uintptr_t)self) {
		if (MUTEX_RECURSIVE(owner)) {
			if (ptm->ptm_recursed == INT_MAX)
				return EAGAIN;
			ptm->ptm_recursed++;
			return 0;
		}
		if (__SIMPLELOCK_LOCKED_P(&ptm->ptm_errorcheck))
			return EDEADLK;
	}

	for (;; owner = ptm->ptm_owner) {
		/* Spin while the owner is running. */
		owner = pthread__mutex_spin(ptm, owner);

		/* If it has become free, try to acquire it again. */
		if (MUTEX_OWNER(owner) == 0) {
			do {
				new = (void *)
				    ((uintptr_t)self | (uintptr_t)owner);
				next = atomic_cas_ptr(&ptm->ptm_owner, owner,
				    new);
				if (next == owner) {
#ifndef PTHREAD__ATOMIC_IS_MEMBAR
					membar_enter();
#endif
					return 0;
				}
				owner = next;
			} while (MUTEX_OWNER(owner) == 0);
			/*
			 * We have lost the race to acquire the mutex.
			 * The new owner could be running on another
			 * CPU, in which case we should spin and avoid
			 * the overhead of blocking.
			 */
			continue;
		}

		/*
		 * Nope, still held.  Add thread to the list of waiters.
		 * Issue a memory barrier to ensure mutexwait/mutexnext
		 * are visible before we enter the waiters list.
		 */
		self->pt_mutexwait = 1;
		for (waiters = ptm->ptm_waiters;; waiters = next) {
			self->pt_mutexnext = waiters;
			membar_producer();
			next = atomic_cas_ptr(&ptm->ptm_waiters, waiters, self);
			if (next == waiters)
			    	break;
		}

		/*
		 * Set the waiters bit and block.
		 *
		 * Note that the mutex can become unlocked before we set
		 * the waiters bit.  If that happens it's not safe to sleep
		 * as we may never be awoken: we must remove the current
		 * thread from the waiters list and try again.
		 *
		 * Because we are doing this atomically, we can't remove
		 * one waiter: we must remove all waiters and awken them,
		 * then sleep in _lwp_park() until we have been awoken. 
		 *
		 * Issue a memory barrier to ensure that we are reading
		 * the value of ptm_owner/pt_mutexwait after we have entered
		 * the waiters list (the CAS itself must be atomic).
		 */
		membar_consumer();
		for (owner = ptm->ptm_owner;; owner = next) {
			if (MUTEX_HAS_WAITERS(owner))
				break;
			if (MUTEX_OWNER(owner) == 0) {
				pthread__mutex_wakeup(self, ptm);
				break;
			}
			new = (void *)((uintptr_t)owner | MUTEX_WAITERS_BIT);
			next = atomic_cas_ptr(&ptm->ptm_owner, owner, new);
			if (next == owner) {
				/*
				 * pthread_mutex_unlock() can do a
				 * non-interlocked CAS.  We cannot
				 * know if our attempt to set the
				 * waiters bit has succeeded while
				 * the holding thread is running.
				 * There are many assumptions; see
				 * sys/kern/kern_mutex.c for details.
				 * In short, we must spin if we see
				 * that the holder is running again.
				 */
				membar_sync();
				next = pthread__mutex_spin(ptm, owner);
			}
		}

		/*
		 * We may have been awoken by the current thread above,
		 * or will be awoken by the current holder of the mutex.
		 * The key requirement is that we must not proceed until
		 * told that we are no longer waiting (via pt_mutexwait
		 * being set to zero).  Otherwise it is unsafe to re-enter
		 * the thread onto the waiters list.
		 */
		while (self->pt_mutexwait) {
			self->pt_blocking++;
			(void)_lwp_park(NULL, self->pt_unpark,
			    __UNVOLATILE(&ptm->ptm_waiters),
			    __UNVOLATILE(&ptm->ptm_waiters));
			self->pt_unpark = 0;
			self->pt_blocking--;
			membar_sync();
		}
	}
}

int
pthread_mutex_trylock(pthread_mutex_t *ptm)
{
	pthread_t self;
	void *val, *new, *next;

	self = pthread__self();
	val = atomic_cas_ptr(&ptm->ptm_owner, NULL, self);
	if (__predict_true(val == NULL)) {
#ifndef PTHREAD__ATOMIC_IS_MEMBAR
		membar_enter();
#endif
		return 0;
	}

	if (MUTEX_RECURSIVE(val)) {
		if (MUTEX_OWNER(val) == 0) {
			new = (void *)((uintptr_t)self | (uintptr_t)val);
			next = atomic_cas_ptr(&ptm->ptm_owner, val, new);
			if (__predict_true(next == val)) {
#ifndef PTHREAD__ATOMIC_IS_MEMBAR
				membar_enter();
#endif
				return 0;
			}
		}
		if (MUTEX_OWNER(val) == (uintptr_t)self) {
			if (ptm->ptm_recursed == INT_MAX)
				return EAGAIN;
			ptm->ptm_recursed++;
			return 0;
		}
	}

	return EBUSY;
}

int
pthread_mutex_unlock(pthread_mutex_t *ptm)
{
	pthread_t self;
	void *value;

	/*
	 * Note this may be a non-interlocked CAS.  See lock_slow()
	 * above and sys/kern/kern_mutex.c for details.
	 */
#ifndef PTHREAD__ATOMIC_IS_MEMBAR
	membar_exit();
#endif
	self = pthread__self();
	value = atomic_cas_ptr_ni(&ptm->ptm_owner, self, NULL);
	if (__predict_true(value == self))
		return 0;
	return pthread__mutex_unlock_slow(ptm);
}

NOINLINE static int
pthread__mutex_unlock_slow(pthread_mutex_t *ptm)
{
	pthread_t self, owner, new;
	int weown, error, deferred;

	pthread__error(EINVAL, "Invalid mutex",
	    ptm->ptm_magic == _PT_MUTEX_MAGIC);

	self = pthread__self();
	owner = ptm->ptm_owner;
	weown = (MUTEX_OWNER(owner) == (uintptr_t)self);
	deferred = (int)((uintptr_t)owner & MUTEX_DEFERRED_BIT);
	error = 0;

	if (__SIMPLELOCK_LOCKED_P(&ptm->ptm_errorcheck)) {
		if (!weown) {
			error = EPERM;
			new = owner;
		} else {
			new = NULL;
		}
	} else if (MUTEX_RECURSIVE(owner)) {
		if (!weown) {
			error = EPERM;
			new = owner;
		} else if (ptm->ptm_recursed) {
			ptm->ptm_recursed--;
			new = owner;
		} else {
			new = (pthread_t)MUTEX_RECURSIVE_BIT;
		}
	} else {
		pthread__error(EPERM,
		    "Unlocking unlocked mutex", (owner != NULL));
		pthread__error(EPERM,
		    "Unlocking mutex owned by another thread", weown);
		new = NULL;
	}

	/*
	 * Release the mutex.  If there appear to be waiters, then
	 * wake them up.
	 */
	if (new != owner) {
		owner = atomic_swap_ptr(&ptm->ptm_owner, new);
		if (MUTEX_HAS_WAITERS(owner) != 0) {
			pthread__mutex_wakeup(self, ptm);
			return 0;
		}
	}

	/*
	 * There were no waiters, but we may have deferred waking
	 * other threads until mutex unlock - we must wake them now.
	 */
	if (!deferred)
		return error;

	if (self->pt_nwaiters == 1) {
		/*
		 * If the calling thread is about to block, defer
		 * unparking the target until _lwp_park() is called.
		 */
		if (self->pt_willpark && self->pt_unpark == 0) {
			self->pt_unpark = self->pt_waiters[0];
		} else {
			(void)_lwp_unpark(self->pt_waiters[0],
			    __UNVOLATILE(&ptm->ptm_waiters));
		}
	} else {
		(void)_lwp_unpark_all(self->pt_waiters, self->pt_nwaiters,
		    __UNVOLATILE(&ptm->ptm_waiters));
	}
	self->pt_nwaiters = 0;

	return error;
}

static void
pthread__mutex_wakeup(pthread_t self, pthread_mutex_t *ptm)
{
	pthread_t thread, next;
	ssize_t n, rv;

	/*
	 * Take ownership of the current set of waiters.  No
	 * need for a memory barrier following this, all loads
	 * are dependent upon 'thread'.
	 */
	thread = atomic_swap_ptr(&ptm->ptm_waiters, NULL);

	for (;;) {
		/*
		 * Pull waiters from the queue and add to our list.
		 * Use a memory barrier to ensure that we safely
		 * read the value of pt_mutexnext before 'thread'
		 * sees pt_mutexwait being cleared.
		 */
		for (n = self->pt_nwaiters, self->pt_nwaiters = 0;
		    n < pthread__unpark_max && thread != NULL;
		    thread = next) {
		    	next = thread->pt_mutexnext;
		    	if (thread != self) {
				self->pt_waiters[n++] = thread->pt_lid;
				membar_sync();
			}
			thread->pt_mutexwait = 0;
			/* No longer safe to touch 'thread' */
		}

		switch (n) {
		case 0:
			return;
		case 1:
			/*
			 * If the calling thread is about to block,
			 * defer unparking the target until _lwp_park()
			 * is called.
			 */
			if (self->pt_willpark && self->pt_unpark == 0) {
				self->pt_unpark = self->pt_waiters[0];
				return;
			}
			rv = (ssize_t)_lwp_unpark(self->pt_waiters[0],
			    __UNVOLATILE(&ptm->ptm_waiters));
			if (rv != 0 && errno != EALREADY && errno != EINTR &&
			    errno != ESRCH) {
				pthread__errorfunc(__FILE__, __LINE__,
				    __func__, "_lwp_unpark failed");
			}
			return;
		default:
			rv = _lwp_unpark_all(self->pt_waiters, (size_t)n,
			    __UNVOLATILE(&ptm->ptm_waiters));
			if (rv != 0 && errno != EINTR) {
				pthread__errorfunc(__FILE__, __LINE__,
				    __func__, "_lwp_unpark_all failed");
			}
			break;
		}
	}
}
int
pthread_mutexattr_init(pthread_mutexattr_t *attr)
{

	attr->ptma_magic = _PT_MUTEXATTR_MAGIC;
	attr->ptma_private = (void *)PTHREAD_MUTEX_DEFAULT;
	return 0;
}

int
pthread_mutexattr_destroy(pthread_mutexattr_t *attr)
{

	pthread__error(EINVAL, "Invalid mutex attribute",
	    attr->ptma_magic == _PT_MUTEXATTR_MAGIC);

	return 0;
}


int
pthread_mutexattr_gettype(const pthread_mutexattr_t *attr, int *typep)
{

	pthread__error(EINVAL, "Invalid mutex attribute",
	    attr->ptma_magic == _PT_MUTEXATTR_MAGIC);

	*typep = (int)(intptr_t)attr->ptma_private;
	return 0;
}


int
pthread_mutexattr_settype(pthread_mutexattr_t *attr, int type)
{

	pthread__error(EINVAL, "Invalid mutex attribute",
	    attr->ptma_magic == _PT_MUTEXATTR_MAGIC);

	switch (type) {
	case PTHREAD_MUTEX_NORMAL:
	case PTHREAD_MUTEX_ERRORCHECK:
	case PTHREAD_MUTEX_RECURSIVE:
		attr->ptma_private = (void *)(intptr_t)type;
		return 0;
	default:
		return EINVAL;
	}
}


static void
once_cleanup(void *closure)
{

       pthread_mutex_unlock((pthread_mutex_t *)closure);
}


int
pthread_once(pthread_once_t *once_control, void (*routine)(void))
{

	if (once_control->pto_done == 0) {
		pthread_mutex_lock(&once_control->pto_mutex);
		pthread_cleanup_push(&once_cleanup, &once_control->pto_mutex);
		if (once_control->pto_done == 0) {
			routine();
			once_control->pto_done = 1;
		}
		pthread_cleanup_pop(1);
	}

	return 0;
}

void
pthread__mutex_deferwake(pthread_t self, pthread_mutex_t *ptm)
{

	if (__predict_false(ptm == NULL ||
	    MUTEX_OWNER(ptm->ptm_owner) != (uintptr_t)self)) {
	    	(void)_lwp_unpark_all(self->pt_waiters, self->pt_nwaiters,
	    	    __UNVOLATILE(&ptm->ptm_waiters));
	    	self->pt_nwaiters = 0;
	} else {
		atomic_or_ulong((volatile unsigned long *)
		    (uintptr_t)&ptm->ptm_owner,
		    (unsigned long)MUTEX_DEFERRED_BIT);
	}
}

int
_pthread_mutex_held_np(pthread_mutex_t *ptm)
{

	return MUTEX_OWNER(ptm->ptm_owner) == (uintptr_t)pthread__self();
}

pthread_t
_pthread_mutex_owner_np(pthread_mutex_t *ptm)
{

	return (pthread_t)MUTEX_OWNER(ptm->ptm_owner);
}