xref: /netbsd-src/sys/kern/kern_condvar.c (revision 6cf6fe02a981b55727c49c3d37b0d8191a98c0ee)
1 /*	$NetBSD: kern_condvar.c,v 1.34 2013/10/25 15:51:36 martin Exp $	*/
2 
3 /*-
4  * Copyright (c) 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 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  *
19  * THIS SOFTWARE IS PROVIDED BY THE NETBSD FOUNDATION, INC. AND CONTRIBUTORS
20  * ``AS IS'' AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED
21  * TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR
22  * PURPOSE ARE DISCLAIMED.  IN NO EVENT SHALL THE FOUNDATION OR CONTRIBUTORS
23  * BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
24  * CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
25  * SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS
26  * INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN
27  * CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE)
28  * ARISING IN ANY WAY OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE
29  * POSSIBILITY OF SUCH DAMAGE.
30  */
31 
32 /*
33  * Kernel condition variable implementation.
34  */
35 
36 #include <sys/cdefs.h>
37 __KERNEL_RCSID(0, "$NetBSD: kern_condvar.c,v 1.34 2013/10/25 15:51:36 martin Exp $");
38 
39 #include <sys/param.h>
40 #include <sys/proc.h>
41 #include <sys/sched.h>
42 #include <sys/systm.h>
43 #include <sys/condvar.h>
44 #include <sys/sleepq.h>
45 #include <sys/lockdebug.h>
46 #include <sys/cpu.h>
47 
48 /*
49  * Accessors for the private contents of the kcondvar_t data type.
50  *
51  *	cv_opaque[0]	sleepq...
52  *	cv_opaque[1]	...pointers
53  *	cv_opaque[2]	description for ps(1)
54  *
55  * cv_opaque[0..1] is protected by the interlock passed to cv_wait() (enqueue
56  * only), and the sleep queue lock acquired with sleeptab_lookup() (enqueue
57  * and dequeue).
58  *
59  * cv_opaque[2] (the wmesg) is static and does not change throughout the life
60  * of the CV.
61  */
62 #define	CV_SLEEPQ(cv)		((sleepq_t *)(cv)->cv_opaque)
63 #define	CV_WMESG(cv)		((const char *)(cv)->cv_opaque[2])
64 #define	CV_SET_WMESG(cv, v) 	(cv)->cv_opaque[2] = __UNCONST(v)
65 
66 #define	CV_DEBUG_P(cv)	(CV_WMESG(cv) != nodebug)
67 #define	CV_RA		((uintptr_t)__builtin_return_address(0))
68 
69 static void	cv_unsleep(lwp_t *, bool);
70 static void	cv_wakeup_one(kcondvar_t *);
71 static void	cv_wakeup_all(kcondvar_t *);
72 
73 static syncobj_t cv_syncobj = {
74 	SOBJ_SLEEPQ_SORTED,
75 	cv_unsleep,
76 	sleepq_changepri,
77 	sleepq_lendpri,
78 	syncobj_noowner,
79 };
80 
81 lockops_t cv_lockops = {
82 	"Condition variable",
83 	LOCKOPS_CV,
84 	NULL
85 };
86 
87 static const char deadcv[] = "deadcv";
88 #ifdef LOCKDEBUG
89 static const char nodebug[] = "nodebug";
90 #endif
91 
92 /*
93  * cv_init:
94  *
95  *	Initialize a condition variable for use.
96  */
97 void
98 cv_init(kcondvar_t *cv, const char *wmesg)
99 {
100 #ifdef LOCKDEBUG
101 	bool dodebug;
102 
103 	dodebug = LOCKDEBUG_ALLOC(cv, &cv_lockops,
104 	    (uintptr_t)__builtin_return_address(0));
105 	if (!dodebug) {
106 		/* XXX This will break vfs_lockf. */
107 		wmesg = nodebug;
108 	}
109 #endif
110 	KASSERT(wmesg != NULL);
111 	CV_SET_WMESG(cv, wmesg);
112 	sleepq_init(CV_SLEEPQ(cv));
113 }
114 
115 /*
116  * cv_destroy:
117  *
118  *	Tear down a condition variable.
119  */
120 void
121 cv_destroy(kcondvar_t *cv)
122 {
123 
124 	LOCKDEBUG_FREE(CV_DEBUG_P(cv), cv);
125 #ifdef DIAGNOSTIC
126 	KASSERT(cv_is_valid(cv));
127 	CV_SET_WMESG(cv, deadcv);
128 #endif
129 }
130 
131 /*
132  * cv_enter:
133  *
134  *	Look up and lock the sleep queue corresponding to the given
135  *	condition variable, and increment the number of waiters.
136  */
137 static inline void
138 cv_enter(kcondvar_t *cv, kmutex_t *mtx, lwp_t *l)
139 {
140 	sleepq_t *sq;
141 	kmutex_t *mp;
142 
143 	KASSERT(cv_is_valid(cv));
144 	KASSERT(!cpu_intr_p());
145 	KASSERT((l->l_pflag & LP_INTR) == 0 || panicstr != NULL);
146 
147 	LOCKDEBUG_LOCKED(CV_DEBUG_P(cv), cv, mtx, CV_RA, 0);
148 
149 	l->l_kpriority = true;
150 	mp = sleepq_hashlock(cv);
151 	sq = CV_SLEEPQ(cv);
152 	sleepq_enter(sq, l, mp);
153 	sleepq_enqueue(sq, cv, CV_WMESG(cv), &cv_syncobj);
154 	mutex_exit(mtx);
155 	KASSERT(cv_has_waiters(cv));
156 }
157 
158 /*
159  * cv_exit:
160  *
161  *	After resuming execution, check to see if we have been restarted
162  *	as a result of cv_signal().  If we have, but cannot take the
163  *	wakeup (because of eg a pending Unix signal or timeout) then try
164  *	to ensure that another LWP sees it.  This is necessary because
165  *	there may be multiple waiters, and at least one should take the
166  *	wakeup if possible.
167  */
168 static inline int
169 cv_exit(kcondvar_t *cv, kmutex_t *mtx, lwp_t *l, const int error)
170 {
171 
172 	mutex_enter(mtx);
173 	if (__predict_false(error != 0))
174 		cv_signal(cv);
175 
176 	LOCKDEBUG_UNLOCKED(CV_DEBUG_P(cv), cv, CV_RA, 0);
177 	KASSERT(cv_is_valid(cv));
178 
179 	return error;
180 }
181 
182 /*
183  * cv_unsleep:
184  *
185  *	Remove an LWP from the condition variable and sleep queue.  This
186  *	is called when the LWP has not been awoken normally but instead
187  *	interrupted: for example, when a signal is received.  Must be
188  *	called with the LWP locked, and must return it unlocked.
189  */
190 static void
191 cv_unsleep(lwp_t *l, bool cleanup)
192 {
193 	kcondvar_t *cv __diagused;
194 
195 	cv = (kcondvar_t *)(uintptr_t)l->l_wchan;
196 
197 	KASSERT(l->l_wchan == (wchan_t)cv);
198 	KASSERT(l->l_sleepq == CV_SLEEPQ(cv));
199 	KASSERT(cv_is_valid(cv));
200 	KASSERT(cv_has_waiters(cv));
201 
202 	sleepq_unsleep(l, cleanup);
203 }
204 
205 /*
206  * cv_wait:
207  *
208  *	Wait non-interruptably on a condition variable until awoken.
209  */
210 void
211 cv_wait(kcondvar_t *cv, kmutex_t *mtx)
212 {
213 	lwp_t *l = curlwp;
214 
215 	KASSERT(mutex_owned(mtx));
216 
217 	cv_enter(cv, mtx, l);
218 	(void)sleepq_block(0, false);
219 	(void)cv_exit(cv, mtx, l, 0);
220 }
221 
222 /*
223  * cv_wait_sig:
224  *
225  *	Wait on a condition variable until a awoken or a signal is received.
226  *	Will also return early if the process is exiting.  Returns zero if
227  *	awoken normally, ERESTART if a signal was received and the system
228  *	call is restartable, or EINTR otherwise.
229  */
230 int
231 cv_wait_sig(kcondvar_t *cv, kmutex_t *mtx)
232 {
233 	lwp_t *l = curlwp;
234 	int error;
235 
236 	KASSERT(mutex_owned(mtx));
237 
238 	cv_enter(cv, mtx, l);
239 	error = sleepq_block(0, true);
240 	return cv_exit(cv, mtx, l, error);
241 }
242 
243 /*
244  * cv_timedwait:
245  *
246  *	Wait on a condition variable until awoken or the specified timeout
247  *	expires.  Returns zero if awoken normally or EWOULDBLOCK if the
248  *	timeout expired.
249  *
250  *	timo is a timeout in ticks.  timo = 0 specifies an infinite timeout.
251  */
252 int
253 cv_timedwait(kcondvar_t *cv, kmutex_t *mtx, int timo)
254 {
255 	lwp_t *l = curlwp;
256 	int error;
257 
258 	KASSERT(mutex_owned(mtx));
259 
260 	cv_enter(cv, mtx, l);
261 	error = sleepq_block(timo, false);
262 	return cv_exit(cv, mtx, l, error);
263 }
264 
265 /*
266  * cv_timedwait_sig:
267  *
268  *	Wait on a condition variable until a timeout expires, awoken or a
269  *	signal is received.  Will also return early if the process is
270  *	exiting.  Returns zero if awoken normally, EWOULDBLOCK if the
271  *	timeout expires, ERESTART if a signal was received and the system
272  *	call is restartable, or EINTR otherwise.
273  *
274  *	timo is a timeout in ticks.  timo = 0 specifies an infinite timeout.
275  */
276 int
277 cv_timedwait_sig(kcondvar_t *cv, kmutex_t *mtx, int timo)
278 {
279 	lwp_t *l = curlwp;
280 	int error;
281 
282 	KASSERT(mutex_owned(mtx));
283 
284 	cv_enter(cv, mtx, l);
285 	error = sleepq_block(timo, true);
286 	return cv_exit(cv, mtx, l, error);
287 }
288 
289 /*
290  * cv_signal:
291  *
292  *	Wake the highest priority LWP waiting on a condition variable.
293  *	Must be called with the interlocking mutex held.
294  */
295 void
296 cv_signal(kcondvar_t *cv)
297 {
298 
299 	/* LOCKDEBUG_WAKEUP(CV_DEBUG_P(cv), cv, CV_RA); */
300 	KASSERT(cv_is_valid(cv));
301 
302 	if (__predict_false(!TAILQ_EMPTY(CV_SLEEPQ(cv))))
303 		cv_wakeup_one(cv);
304 }
305 
306 static void __noinline
307 cv_wakeup_one(kcondvar_t *cv)
308 {
309 	sleepq_t *sq;
310 	kmutex_t *mp;
311 	lwp_t *l;
312 
313 	KASSERT(cv_is_valid(cv));
314 
315 	mp = sleepq_hashlock(cv);
316 	sq = CV_SLEEPQ(cv);
317 	l = TAILQ_FIRST(sq);
318 	if (l == NULL) {
319 		mutex_spin_exit(mp);
320 		return;
321 	}
322 	KASSERT(l->l_sleepq == sq);
323 	KASSERT(l->l_mutex == mp);
324 	KASSERT(l->l_wchan == cv);
325 	sleepq_remove(sq, l);
326 	mutex_spin_exit(mp);
327 
328 	KASSERT(cv_is_valid(cv));
329 }
330 
331 /*
332  * cv_broadcast:
333  *
334  *	Wake all LWPs waiting on a condition variable.  Must be called
335  *	with the interlocking mutex held.
336  */
337 void
338 cv_broadcast(kcondvar_t *cv)
339 {
340 
341 	/* LOCKDEBUG_WAKEUP(CV_DEBUG_P(cv), cv, CV_RA); */
342 	KASSERT(cv_is_valid(cv));
343 
344 	if (__predict_false(!TAILQ_EMPTY(CV_SLEEPQ(cv))))
345 		cv_wakeup_all(cv);
346 }
347 
348 static void __noinline
349 cv_wakeup_all(kcondvar_t *cv)
350 {
351 	sleepq_t *sq;
352 	kmutex_t *mp;
353 	lwp_t *l, *next;
354 
355 	KASSERT(cv_is_valid(cv));
356 
357 	mp = sleepq_hashlock(cv);
358 	sq = CV_SLEEPQ(cv);
359 	for (l = TAILQ_FIRST(sq); l != NULL; l = next) {
360 		KASSERT(l->l_sleepq == sq);
361 		KASSERT(l->l_mutex == mp);
362 		KASSERT(l->l_wchan == cv);
363 		next = TAILQ_NEXT(l, l_sleepchain);
364 		sleepq_remove(sq, l);
365 	}
366 	mutex_spin_exit(mp);
367 
368 	KASSERT(cv_is_valid(cv));
369 }
370 
371 /*
372  * cv_has_waiters:
373  *
374  *	For diagnostic assertions: return non-zero if a condition
375  *	variable has waiters.
376  */
377 bool
378 cv_has_waiters(kcondvar_t *cv)
379 {
380 
381 	return !TAILQ_EMPTY(CV_SLEEPQ(cv));
382 }
383 
384 /*
385  * cv_is_valid:
386  *
387  *	For diagnostic assertions: return non-zero if a condition
388  *	variable appears to be valid.  No locks need be held.
389  */
390 bool
391 cv_is_valid(kcondvar_t *cv)
392 {
393 
394 	return CV_WMESG(cv) != deadcv && CV_WMESG(cv) != NULL;
395 }
396