2 * Copyright (c) 2007 Stephan Uphoff <ups@FreeBSD.org>
5 * Redistribution and use in source and binary forms, with or without
6 * modification, are permitted provided that the following conditions
8 * 1. Redistributions of source code must retain the above copyright
9 * notice, this list of conditions and the following disclaimer.
10 * 2. Redistributions in binary form must reproduce the above copyright
11 * notice, this list of conditions and the following disclaimer in the
12 * documentation and/or other materials provided with the distribution.
13 * 3. Neither the name of the author nor the names of any co-contributors
14 * may be used to endorse or promote products derived from this software
15 * without specific prior written permission.
17 * THIS SOFTWARE IS PROVIDED BY THE AUTHOR AND CONTRIBUTORS ``AS IS'' AND
18 * ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE
19 * IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE
20 * ARE DISCLAIMED. IN NO EVENT SHALL THE AUTHOR OR CONTRIBUTORS BE LIABLE
21 * FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR CONSEQUENTIAL
22 * DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF SUBSTITUTE GOODS
23 * OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR BUSINESS INTERRUPTION)
24 * HOWEVER CAUSED AND ON ANY THEORY OF LIABILITY, WHETHER IN CONTRACT, STRICT
25 * LIABILITY, OR TORT (INCLUDING NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY
26 * OUT OF THE USE OF THIS SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF
31 * Machine independent bits of reader/writer lock implementation.
34 #include <sys/cdefs.h>
35 __FBSDID("$FreeBSD$");
38 #include "opt_kdtrace.h"
40 #include <sys/param.h>
41 #include <sys/systm.h>
43 #include <sys/kernel.h>
47 #include <sys/mutex.h>
49 #include <sys/rmlock.h>
50 #include <sys/sched.h>
52 #include <sys/turnstile.h>
53 #include <sys/lock_profile.h>
54 #include <machine/cpu.h>
61 * A cookie to mark destroyed rmlocks. This is stored in the head of
64 #define RM_DESTROYED ((void *)0xdead)
66 #define rm_destroyed(rm) \
67 (LIST_FIRST(&(rm)->rm_activeReaders) == RM_DESTROYED)
69 #define RMPF_ONQUEUE 1
73 #define _rm_assert(c, what, file, line)
76 static void assert_rm(const struct lock_object *lock, int what);
78 static void db_show_rm(const struct lock_object *lock);
80 static void lock_rm(struct lock_object *lock, uintptr_t how);
82 static int owner_rm(const struct lock_object *lock, struct thread **owner);
84 static uintptr_t unlock_rm(struct lock_object *lock);
86 struct lock_class lock_class_rm = {
88 .lc_flags = LC_SLEEPLOCK | LC_RECURSABLE,
89 .lc_assert = assert_rm,
91 .lc_ddb_show = db_show_rm,
94 .lc_unlock = unlock_rm,
100 struct lock_class lock_class_rm_sleepable = {
101 .lc_name = "sleepable rm",
102 .lc_flags = LC_SLEEPLOCK | LC_SLEEPABLE | LC_RECURSABLE,
103 .lc_assert = assert_rm,
105 .lc_ddb_show = db_show_rm,
108 .lc_unlock = unlock_rm,
110 .lc_owner = owner_rm,
115 assert_rm(const struct lock_object *lock, int what)
118 rm_assert((const struct rmlock *)lock, what);
122 lock_rm(struct lock_object *lock, uintptr_t how)
125 struct rm_priotracker *tracker;
127 rm = (struct rmlock *)lock;
131 tracker = (struct rm_priotracker *)how;
132 rm_rlock(rm, tracker);
137 unlock_rm(struct lock_object *lock)
142 struct rm_queue *queue;
143 struct rm_priotracker *tracker;
146 rm = (struct rmlock *)lock;
149 rm_assert(rm, RA_LOCKED | RA_NOTRECURSED);
154 * Find the right rm_priotracker structure for curthread.
155 * The guarantee about its uniqueness is given by the fact
156 * we already asserted the lock wasn't recursively acquired.
160 pc = pcpu_find(curcpu);
161 for (queue = pc->pc_rm_queue.rmq_next;
162 queue != &pc->pc_rm_queue; queue = queue->rmq_next) {
163 tracker = (struct rm_priotracker *)queue;
164 if ((tracker->rmp_rmlock == rm) &&
165 (tracker->rmp_thread == td)) {
166 how = (uintptr_t)tracker;
170 KASSERT(tracker != NULL,
171 ("rm_priotracker is non-NULL when lock held in read mode"));
173 rm_runlock(rm, tracker);
180 owner_rm(const struct lock_object *lock, struct thread **owner)
182 const struct rmlock *rm;
183 struct lock_class *lc;
185 rm = (const struct rmlock *)lock;
186 lc = LOCK_CLASS(&rm->rm_wlock_object);
187 return (lc->lc_owner(&rm->rm_wlock_object, owner));
191 static struct mtx rm_spinlock;
193 MTX_SYSINIT(rm_spinlock, &rm_spinlock, "rm_spinlock", MTX_SPIN);
196 * Add or remove tracker from per-cpu list.
198 * The per-cpu list can be traversed at any time in forward direction from an
199 * interrupt on the *local* cpu.
202 rm_tracker_add(struct pcpu *pc, struct rm_priotracker *tracker)
204 struct rm_queue *next;
206 /* Initialize all tracker pointers */
207 tracker->rmp_cpuQueue.rmq_prev = &pc->pc_rm_queue;
208 next = pc->pc_rm_queue.rmq_next;
209 tracker->rmp_cpuQueue.rmq_next = next;
211 /* rmq_prev is not used during froward traversal. */
212 next->rmq_prev = &tracker->rmp_cpuQueue;
214 /* Update pointer to first element. */
215 pc->pc_rm_queue.rmq_next = &tracker->rmp_cpuQueue;
219 * Return a count of the number of trackers the thread 'td' already
220 * has on this CPU for the lock 'rm'.
223 rm_trackers_present(const struct pcpu *pc, const struct rmlock *rm,
224 const struct thread *td)
226 struct rm_queue *queue;
227 struct rm_priotracker *tracker;
231 for (queue = pc->pc_rm_queue.rmq_next; queue != &pc->pc_rm_queue;
232 queue = queue->rmq_next) {
233 tracker = (struct rm_priotracker *)queue;
234 if ((tracker->rmp_rmlock == rm) && (tracker->rmp_thread == td))
241 rm_tracker_remove(struct pcpu *pc, struct rm_priotracker *tracker)
243 struct rm_queue *next, *prev;
245 next = tracker->rmp_cpuQueue.rmq_next;
246 prev = tracker->rmp_cpuQueue.rmq_prev;
248 /* Not used during forward traversal. */
249 next->rmq_prev = prev;
251 /* Remove from list. */
252 prev->rmq_next = next;
256 rm_cleanIPI(void *arg)
259 struct rmlock *rm = arg;
260 struct rm_priotracker *tracker;
261 struct rm_queue *queue;
262 pc = pcpu_find(curcpu);
264 for (queue = pc->pc_rm_queue.rmq_next; queue != &pc->pc_rm_queue;
265 queue = queue->rmq_next) {
266 tracker = (struct rm_priotracker *)queue;
267 if (tracker->rmp_rmlock == rm && tracker->rmp_flags == 0) {
268 tracker->rmp_flags = RMPF_ONQUEUE;
269 mtx_lock_spin(&rm_spinlock);
270 LIST_INSERT_HEAD(&rm->rm_activeReaders, tracker,
272 mtx_unlock_spin(&rm_spinlock);
278 rm_init_flags(struct rmlock *rm, const char *name, int opts)
280 struct lock_class *lc;
284 if (!(opts & RM_NOWITNESS))
285 liflags |= LO_WITNESS;
286 if (opts & RM_RECURSE)
287 liflags |= LO_RECURSABLE;
288 rm->rm_writecpus = all_cpus;
289 LIST_INIT(&rm->rm_activeReaders);
290 if (opts & RM_SLEEPABLE) {
291 liflags |= LO_SLEEPABLE;
292 lc = &lock_class_rm_sleepable;
293 sx_init_flags(&rm->rm_lock_sx, "rmlock_sx", SX_NOWITNESS);
296 mtx_init(&rm->rm_lock_mtx, name, "rmlock_mtx", MTX_NOWITNESS);
298 lock_init(&rm->lock_object, lc, name, NULL, liflags);
302 rm_init(struct rmlock *rm, const char *name)
305 rm_init_flags(rm, name, 0);
309 rm_destroy(struct rmlock *rm)
312 rm_assert(rm, RA_UNLOCKED);
313 LIST_FIRST(&rm->rm_activeReaders) = RM_DESTROYED;
314 if (rm->lock_object.lo_flags & LO_SLEEPABLE)
315 sx_destroy(&rm->rm_lock_sx);
317 mtx_destroy(&rm->rm_lock_mtx);
318 lock_destroy(&rm->lock_object);
322 rm_wowned(const struct rmlock *rm)
325 if (rm->lock_object.lo_flags & LO_SLEEPABLE)
326 return (sx_xlocked(&rm->rm_lock_sx));
328 return (mtx_owned(&rm->rm_lock_mtx));
332 rm_sysinit(void *arg)
334 struct rm_args *args = arg;
336 rm_init(args->ra_rm, args->ra_desc);
340 rm_sysinit_flags(void *arg)
342 struct rm_args_flags *args = arg;
344 rm_init_flags(args->ra_rm, args->ra_desc, args->ra_opts);
348 _rm_rlock_hard(struct rmlock *rm, struct rm_priotracker *tracker, int trylock)
353 pc = pcpu_find(curcpu);
355 /* Check if we just need to do a proper critical_exit. */
356 if (!CPU_ISSET(pc->pc_cpuid, &rm->rm_writecpus)) {
361 /* Remove our tracker from the per-cpu list. */
362 rm_tracker_remove(pc, tracker);
364 /* Check to see if the IPI granted us the lock after all. */
365 if (tracker->rmp_flags) {
366 /* Just add back tracker - we hold the lock. */
367 rm_tracker_add(pc, tracker);
373 * We allow readers to aquire a lock even if a writer is blocked if
374 * the lock is recursive and the reader already holds the lock.
376 if ((rm->lock_object.lo_flags & LO_RECURSABLE) != 0) {
378 * Just grant the lock if this thread already has a tracker
379 * for this lock on the per-cpu queue.
381 if (rm_trackers_present(pc, rm, curthread) != 0) {
382 mtx_lock_spin(&rm_spinlock);
383 LIST_INSERT_HEAD(&rm->rm_activeReaders, tracker,
385 tracker->rmp_flags = RMPF_ONQUEUE;
386 mtx_unlock_spin(&rm_spinlock);
387 rm_tracker_add(pc, tracker);
397 if (rm->lock_object.lo_flags & LO_SLEEPABLE) {
398 if (!sx_try_xlock(&rm->rm_lock_sx))
401 if (!mtx_trylock(&rm->rm_lock_mtx))
405 if (rm->lock_object.lo_flags & LO_SLEEPABLE) {
406 THREAD_SLEEPING_OK();
407 sx_xlock(&rm->rm_lock_sx);
408 THREAD_NO_SLEEPING();
410 mtx_lock(&rm->rm_lock_mtx);
414 pc = pcpu_find(curcpu);
415 CPU_CLR(pc->pc_cpuid, &rm->rm_writecpus);
416 rm_tracker_add(pc, tracker);
420 if (rm->lock_object.lo_flags & LO_SLEEPABLE)
421 sx_xunlock(&rm->rm_lock_sx);
423 mtx_unlock(&rm->rm_lock_mtx);
429 _rm_rlock(struct rmlock *rm, struct rm_priotracker *tracker, int trylock)
431 struct thread *td = curthread;
434 if (SCHEDULER_STOPPED())
437 tracker->rmp_flags = 0;
438 tracker->rmp_thread = td;
439 tracker->rmp_rmlock = rm;
441 if (rm->lock_object.lo_flags & LO_SLEEPABLE)
442 THREAD_NO_SLEEPING();
444 td->td_critnest++; /* critical_enter(); */
448 pc = cpuid_to_pcpu[td->td_oncpu]; /* pcpu_find(td->td_oncpu); */
450 rm_tracker_add(pc, tracker);
459 * Fast path to combine two common conditions into a single
462 if (0 == (td->td_owepreempt |
463 CPU_ISSET(pc->pc_cpuid, &rm->rm_writecpus)))
466 /* We do not have a read token and need to acquire one. */
467 return _rm_rlock_hard(rm, tracker, trylock);
471 _rm_unlock_hard(struct thread *td,struct rm_priotracker *tracker)
474 if (td->td_owepreempt) {
479 if (!tracker->rmp_flags)
482 mtx_lock_spin(&rm_spinlock);
483 LIST_REMOVE(tracker, rmp_qentry);
485 if (tracker->rmp_flags & RMPF_SIGNAL) {
487 struct turnstile *ts;
489 rm = tracker->rmp_rmlock;
491 turnstile_chain_lock(&rm->lock_object);
492 mtx_unlock_spin(&rm_spinlock);
494 ts = turnstile_lookup(&rm->lock_object);
496 turnstile_signal(ts, TS_EXCLUSIVE_QUEUE);
497 turnstile_unpend(ts, TS_EXCLUSIVE_LOCK);
498 turnstile_chain_unlock(&rm->lock_object);
500 mtx_unlock_spin(&rm_spinlock);
504 _rm_runlock(struct rmlock *rm, struct rm_priotracker *tracker)
507 struct thread *td = tracker->rmp_thread;
509 if (SCHEDULER_STOPPED())
512 td->td_critnest++; /* critical_enter(); */
513 pc = cpuid_to_pcpu[td->td_oncpu]; /* pcpu_find(td->td_oncpu); */
514 rm_tracker_remove(pc, tracker);
518 if (rm->lock_object.lo_flags & LO_SLEEPABLE)
519 THREAD_SLEEPING_OK();
521 if (0 == (td->td_owepreempt | tracker->rmp_flags))
524 _rm_unlock_hard(td, tracker);
528 _rm_wlock(struct rmlock *rm)
530 struct rm_priotracker *prio;
531 struct turnstile *ts;
534 if (SCHEDULER_STOPPED())
537 if (rm->lock_object.lo_flags & LO_SLEEPABLE)
538 sx_xlock(&rm->rm_lock_sx);
540 mtx_lock(&rm->rm_lock_mtx);
542 if (CPU_CMP(&rm->rm_writecpus, &all_cpus)) {
543 /* Get all read tokens back */
545 CPU_NAND(&readcpus, &rm->rm_writecpus);
546 rm->rm_writecpus = all_cpus;
549 * Assumes rm->rm_writecpus update is visible on other CPUs
550 * before rm_cleanIPI is called.
553 smp_rendezvous_cpus(readcpus,
554 smp_no_rendevous_barrier,
556 smp_no_rendevous_barrier,
563 mtx_lock_spin(&rm_spinlock);
564 while ((prio = LIST_FIRST(&rm->rm_activeReaders)) != NULL) {
565 ts = turnstile_trywait(&rm->lock_object);
566 prio->rmp_flags = RMPF_ONQUEUE | RMPF_SIGNAL;
567 mtx_unlock_spin(&rm_spinlock);
568 turnstile_wait(ts, prio->rmp_thread,
570 mtx_lock_spin(&rm_spinlock);
572 mtx_unlock_spin(&rm_spinlock);
577 _rm_wunlock(struct rmlock *rm)
580 if (rm->lock_object.lo_flags & LO_SLEEPABLE)
581 sx_xunlock(&rm->rm_lock_sx);
583 mtx_unlock(&rm->rm_lock_mtx);
589 _rm_wlock_debug(struct rmlock *rm, const char *file, int line)
592 if (SCHEDULER_STOPPED())
595 KASSERT(kdb_active != 0 || !TD_IS_IDLETHREAD(curthread),
596 ("rm_wlock() by idle thread %p on rmlock %s @ %s:%d",
597 curthread, rm->lock_object.lo_name, file, line));
598 KASSERT(!rm_destroyed(rm),
599 ("rm_wlock() of destroyed rmlock @ %s:%d", file, line));
600 _rm_assert(rm, RA_UNLOCKED, file, line);
602 WITNESS_CHECKORDER(&rm->lock_object, LOP_NEWORDER | LOP_EXCLUSIVE,
607 LOCK_LOG_LOCK("RMWLOCK", &rm->lock_object, 0, 0, file, line);
609 WITNESS_LOCK(&rm->lock_object, LOP_EXCLUSIVE, file, line);
611 curthread->td_locks++;
616 _rm_wunlock_debug(struct rmlock *rm, const char *file, int line)
619 if (SCHEDULER_STOPPED())
622 KASSERT(!rm_destroyed(rm),
623 ("rm_wunlock() of destroyed rmlock @ %s:%d", file, line));
624 _rm_assert(rm, RA_WLOCKED, file, line);
625 WITNESS_UNLOCK(&rm->lock_object, LOP_EXCLUSIVE, file, line);
626 LOCK_LOG_LOCK("RMWUNLOCK", &rm->lock_object, 0, 0, file, line);
628 curthread->td_locks--;
632 _rm_rlock_debug(struct rmlock *rm, struct rm_priotracker *tracker,
633 int trylock, const char *file, int line)
636 if (SCHEDULER_STOPPED())
640 if (!(rm->lock_object.lo_flags & LO_RECURSABLE) && !trylock) {
642 KASSERT(rm_trackers_present(pcpu_find(curcpu), rm,
644 ("rm_rlock: recursed on non-recursive rmlock %s @ %s:%d\n",
645 rm->lock_object.lo_name, file, line));
649 KASSERT(kdb_active != 0 || !TD_IS_IDLETHREAD(curthread),
650 ("rm_rlock() by idle thread %p on rmlock %s @ %s:%d",
651 curthread, rm->lock_object.lo_name, file, line));
652 KASSERT(!rm_destroyed(rm),
653 ("rm_rlock() of destroyed rmlock @ %s:%d", file, line));
655 KASSERT(!rm_wowned(rm),
656 ("rm_rlock: wlock already held for %s @ %s:%d",
657 rm->lock_object.lo_name, file, line));
658 WITNESS_CHECKORDER(&rm->lock_object, LOP_NEWORDER, file, line,
662 if (_rm_rlock(rm, tracker, trylock)) {
664 LOCK_LOG_TRY("RMRLOCK", &rm->lock_object, 0, 1, file,
667 LOCK_LOG_LOCK("RMRLOCK", &rm->lock_object, 0, 0, file,
669 WITNESS_LOCK(&rm->lock_object, 0, file, line);
671 curthread->td_locks++;
675 LOCK_LOG_TRY("RMRLOCK", &rm->lock_object, 0, 0, file, line);
681 _rm_runlock_debug(struct rmlock *rm, struct rm_priotracker *tracker,
682 const char *file, int line)
685 if (SCHEDULER_STOPPED())
688 KASSERT(!rm_destroyed(rm),
689 ("rm_runlock() of destroyed rmlock @ %s:%d", file, line));
690 _rm_assert(rm, RA_RLOCKED, file, line);
691 WITNESS_UNLOCK(&rm->lock_object, 0, file, line);
692 LOCK_LOG_LOCK("RMRUNLOCK", &rm->lock_object, 0, 0, file, line);
693 _rm_runlock(rm, tracker);
694 curthread->td_locks--;
700 * Just strip out file and line arguments if no lock debugging is enabled in
701 * the kernel - we are called from a kernel module.
704 _rm_wlock_debug(struct rmlock *rm, const char *file, int line)
711 _rm_wunlock_debug(struct rmlock *rm, const char *file, int line)
718 _rm_rlock_debug(struct rmlock *rm, struct rm_priotracker *tracker,
719 int trylock, const char *file, int line)
722 return _rm_rlock(rm, tracker, trylock);
726 _rm_runlock_debug(struct rmlock *rm, struct rm_priotracker *tracker,
727 const char *file, int line)
730 _rm_runlock(rm, tracker);
735 #ifdef INVARIANT_SUPPORT
741 * Note that this does not need to use witness_assert() for read lock
742 * assertions since an exact count of read locks held by this thread
746 _rm_assert(const struct rmlock *rm, int what, const char *file, int line)
750 if (panicstr != NULL)
754 case RA_LOCKED | RA_RECURSED:
755 case RA_LOCKED | RA_NOTRECURSED:
757 case RA_RLOCKED | RA_RECURSED:
758 case RA_RLOCKED | RA_NOTRECURSED:
760 * Handle the write-locked case. Unlike other
761 * primitives, writers can never recurse.
764 if (what & RA_RLOCKED)
765 panic("Lock %s exclusively locked @ %s:%d\n",
766 rm->lock_object.lo_name, file, line);
767 if (what & RA_RECURSED)
768 panic("Lock %s not recursed @ %s:%d\n",
769 rm->lock_object.lo_name, file, line);
774 count = rm_trackers_present(pcpu_find(curcpu), rm, curthread);
778 panic("Lock %s not %slocked @ %s:%d\n",
779 rm->lock_object.lo_name, (what & RA_RLOCKED) ?
780 "read " : "", file, line);
782 if (what & RA_NOTRECURSED)
783 panic("Lock %s recursed @ %s:%d\n",
784 rm->lock_object.lo_name, file, line);
785 } else if (what & RA_RECURSED)
786 panic("Lock %s not recursed @ %s:%d\n",
787 rm->lock_object.lo_name, file, line);
791 panic("Lock %s not exclusively locked @ %s:%d\n",
792 rm->lock_object.lo_name, file, line);
796 panic("Lock %s exclusively locked @ %s:%d\n",
797 rm->lock_object.lo_name, file, line);
800 count = rm_trackers_present(pcpu_find(curcpu), rm, curthread);
804 panic("Lock %s read locked @ %s:%d\n",
805 rm->lock_object.lo_name, file, line);
808 panic("Unknown rm lock assertion: %d @ %s:%d", what, file,
812 #endif /* INVARIANT_SUPPORT */
816 print_tracker(struct rm_priotracker *tr)
821 db_printf(" thread %p (tid %d, pid %d, \"%s\") {", td, td->td_tid,
822 td->td_proc->p_pid, td->td_name);
823 if (tr->rmp_flags & RMPF_ONQUEUE) {
824 db_printf("ONQUEUE");
825 if (tr->rmp_flags & RMPF_SIGNAL)
826 db_printf(",SIGNAL");
833 db_show_rm(const struct lock_object *lock)
835 struct rm_priotracker *tr;
836 struct rm_queue *queue;
837 const struct rmlock *rm;
838 struct lock_class *lc;
841 rm = (const struct rmlock *)lock;
842 db_printf(" writecpus: ");
843 ddb_display_cpuset(__DEQUALIFY(const cpuset_t *, &rm->rm_writecpus));
845 db_printf(" per-CPU readers:\n");
846 STAILQ_FOREACH(pc, &cpuhead, pc_allcpu)
847 for (queue = pc->pc_rm_queue.rmq_next;
848 queue != &pc->pc_rm_queue; queue = queue->rmq_next) {
849 tr = (struct rm_priotracker *)queue;
850 if (tr->rmp_rmlock == rm)
853 db_printf(" active readers:\n");
854 LIST_FOREACH(tr, &rm->rm_activeReaders, rmp_qentry)
856 lc = LOCK_CLASS(&rm->rm_wlock_object);
857 db_printf("Backing write-lock (%s):\n", lc->lc_name);
858 lc->lc_ddb_show(&rm->rm_wlock_object);