xref: /netbsd-src/lib/libpthread/pthread_rwlock.c (revision 267197ec1eebfcb9810ea27a89625b6ddf68e3e7)
1 /*	$NetBSD: pthread_rwlock.c,v 1.27 2008/02/10 18:50:55 ad Exp $ */
2 
3 /*-
4  * Copyright (c) 2002, 2006, 2007, 2008 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to The NetBSD Foundation
8  * by Nathan J. Williams, by Jason R. Thorpe, and by Andrew Doran.
9  *
10  * Redistribution and use in source and binary forms, with or without
11  * modification, are permitted provided that the following conditions
12  * are met:
13  * 1. Redistributions of source code must retain the above copyright
14  *    notice, this list of conditions and the following disclaimer.
15  * 2. Redistributions in binary form must reproduce the above copyright
16  *    notice, this list of conditions and the following disclaimer in the
17  *    documentation and/or other materials provided with the distribution.
18  * 3. All advertising materials mentioning features or use of this software
19  *    must display the following acknowledgement:
20  *        This product includes software developed by the NetBSD
21  *        Foundation, Inc. and its contributors.
22  * 4. Neither the name of The NetBSD Foundation nor the names of its
23  *    contributors may be used to endorse or promote products derived
24  *    from this software without specific prior written permission.
25  *
26  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
27  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
28  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
29  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
30  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
31  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
32  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
33  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
34  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
35  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
36  * POSSIBILITY OF SUCH DAMAGE.
37  */
38 
39 #include <sys/cdefs.h>
40 __RCSID("$NetBSD: pthread_rwlock.c,v 1.27 2008/02/10 18:50:55 ad Exp $");
41 
42 #include <errno.h>
43 #include <stddef.h>
44 
45 #include "pthread.h"
46 #include "pthread_int.h"
47 
48 #define	_RW_LOCKED		0
49 #define	_RW_WANT_WRITE		1
50 #define	_RW_WANT_READ		2
51 
52 static int pthread__rwlock_wrlock(pthread_rwlock_t *, const struct timespec *);
53 static int pthread__rwlock_rdlock(pthread_rwlock_t *, const struct timespec *);
54 static void pthread__rwlock_early(void *);
55 
56 int	_pthread_rwlock_held_np(pthread_rwlock_t *);
57 int	_pthread_rwlock_rdheld_np(pthread_rwlock_t *);
58 int	_pthread_rwlock_wrheld_np(pthread_rwlock_t *);
59 
60 #ifndef lint
61 __weak_alias(pthread_rwlock_held_np,_pthread_rwlock_held_np);
62 __weak_alias(pthread_rwlock_rdheld_np,_pthread_rwlock_rdheld_np);
63 __weak_alias(pthread_rwlock_wrheld_np,_pthread_rwlock_wrheld_np);
64 #endif
65 
66 __strong_alias(__libc_rwlock_init,pthread_rwlock_init)
67 __strong_alias(__libc_rwlock_rdlock,pthread_rwlock_rdlock)
68 __strong_alias(__libc_rwlock_wrlock,pthread_rwlock_wrlock)
69 __strong_alias(__libc_rwlock_tryrdlock,pthread_rwlock_tryrdlock)
70 __strong_alias(__libc_rwlock_trywrlock,pthread_rwlock_trywrlock)
71 __strong_alias(__libc_rwlock_unlock,pthread_rwlock_unlock)
72 __strong_alias(__libc_rwlock_destroy,pthread_rwlock_destroy)
73 
74 static inline uintptr_t
75 rw_cas(pthread_rwlock_t *ptr, uintptr_t o, uintptr_t n)
76 {
77 
78 	return (uintptr_t)atomic_cas_ptr(&ptr->ptr_owner, (void *)o,
79 	    (void *)n);
80 }
81 
82 int
83 pthread_rwlock_init(pthread_rwlock_t *ptr,
84 	    const pthread_rwlockattr_t *attr)
85 {
86 
87 	if (attr && (attr->ptra_magic != _PT_RWLOCKATTR_MAGIC))
88 		return EINVAL;
89 	ptr->ptr_magic = _PT_RWLOCK_MAGIC;
90 	pthread_lockinit(&ptr->ptr_interlock);
91 	PTQ_INIT(&ptr->ptr_rblocked);
92 	PTQ_INIT(&ptr->ptr_wblocked);
93 	ptr->ptr_nreaders = 0;
94 	ptr->ptr_owner = NULL;
95 
96 	return 0;
97 }
98 
99 
100 int
101 pthread_rwlock_destroy(pthread_rwlock_t *ptr)
102 {
103 
104 	if ((ptr->ptr_magic != _PT_RWLOCK_MAGIC) ||
105 	    (!PTQ_EMPTY(&ptr->ptr_rblocked)) ||
106 	    (!PTQ_EMPTY(&ptr->ptr_wblocked)) ||
107 	    (ptr->ptr_nreaders != 0) ||
108 	    (ptr->ptr_owner != NULL))
109 		return EINVAL;
110 	ptr->ptr_magic = _PT_RWLOCK_DEAD;
111 
112 	return 0;
113 }
114 
115 static int
116 pthread__rwlock_rdlock(pthread_rwlock_t *ptr, const struct timespec *ts)
117 {
118 	uintptr_t owner, next;
119 	pthread_t self;
120 	int error;
121 
122 	self = pthread__self();
123 
124 #ifdef ERRORCHECK
125 	if (ptr->ptr_magic != _PT_RWLOCK_MAGIC)
126 		return EINVAL;
127 #endif
128 
129 	for (owner = (uintptr_t)ptr->ptr_owner;; owner = next) {
130 		/*
131 		 * Read the lock owner field.  If the need-to-wait
132 		 * indicator is clear, then try to acquire the lock.
133 		 */
134 		if ((owner & (RW_WRITE_LOCKED | RW_WRITE_WANTED)) == 0) {
135 			next = rw_cas(ptr, owner, owner + RW_READ_INCR);
136 			if (owner == next) {
137 				/* Got it! */
138 #ifndef PTHREAD__ATOMIC_IS_MEMBAR
139 				membar_enter();
140 #endif
141 				return 0;
142 			}
143 
144 			/*
145 			 * Didn't get it -- spin around again (we'll
146 			 * probably sleep on the next iteration).
147 			 */
148 			continue;
149 		}
150 
151 		if ((owner & RW_THREAD) == (uintptr_t)self)
152 			return EDEADLK;
153 
154 		/*
155 		 * Grab the interlock.  Once we have that, we
156 		 * can adjust the waiter bits and sleep queue.
157 		 */
158 		pthread__spinlock(self, &ptr->ptr_interlock);
159 
160 		/*
161 		 * Mark the rwlock as having waiters.  If the set fails,
162 		 * then we may not need to sleep and should spin again.
163 		 */
164 		next = rw_cas(ptr, owner, owner | RW_HAS_WAITERS);
165 		if (owner != next) {
166 			pthread__spinunlock(self, &ptr->ptr_interlock);
167 			continue;
168 		}
169 
170 		/* The waiters bit is set - it's safe to sleep. */
171 	    	PTQ_INSERT_HEAD(&ptr->ptr_rblocked, self, pt_sleep);
172 	    	ptr->ptr_nreaders++;
173 		self->pt_rwlocked = _RW_WANT_READ;
174 		self->pt_sleeponq = 1;
175 		self->pt_sleepobj = &ptr->ptr_rblocked;
176 		self->pt_early = pthread__rwlock_early;
177 		pthread__spinunlock(self, &ptr->ptr_interlock);
178 
179 		error = pthread__park(self, &ptr->ptr_interlock,
180 		    &ptr->ptr_rblocked, ts, 0, &ptr->ptr_rblocked);
181 
182 		/* Did we get the lock? */
183 		if (self->pt_rwlocked == _RW_LOCKED) {
184 #ifndef PTHREAD__ATOMIC_IS_MEMBAR
185 			membar_enter();
186 #endif
187 			return 0;
188 		}
189 		if (error != 0)
190 			return error;
191 
192 		pthread__errorfunc(__FILE__, __LINE__, __func__,
193 		    "direct handoff failure");
194 	}
195 }
196 
197 
198 int
199 pthread_rwlock_tryrdlock(pthread_rwlock_t *ptr)
200 {
201 	uintptr_t owner, next;
202 
203 #ifdef ERRORCHECK
204 	if (ptr->ptr_magic != _PT_RWLOCK_MAGIC)
205 		return EINVAL;
206 #endif
207 
208 	/*
209 	 * Don't get a readlock if there is a writer or if there are waiting
210 	 * writers; i.e. prefer writers to readers. This strategy is dictated
211 	 * by SUSv3.
212 	 */
213 	for (owner = (uintptr_t)ptr->ptr_owner;; owner = next) {
214 		if ((owner & (RW_WRITE_LOCKED | RW_WRITE_WANTED)) != 0)
215 			return EBUSY;
216 		next = rw_cas(ptr, owner, owner + RW_READ_INCR);
217 		if (owner == next) {
218 			/* Got it! */
219 #ifndef PTHREAD__ATOMIC_IS_MEMBAR
220 			membar_enter();
221 #endif
222 			return 0;
223 		}
224 	}
225 }
226 
227 static int
228 pthread__rwlock_wrlock(pthread_rwlock_t *ptr, const struct timespec *ts)
229 {
230 	uintptr_t owner, next;
231 	pthread_t self;
232 	int error;
233 
234 	self = pthread__self();
235 
236 #ifdef ERRORCHECK
237 	if (ptr->ptr_magic != _PT_RWLOCK_MAGIC)
238 		return EINVAL;
239 #endif
240 
241 	for (owner = (uintptr_t)ptr->ptr_owner;; owner = next) {
242 		/*
243 		 * Read the lock owner field.  If the need-to-wait
244 		 * indicator is clear, then try to acquire the lock.
245 		 */
246 		if ((owner & RW_THREAD) == 0) {
247 			next = rw_cas(ptr, owner,
248 			    (uintptr_t)self | RW_WRITE_LOCKED);
249 			if (owner == next) {
250 				/* Got it! */
251 #ifndef PTHREAD__ATOMIC_IS_MEMBAR
252 				membar_enter();
253 #endif
254 				return 0;
255 			}
256 
257 			/*
258 			 * Didn't get it -- spin around again (we'll
259 			 * probably sleep on the next iteration).
260 			 */
261 			continue;
262 		}
263 
264 		if ((owner & RW_THREAD) == (uintptr_t)self)
265 			return EDEADLK;
266 
267 		/*
268 		 * Grab the interlock.  Once we have that, we
269 		 * can adjust the waiter bits and sleep queue.
270 		 */
271 		pthread__spinlock(self, &ptr->ptr_interlock);
272 
273 		/*
274 		 * Mark the rwlock as having waiters.  If the set fails,
275 		 * then we may not need to sleep and should spin again.
276 		 */
277 		next = rw_cas(ptr, owner,
278 		    owner | RW_HAS_WAITERS | RW_WRITE_WANTED);
279 		if (owner != next) {
280 			pthread__spinunlock(self, &ptr->ptr_interlock);
281 			continue;
282 		}
283 
284 		/* The waiters bit is set - it's safe to sleep. */
285 	    	PTQ_INSERT_TAIL(&ptr->ptr_wblocked, self, pt_sleep);
286 		self->pt_rwlocked = _RW_WANT_WRITE;
287 		self->pt_sleeponq = 1;
288 		self->pt_sleepobj = &ptr->ptr_wblocked;
289 		self->pt_early = pthread__rwlock_early;
290 		pthread__spinunlock(self, &ptr->ptr_interlock);
291 
292 		error = pthread__park(self, &ptr->ptr_interlock,
293 		    &ptr->ptr_wblocked, ts, 0, &ptr->ptr_wblocked);
294 
295 		/* Did we get the lock? */
296 		if (self->pt_rwlocked == _RW_LOCKED) {
297 #ifndef PTHREAD__ATOMIC_IS_MEMBAR
298 			membar_enter();
299 #endif
300 			return 0;
301 		}
302 		if (error != 0)
303 			return error;
304 
305 		pthread__errorfunc(__FILE__, __LINE__, __func__,
306 		    "direct handoff failure");
307 	}
308 }
309 
310 
311 int
312 pthread_rwlock_trywrlock(pthread_rwlock_t *ptr)
313 {
314 	uintptr_t owner, next;
315 	pthread_t self;
316 
317 #ifdef ERRORCHECK
318 	if (ptr->ptr_magic != _PT_RWLOCK_MAGIC)
319 		return EINVAL;
320 #endif
321 
322 	self = pthread__self();
323 
324 	for (owner = (uintptr_t)ptr->ptr_owner;; owner = next) {
325 		if (owner != 0)
326 			return EBUSY;
327 		next = rw_cas(ptr, owner, (uintptr_t)self | RW_WRITE_LOCKED);
328 		if (owner == next) {
329 			/* Got it! */
330 #ifndef PTHREAD__ATOMIC_IS_MEMBAR
331 			membar_enter();
332 #endif
333 			return 0;
334 		}
335 	}
336 }
337 
338 int
339 pthread_rwlock_rdlock(pthread_rwlock_t *ptr)
340 {
341 
342 	return pthread__rwlock_rdlock(ptr, NULL);
343 }
344 
345 int
346 pthread_rwlock_timedrdlock(pthread_rwlock_t *ptr,
347 			   const struct timespec *abs_timeout)
348 {
349 
350 	if (abs_timeout == NULL)
351 		return EINVAL;
352 	if ((abs_timeout->tv_nsec >= 1000000000) ||
353 	    (abs_timeout->tv_nsec < 0) ||
354 	    (abs_timeout->tv_sec < 0))
355 		return EINVAL;
356 
357 	return pthread__rwlock_rdlock(ptr, abs_timeout);
358 }
359 
360 int
361 pthread_rwlock_wrlock(pthread_rwlock_t *ptr)
362 {
363 
364 	return pthread__rwlock_wrlock(ptr, NULL);
365 }
366 
367 int
368 pthread_rwlock_timedwrlock(pthread_rwlock_t *ptr,
369 			   const struct timespec *abs_timeout)
370 {
371 
372 	if (abs_timeout == NULL)
373 		return EINVAL;
374 	if ((abs_timeout->tv_nsec >= 1000000000) ||
375 	    (abs_timeout->tv_nsec < 0) ||
376 	    (abs_timeout->tv_sec < 0))
377 		return EINVAL;
378 
379 	return pthread__rwlock_wrlock(ptr, abs_timeout);
380 }
381 
382 
383 int
384 pthread_rwlock_unlock(pthread_rwlock_t *ptr)
385 {
386 	uintptr_t owner, decr, new, next;
387 	pthread_t self, thread;
388 
389 #ifdef ERRORCHECK
390 	if ((ptr == NULL) || (ptr->ptr_magic != _PT_RWLOCK_MAGIC))
391 		return EINVAL;
392 #endif
393 
394 #ifndef PTHREAD__ATOMIC_IS_MEMBAR
395 	membar_exit();
396 #endif
397 
398 	/*
399 	 * Since we used an add operation to set the required lock
400 	 * bits, we can use a subtract to clear them, which makes
401 	 * the read-release and write-release path similar.
402 	 */
403 	self = pthread__self();
404 	owner = (uintptr_t)ptr->ptr_owner;
405 	if ((owner & RW_WRITE_LOCKED) != 0) {
406 		decr = (uintptr_t)self | RW_WRITE_LOCKED;
407 		if ((owner & RW_THREAD) != (uintptr_t)self) {
408 			return EPERM;
409 		}
410 	} else {
411 		decr = RW_READ_INCR;
412 		if (owner == 0) {
413 			return EPERM;
414 		}
415 	}
416 
417 	for (;; owner = next) {
418 		/*
419 		 * Compute what we expect the new value of the lock to be.
420 		 * Only proceed to do direct handoff if there are waiters,
421 		 * and if the lock would become unowned.
422 		 */
423 		new = (owner - decr);
424 		if ((new & (RW_THREAD | RW_HAS_WAITERS)) != RW_HAS_WAITERS) {
425 			next = rw_cas(ptr, owner, new);
426 			if (owner == next) {
427 				/* Released! */
428 				return 0;
429 			}
430 			continue;
431 		}
432 
433 		/*
434 		 * Grab the interlock.  Once we have that, we can adjust
435 		 * the waiter bits.  We must check to see if there are
436 		 * still waiters before proceeding.
437 		 */
438 		pthread__spinlock(self, &ptr->ptr_interlock);
439 		owner = (uintptr_t)ptr->ptr_owner;
440 		if ((owner & RW_HAS_WAITERS) == 0) {
441 			pthread__spinunlock(self, &ptr->ptr_interlock);
442 			next = owner;
443 			continue;
444 		}
445 
446 		/*
447 		 * Give the lock away.  SUSv3 dictates that we must give
448 		 * preference to writers.
449 		 */
450 		if ((thread = PTQ_FIRST(&ptr->ptr_wblocked)) != NULL) {
451 			new = (uintptr_t)thread | RW_WRITE_LOCKED;
452 
453 			if (PTQ_NEXT(thread, pt_sleep) != NULL)
454 				new |= RW_HAS_WAITERS | RW_WRITE_WANTED;
455 			else if (ptr->ptr_nreaders != 0)
456 				new |= RW_HAS_WAITERS;
457 
458 			/*
459 			 * Set in the new value.  The lock becomes owned
460 			 * by the writer that we are about to wake.
461 			 */
462 			(void)atomic_swap_ptr(&ptr->ptr_owner, (void *)new);
463 
464 			/* Wake the writer. */
465 			PTQ_REMOVE(&ptr->ptr_wblocked, thread, pt_sleep);
466 			thread->pt_rwlocked = _RW_LOCKED;
467 			pthread__unpark(self, &ptr->ptr_interlock,
468 			    &ptr->ptr_wblocked, thread);
469 		} else {
470 			new = 0;
471 			PTQ_FOREACH(thread, &ptr->ptr_rblocked, pt_sleep) {
472 				/*
473 				 * May have already been handed the lock,
474 				 * since pthread__unpark_all() can release
475 				 * our interlock before awakening all
476 				 * threads.
477 				 */
478 				if (thread->pt_sleepobj == NULL)
479 					continue;
480 				new += RW_READ_INCR;
481 				thread->pt_rwlocked = _RW_LOCKED;
482 			}
483 
484 			/*
485 			 * Set in the new value.  The lock becomes owned
486 			 * by the readers that we are about to wake.
487 			 */
488 			(void)atomic_swap_ptr(&ptr->ptr_owner, (void *)new);
489 
490 			/* Wake up all sleeping readers. */
491 			ptr->ptr_nreaders = 0;
492 			pthread__unpark_all(self, &ptr->ptr_interlock,
493 			    &ptr->ptr_rblocked);
494 		}
495 
496 		return 0;
497 	}
498 }
499 
500 /*
501  * Called when a timedlock awakens early to adjust the waiter bits.
502  * The rwlock's interlock is held on entry, and the caller has been
503  * removed from the waiters lists.
504  */
505 static void
506 pthread__rwlock_early(void *obj)
507 {
508 	uintptr_t owner, set, new, next;
509 	pthread_rwlock_t *ptr;
510 	pthread_t self;
511 	u_int off;
512 
513 	self = pthread__self();
514 
515 	switch (self->pt_rwlocked) {
516 	case _RW_WANT_READ:
517 		off = offsetof(pthread_rwlock_t, ptr_rblocked);
518 		break;
519 	case _RW_WANT_WRITE:
520 		off = offsetof(pthread_rwlock_t, ptr_wblocked);
521 		break;
522 	default:
523 		pthread__errorfunc(__FILE__, __LINE__, __func__,
524 		    "bad value of pt_rwlocked");
525 		off = 0;
526 		/* NOTREACHED */
527 		break;
528 	}
529 
530 	/* LINTED mind your own business */
531 	ptr = (pthread_rwlock_t *)((uint8_t *)obj - off);
532 	owner = (uintptr_t)ptr->ptr_owner;
533 
534 	if ((owner & RW_THREAD) == 0) {
535 		pthread__errorfunc(__FILE__, __LINE__, __func__,
536 		    "lock not held");
537 	}
538 
539 	if (!PTQ_EMPTY(&ptr->ptr_wblocked))
540 		set = RW_HAS_WAITERS | RW_WRITE_WANTED;
541 	else if (ptr->ptr_nreaders != 0)
542 		set = RW_HAS_WAITERS;
543 	else
544 		set = 0;
545 
546 	for (;; owner = next) {
547 		new = (owner & ~(RW_HAS_WAITERS | RW_WRITE_WANTED)) | set;
548 		next = rw_cas(ptr, owner, new);
549 		if (owner == next)
550 			break;
551 	}
552 }
553 
554 int
555 _pthread_rwlock_held_np(pthread_rwlock_t *ptr)
556 {
557 	uintptr_t owner = (uintptr_t)ptr->ptr_owner;
558 
559 	return (owner & RW_THREAD) != 0;
560 }
561 
562 int
563 _pthread_rwlock_rdheld_np(pthread_rwlock_t *ptr)
564 {
565 	uintptr_t owner = (uintptr_t)ptr->ptr_owner;
566 
567 	return (owner & RW_THREAD) != 0 && (owner & RW_WRITE_LOCKED) == 0;
568 }
569 
570 int
571 _pthread_rwlock_wrheld_np(pthread_rwlock_t *ptr)
572 {
573 	uintptr_t owner = (uintptr_t)ptr->ptr_owner;
574 
575 	return (owner & (RW_THREAD | RW_WRITE_LOCKED)) ==
576 	    ((uintptr_t)pthread__self() | RW_WRITE_LOCKED);
577 }
578 
579 int
580 pthread_rwlockattr_init(pthread_rwlockattr_t *attr)
581 {
582 
583 	if (attr == NULL)
584 		return EINVAL;
585 	attr->ptra_magic = _PT_RWLOCKATTR_MAGIC;
586 
587 	return 0;
588 }
589 
590 
591 int
592 pthread_rwlockattr_destroy(pthread_rwlockattr_t *attr)
593 {
594 
595 	if ((attr == NULL) ||
596 	    (attr->ptra_magic != _PT_RWLOCKATTR_MAGIC))
597 		return EINVAL;
598 	attr->ptra_magic = _PT_RWLOCKATTR_DEAD;
599 
600 	return 0;
601 }
602