1/*-
2 * SPDX-License-Identifier: BSD-2-Clause-FreeBSD
3 *
4 * Copyright (c) 2000 Jake Burkholder <jake@freebsd.org>.
5 * All rights reserved.
6 *
7 * Redistribution and use in source and binary forms, with or without
8 * modification, are permitted provided that the following conditions
9 * are met:
10 * 1. Redistributions of source code must retain the above copyright
11 *    notice, this list of conditions and the following disclaimer.
12 * 2. Redistributions in binary form must reproduce the above copyright
13 *    notice, this list of conditions and the following disclaimer in the
14 *    documentation and/or other materials provided with the distribution.
15 *
16 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
17 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
18 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
19 * ARE DISCLAIMED.  IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
20 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
21 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
22 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
23 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
24 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
25 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
26 * SUCH DAMAGE.
27 */
28
29#include <sys/cdefs.h>
30__FBSDID("$FreeBSD$");
31
32#include "opt_ktrace.h"
33
34#include <sys/param.h>
35#include <sys/systm.h>
36#include <sys/limits.h>
37#include <sys/lock.h>
38#include <sys/mutex.h>
39#include <sys/proc.h>
40#include <sys/kernel.h>
41#include <sys/ktr.h>
42#include <sys/condvar.h>
43#include <sys/sched.h>
44#include <sys/signalvar.h>
45#include <sys/sleepqueue.h>
46#include <sys/resourcevar.h>
47#ifdef KTRACE
48#include <sys/uio.h>
49#include <sys/ktrace.h>
50#endif
51
52/*
53 * A bound below which cv_waiters is valid.  Once cv_waiters reaches this bound,
54 * cv_signal must manually check the wait queue for threads.
55 */
56#define	CV_WAITERS_BOUND	INT_MAX
57
58#define	CV_WAITERS_INC(cvp) do {					\
59	if ((cvp)->cv_waiters < CV_WAITERS_BOUND)			\
60		(cvp)->cv_waiters++;					\
61} while (0)
62
63/*
64 * Common sanity checks for cv_wait* functions.
65 */
66#define	CV_ASSERT(cvp, lock, td) do {					\
67	KASSERT((td) != NULL, ("%s: td NULL", __func__));		\
68	KASSERT(TD_IS_RUNNING(td), ("%s: not TDS_RUNNING", __func__));	\
69	KASSERT((cvp) != NULL, ("%s: cvp NULL", __func__));		\
70	KASSERT((lock) != NULL, ("%s: lock NULL", __func__));		\
71} while (0)
72
73/*
74 * Initialize a condition variable.  Must be called before use.
75 */
76void
77cv_init(struct cv *cvp, const char *desc)
78{
79
80	cvp->cv_description = desc;
81	cvp->cv_waiters = 0;
82}
83
84/*
85 * Destroy a condition variable.  The condition variable must be re-initialized
86 * in order to be re-used.
87 */
88void
89cv_destroy(struct cv *cvp)
90{
91#ifdef INVARIANTS
92	struct sleepqueue *sq;
93
94	sleepq_lock(cvp);
95	sq = sleepq_lookup(cvp);
96	sleepq_release(cvp);
97	KASSERT(sq == NULL, ("%s: associated sleep queue non-empty", __func__));
98#endif
99}
100
101/*
102 * Wait on a condition variable.  The current thread is placed on the condition
103 * variable's wait queue and suspended.  A cv_signal or cv_broadcast on the same
104 * condition variable will resume the thread.  The mutex is released before
105 * sleeping and will be held on return.  It is recommended that the mutex be
106 * held when cv_signal or cv_broadcast are called.
107 */
108void
109_cv_wait(struct cv *cvp, struct lock_object *lock)
110{
111	WITNESS_SAVE_DECL(lock_witness);
112	struct lock_class *class;
113	struct thread *td;
114	uintptr_t lock_state;
115
116	td = curthread;
117	lock_state = 0;
118#ifdef KTRACE
119	if (KTRPOINT(td, KTR_CSW))
120		ktrcsw(1, 0, cv_wmesg(cvp));
121#endif
122	CV_ASSERT(cvp, lock, td);
123	WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, lock,
124	    "Waiting on \"%s\"", cvp->cv_description);
125	class = LOCK_CLASS(lock);
126
127	if (SCHEDULER_STOPPED_TD(td))
128		return;
129
130	sleepq_lock(cvp);
131
132	CV_WAITERS_INC(cvp);
133	if (lock == &Giant.lock_object)
134		mtx_assert(&Giant, MA_OWNED);
135	DROP_GIANT();
136
137	sleepq_add(cvp, lock, cvp->cv_description, SLEEPQ_CONDVAR, 0);
138	if (lock != &Giant.lock_object) {
139		if (class->lc_flags & LC_SLEEPABLE)
140			sleepq_release(cvp);
141		WITNESS_SAVE(lock, lock_witness);
142		lock_state = class->lc_unlock(lock);
143		if (class->lc_flags & LC_SLEEPABLE)
144			sleepq_lock(cvp);
145	}
146	sleepq_wait(cvp, 0);
147
148#ifdef KTRACE
149	if (KTRPOINT(td, KTR_CSW))
150		ktrcsw(0, 0, cv_wmesg(cvp));
151#endif
152	PICKUP_GIANT();
153	if (lock != &Giant.lock_object) {
154		class->lc_lock(lock, lock_state);
155		WITNESS_RESTORE(lock, lock_witness);
156	}
157}
158
159/*
160 * Wait on a condition variable.  This function differs from cv_wait by
161 * not acquiring the mutex after condition variable was signaled.
162 */
163void
164_cv_wait_unlock(struct cv *cvp, struct lock_object *lock)
165{
166	struct lock_class *class;
167	struct thread *td;
168
169	td = curthread;
170#ifdef KTRACE
171	if (KTRPOINT(td, KTR_CSW))
172		ktrcsw(1, 0, cv_wmesg(cvp));
173#endif
174	CV_ASSERT(cvp, lock, td);
175	WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, lock,
176	    "Waiting on \"%s\"", cvp->cv_description);
177	KASSERT(lock != &Giant.lock_object,
178	    ("cv_wait_unlock cannot be used with Giant"));
179	class = LOCK_CLASS(lock);
180
181	if (SCHEDULER_STOPPED_TD(td)) {
182		class->lc_unlock(lock);
183		return;
184	}
185
186	sleepq_lock(cvp);
187
188	CV_WAITERS_INC(cvp);
189	DROP_GIANT();
190
191	sleepq_add(cvp, lock, cvp->cv_description, SLEEPQ_CONDVAR, 0);
192	if (class->lc_flags & LC_SLEEPABLE)
193		sleepq_release(cvp);
194	class->lc_unlock(lock);
195	if (class->lc_flags & LC_SLEEPABLE)
196		sleepq_lock(cvp);
197	sleepq_wait(cvp, 0);
198
199#ifdef KTRACE
200	if (KTRPOINT(td, KTR_CSW))
201		ktrcsw(0, 0, cv_wmesg(cvp));
202#endif
203	PICKUP_GIANT();
204}
205
206/*
207 * Wait on a condition variable, allowing interruption by signals.  Return 0 if
208 * the thread was resumed with cv_signal or cv_broadcast, EINTR or ERESTART if
209 * a signal was caught.  If ERESTART is returned the system call should be
210 * restarted if possible.
211 */
212int
213_cv_wait_sig(struct cv *cvp, struct lock_object *lock)
214{
215	WITNESS_SAVE_DECL(lock_witness);
216	struct lock_class *class;
217	struct thread *td;
218	uintptr_t lock_state;
219	int rval;
220
221	td = curthread;
222	lock_state = 0;
223#ifdef KTRACE
224	if (KTRPOINT(td, KTR_CSW))
225		ktrcsw(1, 0, cv_wmesg(cvp));
226#endif
227	CV_ASSERT(cvp, lock, td);
228	WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, lock,
229	    "Waiting on \"%s\"", cvp->cv_description);
230	class = LOCK_CLASS(lock);
231
232	if (SCHEDULER_STOPPED_TD(td))
233		return (0);
234
235	sleepq_lock(cvp);
236
237	CV_WAITERS_INC(cvp);
238	if (lock == &Giant.lock_object)
239		mtx_assert(&Giant, MA_OWNED);
240	DROP_GIANT();
241
242	sleepq_add(cvp, lock, cvp->cv_description, SLEEPQ_CONDVAR |
243	    SLEEPQ_INTERRUPTIBLE, 0);
244	if (lock != &Giant.lock_object) {
245		if (class->lc_flags & LC_SLEEPABLE)
246			sleepq_release(cvp);
247		WITNESS_SAVE(lock, lock_witness);
248		lock_state = class->lc_unlock(lock);
249		if (class->lc_flags & LC_SLEEPABLE)
250			sleepq_lock(cvp);
251	}
252	rval = sleepq_wait_sig(cvp, 0);
253
254#ifdef KTRACE
255	if (KTRPOINT(td, KTR_CSW))
256		ktrcsw(0, 0, cv_wmesg(cvp));
257#endif
258	PICKUP_GIANT();
259	if (lock != &Giant.lock_object) {
260		class->lc_lock(lock, lock_state);
261		WITNESS_RESTORE(lock, lock_witness);
262	}
263
264	return (rval);
265}
266
267/*
268 * Wait on a condition variable for (at most) the value specified in sbt
269 * argument. Returns 0 if the process was resumed by cv_signal or cv_broadcast,
270 * EWOULDBLOCK if the timeout expires.
271 */
272int
273_cv_timedwait_sbt(struct cv *cvp, struct lock_object *lock, sbintime_t sbt,
274    sbintime_t pr, int flags)
275{
276	WITNESS_SAVE_DECL(lock_witness);
277	struct lock_class *class;
278	struct thread *td;
279	int lock_state, rval;
280
281	td = curthread;
282	lock_state = 0;
283#ifdef KTRACE
284	if (KTRPOINT(td, KTR_CSW))
285		ktrcsw(1, 0, cv_wmesg(cvp));
286#endif
287	CV_ASSERT(cvp, lock, td);
288	WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, lock,
289	    "Waiting on \"%s\"", cvp->cv_description);
290	class = LOCK_CLASS(lock);
291
292	if (SCHEDULER_STOPPED_TD(td))
293		return (0);
294
295	sleepq_lock(cvp);
296
297	CV_WAITERS_INC(cvp);
298	if (lock == &Giant.lock_object)
299		mtx_assert(&Giant, MA_OWNED);
300	DROP_GIANT();
301
302	sleepq_add(cvp, lock, cvp->cv_description, SLEEPQ_CONDVAR, 0);
303	sleepq_set_timeout_sbt(cvp, sbt, pr, flags);
304	if (lock != &Giant.lock_object) {
305		if (class->lc_flags & LC_SLEEPABLE)
306			sleepq_release(cvp);
307		WITNESS_SAVE(lock, lock_witness);
308		lock_state = class->lc_unlock(lock);
309		if (class->lc_flags & LC_SLEEPABLE)
310			sleepq_lock(cvp);
311	}
312	rval = sleepq_timedwait(cvp, 0);
313
314#ifdef KTRACE
315	if (KTRPOINT(td, KTR_CSW))
316		ktrcsw(0, 0, cv_wmesg(cvp));
317#endif
318	PICKUP_GIANT();
319	if (lock != &Giant.lock_object) {
320		class->lc_lock(lock, lock_state);
321		WITNESS_RESTORE(lock, lock_witness);
322	}
323
324	return (rval);
325}
326
327/*
328 * Wait on a condition variable for (at most) the value specified in sbt
329 * argument, allowing interruption by signals.
330 * Returns 0 if the thread was resumed by cv_signal or cv_broadcast,
331 * EWOULDBLOCK if the timeout expires, and EINTR or ERESTART if a signal
332 * was caught.
333 */
334int
335_cv_timedwait_sig_sbt(struct cv *cvp, struct lock_object *lock,
336    sbintime_t sbt, sbintime_t pr, int flags)
337{
338	WITNESS_SAVE_DECL(lock_witness);
339	struct lock_class *class;
340	struct thread *td;
341	int lock_state, rval;
342
343	td = curthread;
344	lock_state = 0;
345#ifdef KTRACE
346	if (KTRPOINT(td, KTR_CSW))
347		ktrcsw(1, 0, cv_wmesg(cvp));
348#endif
349	CV_ASSERT(cvp, lock, td);
350	WITNESS_WARN(WARN_GIANTOK | WARN_SLEEPOK, lock,
351	    "Waiting on \"%s\"", cvp->cv_description);
352	class = LOCK_CLASS(lock);
353
354	if (SCHEDULER_STOPPED_TD(td))
355		return (0);
356
357	sleepq_lock(cvp);
358
359	CV_WAITERS_INC(cvp);
360	if (lock == &Giant.lock_object)
361		mtx_assert(&Giant, MA_OWNED);
362	DROP_GIANT();
363
364	sleepq_add(cvp, lock, cvp->cv_description, SLEEPQ_CONDVAR |
365	    SLEEPQ_INTERRUPTIBLE, 0);
366	sleepq_set_timeout_sbt(cvp, sbt, pr, flags);
367	if (lock != &Giant.lock_object) {
368		if (class->lc_flags & LC_SLEEPABLE)
369			sleepq_release(cvp);
370		WITNESS_SAVE(lock, lock_witness);
371		lock_state = class->lc_unlock(lock);
372		if (class->lc_flags & LC_SLEEPABLE)
373			sleepq_lock(cvp);
374	}
375	rval = sleepq_timedwait_sig(cvp, 0);
376
377#ifdef KTRACE
378	if (KTRPOINT(td, KTR_CSW))
379		ktrcsw(0, 0, cv_wmesg(cvp));
380#endif
381	PICKUP_GIANT();
382	if (lock != &Giant.lock_object) {
383		class->lc_lock(lock, lock_state);
384		WITNESS_RESTORE(lock, lock_witness);
385	}
386
387	return (rval);
388}
389
390/*
391 * Signal a condition variable, wakes up one waiting thread.  Will also wakeup
392 * the swapper if the process is not in memory, so that it can bring the
393 * sleeping process in.  Note that this may also result in additional threads
394 * being made runnable.  Should be called with the same mutex as was passed to
395 * cv_wait held.
396 */
397void
398cv_signal(struct cv *cvp)
399{
400	int wakeup_swapper;
401
402	if (cvp->cv_waiters == 0)
403		return;
404	wakeup_swapper = 0;
405	sleepq_lock(cvp);
406	if (cvp->cv_waiters > 0) {
407		if (cvp->cv_waiters == CV_WAITERS_BOUND &&
408		    sleepq_lookup(cvp) == NULL) {
409			cvp->cv_waiters = 0;
410		} else {
411			if (cvp->cv_waiters < CV_WAITERS_BOUND)
412				cvp->cv_waiters--;
413			wakeup_swapper = sleepq_signal(cvp, SLEEPQ_CONDVAR, 0,
414			    0);
415		}
416	}
417	sleepq_release(cvp);
418	if (wakeup_swapper)
419		kick_proc0();
420}
421
422/*
423 * Broadcast a signal to a condition variable.  Wakes up all waiting threads.
424 * Should be called with the same mutex as was passed to cv_wait held.
425 */
426void
427cv_broadcastpri(struct cv *cvp, int pri)
428{
429	int wakeup_swapper;
430
431	if (cvp->cv_waiters == 0)
432		return;
433	/*
434	 * XXX sleepq_broadcast pri argument changed from -1 meaning
435	 * no pri to 0 meaning no pri.
436	 */
437	wakeup_swapper = 0;
438	if (pri == -1)
439		pri = 0;
440	sleepq_lock(cvp);
441	if (cvp->cv_waiters > 0) {
442		cvp->cv_waiters = 0;
443		wakeup_swapper = sleepq_broadcast(cvp, SLEEPQ_CONDVAR, pri, 0);
444	}
445	sleepq_release(cvp);
446	if (wakeup_swapper)
447		kick_proc0();
448}
449