xref: /netbsd/sys/kern/kern_turnstile.c (revision 6550d01e)
1 /*	$NetBSD: kern_turnstile.c,v 1.28 2009/11/18 12:26:22 yamt Exp $	*/
2 
3 /*-
4  * Copyright (c) 2002, 2006, 2007, 2009 The NetBSD Foundation, Inc.
5  * All rights reserved.
6  *
7  * This code is derived from software contributed to The NetBSD Foundation
8  * by Jason R. Thorpe and 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  * Turnstiles are described in detail in:
34  *
35  *	Solaris Internals: Core Kernel Architecture, Jim Mauro and
36  *	    Richard McDougall.
37  *
38  * Turnstiles are kept in a hash table.  There are likely to be many more
39  * synchronisation objects than there are threads.  Since a thread can block
40  * on only one lock at a time, we only need one turnstile per thread, and
41  * so they are allocated at thread creation time.
42  *
43  * When a thread decides it needs to block on a lock, it looks up the
44  * active turnstile for that lock.  If no active turnstile exists, then
45  * the process lends its turnstile to the lock.  If there is already an
46  * active turnstile for the lock, the thread places its turnstile on a
47  * list of free turnstiles, and references the active one instead.
48  *
49  * The act of looking up the turnstile acquires an interlock on the sleep
50  * queue.  If a thread decides it doesn't need to block after all, then this
51  * interlock must be released by explicitly aborting the turnstile
52  * operation.
53  *
54  * When a thread is awakened, it needs to get its turnstile back.  If there
55  * are still other threads waiting in the active turnstile, the thread
56  * grabs a free turnstile off the free list.  Otherwise, it can take back
57  * the active turnstile from the lock (thus deactivating the turnstile).
58  *
59  * Turnstiles are the place to do priority inheritence.
60  */
61 
62 #include <sys/cdefs.h>
63 __KERNEL_RCSID(0, "$NetBSD: kern_turnstile.c,v 1.28 2009/11/18 12:26:22 yamt Exp $");
64 
65 #include <sys/param.h>
66 #include <sys/lockdebug.h>
67 #include <sys/pool.h>
68 #include <sys/proc.h>
69 #include <sys/sleepq.h>
70 #include <sys/systm.h>
71 
72 #include <uvm/uvm_extern.h>
73 
74 #define	TS_HASH_SIZE	64
75 #define	TS_HASH_MASK	(TS_HASH_SIZE - 1)
76 #define	TS_HASH(obj)	(((uintptr_t)(obj) >> 3) & TS_HASH_MASK)
77 
78 tschain_t	turnstile_tab[TS_HASH_SIZE];
79 pool_cache_t	turnstile_cache;
80 
81 int	turnstile_ctor(void *, void *, int);
82 
83 extern turnstile_t turnstile0;
84 
85 /*
86  * turnstile_init:
87  *
88  *	Initialize the turnstile mechanism.
89  */
90 void
91 turnstile_init(void)
92 {
93 	tschain_t *tc;
94 	int i;
95 
96 	for (i = 0; i < TS_HASH_SIZE; i++) {
97 		tc = &turnstile_tab[i];
98 		LIST_INIT(&tc->tc_chain);
99 		tc->tc_mutex = mutex_obj_alloc(MUTEX_DEFAULT, IPL_SCHED);
100 	}
101 
102 	turnstile_cache = pool_cache_init(sizeof(turnstile_t), 0, 0, 0,
103 	    "tstilepl", NULL, IPL_NONE, turnstile_ctor, NULL, NULL);
104 	KASSERT(turnstile_cache != NULL);
105 
106 	(void)turnstile_ctor(NULL, &turnstile0, 0);
107 }
108 
109 /*
110  * turnstile_ctor:
111  *
112  *	Constructor for turnstiles.
113  */
114 int
115 turnstile_ctor(void *arg, void *obj, int flags)
116 {
117 	turnstile_t *ts = obj;
118 
119 	memset(ts, 0, sizeof(*ts));
120 	sleepq_init(&ts->ts_sleepq[TS_READER_Q]);
121 	sleepq_init(&ts->ts_sleepq[TS_WRITER_Q]);
122 	return (0);
123 }
124 
125 /*
126  * turnstile_remove:
127  *
128  *	Remove an LWP from a turnstile sleep queue and wake it.
129  */
130 static inline void
131 turnstile_remove(turnstile_t *ts, lwp_t *l, int q)
132 {
133 	turnstile_t *nts;
134 
135 	KASSERT(l->l_ts == ts);
136 
137 	/*
138 	 * This process is no longer using the active turnstile.
139 	 * Find an inactive one on the free list to give to it.
140 	 */
141 	if ((nts = ts->ts_free) != NULL) {
142 		KASSERT(TS_ALL_WAITERS(ts) > 1);
143 		l->l_ts = nts;
144 		ts->ts_free = nts->ts_free;
145 		nts->ts_free = NULL;
146 	} else {
147 		/*
148 		 * If the free list is empty, this is the last
149 		 * waiter.
150 		 */
151 		KASSERT(TS_ALL_WAITERS(ts) == 1);
152 		LIST_REMOVE(ts, ts_chain);
153 	}
154 
155 	ts->ts_waiters[q]--;
156 	sleepq_remove(&ts->ts_sleepq[q], l);
157 }
158 
159 /*
160  * turnstile_lookup:
161  *
162  *	Look up the turnstile for the specified lock.  This acquires and
163  *	holds the turnstile chain lock (sleep queue interlock).
164  */
165 turnstile_t *
166 turnstile_lookup(wchan_t obj)
167 {
168 	turnstile_t *ts;
169 	tschain_t *tc;
170 
171 	tc = &turnstile_tab[TS_HASH(obj)];
172 	mutex_spin_enter(tc->tc_mutex);
173 
174 	LIST_FOREACH(ts, &tc->tc_chain, ts_chain)
175 		if (ts->ts_obj == obj)
176 			return (ts);
177 
178 	/*
179 	 * No turnstile yet for this lock.  No problem, turnstile_block()
180 	 * handles this by fetching the turnstile from the blocking thread.
181 	 */
182 	return (NULL);
183 }
184 
185 /*
186  * turnstile_exit:
187  *
188  *	Abort a turnstile operation.
189  */
190 void
191 turnstile_exit(wchan_t obj)
192 {
193 	tschain_t *tc;
194 
195 	tc = &turnstile_tab[TS_HASH(obj)];
196 	mutex_spin_exit(tc->tc_mutex);
197 }
198 
199 /*
200  * turnstile_block:
201  *
202  *	 Enter an object into the turnstile chain and prepare the current
203  *	 LWP for sleep.
204  */
205 void
206 turnstile_block(turnstile_t *ts, int q, wchan_t obj, syncobj_t *sobj)
207 {
208 	lwp_t *l;
209 	lwp_t *cur; /* cached curlwp */
210 	lwp_t *owner;
211 	turnstile_t *ots;
212 	tschain_t *tc;
213 	sleepq_t *sq;
214 	pri_t prio, obase;
215 
216 	tc = &turnstile_tab[TS_HASH(obj)];
217 	l = cur = curlwp;
218 
219 	KASSERT(q == TS_READER_Q || q == TS_WRITER_Q);
220 	KASSERT(mutex_owned(tc->tc_mutex));
221 	KASSERT(l != NULL && l->l_ts != NULL);
222 
223 	if (ts == NULL) {
224 		/*
225 		 * We are the first thread to wait for this object;
226 		 * lend our turnstile to it.
227 		 */
228 		ts = l->l_ts;
229 		KASSERT(TS_ALL_WAITERS(ts) == 0);
230 		KASSERT(TAILQ_EMPTY(&ts->ts_sleepq[TS_READER_Q]) &&
231 			TAILQ_EMPTY(&ts->ts_sleepq[TS_WRITER_Q]));
232 		ts->ts_obj = obj;
233 		ts->ts_inheritor = NULL;
234 		LIST_INSERT_HEAD(&tc->tc_chain, ts, ts_chain);
235 	} else {
236 		/*
237 		 * Object already has a turnstile.  Put our turnstile
238 		 * onto the free list, and reference the existing
239 		 * turnstile instead.
240 		 */
241 		ots = l->l_ts;
242 		KASSERT(ots->ts_free == NULL);
243 		ots->ts_free = ts->ts_free;
244 		ts->ts_free = ots;
245 		l->l_ts = ts;
246 
247 		KASSERT(ts->ts_obj == obj);
248 		KASSERT(TS_ALL_WAITERS(ts) != 0);
249 		KASSERT(!TAILQ_EMPTY(&ts->ts_sleepq[TS_READER_Q]) ||
250 			!TAILQ_EMPTY(&ts->ts_sleepq[TS_WRITER_Q]));
251 	}
252 
253 	sq = &ts->ts_sleepq[q];
254 	ts->ts_waiters[q]++;
255 	sleepq_enter(sq, l, tc->tc_mutex);
256 	LOCKDEBUG_BARRIER(tc->tc_mutex, 1);
257 	l->l_kpriority = true;
258 	obase = l->l_kpribase;
259 	if (obase < PRI_KTHREAD)
260 		l->l_kpribase = PRI_KTHREAD;
261 	sleepq_enqueue(sq, obj, "tstile", sobj);
262 
263 	/*
264 	 * Disable preemption across this entire block, as we may drop
265 	 * scheduler locks (allowing preemption), and would prefer not
266 	 * to be interrupted while in a state of flux.
267 	 */
268 	KPREEMPT_DISABLE(l);
269 
270 	/*
271 	 * Lend our priority to lwps on the blocking chain.
272 	 *
273 	 * NOTE: if you get a panic in this code block, it is likely that
274 	 * a lock has been destroyed or corrupted while still in use.  Try
275 	 * compiling a kernel with LOCKDEBUG to pinpoint the problem.
276 	 */
277 	prio = lwp_eprio(l);
278 	KASSERT(cur == l);
279 	KASSERT(tc->tc_mutex == cur->l_mutex);
280 	for (;;) {
281 		bool dolock;
282 
283 		if (l->l_wchan == NULL)
284 			break;
285 
286 		owner = (*l->l_syncobj->sobj_owner)(l->l_wchan);
287 		if (owner == NULL)
288 			break;
289 
290 		/* The owner may have changed as we have dropped the tc lock */
291 		if (cur == owner) {
292 			/*
293 			 * we own the lock: stop here, sleepq_block()
294 			 * should wake up immediatly
295 			 */
296 			break;
297 		}
298 		if (l->l_mutex != owner->l_mutex)
299 			dolock = true;
300 		else
301 			dolock = false;
302 		if (l == owner || (dolock && !lwp_trylock(owner))) {
303 			/*
304 			 * restart from curlwp.
305 			 * Note that there may be a livelock here:
306 			 * the owner may try grabing cur's lock (which is
307 			 * the tc lock) while we're trying to grab
308 			 * the owner's lock.
309 			 */
310 			lwp_unlock(l);
311 			l = cur;
312 			lwp_lock(l);
313 			prio = lwp_eprio(l);
314 			continue;
315 		}
316 		if (prio <= lwp_eprio(owner)) {
317 			if (dolock)
318 				lwp_unlock(owner);
319 			break;
320 		}
321 		ts = l->l_ts;
322 		KASSERT(ts->ts_inheritor == owner || ts->ts_inheritor == NULL);
323 		if (ts->ts_inheritor == NULL) {
324 			ts->ts_inheritor = owner;
325 			ts->ts_eprio = prio;
326 			SLIST_INSERT_HEAD(&owner->l_pi_lenders, ts, ts_pichain);
327 			lwp_lendpri(owner, prio);
328 		} else if (prio > ts->ts_eprio) {
329 			ts->ts_eprio = prio;
330 			lwp_lendpri(owner, prio);
331 		}
332 		if (dolock)
333 			lwp_unlock(l);
334 		l = owner;
335 	}
336 	LOCKDEBUG_BARRIER(l->l_mutex, 1);
337 	if (cur->l_mutex != l->l_mutex) {
338 		lwp_unlock(l);
339 		lwp_lock(cur);
340 	}
341 	LOCKDEBUG_BARRIER(cur->l_mutex, 1);
342 
343 	sleepq_block(0, false);
344 	cur->l_kpribase = obase;
345 	KPREEMPT_ENABLE(cur);
346 }
347 
348 /*
349  * turnstile_wakeup:
350  *
351  *	Wake up the specified number of threads that are blocked
352  *	in a turnstile.
353  */
354 void
355 turnstile_wakeup(turnstile_t *ts, int q, int count, lwp_t *nl)
356 {
357 	sleepq_t *sq;
358 	tschain_t *tc;
359 	lwp_t *l;
360 
361 	tc = &turnstile_tab[TS_HASH(ts->ts_obj)];
362 	sq = &ts->ts_sleepq[q];
363 
364 	KASSERT(q == TS_READER_Q || q == TS_WRITER_Q);
365 	KASSERT(count > 0 && count <= TS_WAITERS(ts, q));
366 	KASSERT(mutex_owned(tc->tc_mutex));
367 	KASSERT(ts->ts_inheritor == curlwp || ts->ts_inheritor == NULL);
368 
369 	/*
370 	 * restore inherited priority if necessary.
371 	 */
372 
373 	if (ts->ts_inheritor != NULL) {
374 		turnstile_t *iter;
375 		turnstile_t *next;
376 		turnstile_t *prev = NULL;
377 		pri_t prio;
378 		bool dolock;
379 
380 		ts->ts_inheritor = NULL;
381 		l = curlwp;
382 
383 		dolock = l->l_mutex == l->l_cpu->ci_schedstate.spc_lwplock;
384 		if (dolock) {
385 			lwp_lock(l);
386 		}
387 
388 		/*
389 		 * the following loop does two things.
390 		 *
391 		 * - remove ts from the list.
392 		 *
393 		 * - from the rest of the list, find the highest priority.
394 		 */
395 
396 		prio = -1;
397 		KASSERT(!SLIST_EMPTY(&l->l_pi_lenders));
398 		for (iter = SLIST_FIRST(&l->l_pi_lenders);
399 		    iter != NULL; iter = next) {
400 			KASSERT(lwp_eprio(l) >= ts->ts_eprio);
401 			next = SLIST_NEXT(iter, ts_pichain);
402 			if (iter == ts) {
403 				if (prev == NULL) {
404 					SLIST_REMOVE_HEAD(&l->l_pi_lenders,
405 					    ts_pichain);
406 				} else {
407 					SLIST_REMOVE_AFTER(prev, ts_pichain);
408 				}
409 			} else if (prio < iter->ts_eprio) {
410 				prio = iter->ts_eprio;
411 			}
412 			prev = iter;
413 		}
414 
415 		lwp_lendpri(l, prio);
416 
417 		if (dolock) {
418 			lwp_unlock(l);
419 		}
420 	}
421 
422 	if (nl != NULL) {
423 #if defined(DEBUG) || defined(LOCKDEBUG)
424 		TAILQ_FOREACH(l, sq, l_sleepchain) {
425 			if (l == nl)
426 				break;
427 		}
428 		if (l == NULL)
429 			panic("turnstile_wakeup: nl not on sleepq");
430 #endif
431 		turnstile_remove(ts, nl, q);
432 	} else {
433 		while (count-- > 0) {
434 			l = TAILQ_FIRST(sq);
435 			KASSERT(l != NULL);
436 			turnstile_remove(ts, l, q);
437 		}
438 	}
439 	mutex_spin_exit(tc->tc_mutex);
440 }
441 
442 /*
443  * turnstile_unsleep:
444  *
445  *	Remove an LWP from the turnstile.  This is called when the LWP has
446  *	not been awoken normally but instead interrupted: for example, if it
447  *	has received a signal.  It's not a valid action for turnstiles,
448  *	since LWPs blocking on a turnstile are not interruptable.
449  */
450 void
451 turnstile_unsleep(lwp_t *l, bool cleanup)
452 {
453 
454 	lwp_unlock(l);
455 	panic("turnstile_unsleep");
456 }
457 
458 /*
459  * turnstile_changepri:
460  *
461  *	Adjust the priority of an LWP residing on a turnstile.
462  */
463 void
464 turnstile_changepri(lwp_t *l, pri_t pri)
465 {
466 
467 	/* XXX priority inheritance */
468 	sleepq_changepri(l, pri);
469 }
470 
471 #if defined(LOCKDEBUG)
472 /*
473  * turnstile_print:
474  *
475  *	Given the address of a lock object, print the contents of a
476  *	turnstile.
477  */
478 void
479 turnstile_print(volatile void *obj, void (*pr)(const char *, ...))
480 {
481 	turnstile_t *ts;
482 	tschain_t *tc;
483 	sleepq_t *rsq, *wsq;
484 	lwp_t *l;
485 
486 	tc = &turnstile_tab[TS_HASH(obj)];
487 
488 	LIST_FOREACH(ts, &tc->tc_chain, ts_chain)
489 		if (ts->ts_obj == obj)
490 			break;
491 
492 	(*pr)("Turnstile chain at %p.\n", tc);
493 	if (ts == NULL) {
494 		(*pr)("=> No active turnstile for this lock.\n");
495 		return;
496 	}
497 
498 	rsq = &ts->ts_sleepq[TS_READER_Q];
499 	wsq = &ts->ts_sleepq[TS_WRITER_Q];
500 
501 	(*pr)("=> Turnstile at %p (wrq=%p, rdq=%p).\n", ts, rsq, wsq);
502 
503 	(*pr)("=> %d waiting readers:", TS_WAITERS(ts, TS_READER_Q));
504 	TAILQ_FOREACH(l, rsq, l_sleepchain) {
505 		(*pr)(" %p", l);
506 	}
507 	(*pr)("\n");
508 
509 	(*pr)("=> %d waiting writers:", TS_WAITERS(ts, TS_WRITER_Q));
510 	TAILQ_FOREACH(l, wsq, l_sleepchain) {
511 		(*pr)(" %p", l);
512 	}
513 	(*pr)("\n");
514 }
515 #endif	/* LOCKDEBUG */
516