subr_witness.c revision 116182
1/*-
2 * Copyright (c) 1998 Berkeley Software Design, Inc. All rights reserved.
3 *
4 * Redistribution and use in source and binary forms, with or without
5 * modification, are permitted provided that the following conditions
6 * are met:
7 * 1. Redistributions of source code must retain the above copyright
8 *    notice, this list of conditions and the following disclaimer.
9 * 2. Redistributions in binary form must reproduce the above copyright
10 *    notice, this list of conditions and the following disclaimer in the
11 *    documentation and/or other materials provided with the distribution.
12 * 3. Berkeley Software Design Inc's name may not be used to endorse or
13 *    promote products derived from this software without specific prior
14 *    written permission.
15 *
16 * THIS SOFTWARE IS PROVIDED BY BERKELEY SOFTWARE DESIGN INC ``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 BERKELEY SOFTWARE DESIGN INC 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 *	from BSDI $Id: mutex_witness.c,v 1.1.2.20 2000/04/27 03:10:27 cp Exp $
29 *	and BSDI $Id: synch_machdep.c,v 2.3.2.39 2000/04/27 03:10:25 cp Exp $
30 */
31
32/*
33 * Implementation of the `witness' lock verifier.  Originally implemented for
34 * mutexes in BSD/OS.  Extended to handle generic lock objects and lock
35 * classes in FreeBSD.
36 */
37
38/*
39 *	Main Entry: witness
40 *	Pronunciation: 'wit-n&s
41 *	Function: noun
42 *	Etymology: Middle English witnesse, from Old English witnes knowledge,
43 *	    testimony, witness, from 2wit
44 *	Date: before 12th century
45 *	1 : attestation of a fact or event : TESTIMONY
46 *	2 : one that gives evidence; specifically : one who testifies in
47 *	    a cause or before a judicial tribunal
48 *	3 : one asked to be present at a transaction so as to be able to
49 *	    testify to its having taken place
50 *	4 : one who has personal knowledge of something
51 *	5 a : something serving as evidence or proof : SIGN
52 *	  b : public affirmation by word or example of usually
53 *	      religious faith or conviction <the heroic witness to divine
54 *	      life -- Pilot>
55 *	6 capitalized : a member of the Jehovah's Witnesses
56 */
57
58/*
59 * Special rules concerning Giant and lock orders:
60 *
61 * 1) Giant must be acquired before any other mutexes.  Stated another way,
62 *    no other mutex may be held when Giant is acquired.
63 *
64 * 2) Giant must be released when blocking on a sleepable lock.
65 *
66 * This rule is less obvious, but is a result of Giant providing the same
67 * semantics as spl().  Basically, when a thread sleeps, it must release
68 * Giant.  When a thread blocks on a sleepable lock, it sleeps.  Hence rule
69 * 2).
70 *
71 * 3) Giant may be acquired before or after sleepable locks.
72 *
73 * This rule is also not quite as obvious.  Giant may be acquired after
74 * a sleepable lock because it is a non-sleepable lock and non-sleepable
75 * locks may always be acquired while holding a sleepable lock.  The second
76 * case, Giant before a sleepable lock, follows from rule 2) above.  Suppose
77 * you have two threads T1 and T2 and a sleepable lock X.  Suppose that T1
78 * acquires X and blocks on Giant.  Then suppose that T2 acquires Giant and
79 * blocks on X.  When T2 blocks on X, T2 will release Giant allowing T1 to
80 * execute.  Thus, acquiring Giant both before and after a sleepable lock
81 * will not result in a lock order reversal.
82 */
83
84#include <sys/cdefs.h>
85__FBSDID("$FreeBSD: head/sys/kern/subr_witness.c 116182 2003-06-11 00:56:59Z obrien $");
86
87#include "opt_ddb.h"
88#include "opt_witness.h"
89#ifdef __i386__
90#include "opt_swtch.h"
91#endif
92
93#include <sys/param.h>
94#include <sys/bus.h>
95#include <sys/kernel.h>
96#include <sys/ktr.h>
97#include <sys/lock.h>
98#include <sys/malloc.h>
99#include <sys/mutex.h>
100#include <sys/proc.h>
101#include <sys/sysctl.h>
102#include <sys/systm.h>
103
104#include <ddb/ddb.h>
105
106#include <machine/stdarg.h>
107
108/* Define this to check for blessed mutexes */
109#undef BLESSING
110
111#define WITNESS_COUNT 200
112#define WITNESS_CHILDCOUNT (WITNESS_COUNT * 4)
113/*
114 * XXX: This is somewhat bogus, as we assume here that at most 1024 threads
115 * will hold LOCK_NCHILDREN * 2 locks.  We handle failure ok, and we should
116 * probably be safe for the most part, but it's still a SWAG.
117 */
118#define LOCK_CHILDCOUNT (MAXCPU + 1024) * 2
119
120#define	WITNESS_NCHILDREN 6
121
122struct witness_child_list_entry;
123
124struct witness {
125	const	char *w_name;
126	struct	lock_class *w_class;
127	STAILQ_ENTRY(witness) w_list;		/* List of all witnesses. */
128	STAILQ_ENTRY(witness) w_typelist;	/* Witnesses of a type. */
129	struct	witness_child_list_entry *w_children;	/* Great evilness... */
130	const	char *w_file;
131	int	w_line;
132	u_int	w_level;
133	u_int	w_refcount;
134	u_char	w_Giant_squawked:1;
135	u_char	w_other_squawked:1;
136	u_char	w_same_squawked:1;
137	u_char	w_displayed:1;
138};
139
140struct witness_child_list_entry {
141	struct	witness_child_list_entry *wcl_next;
142	struct	witness *wcl_children[WITNESS_NCHILDREN];
143	u_int	wcl_count;
144};
145
146STAILQ_HEAD(witness_list, witness);
147
148#ifdef BLESSING
149struct witness_blessed {
150	const	char *b_lock1;
151	const	char *b_lock2;
152};
153#endif
154
155struct witness_order_list_entry {
156	const	char *w_name;
157	struct	lock_class *w_class;
158};
159
160#ifdef BLESSING
161static int	blessed(struct witness *, struct witness *);
162#endif
163static int	depart(struct witness *w);
164static struct	witness *enroll(const char *description,
165				struct lock_class *lock_class);
166static int	insertchild(struct witness *parent, struct witness *child);
167static int	isitmychild(struct witness *parent, struct witness *child);
168static int	isitmydescendant(struct witness *parent, struct witness *child);
169static int	itismychild(struct witness *parent, struct witness *child);
170static int	rebalancetree(struct witness_list *list);
171static void	removechild(struct witness *parent, struct witness *child);
172static int	reparentchildren(struct witness *newparent,
173		    struct witness *oldparent);
174static int	sysctl_debug_witness_watch(SYSCTL_HANDLER_ARGS);
175static void	witness_displaydescendants(void(*)(const char *fmt, ...),
176					   struct witness *, int indent);
177static const char *fixup_filename(const char *file);
178static void	witness_leveldescendents(struct witness *parent, int level);
179static void	witness_levelall(void);
180static struct	witness *witness_get(void);
181static void	witness_free(struct witness *m);
182static struct	witness_child_list_entry *witness_child_get(void);
183static void	witness_child_free(struct witness_child_list_entry *wcl);
184static struct	lock_list_entry *witness_lock_list_get(void);
185static void	witness_lock_list_free(struct lock_list_entry *lle);
186static struct	lock_instance *find_instance(struct lock_list_entry *lock_list,
187					     struct lock_object *lock);
188static void	witness_list_lock(struct lock_instance *instance);
189#ifdef DDB
190static void	witness_list(struct thread *td);
191static void	witness_display_list(void(*prnt)(const char *fmt, ...),
192				     struct witness_list *list);
193static void	witness_display(void(*)(const char *fmt, ...));
194#endif
195
196MALLOC_DEFINE(M_WITNESS, "witness", "witness structure");
197
198/*
199 * If set to 0, witness is disabled.  If set to 1, witness performs full lock
200 * order checking for all locks.  If set to 2 or higher, then witness skips
201 * the full lock order check if the lock being acquired is at a higher level
202 * (i.e. farther down in the tree) than the current lock.  This last mode is
203 * somewhat experimental and not considered fully safe.  At runtime, this
204 * value may be set to 0 to turn off witness.  witness is not allowed be
205 * turned on once it is turned off, however.
206 */
207static int witness_watch = 1;
208TUNABLE_INT("debug.witness_watch", &witness_watch);
209SYSCTL_PROC(_debug, OID_AUTO, witness_watch, CTLFLAG_RW | CTLTYPE_INT, NULL, 0,
210    sysctl_debug_witness_watch, "I", "witness is watching lock operations");
211
212#ifdef DDB
213/*
214 * When DDB is enabled and witness_ddb is set to 1, it will cause the system to
215 * drop into kdebug() when:
216 *	- a lock heirarchy violation occurs
217 *	- locks are held when going to sleep.
218 */
219#ifdef WITNESS_DDB
220int	witness_ddb = 1;
221#else
222int	witness_ddb = 0;
223#endif
224TUNABLE_INT("debug.witness_ddb", &witness_ddb);
225SYSCTL_INT(_debug, OID_AUTO, witness_ddb, CTLFLAG_RW, &witness_ddb, 0, "");
226
227/*
228 * When DDB is enabled and witness_trace is set to 1, it will cause the system
229 * to print a stack trace:
230 *	- a lock heirarchy violation occurs
231 *	- locks are held when going to sleep.
232 */
233int	witness_trace = 1;
234TUNABLE_INT("debug.witness_trace", &witness_trace);
235SYSCTL_INT(_debug, OID_AUTO, witness_trace, CTLFLAG_RW, &witness_trace, 0, "");
236#endif /* DDB */
237
238#ifdef WITNESS_SKIPSPIN
239int	witness_skipspin = 1;
240#else
241int	witness_skipspin = 0;
242#endif
243TUNABLE_INT("debug.witness_skipspin", &witness_skipspin);
244SYSCTL_INT(_debug, OID_AUTO, witness_skipspin, CTLFLAG_RD, &witness_skipspin, 0,
245    "");
246
247static struct mtx w_mtx;
248static struct witness_list w_free = STAILQ_HEAD_INITIALIZER(w_free);
249static struct witness_list w_all = STAILQ_HEAD_INITIALIZER(w_all);
250static struct witness_list w_spin = STAILQ_HEAD_INITIALIZER(w_spin);
251static struct witness_list w_sleep = STAILQ_HEAD_INITIALIZER(w_sleep);
252static struct witness_child_list_entry *w_child_free = NULL;
253static struct lock_list_entry *w_lock_list_free = NULL;
254
255static struct witness w_data[WITNESS_COUNT];
256static struct witness_child_list_entry w_childdata[WITNESS_CHILDCOUNT];
257static struct lock_list_entry w_locklistdata[LOCK_CHILDCOUNT];
258
259static struct witness_order_list_entry order_lists[] = {
260	{ "proctree", &lock_class_sx },
261	{ "allproc", &lock_class_sx },
262	{ "Giant", &lock_class_mtx_sleep },
263	{ "filedesc structure", &lock_class_mtx_sleep },
264	{ "pipe mutex", &lock_class_mtx_sleep },
265	{ "sigio lock", &lock_class_mtx_sleep },
266	{ "process group", &lock_class_mtx_sleep },
267	{ "process lock", &lock_class_mtx_sleep },
268	{ "session", &lock_class_mtx_sleep },
269	{ "uidinfo hash", &lock_class_mtx_sleep },
270	{ "uidinfo struct", &lock_class_mtx_sleep },
271	{ "allprison", &lock_class_mtx_sleep },
272	{ NULL, NULL },
273	/*
274	 * spin locks
275	 */
276#ifdef SMP
277	{ "ap boot", &lock_class_mtx_spin },
278#ifdef __i386__
279	{ "com", &lock_class_mtx_spin },
280#endif
281#endif
282	{ "sio", &lock_class_mtx_spin },
283#ifdef __i386__
284	{ "cy", &lock_class_mtx_spin },
285#endif
286	{ "sabtty", &lock_class_mtx_spin },
287	{ "zstty", &lock_class_mtx_spin },
288	{ "ng_node", &lock_class_mtx_spin },
289	{ "ng_worklist", &lock_class_mtx_spin },
290	{ "ithread table lock", &lock_class_mtx_spin },
291	{ "sched lock", &lock_class_mtx_spin },
292	{ "callout", &lock_class_mtx_spin },
293	/*
294	 * leaf locks
295	 */
296	{ "allpmaps", &lock_class_mtx_spin },
297	{ "vm page queue free mutex", &lock_class_mtx_spin },
298	{ "icu", &lock_class_mtx_spin },
299#ifdef SMP
300	{ "smp rendezvous", &lock_class_mtx_spin },
301#if defined(__i386__) && defined(APIC_IO)
302	{ "tlb", &lock_class_mtx_spin },
303#endif
304#if defined(__i386__) && defined(LAZY_SWITCH)
305	{ "lazypmap", &lock_class_mtx_spin },
306#endif
307#ifdef __sparc64__
308	{ "ipi", &lock_class_mtx_spin },
309#endif
310#endif
311	{ "clk", &lock_class_mtx_spin },
312	{ "mutex profiling lock", &lock_class_mtx_spin },
313	{ "kse zombie lock", &lock_class_mtx_spin },
314	{ "ALD Queue", &lock_class_mtx_spin },
315#ifdef __ia64__
316	{ "MCA spin lock", &lock_class_mtx_spin },
317#endif
318#if defined(__i386__) || defined(__amd64__)
319	{ "pcicfg", &lock_class_mtx_spin },
320#endif
321	{ NULL, NULL },
322	{ NULL, NULL }
323};
324
325#ifdef BLESSING
326/*
327 * Pairs of locks which have been blessed
328 * Don't complain about order problems with blessed locks
329 */
330static struct witness_blessed blessed_list[] = {
331};
332static int blessed_count =
333	sizeof(blessed_list) / sizeof(struct witness_blessed);
334#endif
335
336/*
337 * List of all locks in the system.
338 */
339TAILQ_HEAD(, lock_object) all_locks = TAILQ_HEAD_INITIALIZER(all_locks);
340
341static struct mtx all_mtx = {
342	{ &lock_class_mtx_sleep,	/* mtx_object.lo_class */
343	  "All locks list",		/* mtx_object.lo_name */
344	  "All locks list",		/* mtx_object.lo_type */
345	  LO_INITIALIZED,		/* mtx_object.lo_flags */
346	  { NULL, NULL },		/* mtx_object.lo_list */
347	  NULL },			/* mtx_object.lo_witness */
348	MTX_UNOWNED, 0,			/* mtx_lock, mtx_recurse */
349	TAILQ_HEAD_INITIALIZER(all_mtx.mtx_blocked),
350	{ NULL, NULL }			/* mtx_contested */
351};
352
353/*
354 * This global is set to 0 once it becomes safe to use the witness code.
355 */
356static int witness_cold = 1;
357
358/*
359 * Global variables for book keeping.
360 */
361static int lock_cur_cnt;
362static int lock_max_cnt;
363
364/*
365 * The WITNESS-enabled diagnostic code.
366 */
367static void
368witness_initialize(void *dummy __unused)
369{
370	struct lock_object *lock;
371	struct witness_order_list_entry *order;
372	struct witness *w, *w1;
373	int i;
374
375	/*
376	 * We have to release Giant before initializing its witness
377	 * structure so that WITNESS doesn't get confused.
378	 */
379	mtx_unlock(&Giant);
380	mtx_assert(&Giant, MA_NOTOWNED);
381
382	CTR1(KTR_WITNESS, "%s: initializing witness", __func__);
383	TAILQ_INSERT_HEAD(&all_locks, &all_mtx.mtx_object, lo_list);
384	mtx_init(&w_mtx, "witness lock", NULL, MTX_SPIN | MTX_QUIET |
385	    MTX_NOWITNESS);
386	for (i = 0; i < WITNESS_COUNT; i++)
387		witness_free(&w_data[i]);
388	for (i = 0; i < WITNESS_CHILDCOUNT; i++)
389		witness_child_free(&w_childdata[i]);
390	for (i = 0; i < LOCK_CHILDCOUNT; i++)
391		witness_lock_list_free(&w_locklistdata[i]);
392
393	/* First add in all the specified order lists. */
394	for (order = order_lists; order->w_name != NULL; order++) {
395		w = enroll(order->w_name, order->w_class);
396		if (w == NULL)
397			continue;
398		w->w_file = "order list";
399		for (order++; order->w_name != NULL; order++) {
400			w1 = enroll(order->w_name, order->w_class);
401			if (w1 == NULL)
402				continue;
403			w1->w_file = "order list";
404			if (!itismychild(w, w1))
405				panic("Not enough memory for static orders!");
406			w = w1;
407		}
408	}
409
410	/* Iterate through all locks and add them to witness. */
411	mtx_lock(&all_mtx);
412	TAILQ_FOREACH(lock, &all_locks, lo_list) {
413		if (lock->lo_flags & LO_WITNESS)
414			lock->lo_witness = enroll(lock->lo_type,
415			    lock->lo_class);
416		else
417			lock->lo_witness = NULL;
418	}
419	mtx_unlock(&all_mtx);
420
421	/* Mark the witness code as being ready for use. */
422	atomic_store_rel_int(&witness_cold, 0);
423
424	mtx_lock(&Giant);
425}
426SYSINIT(witness_init, SI_SUB_WITNESS, SI_ORDER_FIRST, witness_initialize, NULL)
427
428static int
429sysctl_debug_witness_watch(SYSCTL_HANDLER_ARGS)
430{
431	int error, value;
432
433	value = witness_watch;
434	error = sysctl_handle_int(oidp, &value, 0, req);
435	if (error != 0 || req->newptr == NULL)
436		return (error);
437	error = suser(req->td);
438	if (error != 0)
439		return (error);
440	if (value == witness_watch)
441		return (0);
442	if (value != 0)
443		return (EINVAL);
444	witness_watch = 0;
445	return (0);
446}
447
448void
449witness_init(struct lock_object *lock)
450{
451	struct lock_class *class;
452
453	class = lock->lo_class;
454	if (lock->lo_flags & LO_INITIALIZED)
455		panic("%s: lock (%s) %s is already initialized", __func__,
456		    class->lc_name, lock->lo_name);
457	if ((lock->lo_flags & LO_RECURSABLE) != 0 &&
458	    (class->lc_flags & LC_RECURSABLE) == 0)
459		panic("%s: lock (%s) %s can not be recursable", __func__,
460		    class->lc_name, lock->lo_name);
461	if ((lock->lo_flags & LO_SLEEPABLE) != 0 &&
462	    (class->lc_flags & LC_SLEEPABLE) == 0)
463		panic("%s: lock (%s) %s can not be sleepable", __func__,
464		    class->lc_name, lock->lo_name);
465	if ((lock->lo_flags & LO_UPGRADABLE) != 0 &&
466	    (class->lc_flags & LC_UPGRADABLE) == 0)
467		panic("%s: lock (%s) %s can not be upgradable", __func__,
468		    class->lc_name, lock->lo_name);
469
470	mtx_lock(&all_mtx);
471	TAILQ_INSERT_TAIL(&all_locks, lock, lo_list);
472	lock->lo_flags |= LO_INITIALIZED;
473	lock_cur_cnt++;
474	if (lock_cur_cnt > lock_max_cnt)
475		lock_max_cnt = lock_cur_cnt;
476	mtx_unlock(&all_mtx);
477	if (!witness_cold && witness_watch != 0 && panicstr == NULL &&
478	    (lock->lo_flags & LO_WITNESS) != 0)
479		lock->lo_witness = enroll(lock->lo_type, class);
480	else
481		lock->lo_witness = NULL;
482}
483
484void
485witness_destroy(struct lock_object *lock)
486{
487	struct witness *w;
488
489	if (witness_cold)
490		panic("lock (%s) %s destroyed while witness_cold",
491		    lock->lo_class->lc_name, lock->lo_name);
492	if ((lock->lo_flags & LO_INITIALIZED) == 0)
493		panic("%s: lock (%s) %s is not initialized", __func__,
494		    lock->lo_class->lc_name, lock->lo_name);
495
496	/* XXX: need to verify that no one holds the lock */
497	w = lock->lo_witness;
498	if (w != NULL) {
499		mtx_lock_spin(&w_mtx);
500		MPASS(w->w_refcount > 0);
501		w->w_refcount--;
502
503		/*
504		 * Lock is already released if we have an allocation failure
505		 * and depart() fails.
506		 */
507		if (w->w_refcount != 0 || depart(w))
508			mtx_unlock_spin(&w_mtx);
509	}
510
511	mtx_lock(&all_mtx);
512	lock_cur_cnt--;
513	TAILQ_REMOVE(&all_locks, lock, lo_list);
514	lock->lo_flags &= ~LO_INITIALIZED;
515	mtx_unlock(&all_mtx);
516}
517
518#ifdef DDB
519static void
520witness_display_list(void(*prnt)(const char *fmt, ...),
521		     struct witness_list *list)
522{
523	struct witness *w;
524
525	STAILQ_FOREACH(w, list, w_typelist) {
526		if (w->w_file == NULL || w->w_level > 0)
527			continue;
528		/*
529		 * This lock has no anscestors, display its descendants.
530		 */
531		witness_displaydescendants(prnt, w, 0);
532	}
533}
534
535static void
536witness_display(void(*prnt)(const char *fmt, ...))
537{
538	struct witness *w;
539
540	KASSERT(!witness_cold, ("%s: witness_cold", __func__));
541	witness_levelall();
542
543	/* Clear all the displayed flags. */
544	STAILQ_FOREACH(w, &w_all, w_list) {
545		w->w_displayed = 0;
546	}
547
548	/*
549	 * First, handle sleep locks which have been acquired at least
550	 * once.
551	 */
552	prnt("Sleep locks:\n");
553	witness_display_list(prnt, &w_sleep);
554
555	/*
556	 * Now do spin locks which have been acquired at least once.
557	 */
558	prnt("\nSpin locks:\n");
559	witness_display_list(prnt, &w_spin);
560
561	/*
562	 * Finally, any locks which have not been acquired yet.
563	 */
564	prnt("\nLocks which were never acquired:\n");
565	STAILQ_FOREACH(w, &w_all, w_list) {
566		if (w->w_file != NULL || w->w_refcount == 0)
567			continue;
568		prnt("%s\n", w->w_name);
569	}
570}
571#endif /* DDB */
572
573/* Trim useless garbage from filenames. */
574static const char *
575fixup_filename(const char *file)
576{
577
578	if (file == NULL)
579		return (NULL);
580	while (strncmp(file, "../", 3) == 0)
581		file += 3;
582	return (file);
583}
584
585void
586witness_lock(struct lock_object *lock, int flags, const char *file, int line)
587{
588	struct lock_list_entry **lock_list, *lle;
589	struct lock_instance *lock1, *lock2;
590	struct lock_class *class;
591	struct witness *w, *w1;
592	struct thread *td;
593	int i, j;
594#ifdef DDB
595	int go_into_ddb = 0;
596#endif
597
598	if (witness_cold || witness_watch == 0 || lock->lo_witness == NULL ||
599	    panicstr != NULL)
600		return;
601	w = lock->lo_witness;
602	class = lock->lo_class;
603	td = curthread;
604	file = fixup_filename(file);
605
606	if (class->lc_flags & LC_SLEEPLOCK) {
607		/*
608		 * Since spin locks include a critical section, this check
609		 * impliclty enforces a lock order of all sleep locks before
610		 * all spin locks.
611		 */
612		if (td->td_critnest != 0 && (flags & LOP_TRYLOCK) == 0)
613			panic("blockable sleep lock (%s) %s @ %s:%d",
614			    class->lc_name, lock->lo_name, file, line);
615		lock_list = &td->td_sleeplocks;
616	} else
617		lock_list = PCPU_PTR(spinlocks);
618
619	/*
620	 * Is this the first lock acquired?  If so, then no order checking
621	 * is needed.
622	 */
623	if (*lock_list == NULL)
624		goto out;
625
626	/*
627	 * Check to see if we are recursing on a lock we already own.
628	 */
629	lock1 = find_instance(*lock_list, lock);
630	if (lock1 != NULL) {
631		if ((lock1->li_flags & LI_EXCLUSIVE) != 0 &&
632		    (flags & LOP_EXCLUSIVE) == 0) {
633			printf("shared lock of (%s) %s @ %s:%d\n",
634			    class->lc_name, lock->lo_name, file, line);
635			printf("while exclusively locked from %s:%d\n",
636			    lock1->li_file, lock1->li_line);
637			panic("share->excl");
638		}
639		if ((lock1->li_flags & LI_EXCLUSIVE) == 0 &&
640		    (flags & LOP_EXCLUSIVE) != 0) {
641			printf("exclusive lock of (%s) %s @ %s:%d\n",
642			    class->lc_name, lock->lo_name, file, line);
643			printf("while share locked from %s:%d\n",
644			    lock1->li_file, lock1->li_line);
645			panic("excl->share");
646		}
647		lock1->li_flags++;
648		if ((lock->lo_flags & LO_RECURSABLE) == 0) {
649			printf(
650			"recursed on non-recursive lock (%s) %s @ %s:%d\n",
651			    class->lc_name, lock->lo_name, file, line);
652			printf("first acquired @ %s:%d\n", lock1->li_file,
653			    lock1->li_line);
654			panic("recurse");
655		}
656		CTR4(KTR_WITNESS, "%s: pid %d recursed on %s r=%d", __func__,
657		    td->td_proc->p_pid, lock->lo_name,
658		    lock1->li_flags & LI_RECURSEMASK);
659		lock1->li_file = file;
660		lock1->li_line = line;
661		return;
662	}
663
664	/*
665	 * Try locks do not block if they fail to acquire the lock, thus
666	 * there is no danger of deadlocks or of switching while holding a
667	 * spin lock if we acquire a lock via a try operation.
668	 */
669	if (flags & LOP_TRYLOCK)
670		goto out;
671
672	/*
673	 * Check for duplicate locks of the same type.  Note that we only
674	 * have to check for this on the last lock we just acquired.  Any
675	 * other cases will be caught as lock order violations.
676	 */
677	lock1 = &(*lock_list)->ll_children[(*lock_list)->ll_count - 1];
678	w1 = lock1->li_lock->lo_witness;
679	if (w1 == w) {
680		if (w->w_same_squawked || (lock->lo_flags & LO_DUPOK))
681			goto out;
682		w->w_same_squawked = 1;
683		printf("acquiring duplicate lock of same type: \"%s\"\n",
684			lock->lo_type);
685		printf(" 1st %s @ %s:%d\n", lock1->li_lock->lo_name,
686		    lock1->li_file, lock1->li_line);
687		printf(" 2nd %s @ %s:%d\n", lock->lo_name, file, line);
688#ifdef DDB
689		go_into_ddb = 1;
690#endif
691		goto out;
692	}
693	MPASS(!mtx_owned(&w_mtx));
694	mtx_lock_spin(&w_mtx);
695	/*
696	 * If we have a known higher number just say ok
697	 */
698	if (witness_watch > 1 && w->w_level > w1->w_level) {
699		mtx_unlock_spin(&w_mtx);
700		goto out;
701	}
702	/*
703	 * If we know that the the lock we are acquiring comes after
704	 * the lock we most recently acquired in the lock order tree,
705	 * then there is no need for any further checks.
706	 */
707	if (isitmydescendant(w1, w)) {
708		mtx_unlock_spin(&w_mtx);
709		goto out;
710	}
711	for (j = 0, lle = *lock_list; lle != NULL; lle = lle->ll_next) {
712		for (i = lle->ll_count - 1; i >= 0; i--, j++) {
713
714			MPASS(j < WITNESS_COUNT);
715			lock1 = &lle->ll_children[i];
716			w1 = lock1->li_lock->lo_witness;
717
718			/*
719			 * If this lock doesn't undergo witness checking,
720			 * then skip it.
721			 */
722			if (w1 == NULL) {
723				KASSERT((lock1->li_lock->lo_flags & LO_WITNESS) == 0,
724				    ("lock missing witness structure"));
725				continue;
726			}
727			/*
728			 * If we are locking Giant and this is a sleepable
729			 * lock, then skip it.
730			 */
731			if ((lock1->li_lock->lo_flags & LO_SLEEPABLE) != 0 &&
732			    lock == &Giant.mtx_object)
733				continue;
734			/*
735			 * If we are locking a sleepable lock and this lock
736			 * is Giant, then skip it.
737			 */
738			if ((lock->lo_flags & LO_SLEEPABLE) != 0 &&
739			    lock1->li_lock == &Giant.mtx_object)
740				continue;
741			/*
742			 * If we are locking a sleepable lock and this lock
743			 * isn't sleepable, we want to treat it as a lock
744			 * order violation to enfore a general lock order of
745			 * sleepable locks before non-sleepable locks.
746			 */
747			if (!((lock->lo_flags & LO_SLEEPABLE) != 0 &&
748			    (lock1->li_lock->lo_flags & LO_SLEEPABLE) == 0))
749			    /*
750			     * Check the lock order hierarchy for a reveresal.
751			     */
752			    if (!isitmydescendant(w, w1))
753				continue;
754			/*
755			 * We have a lock order violation, check to see if it
756			 * is allowed or has already been yelled about.
757			 */
758			mtx_unlock_spin(&w_mtx);
759#ifdef BLESSING
760			if (blessed(w, w1))
761				goto out;
762#endif
763			if (lock1->li_lock == &Giant.mtx_object) {
764				if (w1->w_Giant_squawked)
765					goto out;
766				else
767					w1->w_Giant_squawked = 1;
768			} else {
769				if (w1->w_other_squawked)
770					goto out;
771				else
772					w1->w_other_squawked = 1;
773			}
774			/*
775			 * Ok, yell about it.
776			 */
777			printf("lock order reversal\n");
778			/*
779			 * Try to locate an earlier lock with
780			 * witness w in our list.
781			 */
782			do {
783				lock2 = &lle->ll_children[i];
784				MPASS(lock2->li_lock != NULL);
785				if (lock2->li_lock->lo_witness == w)
786					break;
787				i--;
788				if (i == 0 && lle->ll_next != NULL) {
789					lle = lle->ll_next;
790					i = lle->ll_count - 1;
791					MPASS(i >= 0 && i < LOCK_NCHILDREN);
792				}
793			} while (i >= 0);
794			if (i < 0) {
795				printf(" 1st %p %s (%s) @ %s:%d\n",
796				    lock1->li_lock, lock1->li_lock->lo_name,
797				    lock1->li_lock->lo_type, lock1->li_file,
798				    lock1->li_line);
799				printf(" 2nd %p %s (%s) @ %s:%d\n", lock,
800				    lock->lo_name, lock->lo_type, file, line);
801			} else {
802				printf(" 1st %p %s (%s) @ %s:%d\n",
803				    lock2->li_lock, lock2->li_lock->lo_name,
804				    lock2->li_lock->lo_type, lock2->li_file,
805				    lock2->li_line);
806				printf(" 2nd %p %s (%s) @ %s:%d\n",
807				    lock1->li_lock, lock1->li_lock->lo_name,
808				    lock1->li_lock->lo_type, lock1->li_file,
809				    lock1->li_line);
810				printf(" 3rd %p %s (%s) @ %s:%d\n", lock,
811				    lock->lo_name, lock->lo_type, file, line);
812			}
813#ifdef DDB
814			go_into_ddb = 1;
815#endif
816			goto out;
817		}
818	}
819	lock1 = &(*lock_list)->ll_children[(*lock_list)->ll_count - 1];
820	/*
821	 * Don't build a new relationship between a sleepable lock and
822	 * Giant if it is the wrong direction.  The real lock order is that
823	 * sleepable locks come before Giant.
824	 */
825	if (!(lock1->li_lock == &Giant.mtx_object &&
826	    (lock->lo_flags & LO_SLEEPABLE) != 0)) {
827		CTR3(KTR_WITNESS, "%s: adding %s as a child of %s", __func__,
828		    lock->lo_type, lock1->li_lock->lo_type);
829		if (!itismychild(lock1->li_lock->lo_witness, w))
830			/* Witness is dead. */
831			return;
832	}
833	mtx_unlock_spin(&w_mtx);
834
835out:
836#ifdef DDB
837	if (go_into_ddb) {
838		if (witness_trace)
839			backtrace();
840		if (witness_ddb)
841			Debugger(__func__);
842	}
843#endif
844	w->w_file = file;
845	w->w_line = line;
846
847	lle = *lock_list;
848	if (lle == NULL || lle->ll_count == LOCK_NCHILDREN) {
849		lle = witness_lock_list_get();
850		if (lle == NULL)
851			return;
852		lle->ll_next = *lock_list;
853		CTR3(KTR_WITNESS, "%s: pid %d added lle %p", __func__,
854		    td->td_proc->p_pid, lle);
855		*lock_list = lle;
856	}
857	lock1 = &lle->ll_children[lle->ll_count++];
858	lock1->li_lock = lock;
859	lock1->li_line = line;
860	lock1->li_file = file;
861	if ((flags & LOP_EXCLUSIVE) != 0)
862		lock1->li_flags = LI_EXCLUSIVE;
863	else
864		lock1->li_flags = 0;
865	CTR4(KTR_WITNESS, "%s: pid %d added %s as lle[%d]", __func__,
866	    td->td_proc->p_pid, lock->lo_name, lle->ll_count - 1);
867}
868
869void
870witness_upgrade(struct lock_object *lock, int flags, const char *file, int line)
871{
872	struct lock_instance *instance;
873	struct lock_class *class;
874
875	KASSERT(!witness_cold, ("%s: witness_cold", __func__));
876	if (lock->lo_witness == NULL || witness_watch == 0 || panicstr != NULL)
877		return;
878	class = lock->lo_class;
879	file = fixup_filename(file);
880	if ((lock->lo_flags & LO_UPGRADABLE) == 0)
881		panic("upgrade of non-upgradable lock (%s) %s @ %s:%d",
882		    class->lc_name, lock->lo_name, file, line);
883	if ((flags & LOP_TRYLOCK) == 0)
884		panic("non-try upgrade of lock (%s) %s @ %s:%d", class->lc_name,
885		    lock->lo_name, file, line);
886	if ((lock->lo_class->lc_flags & LC_SLEEPLOCK) == 0)
887		panic("upgrade of non-sleep lock (%s) %s @ %s:%d",
888		    class->lc_name, lock->lo_name, file, line);
889	instance = find_instance(curthread->td_sleeplocks, lock);
890	if (instance == NULL)
891		panic("upgrade of unlocked lock (%s) %s @ %s:%d",
892		    class->lc_name, lock->lo_name, file, line);
893	if ((instance->li_flags & LI_EXCLUSIVE) != 0)
894		panic("upgrade of exclusive lock (%s) %s @ %s:%d",
895		    class->lc_name, lock->lo_name, file, line);
896	if ((instance->li_flags & LI_RECURSEMASK) != 0)
897		panic("upgrade of recursed lock (%s) %s r=%d @ %s:%d",
898		    class->lc_name, lock->lo_name,
899		    instance->li_flags & LI_RECURSEMASK, file, line);
900	instance->li_flags |= LI_EXCLUSIVE;
901}
902
903void
904witness_downgrade(struct lock_object *lock, int flags, const char *file,
905    int line)
906{
907	struct lock_instance *instance;
908	struct lock_class *class;
909
910	KASSERT(!witness_cold, ("%s: witness_cold", __func__));
911	if (lock->lo_witness == NULL || witness_watch == 0 || panicstr != NULL)
912		return;
913	class = lock->lo_class;
914	file = fixup_filename(file);
915	if ((lock->lo_flags & LO_UPGRADABLE) == 0)
916		panic("downgrade of non-upgradable lock (%s) %s @ %s:%d",
917		    class->lc_name, lock->lo_name, file, line);
918	if ((lock->lo_class->lc_flags & LC_SLEEPLOCK) == 0)
919		panic("downgrade of non-sleep lock (%s) %s @ %s:%d",
920		    class->lc_name, lock->lo_name, file, line);
921	instance = find_instance(curthread->td_sleeplocks, lock);
922	if (instance == NULL)
923		panic("downgrade of unlocked lock (%s) %s @ %s:%d",
924		    class->lc_name, lock->lo_name, file, line);
925	if ((instance->li_flags & LI_EXCLUSIVE) == 0)
926		panic("downgrade of shared lock (%s) %s @ %s:%d",
927		    class->lc_name, lock->lo_name, file, line);
928	if ((instance->li_flags & LI_RECURSEMASK) != 0)
929		panic("downgrade of recursed lock (%s) %s r=%d @ %s:%d",
930		    class->lc_name, lock->lo_name,
931		    instance->li_flags & LI_RECURSEMASK, file, line);
932	instance->li_flags &= ~LI_EXCLUSIVE;
933}
934
935void
936witness_unlock(struct lock_object *lock, int flags, const char *file, int line)
937{
938	struct lock_list_entry **lock_list, *lle;
939	struct lock_instance *instance;
940	struct lock_class *class;
941	struct thread *td;
942	register_t s;
943	int i, j;
944
945	if (witness_cold || witness_watch == 0 || lock->lo_witness == NULL ||
946	    panicstr != NULL)
947		return;
948	td = curthread;
949	class = lock->lo_class;
950	file = fixup_filename(file);
951	if (class->lc_flags & LC_SLEEPLOCK)
952		lock_list = &td->td_sleeplocks;
953	else
954		lock_list = PCPU_PTR(spinlocks);
955	for (; *lock_list != NULL; lock_list = &(*lock_list)->ll_next)
956		for (i = 0; i < (*lock_list)->ll_count; i++) {
957			instance = &(*lock_list)->ll_children[i];
958			if (instance->li_lock == lock) {
959				if ((instance->li_flags & LI_EXCLUSIVE) != 0 &&
960				    (flags & LOP_EXCLUSIVE) == 0) {
961					printf(
962					"shared unlock of (%s) %s @ %s:%d\n",
963					    class->lc_name, lock->lo_name,
964					    file, line);
965					printf(
966					"while exclusively locked from %s:%d\n",
967					    instance->li_file,
968					    instance->li_line);
969					panic("excl->ushare");
970				}
971				if ((instance->li_flags & LI_EXCLUSIVE) == 0 &&
972				    (flags & LOP_EXCLUSIVE) != 0) {
973					printf(
974					"exclusive unlock of (%s) %s @ %s:%d\n",
975					    class->lc_name, lock->lo_name,
976					    file, line);
977					printf(
978					"while share locked from %s:%d\n",
979					    instance->li_file,
980					    instance->li_line);
981					panic("share->uexcl");
982				}
983				/* If we are recursed, unrecurse. */
984				if ((instance->li_flags & LI_RECURSEMASK) > 0) {
985					CTR4(KTR_WITNESS,
986				    "%s: pid %d unrecursed on %s r=%d", __func__,
987					    td->td_proc->p_pid,
988					    instance->li_lock->lo_name,
989					    instance->li_flags);
990					instance->li_flags--;
991					return;
992				}
993				s = intr_disable();
994				CTR4(KTR_WITNESS,
995				    "%s: pid %d removed %s from lle[%d]", __func__,
996				    td->td_proc->p_pid,
997				    instance->li_lock->lo_name,
998				    (*lock_list)->ll_count - 1);
999				for (j = i; j < (*lock_list)->ll_count - 1; j++)
1000					(*lock_list)->ll_children[j] =
1001					    (*lock_list)->ll_children[j + 1];
1002				(*lock_list)->ll_count--;
1003				intr_restore(s);
1004				if ((*lock_list)->ll_count == 0) {
1005					lle = *lock_list;
1006					*lock_list = lle->ll_next;
1007					CTR3(KTR_WITNESS,
1008					    "%s: pid %d removed lle %p", __func__,
1009					    td->td_proc->p_pid, lle);
1010					witness_lock_list_free(lle);
1011				}
1012				return;
1013			}
1014		}
1015	panic("lock (%s) %s not locked @ %s:%d", class->lc_name, lock->lo_name,
1016	    file, line);
1017}
1018
1019/*
1020 * Warn if any locks other than 'lock' are held.  Flags can be passed in to
1021 * exempt Giant and sleepable locks from the checks as well.  If any
1022 * non-exempt locks are held, then a supplied message is printed to the
1023 * console along with a list of the offending locks.  If indicated in the
1024 * flags then a failure results in a panic as well.
1025 */
1026int
1027witness_warn(int flags, struct lock_object *lock, const char *fmt, ...)
1028{
1029	struct lock_list_entry *lle;
1030	struct lock_instance *lock1;
1031	struct thread *td;
1032	va_list ap;
1033	int i, n;
1034
1035	if (witness_cold || witness_watch == 0 || panicstr != NULL)
1036		return (0);
1037	n = 0;
1038	td = curthread;
1039	for (lle = td->td_sleeplocks; lle != NULL; lle = lle->ll_next)
1040		for (i = lle->ll_count - 1; i >= 0; i--) {
1041			lock1 = &lle->ll_children[i];
1042			if (lock1->li_lock == lock)
1043				continue;
1044			if (flags & WARN_GIANTOK &&
1045			    lock1->li_lock == &Giant.mtx_object)
1046				continue;
1047			if (flags & WARN_SLEEPOK &&
1048			    (lock1->li_lock->lo_flags & LO_SLEEPABLE) != 0)
1049				continue;
1050			if (n == 0) {
1051				va_start(ap, fmt);
1052				vprintf(fmt, ap);
1053				va_end(ap);
1054				printf(" with the following");
1055				if (flags & WARN_SLEEPOK)
1056					printf(" non-sleepable");
1057				printf("locks held:\n");
1058			}
1059			n++;
1060			witness_list_lock(lock1);
1061		}
1062	if (PCPU_GET(spinlocks) != NULL) {
1063		/*
1064		 * Since we already hold a spinlock preemption is
1065		 * already blocked.
1066		 */
1067		if (n == 0) {
1068			va_start(ap, fmt);
1069			vprintf(fmt, ap);
1070			va_end(ap);
1071			printf(" with the following");
1072			if (flags & WARN_SLEEPOK)
1073				printf(" non-sleepable");
1074			printf("locks held:\n");
1075		}
1076		n += witness_list_locks(PCPU_PTR(spinlocks));
1077	}
1078	if (flags & WARN_PANIC && n)
1079		panic("witness_warn");
1080#ifdef DDB
1081	else if (witness_ddb && n)
1082		Debugger(__func__);
1083#endif
1084	return (n);
1085}
1086
1087const char *
1088witness_file(struct lock_object *lock)
1089{
1090	struct witness *w;
1091
1092	if (witness_cold || witness_watch == 0 || lock->lo_witness == NULL)
1093		return ("?");
1094	w = lock->lo_witness;
1095	return (w->w_file);
1096}
1097
1098int
1099witness_line(struct lock_object *lock)
1100{
1101	struct witness *w;
1102
1103	if (witness_cold || witness_watch == 0 || lock->lo_witness == NULL)
1104		return (0);
1105	w = lock->lo_witness;
1106	return (w->w_line);
1107}
1108
1109static struct witness *
1110enroll(const char *description, struct lock_class *lock_class)
1111{
1112	struct witness *w;
1113
1114	if (!witness_watch || witness_watch == 0 || panicstr != NULL)
1115		return (NULL);
1116	if ((lock_class->lc_flags & LC_SPINLOCK) && witness_skipspin)
1117		return (NULL);
1118	mtx_lock_spin(&w_mtx);
1119	STAILQ_FOREACH(w, &w_all, w_list) {
1120		if (w->w_name == description || (w->w_refcount > 0 &&
1121		    strcmp(description, w->w_name) == 0)) {
1122			w->w_refcount++;
1123			mtx_unlock_spin(&w_mtx);
1124			if (lock_class != w->w_class)
1125				panic(
1126				"lock (%s) %s does not match earlier (%s) lock",
1127				    description, lock_class->lc_name,
1128				    w->w_class->lc_name);
1129			return (w);
1130		}
1131	}
1132	/*
1133	 * This isn't quite right, as witness_cold is still 0 while we
1134	 * enroll all the locks initialized before witness_initialize().
1135	 */
1136	if ((lock_class->lc_flags & LC_SPINLOCK) && !witness_cold) {
1137		mtx_unlock_spin(&w_mtx);
1138		panic("spin lock %s not in order list", description);
1139	}
1140	if ((w = witness_get()) == NULL)
1141		return (NULL);
1142	w->w_name = description;
1143	w->w_class = lock_class;
1144	w->w_refcount = 1;
1145	STAILQ_INSERT_HEAD(&w_all, w, w_list);
1146	if (lock_class->lc_flags & LC_SPINLOCK)
1147		STAILQ_INSERT_HEAD(&w_spin, w, w_typelist);
1148	else if (lock_class->lc_flags & LC_SLEEPLOCK)
1149		STAILQ_INSERT_HEAD(&w_sleep, w, w_typelist);
1150	else {
1151		mtx_unlock_spin(&w_mtx);
1152		panic("lock class %s is not sleep or spin",
1153		    lock_class->lc_name);
1154	}
1155	mtx_unlock_spin(&w_mtx);
1156	return (w);
1157}
1158
1159/* Don't let the door bang you on the way out... */
1160static int
1161depart(struct witness *w)
1162{
1163	struct witness_child_list_entry *wcl, *nwcl;
1164	struct witness_list *list;
1165	struct witness *parent;
1166
1167	MPASS(w->w_refcount == 0);
1168	if (w->w_class->lc_flags & LC_SLEEPLOCK)
1169		list = &w_sleep;
1170	else
1171		list = &w_spin;
1172	/*
1173	 * First, we run through the entire tree looking for any
1174	 * witnesses that the outgoing witness is a child of.  For
1175	 * each parent that we find, we reparent all the direct
1176	 * children of the outgoing witness to its parent.
1177	 */
1178	STAILQ_FOREACH(parent, list, w_typelist) {
1179		if (!isitmychild(parent, w))
1180			continue;
1181		removechild(parent, w);
1182		if (!reparentchildren(parent, w))
1183			return (0);
1184	}
1185
1186	/*
1187	 * Now we go through and free up the child list of the
1188	 * outgoing witness.
1189	 */
1190	for (wcl = w->w_children; wcl != NULL; wcl = nwcl) {
1191		nwcl = wcl->wcl_next;
1192		witness_child_free(wcl);
1193	}
1194
1195	/*
1196	 * Detach from various lists and free.
1197	 */
1198	STAILQ_REMOVE(list, w, witness, w_typelist);
1199	STAILQ_REMOVE(&w_all, w, witness, w_list);
1200	witness_free(w);
1201
1202	/* Finally, fixup the tree. */
1203	return (rebalancetree(list));
1204}
1205
1206/*
1207 * Prune an entire lock order tree.  We look for cases where a lock
1208 * is now both a descendant and a direct child of a given lock.  In
1209 * that case, we want to remove the direct child link from the tree.
1210 *
1211 * Returns false if insertchild() fails.
1212 */
1213static int
1214rebalancetree(struct witness_list *list)
1215{
1216	struct witness *child, *parent;
1217
1218	STAILQ_FOREACH(child, list, w_typelist) {
1219		STAILQ_FOREACH(parent, list, w_typelist) {
1220			if (!isitmychild(parent, child))
1221				continue;
1222			removechild(parent, child);
1223			if (isitmydescendant(parent, child))
1224				continue;
1225			if (!insertchild(parent, child))
1226				return (0);
1227		}
1228	}
1229	witness_levelall();
1230	return (1);
1231}
1232
1233/*
1234 * Add "child" as a direct child of "parent".  Returns false if
1235 * we fail due to out of memory.
1236 */
1237static int
1238insertchild(struct witness *parent, struct witness *child)
1239{
1240	struct witness_child_list_entry **wcl;
1241
1242	MPASS(child != NULL && parent != NULL);
1243
1244	/*
1245	 * Insert "child" after "parent"
1246	 */
1247	wcl = &parent->w_children;
1248	while (*wcl != NULL && (*wcl)->wcl_count == WITNESS_NCHILDREN)
1249		wcl = &(*wcl)->wcl_next;
1250	if (*wcl == NULL) {
1251		*wcl = witness_child_get();
1252		if (*wcl == NULL)
1253			return (0);
1254	}
1255	(*wcl)->wcl_children[(*wcl)->wcl_count++] = child;
1256
1257	return (1);
1258}
1259
1260/*
1261 * Make all the direct descendants of oldparent be direct descendants
1262 * of newparent.
1263 */
1264static int
1265reparentchildren(struct witness *newparent, struct witness *oldparent)
1266{
1267	struct witness_child_list_entry *wcl;
1268	int i;
1269
1270	/* Avoid making a witness a child of itself. */
1271	MPASS(!isitmychild(oldparent, newparent));
1272
1273	for (wcl = oldparent->w_children; wcl != NULL; wcl = wcl->wcl_next)
1274		for (i = 0; i < wcl->wcl_count; i++)
1275			if (!insertchild(newparent, wcl->wcl_children[i]))
1276				return (0);
1277	return (1);
1278}
1279
1280static int
1281itismychild(struct witness *parent, struct witness *child)
1282{
1283	struct witness_list *list;
1284
1285	MPASS(child != NULL && parent != NULL);
1286	if ((parent->w_class->lc_flags & (LC_SLEEPLOCK | LC_SPINLOCK)) !=
1287	    (child->w_class->lc_flags & (LC_SLEEPLOCK | LC_SPINLOCK)))
1288		panic(
1289		"%s: parent (%s) and child (%s) are not the same lock type",
1290		    __func__, parent->w_class->lc_name,
1291		    child->w_class->lc_name);
1292
1293	if (!insertchild(parent, child))
1294		return (0);
1295
1296	if (parent->w_class->lc_flags & LC_SLEEPLOCK)
1297		list = &w_sleep;
1298	else
1299		list = &w_spin;
1300	return (rebalancetree(list));
1301}
1302
1303static void
1304removechild(struct witness *parent, struct witness *child)
1305{
1306	struct witness_child_list_entry **wcl, *wcl1;
1307	int i;
1308
1309	for (wcl = &parent->w_children; *wcl != NULL; wcl = &(*wcl)->wcl_next)
1310		for (i = 0; i < (*wcl)->wcl_count; i++)
1311			if ((*wcl)->wcl_children[i] == child)
1312				goto found;
1313	return;
1314found:
1315	(*wcl)->wcl_count--;
1316	if ((*wcl)->wcl_count > i)
1317		(*wcl)->wcl_children[i] =
1318		    (*wcl)->wcl_children[(*wcl)->wcl_count];
1319	MPASS((*wcl)->wcl_children[i] != NULL);
1320	if ((*wcl)->wcl_count != 0)
1321		return;
1322	wcl1 = *wcl;
1323	*wcl = wcl1->wcl_next;
1324	witness_child_free(wcl1);
1325}
1326
1327static int
1328isitmychild(struct witness *parent, struct witness *child)
1329{
1330	struct witness_child_list_entry *wcl;
1331	int i;
1332
1333	for (wcl = parent->w_children; wcl != NULL; wcl = wcl->wcl_next) {
1334		for (i = 0; i < wcl->wcl_count; i++) {
1335			if (wcl->wcl_children[i] == child)
1336				return (1);
1337		}
1338	}
1339	return (0);
1340}
1341
1342static int
1343isitmydescendant(struct witness *parent, struct witness *child)
1344{
1345	struct witness_child_list_entry *wcl;
1346	int i, j;
1347
1348	if (isitmychild(parent, child))
1349		return (1);
1350	j = 0;
1351	for (wcl = parent->w_children; wcl != NULL; wcl = wcl->wcl_next) {
1352		MPASS(j < 1000);
1353		for (i = 0; i < wcl->wcl_count; i++) {
1354			if (isitmydescendant(wcl->wcl_children[i], child))
1355				return (1);
1356		}
1357		j++;
1358	}
1359	return (0);
1360}
1361
1362static void
1363witness_levelall (void)
1364{
1365	struct witness_list *list;
1366	struct witness *w, *w1;
1367
1368	/*
1369	 * First clear all levels.
1370	 */
1371	STAILQ_FOREACH(w, &w_all, w_list) {
1372		w->w_level = 0;
1373	}
1374
1375	/*
1376	 * Look for locks with no parent and level all their descendants.
1377	 */
1378	STAILQ_FOREACH(w, &w_all, w_list) {
1379		/*
1380		 * This is just an optimization, technically we could get
1381		 * away just walking the all list each time.
1382		 */
1383		if (w->w_class->lc_flags & LC_SLEEPLOCK)
1384			list = &w_sleep;
1385		else
1386			list = &w_spin;
1387		STAILQ_FOREACH(w1, list, w_typelist) {
1388			if (isitmychild(w1, w))
1389				goto skip;
1390		}
1391		witness_leveldescendents(w, 0);
1392	skip:
1393		;	/* silence GCC 3.x */
1394	}
1395}
1396
1397static void
1398witness_leveldescendents(struct witness *parent, int level)
1399{
1400	struct witness_child_list_entry *wcl;
1401	int i;
1402
1403	if (parent->w_level < level)
1404		parent->w_level = level;
1405	level++;
1406	for (wcl = parent->w_children; wcl != NULL; wcl = wcl->wcl_next)
1407		for (i = 0; i < wcl->wcl_count; i++)
1408			witness_leveldescendents(wcl->wcl_children[i], level);
1409}
1410
1411static void
1412witness_displaydescendants(void(*prnt)(const char *fmt, ...),
1413			   struct witness *parent, int indent)
1414{
1415	struct witness_child_list_entry *wcl;
1416	int i, level;
1417
1418	level = parent->w_level;
1419	prnt("%-2d", level);
1420	for (i = 0; i < indent; i++)
1421		prnt(" ");
1422	if (parent->w_refcount > 0)
1423		prnt("%s", parent->w_name);
1424	else
1425		prnt("(dead)");
1426	if (parent->w_displayed) {
1427		prnt(" -- (already displayed)\n");
1428		return;
1429	}
1430	parent->w_displayed = 1;
1431	if (parent->w_refcount > 0) {
1432		if (parent->w_file != NULL)
1433			prnt(" -- last acquired @ %s:%d", parent->w_file,
1434			    parent->w_line);
1435	}
1436	prnt("\n");
1437	for (wcl = parent->w_children; wcl != NULL; wcl = wcl->wcl_next)
1438		for (i = 0; i < wcl->wcl_count; i++)
1439			    witness_displaydescendants(prnt,
1440				wcl->wcl_children[i], indent + 1);
1441}
1442
1443#ifdef BLESSING
1444static int
1445blessed(struct witness *w1, struct witness *w2)
1446{
1447	int i;
1448	struct witness_blessed *b;
1449
1450	for (i = 0; i < blessed_count; i++) {
1451		b = &blessed_list[i];
1452		if (strcmp(w1->w_name, b->b_lock1) == 0) {
1453			if (strcmp(w2->w_name, b->b_lock2) == 0)
1454				return (1);
1455			continue;
1456		}
1457		if (strcmp(w1->w_name, b->b_lock2) == 0)
1458			if (strcmp(w2->w_name, b->b_lock1) == 0)
1459				return (1);
1460	}
1461	return (0);
1462}
1463#endif
1464
1465static struct witness *
1466witness_get(void)
1467{
1468	struct witness *w;
1469
1470	if (witness_watch == 0) {
1471		mtx_unlock_spin(&w_mtx);
1472		return (NULL);
1473	}
1474	if (STAILQ_EMPTY(&w_free)) {
1475		witness_watch = 0;
1476		mtx_unlock_spin(&w_mtx);
1477		printf("%s: witness exhausted\n", __func__);
1478		return (NULL);
1479	}
1480	w = STAILQ_FIRST(&w_free);
1481	STAILQ_REMOVE_HEAD(&w_free, w_list);
1482	bzero(w, sizeof(*w));
1483	return (w);
1484}
1485
1486static void
1487witness_free(struct witness *w)
1488{
1489
1490	STAILQ_INSERT_HEAD(&w_free, w, w_list);
1491}
1492
1493static struct witness_child_list_entry *
1494witness_child_get(void)
1495{
1496	struct witness_child_list_entry *wcl;
1497
1498	if (witness_watch == 0) {
1499		mtx_unlock_spin(&w_mtx);
1500		return (NULL);
1501	}
1502	wcl = w_child_free;
1503	if (wcl == NULL) {
1504		witness_watch = 0;
1505		mtx_unlock_spin(&w_mtx);
1506		printf("%s: witness exhausted\n", __func__);
1507		return (NULL);
1508	}
1509	w_child_free = wcl->wcl_next;
1510	bzero(wcl, sizeof(*wcl));
1511	return (wcl);
1512}
1513
1514static void
1515witness_child_free(struct witness_child_list_entry *wcl)
1516{
1517
1518	wcl->wcl_next = w_child_free;
1519	w_child_free = wcl;
1520}
1521
1522static struct lock_list_entry *
1523witness_lock_list_get(void)
1524{
1525	struct lock_list_entry *lle;
1526
1527	if (witness_watch == 0)
1528		return (NULL);
1529	mtx_lock_spin(&w_mtx);
1530	lle = w_lock_list_free;
1531	if (lle == NULL) {
1532		witness_watch = 0;
1533		mtx_unlock_spin(&w_mtx);
1534		printf("%s: witness exhausted\n", __func__);
1535		return (NULL);
1536	}
1537	w_lock_list_free = lle->ll_next;
1538	mtx_unlock_spin(&w_mtx);
1539	bzero(lle, sizeof(*lle));
1540	return (lle);
1541}
1542
1543static void
1544witness_lock_list_free(struct lock_list_entry *lle)
1545{
1546
1547	mtx_lock_spin(&w_mtx);
1548	lle->ll_next = w_lock_list_free;
1549	w_lock_list_free = lle;
1550	mtx_unlock_spin(&w_mtx);
1551}
1552
1553static struct lock_instance *
1554find_instance(struct lock_list_entry *lock_list, struct lock_object *lock)
1555{
1556	struct lock_list_entry *lle;
1557	struct lock_instance *instance;
1558	int i;
1559
1560	for (lle = lock_list; lle != NULL; lle = lle->ll_next)
1561		for (i = lle->ll_count - 1; i >= 0; i--) {
1562			instance = &lle->ll_children[i];
1563			if (instance->li_lock == lock)
1564				return (instance);
1565		}
1566	return (NULL);
1567}
1568
1569static void
1570witness_list_lock(struct lock_instance *instance)
1571{
1572	struct lock_object *lock;
1573
1574	lock = instance->li_lock;
1575	printf("%s %s %s", (instance->li_flags & LI_EXCLUSIVE) != 0 ?
1576	    "exclusive" : "shared", lock->lo_class->lc_name, lock->lo_name);
1577	if (lock->lo_type != lock->lo_name)
1578		printf(" (%s)", lock->lo_type);
1579	printf(" r = %d (%p) locked @ %s:%d\n",
1580	    instance->li_flags & LI_RECURSEMASK, lock, instance->li_file,
1581	    instance->li_line);
1582}
1583
1584int
1585witness_list_locks(struct lock_list_entry **lock_list)
1586{
1587	struct lock_list_entry *lle;
1588	int i, nheld;
1589
1590	nheld = 0;
1591	for (lle = *lock_list; lle != NULL; lle = lle->ll_next)
1592		for (i = lle->ll_count - 1; i >= 0; i--) {
1593			witness_list_lock(&lle->ll_children[i]);
1594			nheld++;
1595		}
1596	return (nheld);
1597}
1598
1599void
1600witness_save(struct lock_object *lock, const char **filep, int *linep)
1601{
1602	struct lock_instance *instance;
1603
1604	KASSERT(!witness_cold, ("%s: witness_cold", __func__));
1605	if (lock->lo_witness == NULL || witness_watch == 0 || panicstr != NULL)
1606		return;
1607	if ((lock->lo_class->lc_flags & LC_SLEEPLOCK) == 0)
1608		panic("%s: lock (%s) %s is not a sleep lock", __func__,
1609		    lock->lo_class->lc_name, lock->lo_name);
1610	instance = find_instance(curthread->td_sleeplocks, lock);
1611	if (instance == NULL)
1612		panic("%s: lock (%s) %s not locked", __func__,
1613		    lock->lo_class->lc_name, lock->lo_name);
1614	*filep = instance->li_file;
1615	*linep = instance->li_line;
1616}
1617
1618void
1619witness_restore(struct lock_object *lock, const char *file, int line)
1620{
1621	struct lock_instance *instance;
1622
1623	KASSERT(!witness_cold, ("%s: witness_cold", __func__));
1624	if (lock->lo_witness == NULL || witness_watch == 0 || panicstr != NULL)
1625		return;
1626	if ((lock->lo_class->lc_flags & LC_SLEEPLOCK) == 0)
1627		panic("%s: lock (%s) %s is not a sleep lock", __func__,
1628		    lock->lo_class->lc_name, lock->lo_name);
1629	instance = find_instance(curthread->td_sleeplocks, lock);
1630	if (instance == NULL)
1631		panic("%s: lock (%s) %s not locked", __func__,
1632		    lock->lo_class->lc_name, lock->lo_name);
1633	lock->lo_witness->w_file = file;
1634	lock->lo_witness->w_line = line;
1635	instance->li_file = file;
1636	instance->li_line = line;
1637}
1638
1639void
1640witness_assert(struct lock_object *lock, int flags, const char *file, int line)
1641{
1642#ifdef INVARIANT_SUPPORT
1643	struct lock_instance *instance;
1644
1645	if (lock->lo_witness == NULL || witness_watch == 0 || panicstr != NULL)
1646		return;
1647	if ((lock->lo_class->lc_flags & LC_SLEEPLOCK) != 0)
1648		instance = find_instance(curthread->td_sleeplocks, lock);
1649	else if ((lock->lo_class->lc_flags & LC_SPINLOCK) != 0)
1650		instance = find_instance(PCPU_GET(spinlocks), lock);
1651	else {
1652		panic("Lock (%s) %s is not sleep or spin!",
1653		    lock->lo_class->lc_name, lock->lo_name);
1654	}
1655	file = fixup_filename(file);
1656	switch (flags) {
1657	case LA_UNLOCKED:
1658		if (instance != NULL)
1659			panic("Lock (%s) %s locked @ %s:%d.",
1660			    lock->lo_class->lc_name, lock->lo_name, file, line);
1661		break;
1662	case LA_LOCKED:
1663	case LA_LOCKED | LA_RECURSED:
1664	case LA_LOCKED | LA_NOTRECURSED:
1665	case LA_SLOCKED:
1666	case LA_SLOCKED | LA_RECURSED:
1667	case LA_SLOCKED | LA_NOTRECURSED:
1668	case LA_XLOCKED:
1669	case LA_XLOCKED | LA_RECURSED:
1670	case LA_XLOCKED | LA_NOTRECURSED:
1671		if (instance == NULL) {
1672			panic("Lock (%s) %s not locked @ %s:%d.",
1673			    lock->lo_class->lc_name, lock->lo_name, file, line);
1674			break;
1675		}
1676		if ((flags & LA_XLOCKED) != 0 &&
1677		    (instance->li_flags & LI_EXCLUSIVE) == 0)
1678			panic("Lock (%s) %s not exclusively locked @ %s:%d.",
1679			    lock->lo_class->lc_name, lock->lo_name, file, line);
1680		if ((flags & LA_SLOCKED) != 0 &&
1681		    (instance->li_flags & LI_EXCLUSIVE) != 0)
1682			panic("Lock (%s) %s exclusively locked @ %s:%d.",
1683			    lock->lo_class->lc_name, lock->lo_name, file, line);
1684		if ((flags & LA_RECURSED) != 0 &&
1685		    (instance->li_flags & LI_RECURSEMASK) == 0)
1686			panic("Lock (%s) %s not recursed @ %s:%d.",
1687			    lock->lo_class->lc_name, lock->lo_name, file, line);
1688		if ((flags & LA_NOTRECURSED) != 0 &&
1689		    (instance->li_flags & LI_RECURSEMASK) != 0)
1690			panic("Lock (%s) %s recursed @ %s:%d.",
1691			    lock->lo_class->lc_name, lock->lo_name, file, line);
1692		break;
1693	default:
1694		panic("Invalid lock assertion at %s:%d.", file, line);
1695
1696	}
1697#endif	/* INVARIANT_SUPPORT */
1698}
1699
1700#ifdef DDB
1701static void
1702witness_list(struct thread *td)
1703{
1704
1705	KASSERT(!witness_cold, ("%s: witness_cold", __func__));
1706	KASSERT(db_active, ("%s: not in the debugger", __func__));
1707
1708	if (witness_watch == 0)
1709		return;
1710
1711	witness_list_locks(&td->td_sleeplocks);
1712
1713	/*
1714	 * We only handle spinlocks if td == curthread.  This is somewhat broken
1715	 * if td is currently executing on some other CPU and holds spin locks
1716	 * as we won't display those locks.  If we had a MI way of getting
1717	 * the per-cpu data for a given cpu then we could use
1718	 * td->td_oncpu to get the list of spinlocks for this thread
1719	 * and "fix" this.
1720	 *
1721	 * That still wouldn't really fix this unless we locked sched_lock
1722	 * or stopped the other CPU to make sure it wasn't changing the list
1723	 * out from under us.  It is probably best to just not try to handle
1724	 * threads on other CPU's for now.
1725	 */
1726	if (td == curthread && PCPU_GET(spinlocks) != NULL)
1727		witness_list_locks(PCPU_PTR(spinlocks));
1728}
1729
1730DB_SHOW_COMMAND(locks, db_witness_list)
1731{
1732	struct thread *td;
1733	pid_t pid;
1734	struct proc *p;
1735
1736	if (have_addr) {
1737		pid = (addr % 16) + ((addr >> 4) % 16) * 10 +
1738		    ((addr >> 8) % 16) * 100 + ((addr >> 12) % 16) * 1000 +
1739		    ((addr >> 16) % 16) * 10000;
1740		/* sx_slock(&allproc_lock); */
1741		FOREACH_PROC_IN_SYSTEM(p) {
1742			if (p->p_pid == pid)
1743				break;
1744		}
1745		/* sx_sunlock(&allproc_lock); */
1746		if (p == NULL) {
1747			db_printf("pid %d not found\n", pid);
1748			return;
1749		}
1750		FOREACH_THREAD_IN_PROC(p, td) {
1751			witness_list(td);
1752		}
1753	} else {
1754		td = curthread;
1755		witness_list(td);
1756	}
1757}
1758
1759DB_SHOW_COMMAND(witness, db_witness_display)
1760{
1761
1762	witness_display(db_printf);
1763}
1764#endif
1765