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("$FreeBSD$"); 86 87 #include "opt_ddb.h" 88 #include "opt_witness.h" 89 90 #include <sys/param.h> 91 #include <sys/bus.h> 92 #include <sys/kernel.h> 93 #include <sys/ktr.h> 94 #include <sys/lock.h> 95 #include <sys/malloc.h> 96 #include <sys/mutex.h> 97 #include <sys/proc.h> 98 #include <sys/sysctl.h> 99 #include <sys/systm.h> 100 101 #include <ddb/ddb.h> 102 103 #include <machine/stdarg.h> 104 105 /* Define this to check for blessed mutexes */ 106 #undef BLESSING 107 108 #define WITNESS_COUNT 200 109 #define WITNESS_CHILDCOUNT (WITNESS_COUNT * 4) 110 /* 111 * XXX: This is somewhat bogus, as we assume here that at most 1024 threads 112 * will hold LOCK_NCHILDREN * 2 locks. We handle failure ok, and we should 113 * probably be safe for the most part, but it's still a SWAG. 114 */ 115 #define LOCK_CHILDCOUNT (MAXCPU + 1024) * 2 116 117 #define WITNESS_NCHILDREN 6 118 119 struct witness_child_list_entry; 120 121 struct witness { 122 const char *w_name; 123 struct lock_class *w_class; 124 STAILQ_ENTRY(witness) w_list; /* List of all witnesses. */ 125 STAILQ_ENTRY(witness) w_typelist; /* Witnesses of a type. */ 126 struct witness_child_list_entry *w_children; /* Great evilness... */ 127 const char *w_file; 128 int w_line; 129 u_int w_level; 130 u_int w_refcount; 131 u_char w_Giant_squawked:1; 132 u_char w_other_squawked:1; 133 u_char w_same_squawked:1; 134 u_char w_displayed:1; 135 }; 136 137 struct witness_child_list_entry { 138 struct witness_child_list_entry *wcl_next; 139 struct witness *wcl_children[WITNESS_NCHILDREN]; 140 u_int wcl_count; 141 }; 142 143 STAILQ_HEAD(witness_list, witness); 144 145 #ifdef BLESSING 146 struct witness_blessed { 147 const char *b_lock1; 148 const char *b_lock2; 149 }; 150 #endif 151 152 struct witness_order_list_entry { 153 const char *w_name; 154 struct lock_class *w_class; 155 }; 156 157 #ifdef BLESSING 158 static int blessed(struct witness *, struct witness *); 159 #endif 160 static int depart(struct witness *w); 161 static struct witness *enroll(const char *description, 162 struct lock_class *lock_class); 163 static int insertchild(struct witness *parent, struct witness *child); 164 static int isitmychild(struct witness *parent, struct witness *child); 165 static int isitmydescendant(struct witness *parent, struct witness *child); 166 static int itismychild(struct witness *parent, struct witness *child); 167 static int rebalancetree(struct witness_list *list); 168 static void removechild(struct witness *parent, struct witness *child); 169 static int reparentchildren(struct witness *newparent, 170 struct witness *oldparent); 171 static int sysctl_debug_witness_watch(SYSCTL_HANDLER_ARGS); 172 static void witness_displaydescendants(void(*)(const char *fmt, ...), 173 struct witness *, int indent); 174 static const char *fixup_filename(const char *file); 175 static void witness_leveldescendents(struct witness *parent, int level); 176 static void witness_levelall(void); 177 static struct witness *witness_get(void); 178 static void witness_free(struct witness *m); 179 static struct witness_child_list_entry *witness_child_get(void); 180 static void witness_child_free(struct witness_child_list_entry *wcl); 181 static struct lock_list_entry *witness_lock_list_get(void); 182 static void witness_lock_list_free(struct lock_list_entry *lle); 183 static struct lock_instance *find_instance(struct lock_list_entry *lock_list, 184 struct lock_object *lock); 185 static void witness_list_lock(struct lock_instance *instance); 186 #ifdef DDB 187 static void witness_list(struct thread *td); 188 static void witness_display_list(void(*prnt)(const char *fmt, ...), 189 struct witness_list *list); 190 static void witness_display(void(*)(const char *fmt, ...)); 191 #endif 192 193 MALLOC_DEFINE(M_WITNESS, "witness", "witness structure"); 194 195 /* 196 * If set to 0, witness is disabled. If set to 1, witness performs full lock 197 * order checking for all locks. If set to 2 or higher, then witness skips 198 * the full lock order check if the lock being acquired is at a higher level 199 * (i.e. farther down in the tree) than the current lock. This last mode is 200 * somewhat experimental and not considered fully safe. At runtime, this 201 * value may be set to 0 to turn off witness. witness is not allowed be 202 * turned on once it is turned off, however. 203 */ 204 static int witness_watch = 1; 205 TUNABLE_INT("debug.witness_watch", &witness_watch); 206 SYSCTL_PROC(_debug, OID_AUTO, witness_watch, CTLFLAG_RW | CTLTYPE_INT, NULL, 0, 207 sysctl_debug_witness_watch, "I", "witness is watching lock operations"); 208 209 #ifdef DDB 210 /* 211 * When DDB is enabled and witness_ddb is set to 1, it will cause the system to 212 * drop into kdebug() when: 213 * - a lock heirarchy violation occurs 214 * - locks are held when going to sleep. 215 */ 216 #ifdef WITNESS_DDB 217 int witness_ddb = 1; 218 #else 219 int witness_ddb = 0; 220 #endif 221 TUNABLE_INT("debug.witness_ddb", &witness_ddb); 222 SYSCTL_INT(_debug, OID_AUTO, witness_ddb, CTLFLAG_RW, &witness_ddb, 0, ""); 223 224 /* 225 * When DDB is enabled and witness_trace is set to 1, it will cause the system 226 * to print a stack trace: 227 * - a lock heirarchy violation occurs 228 * - locks are held when going to sleep. 229 */ 230 int witness_trace = 1; 231 TUNABLE_INT("debug.witness_trace", &witness_trace); 232 SYSCTL_INT(_debug, OID_AUTO, witness_trace, CTLFLAG_RW, &witness_trace, 0, ""); 233 #endif /* DDB */ 234 235 #ifdef WITNESS_SKIPSPIN 236 int witness_skipspin = 1; 237 #else 238 int witness_skipspin = 0; 239 #endif 240 TUNABLE_INT("debug.witness_skipspin", &witness_skipspin); 241 SYSCTL_INT(_debug, OID_AUTO, witness_skipspin, CTLFLAG_RDTUN, &witness_skipspin, 0, 242 ""); 243 244 static struct mtx w_mtx; 245 static struct witness_list w_free = STAILQ_HEAD_INITIALIZER(w_free); 246 static struct witness_list w_all = STAILQ_HEAD_INITIALIZER(w_all); 247 static struct witness_list w_spin = STAILQ_HEAD_INITIALIZER(w_spin); 248 static struct witness_list w_sleep = STAILQ_HEAD_INITIALIZER(w_sleep); 249 static struct witness_child_list_entry *w_child_free = NULL; 250 static struct lock_list_entry *w_lock_list_free = NULL; 251 252 static struct witness w_data[WITNESS_COUNT]; 253 static struct witness_child_list_entry w_childdata[WITNESS_CHILDCOUNT]; 254 static struct lock_list_entry w_locklistdata[LOCK_CHILDCOUNT]; 255 256 static struct witness_order_list_entry order_lists[] = { 257 { "proctree", &lock_class_sx }, 258 { "allproc", &lock_class_sx }, 259 { "Giant", &lock_class_mtx_sleep }, 260 { "filedesc structure", &lock_class_mtx_sleep }, 261 { "pipe mutex", &lock_class_mtx_sleep }, 262 { "sigio lock", &lock_class_mtx_sleep }, 263 { "process group", &lock_class_mtx_sleep }, 264 { "process lock", &lock_class_mtx_sleep }, 265 { "session", &lock_class_mtx_sleep }, 266 { "uidinfo hash", &lock_class_mtx_sleep }, 267 { "uidinfo struct", &lock_class_mtx_sleep }, 268 { "allprison", &lock_class_mtx_sleep }, 269 { NULL, NULL }, 270 /* 271 * spin locks 272 */ 273 #ifdef SMP 274 { "ap boot", &lock_class_mtx_spin }, 275 #endif 276 { "sio", &lock_class_mtx_spin }, 277 #ifdef __i386__ 278 { "cy", &lock_class_mtx_spin }, 279 #endif 280 { "sabtty", &lock_class_mtx_spin }, 281 { "zstty", &lock_class_mtx_spin }, 282 { "ng_node", &lock_class_mtx_spin }, 283 { "ng_worklist", &lock_class_mtx_spin }, 284 { "taskqueue_fast", &lock_class_mtx_spin }, 285 { "intr table", &lock_class_mtx_spin }, 286 { "ithread table lock", &lock_class_mtx_spin }, 287 { "sched lock", &lock_class_mtx_spin }, 288 { "turnstile chain", &lock_class_mtx_spin }, 289 { "td_contested", &lock_class_mtx_spin }, 290 { "callout", &lock_class_mtx_spin }, 291 { "entropy harvest", &lock_class_mtx_spin }, 292 { "entropy harvest buffers", &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(__amd64__) 302 { "tlb", &lock_class_mtx_spin }, 303 { "lazypmap", &lock_class_mtx_spin }, 304 #endif 305 #ifdef __sparc64__ 306 { "ipi", &lock_class_mtx_spin }, 307 #endif 308 #endif 309 { "clk", &lock_class_mtx_spin }, 310 { "mutex profiling lock", &lock_class_mtx_spin }, 311 { "kse zombie lock", &lock_class_mtx_spin }, 312 { "ALD Queue", &lock_class_mtx_spin }, 313 #ifdef __ia64__ 314 { "MCA spin lock", &lock_class_mtx_spin }, 315 #endif 316 #if defined(__i386__) || defined(__amd64__) 317 { "pcicfg", &lock_class_mtx_spin }, 318 #endif 319 { NULL, NULL }, 320 { NULL, NULL } 321 }; 322 323 #ifdef BLESSING 324 /* 325 * Pairs of locks which have been blessed 326 * Don't complain about order problems with blessed locks 327 */ 328 static struct witness_blessed blessed_list[] = { 329 }; 330 static int blessed_count = 331 sizeof(blessed_list) / sizeof(struct witness_blessed); 332 #endif 333 334 /* 335 * List of all locks in the system. 336 */ 337 TAILQ_HEAD(, lock_object) all_locks = TAILQ_HEAD_INITIALIZER(all_locks); 338 339 static struct mtx all_mtx = { 340 { &lock_class_mtx_sleep, /* mtx_object.lo_class */ 341 "All locks list", /* mtx_object.lo_name */ 342 "All locks list", /* mtx_object.lo_type */ 343 LO_INITIALIZED, /* mtx_object.lo_flags */ 344 { NULL, NULL }, /* mtx_object.lo_list */ 345 NULL }, /* mtx_object.lo_witness */ 346 MTX_UNOWNED, 0 /* mtx_lock, mtx_recurse */ 347 }; 348 349 /* 350 * This global is set to 0 once it becomes safe to use the witness code. 351 */ 352 static int witness_cold = 1; 353 354 /* 355 * Global variables for book keeping. 356 */ 357 static int lock_cur_cnt; 358 static int lock_max_cnt; 359 360 /* 361 * The WITNESS-enabled diagnostic code. 362 */ 363 static void 364 witness_initialize(void *dummy __unused) 365 { 366 struct lock_object *lock; 367 struct witness_order_list_entry *order; 368 struct witness *w, *w1; 369 int i; 370 371 /* 372 * We have to release Giant before initializing its witness 373 * structure so that WITNESS doesn't get confused. 374 */ 375 mtx_unlock(&Giant); 376 mtx_assert(&Giant, MA_NOTOWNED); 377 378 CTR1(KTR_WITNESS, "%s: initializing witness", __func__); 379 TAILQ_INSERT_HEAD(&all_locks, &all_mtx.mtx_object, lo_list); 380 mtx_init(&w_mtx, "witness lock", NULL, MTX_SPIN | MTX_QUIET | 381 MTX_NOWITNESS); 382 for (i = 0; i < WITNESS_COUNT; i++) 383 witness_free(&w_data[i]); 384 for (i = 0; i < WITNESS_CHILDCOUNT; i++) 385 witness_child_free(&w_childdata[i]); 386 for (i = 0; i < LOCK_CHILDCOUNT; i++) 387 witness_lock_list_free(&w_locklistdata[i]); 388 389 /* First add in all the specified order lists. */ 390 for (order = order_lists; order->w_name != NULL; order++) { 391 w = enroll(order->w_name, order->w_class); 392 if (w == NULL) 393 continue; 394 w->w_file = "order list"; 395 for (order++; order->w_name != NULL; order++) { 396 w1 = enroll(order->w_name, order->w_class); 397 if (w1 == NULL) 398 continue; 399 w1->w_file = "order list"; 400 if (!itismychild(w, w1)) 401 panic("Not enough memory for static orders!"); 402 w = w1; 403 } 404 } 405 406 /* Iterate through all locks and add them to witness. */ 407 mtx_lock(&all_mtx); 408 TAILQ_FOREACH(lock, &all_locks, lo_list) { 409 if (lock->lo_flags & LO_WITNESS) 410 lock->lo_witness = enroll(lock->lo_type, 411 lock->lo_class); 412 else 413 lock->lo_witness = NULL; 414 } 415 mtx_unlock(&all_mtx); 416 417 /* Mark the witness code as being ready for use. */ 418 atomic_store_rel_int(&witness_cold, 0); 419 420 mtx_lock(&Giant); 421 } 422 SYSINIT(witness_init, SI_SUB_WITNESS, SI_ORDER_FIRST, witness_initialize, NULL) 423 424 static int 425 sysctl_debug_witness_watch(SYSCTL_HANDLER_ARGS) 426 { 427 int error, value; 428 429 value = witness_watch; 430 error = sysctl_handle_int(oidp, &value, 0, req); 431 if (error != 0 || req->newptr == NULL) 432 return (error); 433 error = suser(req->td); 434 if (error != 0) 435 return (error); 436 if (value == witness_watch) 437 return (0); 438 if (value != 0) 439 return (EINVAL); 440 witness_watch = 0; 441 return (0); 442 } 443 444 void 445 witness_init(struct lock_object *lock) 446 { 447 struct lock_class *class; 448 449 class = lock->lo_class; 450 if (lock->lo_flags & LO_INITIALIZED) 451 panic("%s: lock (%s) %s is already initialized", __func__, 452 class->lc_name, lock->lo_name); 453 if ((lock->lo_flags & LO_RECURSABLE) != 0 && 454 (class->lc_flags & LC_RECURSABLE) == 0) 455 panic("%s: lock (%s) %s can not be recursable", __func__, 456 class->lc_name, lock->lo_name); 457 if ((lock->lo_flags & LO_SLEEPABLE) != 0 && 458 (class->lc_flags & LC_SLEEPABLE) == 0) 459 panic("%s: lock (%s) %s can not be sleepable", __func__, 460 class->lc_name, lock->lo_name); 461 if ((lock->lo_flags & LO_UPGRADABLE) != 0 && 462 (class->lc_flags & LC_UPGRADABLE) == 0) 463 panic("%s: lock (%s) %s can not be upgradable", __func__, 464 class->lc_name, lock->lo_name); 465 466 mtx_lock(&all_mtx); 467 TAILQ_INSERT_TAIL(&all_locks, lock, lo_list); 468 lock->lo_flags |= LO_INITIALIZED; 469 lock_cur_cnt++; 470 if (lock_cur_cnt > lock_max_cnt) 471 lock_max_cnt = lock_cur_cnt; 472 mtx_unlock(&all_mtx); 473 if (!witness_cold && witness_watch != 0 && panicstr == NULL && 474 (lock->lo_flags & LO_WITNESS) != 0) 475 lock->lo_witness = enroll(lock->lo_type, class); 476 else 477 lock->lo_witness = NULL; 478 } 479 480 void 481 witness_destroy(struct lock_object *lock) 482 { 483 struct witness *w; 484 485 if (witness_cold) 486 panic("lock (%s) %s destroyed while witness_cold", 487 lock->lo_class->lc_name, lock->lo_name); 488 if ((lock->lo_flags & LO_INITIALIZED) == 0) 489 panic("%s: lock (%s) %s is not initialized", __func__, 490 lock->lo_class->lc_name, lock->lo_name); 491 492 /* XXX: need to verify that no one holds the lock */ 493 w = lock->lo_witness; 494 if (w != NULL) { 495 mtx_lock_spin(&w_mtx); 496 MPASS(w->w_refcount > 0); 497 w->w_refcount--; 498 499 /* 500 * Lock is already released if we have an allocation failure 501 * and depart() fails. 502 */ 503 if (w->w_refcount != 0 || depart(w)) 504 mtx_unlock_spin(&w_mtx); 505 } 506 507 mtx_lock(&all_mtx); 508 lock_cur_cnt--; 509 TAILQ_REMOVE(&all_locks, lock, lo_list); 510 lock->lo_flags &= ~LO_INITIALIZED; 511 mtx_unlock(&all_mtx); 512 } 513 514 #ifdef DDB 515 static void 516 witness_display_list(void(*prnt)(const char *fmt, ...), 517 struct witness_list *list) 518 { 519 struct witness *w; 520 521 STAILQ_FOREACH(w, list, w_typelist) { 522 if (w->w_file == NULL || w->w_level > 0) 523 continue; 524 /* 525 * This lock has no anscestors, display its descendants. 526 */ 527 witness_displaydescendants(prnt, w, 0); 528 } 529 } 530 531 static void 532 witness_display(void(*prnt)(const char *fmt, ...)) 533 { 534 struct witness *w; 535 536 KASSERT(!witness_cold, ("%s: witness_cold", __func__)); 537 witness_levelall(); 538 539 /* Clear all the displayed flags. */ 540 STAILQ_FOREACH(w, &w_all, w_list) { 541 w->w_displayed = 0; 542 } 543 544 /* 545 * First, handle sleep locks which have been acquired at least 546 * once. 547 */ 548 prnt("Sleep locks:\n"); 549 witness_display_list(prnt, &w_sleep); 550 551 /* 552 * Now do spin locks which have been acquired at least once. 553 */ 554 prnt("\nSpin locks:\n"); 555 witness_display_list(prnt, &w_spin); 556 557 /* 558 * Finally, any locks which have not been acquired yet. 559 */ 560 prnt("\nLocks which were never acquired:\n"); 561 STAILQ_FOREACH(w, &w_all, w_list) { 562 if (w->w_file != NULL || w->w_refcount == 0) 563 continue; 564 prnt("%s\n", w->w_name); 565 } 566 } 567 #endif /* DDB */ 568 569 /* Trim useless garbage from filenames. */ 570 static const char * 571 fixup_filename(const char *file) 572 { 573 574 if (file == NULL) 575 return (NULL); 576 while (strncmp(file, "../", 3) == 0) 577 file += 3; 578 return (file); 579 } 580 581 void 582 witness_lock(struct lock_object *lock, int flags, const char *file, int line) 583 { 584 struct lock_list_entry **lock_list, *lle; 585 struct lock_instance *lock1, *lock2; 586 struct lock_class *class; 587 struct witness *w, *w1; 588 struct thread *td; 589 int i, j; 590 #ifdef DDB 591 int go_into_ddb = 0; 592 #endif 593 594 if (witness_cold || witness_watch == 0 || lock->lo_witness == NULL || 595 panicstr != NULL) 596 return; 597 w = lock->lo_witness; 598 class = lock->lo_class; 599 td = curthread; 600 file = fixup_filename(file); 601 602 if (class->lc_flags & LC_SLEEPLOCK) { 603 /* 604 * Since spin locks include a critical section, this check 605 * impliclty enforces a lock order of all sleep locks before 606 * all spin locks. 607 */ 608 if (td->td_critnest != 0 && (flags & LOP_TRYLOCK) == 0) 609 panic("blockable sleep lock (%s) %s @ %s:%d", 610 class->lc_name, lock->lo_name, file, line); 611 lock_list = &td->td_sleeplocks; 612 } else 613 lock_list = PCPU_PTR(spinlocks); 614 615 /* 616 * Is this the first lock acquired? If so, then no order checking 617 * is needed. 618 */ 619 if (*lock_list == NULL) 620 goto out; 621 622 /* 623 * Check to see if we are recursing on a lock we already own. 624 */ 625 lock1 = find_instance(*lock_list, lock); 626 if (lock1 != NULL) { 627 if ((lock1->li_flags & LI_EXCLUSIVE) != 0 && 628 (flags & LOP_EXCLUSIVE) == 0) { 629 printf("shared lock of (%s) %s @ %s:%d\n", 630 class->lc_name, lock->lo_name, file, line); 631 printf("while exclusively locked from %s:%d\n", 632 lock1->li_file, lock1->li_line); 633 panic("share->excl"); 634 } 635 if ((lock1->li_flags & LI_EXCLUSIVE) == 0 && 636 (flags & LOP_EXCLUSIVE) != 0) { 637 printf("exclusive lock of (%s) %s @ %s:%d\n", 638 class->lc_name, lock->lo_name, file, line); 639 printf("while share locked from %s:%d\n", 640 lock1->li_file, lock1->li_line); 641 panic("excl->share"); 642 } 643 lock1->li_flags++; 644 if ((lock->lo_flags & LO_RECURSABLE) == 0) { 645 printf( 646 "recursed on non-recursive lock (%s) %s @ %s:%d\n", 647 class->lc_name, lock->lo_name, file, line); 648 printf("first acquired @ %s:%d\n", lock1->li_file, 649 lock1->li_line); 650 panic("recurse"); 651 } 652 CTR4(KTR_WITNESS, "%s: pid %d recursed on %s r=%d", __func__, 653 td->td_proc->p_pid, lock->lo_name, 654 lock1->li_flags & LI_RECURSEMASK); 655 lock1->li_file = file; 656 lock1->li_line = line; 657 return; 658 } 659 660 /* 661 * Try locks do not block if they fail to acquire the lock, thus 662 * there is no danger of deadlocks or of switching while holding a 663 * spin lock if we acquire a lock via a try operation. 664 */ 665 if (flags & LOP_TRYLOCK) 666 goto out; 667 668 /* 669 * Check for duplicate locks of the same type. Note that we only 670 * have to check for this on the last lock we just acquired. Any 671 * other cases will be caught as lock order violations. 672 */ 673 lock1 = &(*lock_list)->ll_children[(*lock_list)->ll_count - 1]; 674 w1 = lock1->li_lock->lo_witness; 675 if (w1 == w) { 676 if (w->w_same_squawked || (lock->lo_flags & LO_DUPOK)) 677 goto out; 678 w->w_same_squawked = 1; 679 printf("acquiring duplicate lock of same type: \"%s\"\n", 680 lock->lo_type); 681 printf(" 1st %s @ %s:%d\n", lock1->li_lock->lo_name, 682 lock1->li_file, lock1->li_line); 683 printf(" 2nd %s @ %s:%d\n", lock->lo_name, file, line); 684 #ifdef DDB 685 go_into_ddb = 1; 686 #endif 687 goto out; 688 } 689 MPASS(!mtx_owned(&w_mtx)); 690 mtx_lock_spin(&w_mtx); 691 /* 692 * If we have a known higher number just say ok 693 */ 694 if (witness_watch > 1 && w->w_level > w1->w_level) { 695 mtx_unlock_spin(&w_mtx); 696 goto out; 697 } 698 /* 699 * If we know that the the lock we are acquiring comes after 700 * the lock we most recently acquired in the lock order tree, 701 * then there is no need for any further checks. 702 */ 703 if (isitmydescendant(w1, w)) { 704 mtx_unlock_spin(&w_mtx); 705 goto out; 706 } 707 for (j = 0, lle = *lock_list; lle != NULL; lle = lle->ll_next) { 708 for (i = lle->ll_count - 1; i >= 0; i--, j++) { 709 710 MPASS(j < WITNESS_COUNT); 711 lock1 = &lle->ll_children[i]; 712 w1 = lock1->li_lock->lo_witness; 713 714 /* 715 * If this lock doesn't undergo witness checking, 716 * then skip it. 717 */ 718 if (w1 == NULL) { 719 KASSERT((lock1->li_lock->lo_flags & LO_WITNESS) == 0, 720 ("lock missing witness structure")); 721 continue; 722 } 723 /* 724 * If we are locking Giant and this is a sleepable 725 * lock, then skip it. 726 */ 727 if ((lock1->li_lock->lo_flags & LO_SLEEPABLE) != 0 && 728 lock == &Giant.mtx_object) 729 continue; 730 /* 731 * If we are locking a sleepable lock and this lock 732 * is Giant, then skip it. 733 */ 734 if ((lock->lo_flags & LO_SLEEPABLE) != 0 && 735 lock1->li_lock == &Giant.mtx_object) 736 continue; 737 /* 738 * If we are locking a sleepable lock and this lock 739 * isn't sleepable, we want to treat it as a lock 740 * order violation to enfore a general lock order of 741 * sleepable locks before non-sleepable locks. 742 */ 743 if (!((lock->lo_flags & LO_SLEEPABLE) != 0 && 744 (lock1->li_lock->lo_flags & LO_SLEEPABLE) == 0)) 745 /* 746 * Check the lock order hierarchy for a reveresal. 747 */ 748 if (!isitmydescendant(w, w1)) 749 continue; 750 /* 751 * We have a lock order violation, check to see if it 752 * is allowed or has already been yelled about. 753 */ 754 mtx_unlock_spin(&w_mtx); 755 #ifdef BLESSING 756 if (blessed(w, w1)) 757 goto out; 758 #endif 759 if (lock1->li_lock == &Giant.mtx_object) { 760 if (w1->w_Giant_squawked) 761 goto out; 762 else 763 w1->w_Giant_squawked = 1; 764 } else { 765 if (w1->w_other_squawked) 766 goto out; 767 else 768 w1->w_other_squawked = 1; 769 } 770 /* 771 * Ok, yell about it. 772 */ 773 printf("lock order reversal\n"); 774 /* 775 * Try to locate an earlier lock with 776 * witness w in our list. 777 */ 778 do { 779 lock2 = &lle->ll_children[i]; 780 MPASS(lock2->li_lock != NULL); 781 if (lock2->li_lock->lo_witness == w) 782 break; 783 i--; 784 if (i == 0 && lle->ll_next != NULL) { 785 lle = lle->ll_next; 786 i = lle->ll_count - 1; 787 MPASS(i >= 0 && i < LOCK_NCHILDREN); 788 } 789 } while (i >= 0); 790 if (i < 0) { 791 printf(" 1st %p %s (%s) @ %s:%d\n", 792 lock1->li_lock, lock1->li_lock->lo_name, 793 lock1->li_lock->lo_type, lock1->li_file, 794 lock1->li_line); 795 printf(" 2nd %p %s (%s) @ %s:%d\n", lock, 796 lock->lo_name, lock->lo_type, file, line); 797 } else { 798 printf(" 1st %p %s (%s) @ %s:%d\n", 799 lock2->li_lock, lock2->li_lock->lo_name, 800 lock2->li_lock->lo_type, lock2->li_file, 801 lock2->li_line); 802 printf(" 2nd %p %s (%s) @ %s:%d\n", 803 lock1->li_lock, lock1->li_lock->lo_name, 804 lock1->li_lock->lo_type, lock1->li_file, 805 lock1->li_line); 806 printf(" 3rd %p %s (%s) @ %s:%d\n", lock, 807 lock->lo_name, lock->lo_type, file, line); 808 } 809 #ifdef DDB 810 go_into_ddb = 1; 811 #endif 812 goto out; 813 } 814 } 815 lock1 = &(*lock_list)->ll_children[(*lock_list)->ll_count - 1]; 816 /* 817 * Don't build a new relationship between a sleepable lock and 818 * Giant if it is the wrong direction. The real lock order is that 819 * sleepable locks come before Giant. 820 */ 821 if (!(lock1->li_lock == &Giant.mtx_object && 822 (lock->lo_flags & LO_SLEEPABLE) != 0)) { 823 CTR3(KTR_WITNESS, "%s: adding %s as a child of %s", __func__, 824 lock->lo_type, lock1->li_lock->lo_type); 825 if (!itismychild(lock1->li_lock->lo_witness, w)) 826 /* Witness is dead. */ 827 return; 828 } 829 mtx_unlock_spin(&w_mtx); 830 831 out: 832 #ifdef DDB 833 if (go_into_ddb) { 834 if (witness_trace) 835 backtrace(); 836 if (witness_ddb) 837 Debugger(__func__); 838 } 839 #endif 840 w->w_file = file; 841 w->w_line = line; 842 843 lle = *lock_list; 844 if (lle == NULL || lle->ll_count == LOCK_NCHILDREN) { 845 lle = witness_lock_list_get(); 846 if (lle == NULL) 847 return; 848 lle->ll_next = *lock_list; 849 CTR3(KTR_WITNESS, "%s: pid %d added lle %p", __func__, 850 td->td_proc->p_pid, lle); 851 *lock_list = lle; 852 } 853 lock1 = &lle->ll_children[lle->ll_count++]; 854 lock1->li_lock = lock; 855 lock1->li_line = line; 856 lock1->li_file = file; 857 if ((flags & LOP_EXCLUSIVE) != 0) 858 lock1->li_flags = LI_EXCLUSIVE; 859 else 860 lock1->li_flags = 0; 861 CTR4(KTR_WITNESS, "%s: pid %d added %s as lle[%d]", __func__, 862 td->td_proc->p_pid, lock->lo_name, lle->ll_count - 1); 863 } 864 865 void 866 witness_upgrade(struct lock_object *lock, int flags, const char *file, int line) 867 { 868 struct lock_instance *instance; 869 struct lock_class *class; 870 871 KASSERT(!witness_cold, ("%s: witness_cold", __func__)); 872 if (lock->lo_witness == NULL || witness_watch == 0 || panicstr != NULL) 873 return; 874 class = lock->lo_class; 875 file = fixup_filename(file); 876 if ((lock->lo_flags & LO_UPGRADABLE) == 0) 877 panic("upgrade of non-upgradable lock (%s) %s @ %s:%d", 878 class->lc_name, lock->lo_name, file, line); 879 if ((flags & LOP_TRYLOCK) == 0) 880 panic("non-try upgrade of lock (%s) %s @ %s:%d", class->lc_name, 881 lock->lo_name, file, line); 882 if ((lock->lo_class->lc_flags & LC_SLEEPLOCK) == 0) 883 panic("upgrade of non-sleep lock (%s) %s @ %s:%d", 884 class->lc_name, lock->lo_name, file, line); 885 instance = find_instance(curthread->td_sleeplocks, lock); 886 if (instance == NULL) 887 panic("upgrade of unlocked lock (%s) %s @ %s:%d", 888 class->lc_name, lock->lo_name, file, line); 889 if ((instance->li_flags & LI_EXCLUSIVE) != 0) 890 panic("upgrade of exclusive lock (%s) %s @ %s:%d", 891 class->lc_name, lock->lo_name, file, line); 892 if ((instance->li_flags & LI_RECURSEMASK) != 0) 893 panic("upgrade of recursed lock (%s) %s r=%d @ %s:%d", 894 class->lc_name, lock->lo_name, 895 instance->li_flags & LI_RECURSEMASK, file, line); 896 instance->li_flags |= LI_EXCLUSIVE; 897 } 898 899 void 900 witness_downgrade(struct lock_object *lock, int flags, const char *file, 901 int line) 902 { 903 struct lock_instance *instance; 904 struct lock_class *class; 905 906 KASSERT(!witness_cold, ("%s: witness_cold", __func__)); 907 if (lock->lo_witness == NULL || witness_watch == 0 || panicstr != NULL) 908 return; 909 class = lock->lo_class; 910 file = fixup_filename(file); 911 if ((lock->lo_flags & LO_UPGRADABLE) == 0) 912 panic("downgrade of non-upgradable lock (%s) %s @ %s:%d", 913 class->lc_name, lock->lo_name, file, line); 914 if ((lock->lo_class->lc_flags & LC_SLEEPLOCK) == 0) 915 panic("downgrade of non-sleep lock (%s) %s @ %s:%d", 916 class->lc_name, lock->lo_name, file, line); 917 instance = find_instance(curthread->td_sleeplocks, lock); 918 if (instance == NULL) 919 panic("downgrade of unlocked lock (%s) %s @ %s:%d", 920 class->lc_name, lock->lo_name, file, line); 921 if ((instance->li_flags & LI_EXCLUSIVE) == 0) 922 panic("downgrade of shared lock (%s) %s @ %s:%d", 923 class->lc_name, lock->lo_name, file, line); 924 if ((instance->li_flags & LI_RECURSEMASK) != 0) 925 panic("downgrade of recursed lock (%s) %s r=%d @ %s:%d", 926 class->lc_name, lock->lo_name, 927 instance->li_flags & LI_RECURSEMASK, file, line); 928 instance->li_flags &= ~LI_EXCLUSIVE; 929 } 930 931 void 932 witness_unlock(struct lock_object *lock, int flags, const char *file, int line) 933 { 934 struct lock_list_entry **lock_list, *lle; 935 struct lock_instance *instance; 936 struct lock_class *class; 937 struct thread *td; 938 register_t s; 939 int i, j; 940 941 if (witness_cold || witness_watch == 0 || lock->lo_witness == NULL || 942 panicstr != NULL) 943 return; 944 td = curthread; 945 class = lock->lo_class; 946 file = fixup_filename(file); 947 if (class->lc_flags & LC_SLEEPLOCK) 948 lock_list = &td->td_sleeplocks; 949 else 950 lock_list = PCPU_PTR(spinlocks); 951 for (; *lock_list != NULL; lock_list = &(*lock_list)->ll_next) 952 for (i = 0; i < (*lock_list)->ll_count; i++) { 953 instance = &(*lock_list)->ll_children[i]; 954 if (instance->li_lock == lock) { 955 if ((instance->li_flags & LI_EXCLUSIVE) != 0 && 956 (flags & LOP_EXCLUSIVE) == 0) { 957 printf( 958 "shared unlock of (%s) %s @ %s:%d\n", 959 class->lc_name, lock->lo_name, 960 file, line); 961 printf( 962 "while exclusively locked from %s:%d\n", 963 instance->li_file, 964 instance->li_line); 965 panic("excl->ushare"); 966 } 967 if ((instance->li_flags & LI_EXCLUSIVE) == 0 && 968 (flags & LOP_EXCLUSIVE) != 0) { 969 printf( 970 "exclusive unlock of (%s) %s @ %s:%d\n", 971 class->lc_name, lock->lo_name, 972 file, line); 973 printf( 974 "while share locked from %s:%d\n", 975 instance->li_file, 976 instance->li_line); 977 panic("share->uexcl"); 978 } 979 /* If we are recursed, unrecurse. */ 980 if ((instance->li_flags & LI_RECURSEMASK) > 0) { 981 CTR4(KTR_WITNESS, 982 "%s: pid %d unrecursed on %s r=%d", __func__, 983 td->td_proc->p_pid, 984 instance->li_lock->lo_name, 985 instance->li_flags); 986 instance->li_flags--; 987 return; 988 } 989 s = intr_disable(); 990 CTR4(KTR_WITNESS, 991 "%s: pid %d removed %s from lle[%d]", __func__, 992 td->td_proc->p_pid, 993 instance->li_lock->lo_name, 994 (*lock_list)->ll_count - 1); 995 for (j = i; j < (*lock_list)->ll_count - 1; j++) 996 (*lock_list)->ll_children[j] = 997 (*lock_list)->ll_children[j + 1]; 998 (*lock_list)->ll_count--; 999 intr_restore(s); 1000 if ((*lock_list)->ll_count == 0) { 1001 lle = *lock_list; 1002 *lock_list = lle->ll_next; 1003 CTR3(KTR_WITNESS, 1004 "%s: pid %d removed lle %p", __func__, 1005 td->td_proc->p_pid, lle); 1006 witness_lock_list_free(lle); 1007 } 1008 return; 1009 } 1010 } 1011 panic("lock (%s) %s not locked @ %s:%d", class->lc_name, lock->lo_name, 1012 file, line); 1013 } 1014 1015 /* 1016 * Warn if any locks other than 'lock' are held. Flags can be passed in to 1017 * exempt Giant and sleepable locks from the checks as well. If any 1018 * non-exempt locks are held, then a supplied message is printed to the 1019 * console along with a list of the offending locks. If indicated in the 1020 * flags then a failure results in a panic as well. 1021 */ 1022 int 1023 witness_warn(int flags, struct lock_object *lock, const char *fmt, ...) 1024 { 1025 struct lock_list_entry *lle; 1026 struct lock_instance *lock1; 1027 struct thread *td; 1028 va_list ap; 1029 int i, n; 1030 1031 if (witness_cold || witness_watch == 0 || panicstr != NULL) 1032 return (0); 1033 n = 0; 1034 td = curthread; 1035 for (lle = td->td_sleeplocks; lle != NULL; lle = lle->ll_next) 1036 for (i = lle->ll_count - 1; i >= 0; i--) { 1037 lock1 = &lle->ll_children[i]; 1038 if (lock1->li_lock == lock) 1039 continue; 1040 if (flags & WARN_GIANTOK && 1041 lock1->li_lock == &Giant.mtx_object) 1042 continue; 1043 if (flags & WARN_SLEEPOK && 1044 (lock1->li_lock->lo_flags & LO_SLEEPABLE) != 0) 1045 continue; 1046 if (n == 0) { 1047 va_start(ap, fmt); 1048 vprintf(fmt, ap); 1049 va_end(ap); 1050 printf(" with the following"); 1051 if (flags & WARN_SLEEPOK) 1052 printf(" non-sleepable"); 1053 printf(" locks held:\n"); 1054 } 1055 n++; 1056 witness_list_lock(lock1); 1057 } 1058 if (PCPU_GET(spinlocks) != NULL) { 1059 /* 1060 * Since we already hold a spinlock preemption is 1061 * already blocked. 1062 */ 1063 if (n == 0) { 1064 va_start(ap, fmt); 1065 vprintf(fmt, ap); 1066 va_end(ap); 1067 printf(" with the following"); 1068 if (flags & WARN_SLEEPOK) 1069 printf(" non-sleepable"); 1070 printf(" locks held:\n"); 1071 } 1072 n += witness_list_locks(PCPU_PTR(spinlocks)); 1073 } 1074 if (flags & WARN_PANIC && n) 1075 panic("witness_warn"); 1076 #ifdef DDB 1077 else if (witness_ddb && n) 1078 Debugger(__func__); 1079 #endif 1080 return (n); 1081 } 1082 1083 const char * 1084 witness_file(struct lock_object *lock) 1085 { 1086 struct witness *w; 1087 1088 if (witness_cold || witness_watch == 0 || lock->lo_witness == NULL) 1089 return ("?"); 1090 w = lock->lo_witness; 1091 return (w->w_file); 1092 } 1093 1094 int 1095 witness_line(struct lock_object *lock) 1096 { 1097 struct witness *w; 1098 1099 if (witness_cold || witness_watch == 0 || lock->lo_witness == NULL) 1100 return (0); 1101 w = lock->lo_witness; 1102 return (w->w_line); 1103 } 1104 1105 static struct witness * 1106 enroll(const char *description, struct lock_class *lock_class) 1107 { 1108 struct witness *w; 1109 1110 if (!witness_watch || witness_watch == 0 || panicstr != NULL) 1111 return (NULL); 1112 if ((lock_class->lc_flags & LC_SPINLOCK) && witness_skipspin) 1113 return (NULL); 1114 mtx_lock_spin(&w_mtx); 1115 STAILQ_FOREACH(w, &w_all, w_list) { 1116 if (w->w_name == description || (w->w_refcount > 0 && 1117 strcmp(description, w->w_name) == 0)) { 1118 w->w_refcount++; 1119 mtx_unlock_spin(&w_mtx); 1120 if (lock_class != w->w_class) 1121 panic( 1122 "lock (%s) %s does not match earlier (%s) lock", 1123 description, lock_class->lc_name, 1124 w->w_class->lc_name); 1125 return (w); 1126 } 1127 } 1128 /* 1129 * This isn't quite right, as witness_cold is still 0 while we 1130 * enroll all the locks initialized before witness_initialize(). 1131 */ 1132 if ((lock_class->lc_flags & LC_SPINLOCK) && !witness_cold) { 1133 mtx_unlock_spin(&w_mtx); 1134 panic("spin lock %s not in order list", description); 1135 } 1136 if ((w = witness_get()) == NULL) 1137 return (NULL); 1138 w->w_name = description; 1139 w->w_class = lock_class; 1140 w->w_refcount = 1; 1141 STAILQ_INSERT_HEAD(&w_all, w, w_list); 1142 if (lock_class->lc_flags & LC_SPINLOCK) 1143 STAILQ_INSERT_HEAD(&w_spin, w, w_typelist); 1144 else if (lock_class->lc_flags & LC_SLEEPLOCK) 1145 STAILQ_INSERT_HEAD(&w_sleep, w, w_typelist); 1146 else { 1147 mtx_unlock_spin(&w_mtx); 1148 panic("lock class %s is not sleep or spin", 1149 lock_class->lc_name); 1150 } 1151 mtx_unlock_spin(&w_mtx); 1152 return (w); 1153 } 1154 1155 /* Don't let the door bang you on the way out... */ 1156 static int 1157 depart(struct witness *w) 1158 { 1159 struct witness_child_list_entry *wcl, *nwcl; 1160 struct witness_list *list; 1161 struct witness *parent; 1162 1163 MPASS(w->w_refcount == 0); 1164 if (w->w_class->lc_flags & LC_SLEEPLOCK) 1165 list = &w_sleep; 1166 else 1167 list = &w_spin; 1168 /* 1169 * First, we run through the entire tree looking for any 1170 * witnesses that the outgoing witness is a child of. For 1171 * each parent that we find, we reparent all the direct 1172 * children of the outgoing witness to its parent. 1173 */ 1174 STAILQ_FOREACH(parent, list, w_typelist) { 1175 if (!isitmychild(parent, w)) 1176 continue; 1177 removechild(parent, w); 1178 if (!reparentchildren(parent, w)) 1179 return (0); 1180 } 1181 1182 /* 1183 * Now we go through and free up the child list of the 1184 * outgoing witness. 1185 */ 1186 for (wcl = w->w_children; wcl != NULL; wcl = nwcl) { 1187 nwcl = wcl->wcl_next; 1188 witness_child_free(wcl); 1189 } 1190 1191 /* 1192 * Detach from various lists and free. 1193 */ 1194 STAILQ_REMOVE(list, w, witness, w_typelist); 1195 STAILQ_REMOVE(&w_all, w, witness, w_list); 1196 witness_free(w); 1197 1198 /* Finally, fixup the tree. */ 1199 return (rebalancetree(list)); 1200 } 1201 1202 /* 1203 * Prune an entire lock order tree. We look for cases where a lock 1204 * is now both a descendant and a direct child of a given lock. In 1205 * that case, we want to remove the direct child link from the tree. 1206 * 1207 * Returns false if insertchild() fails. 1208 */ 1209 static int 1210 rebalancetree(struct witness_list *list) 1211 { 1212 struct witness *child, *parent; 1213 1214 STAILQ_FOREACH(child, list, w_typelist) { 1215 STAILQ_FOREACH(parent, list, w_typelist) { 1216 if (!isitmychild(parent, child)) 1217 continue; 1218 removechild(parent, child); 1219 if (isitmydescendant(parent, child)) 1220 continue; 1221 if (!insertchild(parent, child)) 1222 return (0); 1223 } 1224 } 1225 witness_levelall(); 1226 return (1); 1227 } 1228 1229 /* 1230 * Add "child" as a direct child of "parent". Returns false if 1231 * we fail due to out of memory. 1232 */ 1233 static int 1234 insertchild(struct witness *parent, struct witness *child) 1235 { 1236 struct witness_child_list_entry **wcl; 1237 1238 MPASS(child != NULL && parent != NULL); 1239 1240 /* 1241 * Insert "child" after "parent" 1242 */ 1243 wcl = &parent->w_children; 1244 while (*wcl != NULL && (*wcl)->wcl_count == WITNESS_NCHILDREN) 1245 wcl = &(*wcl)->wcl_next; 1246 if (*wcl == NULL) { 1247 *wcl = witness_child_get(); 1248 if (*wcl == NULL) 1249 return (0); 1250 } 1251 (*wcl)->wcl_children[(*wcl)->wcl_count++] = child; 1252 1253 return (1); 1254 } 1255 1256 /* 1257 * Make all the direct descendants of oldparent be direct descendants 1258 * of newparent. 1259 */ 1260 static int 1261 reparentchildren(struct witness *newparent, struct witness *oldparent) 1262 { 1263 struct witness_child_list_entry *wcl; 1264 int i; 1265 1266 /* Avoid making a witness a child of itself. */ 1267 MPASS(!isitmychild(oldparent, newparent)); 1268 1269 for (wcl = oldparent->w_children; wcl != NULL; wcl = wcl->wcl_next) 1270 for (i = 0; i < wcl->wcl_count; i++) 1271 if (!insertchild(newparent, wcl->wcl_children[i])) 1272 return (0); 1273 return (1); 1274 } 1275 1276 static int 1277 itismychild(struct witness *parent, struct witness *child) 1278 { 1279 struct witness_list *list; 1280 1281 MPASS(child != NULL && parent != NULL); 1282 if ((parent->w_class->lc_flags & (LC_SLEEPLOCK | LC_SPINLOCK)) != 1283 (child->w_class->lc_flags & (LC_SLEEPLOCK | LC_SPINLOCK))) 1284 panic( 1285 "%s: parent (%s) and child (%s) are not the same lock type", 1286 __func__, parent->w_class->lc_name, 1287 child->w_class->lc_name); 1288 1289 if (!insertchild(parent, child)) 1290 return (0); 1291 1292 if (parent->w_class->lc_flags & LC_SLEEPLOCK) 1293 list = &w_sleep; 1294 else 1295 list = &w_spin; 1296 return (rebalancetree(list)); 1297 } 1298 1299 static void 1300 removechild(struct witness *parent, struct witness *child) 1301 { 1302 struct witness_child_list_entry **wcl, *wcl1; 1303 int i; 1304 1305 for (wcl = &parent->w_children; *wcl != NULL; wcl = &(*wcl)->wcl_next) 1306 for (i = 0; i < (*wcl)->wcl_count; i++) 1307 if ((*wcl)->wcl_children[i] == child) 1308 goto found; 1309 return; 1310 found: 1311 (*wcl)->wcl_count--; 1312 if ((*wcl)->wcl_count > i) 1313 (*wcl)->wcl_children[i] = 1314 (*wcl)->wcl_children[(*wcl)->wcl_count]; 1315 MPASS((*wcl)->wcl_children[i] != NULL); 1316 if ((*wcl)->wcl_count != 0) 1317 return; 1318 wcl1 = *wcl; 1319 *wcl = wcl1->wcl_next; 1320 witness_child_free(wcl1); 1321 } 1322 1323 static int 1324 isitmychild(struct witness *parent, struct witness *child) 1325 { 1326 struct witness_child_list_entry *wcl; 1327 int i; 1328 1329 for (wcl = parent->w_children; wcl != NULL; wcl = wcl->wcl_next) { 1330 for (i = 0; i < wcl->wcl_count; i++) { 1331 if (wcl->wcl_children[i] == child) 1332 return (1); 1333 } 1334 } 1335 return (0); 1336 } 1337 1338 static int 1339 isitmydescendant(struct witness *parent, struct witness *child) 1340 { 1341 struct witness_child_list_entry *wcl; 1342 int i, j; 1343 1344 if (isitmychild(parent, child)) 1345 return (1); 1346 j = 0; 1347 for (wcl = parent->w_children; wcl != NULL; wcl = wcl->wcl_next) { 1348 MPASS(j < 1000); 1349 for (i = 0; i < wcl->wcl_count; i++) { 1350 if (isitmydescendant(wcl->wcl_children[i], child)) 1351 return (1); 1352 } 1353 j++; 1354 } 1355 return (0); 1356 } 1357 1358 static void 1359 witness_levelall (void) 1360 { 1361 struct witness_list *list; 1362 struct witness *w, *w1; 1363 1364 /* 1365 * First clear all levels. 1366 */ 1367 STAILQ_FOREACH(w, &w_all, w_list) { 1368 w->w_level = 0; 1369 } 1370 1371 /* 1372 * Look for locks with no parent and level all their descendants. 1373 */ 1374 STAILQ_FOREACH(w, &w_all, w_list) { 1375 /* 1376 * This is just an optimization, technically we could get 1377 * away just walking the all list each time. 1378 */ 1379 if (w->w_class->lc_flags & LC_SLEEPLOCK) 1380 list = &w_sleep; 1381 else 1382 list = &w_spin; 1383 STAILQ_FOREACH(w1, list, w_typelist) { 1384 if (isitmychild(w1, w)) 1385 goto skip; 1386 } 1387 witness_leveldescendents(w, 0); 1388 skip: 1389 ; /* silence GCC 3.x */ 1390 } 1391 } 1392 1393 static void 1394 witness_leveldescendents(struct witness *parent, int level) 1395 { 1396 struct witness_child_list_entry *wcl; 1397 int i; 1398 1399 if (parent->w_level < level) 1400 parent->w_level = level; 1401 level++; 1402 for (wcl = parent->w_children; wcl != NULL; wcl = wcl->wcl_next) 1403 for (i = 0; i < wcl->wcl_count; i++) 1404 witness_leveldescendents(wcl->wcl_children[i], level); 1405 } 1406 1407 static void 1408 witness_displaydescendants(void(*prnt)(const char *fmt, ...), 1409 struct witness *parent, int indent) 1410 { 1411 struct witness_child_list_entry *wcl; 1412 int i, level; 1413 1414 level = parent->w_level; 1415 prnt("%-2d", level); 1416 for (i = 0; i < indent; i++) 1417 prnt(" "); 1418 if (parent->w_refcount > 0) 1419 prnt("%s", parent->w_name); 1420 else 1421 prnt("(dead)"); 1422 if (parent->w_displayed) { 1423 prnt(" -- (already displayed)\n"); 1424 return; 1425 } 1426 parent->w_displayed = 1; 1427 if (parent->w_refcount > 0) { 1428 if (parent->w_file != NULL) 1429 prnt(" -- last acquired @ %s:%d", parent->w_file, 1430 parent->w_line); 1431 } 1432 prnt("\n"); 1433 for (wcl = parent->w_children; wcl != NULL; wcl = wcl->wcl_next) 1434 for (i = 0; i < wcl->wcl_count; i++) 1435 witness_displaydescendants(prnt, 1436 wcl->wcl_children[i], indent + 1); 1437 } 1438 1439 #ifdef BLESSING 1440 static int 1441 blessed(struct witness *w1, struct witness *w2) 1442 { 1443 int i; 1444 struct witness_blessed *b; 1445 1446 for (i = 0; i < blessed_count; i++) { 1447 b = &blessed_list[i]; 1448 if (strcmp(w1->w_name, b->b_lock1) == 0) { 1449 if (strcmp(w2->w_name, b->b_lock2) == 0) 1450 return (1); 1451 continue; 1452 } 1453 if (strcmp(w1->w_name, b->b_lock2) == 0) 1454 if (strcmp(w2->w_name, b->b_lock1) == 0) 1455 return (1); 1456 } 1457 return (0); 1458 } 1459 #endif 1460 1461 static struct witness * 1462 witness_get(void) 1463 { 1464 struct witness *w; 1465 1466 if (witness_watch == 0) { 1467 mtx_unlock_spin(&w_mtx); 1468 return (NULL); 1469 } 1470 if (STAILQ_EMPTY(&w_free)) { 1471 witness_watch = 0; 1472 mtx_unlock_spin(&w_mtx); 1473 printf("%s: witness exhausted\n", __func__); 1474 return (NULL); 1475 } 1476 w = STAILQ_FIRST(&w_free); 1477 STAILQ_REMOVE_HEAD(&w_free, w_list); 1478 bzero(w, sizeof(*w)); 1479 return (w); 1480 } 1481 1482 static void 1483 witness_free(struct witness *w) 1484 { 1485 1486 STAILQ_INSERT_HEAD(&w_free, w, w_list); 1487 } 1488 1489 static struct witness_child_list_entry * 1490 witness_child_get(void) 1491 { 1492 struct witness_child_list_entry *wcl; 1493 1494 if (witness_watch == 0) { 1495 mtx_unlock_spin(&w_mtx); 1496 return (NULL); 1497 } 1498 wcl = w_child_free; 1499 if (wcl == NULL) { 1500 witness_watch = 0; 1501 mtx_unlock_spin(&w_mtx); 1502 printf("%s: witness exhausted\n", __func__); 1503 return (NULL); 1504 } 1505 w_child_free = wcl->wcl_next; 1506 bzero(wcl, sizeof(*wcl)); 1507 return (wcl); 1508 } 1509 1510 static void 1511 witness_child_free(struct witness_child_list_entry *wcl) 1512 { 1513 1514 wcl->wcl_next = w_child_free; 1515 w_child_free = wcl; 1516 } 1517 1518 static struct lock_list_entry * 1519 witness_lock_list_get(void) 1520 { 1521 struct lock_list_entry *lle; 1522 1523 if (witness_watch == 0) 1524 return (NULL); 1525 mtx_lock_spin(&w_mtx); 1526 lle = w_lock_list_free; 1527 if (lle == NULL) { 1528 witness_watch = 0; 1529 mtx_unlock_spin(&w_mtx); 1530 printf("%s: witness exhausted\n", __func__); 1531 return (NULL); 1532 } 1533 w_lock_list_free = lle->ll_next; 1534 mtx_unlock_spin(&w_mtx); 1535 bzero(lle, sizeof(*lle)); 1536 return (lle); 1537 } 1538 1539 static void 1540 witness_lock_list_free(struct lock_list_entry *lle) 1541 { 1542 1543 mtx_lock_spin(&w_mtx); 1544 lle->ll_next = w_lock_list_free; 1545 w_lock_list_free = lle; 1546 mtx_unlock_spin(&w_mtx); 1547 } 1548 1549 static struct lock_instance * 1550 find_instance(struct lock_list_entry *lock_list, struct lock_object *lock) 1551 { 1552 struct lock_list_entry *lle; 1553 struct lock_instance *instance; 1554 int i; 1555 1556 for (lle = lock_list; lle != NULL; lle = lle->ll_next) 1557 for (i = lle->ll_count - 1; i >= 0; i--) { 1558 instance = &lle->ll_children[i]; 1559 if (instance->li_lock == lock) 1560 return (instance); 1561 } 1562 return (NULL); 1563 } 1564 1565 static void 1566 witness_list_lock(struct lock_instance *instance) 1567 { 1568 struct lock_object *lock; 1569 1570 lock = instance->li_lock; 1571 printf("%s %s %s", (instance->li_flags & LI_EXCLUSIVE) != 0 ? 1572 "exclusive" : "shared", lock->lo_class->lc_name, lock->lo_name); 1573 if (lock->lo_type != lock->lo_name) 1574 printf(" (%s)", lock->lo_type); 1575 printf(" r = %d (%p) locked @ %s:%d\n", 1576 instance->li_flags & LI_RECURSEMASK, lock, instance->li_file, 1577 instance->li_line); 1578 } 1579 1580 int 1581 witness_list_locks(struct lock_list_entry **lock_list) 1582 { 1583 struct lock_list_entry *lle; 1584 int i, nheld; 1585 1586 nheld = 0; 1587 for (lle = *lock_list; lle != NULL; lle = lle->ll_next) 1588 for (i = lle->ll_count - 1; i >= 0; i--) { 1589 witness_list_lock(&lle->ll_children[i]); 1590 nheld++; 1591 } 1592 return (nheld); 1593 } 1594 1595 /* 1596 * This is a bit risky at best. We call this function when we have timed 1597 * out acquiring a spin lock, and we assume that the other CPU is stuck 1598 * with this lock held. So, we go groveling around in the other CPU's 1599 * per-cpu data to try to find the lock instance for this spin lock to 1600 * see when it was last acquired. 1601 */ 1602 void 1603 witness_display_spinlock(struct lock_object *lock, struct thread *owner) 1604 { 1605 struct lock_instance *instance; 1606 struct pcpu *pc; 1607 1608 if (owner->td_critnest == 0 || owner->td_oncpu == NOCPU) 1609 return; 1610 pc = pcpu_find(owner->td_oncpu); 1611 instance = find_instance(pc->pc_spinlocks, lock); 1612 if (instance != NULL) 1613 witness_list_lock(instance); 1614 } 1615 1616 void 1617 witness_save(struct lock_object *lock, const char **filep, int *linep) 1618 { 1619 struct lock_instance *instance; 1620 1621 KASSERT(!witness_cold, ("%s: witness_cold", __func__)); 1622 if (lock->lo_witness == NULL || witness_watch == 0 || panicstr != NULL) 1623 return; 1624 if ((lock->lo_class->lc_flags & LC_SLEEPLOCK) == 0) 1625 panic("%s: lock (%s) %s is not a sleep lock", __func__, 1626 lock->lo_class->lc_name, lock->lo_name); 1627 instance = find_instance(curthread->td_sleeplocks, lock); 1628 if (instance == NULL) 1629 panic("%s: lock (%s) %s not locked", __func__, 1630 lock->lo_class->lc_name, lock->lo_name); 1631 *filep = instance->li_file; 1632 *linep = instance->li_line; 1633 } 1634 1635 void 1636 witness_restore(struct lock_object *lock, const char *file, int line) 1637 { 1638 struct lock_instance *instance; 1639 1640 KASSERT(!witness_cold, ("%s: witness_cold", __func__)); 1641 if (lock->lo_witness == NULL || witness_watch == 0 || panicstr != NULL) 1642 return; 1643 if ((lock->lo_class->lc_flags & LC_SLEEPLOCK) == 0) 1644 panic("%s: lock (%s) %s is not a sleep lock", __func__, 1645 lock->lo_class->lc_name, lock->lo_name); 1646 instance = find_instance(curthread->td_sleeplocks, lock); 1647 if (instance == NULL) 1648 panic("%s: lock (%s) %s not locked", __func__, 1649 lock->lo_class->lc_name, lock->lo_name); 1650 lock->lo_witness->w_file = file; 1651 lock->lo_witness->w_line = line; 1652 instance->li_file = file; 1653 instance->li_line = line; 1654 } 1655 1656 void 1657 witness_assert(struct lock_object *lock, int flags, const char *file, int line) 1658 { 1659 #ifdef INVARIANT_SUPPORT 1660 struct lock_instance *instance; 1661 1662 if (lock->lo_witness == NULL || witness_watch == 0 || panicstr != NULL) 1663 return; 1664 if ((lock->lo_class->lc_flags & LC_SLEEPLOCK) != 0) 1665 instance = find_instance(curthread->td_sleeplocks, lock); 1666 else if ((lock->lo_class->lc_flags & LC_SPINLOCK) != 0) 1667 instance = find_instance(PCPU_GET(spinlocks), lock); 1668 else { 1669 panic("Lock (%s) %s is not sleep or spin!", 1670 lock->lo_class->lc_name, lock->lo_name); 1671 } 1672 file = fixup_filename(file); 1673 switch (flags) { 1674 case LA_UNLOCKED: 1675 if (instance != NULL) 1676 panic("Lock (%s) %s locked @ %s:%d.", 1677 lock->lo_class->lc_name, lock->lo_name, file, line); 1678 break; 1679 case LA_LOCKED: 1680 case LA_LOCKED | LA_RECURSED: 1681 case LA_LOCKED | LA_NOTRECURSED: 1682 case LA_SLOCKED: 1683 case LA_SLOCKED | LA_RECURSED: 1684 case LA_SLOCKED | LA_NOTRECURSED: 1685 case LA_XLOCKED: 1686 case LA_XLOCKED | LA_RECURSED: 1687 case LA_XLOCKED | LA_NOTRECURSED: 1688 if (instance == NULL) { 1689 panic("Lock (%s) %s not locked @ %s:%d.", 1690 lock->lo_class->lc_name, lock->lo_name, file, line); 1691 break; 1692 } 1693 if ((flags & LA_XLOCKED) != 0 && 1694 (instance->li_flags & LI_EXCLUSIVE) == 0) 1695 panic("Lock (%s) %s not exclusively locked @ %s:%d.", 1696 lock->lo_class->lc_name, lock->lo_name, file, line); 1697 if ((flags & LA_SLOCKED) != 0 && 1698 (instance->li_flags & LI_EXCLUSIVE) != 0) 1699 panic("Lock (%s) %s exclusively locked @ %s:%d.", 1700 lock->lo_class->lc_name, lock->lo_name, file, line); 1701 if ((flags & LA_RECURSED) != 0 && 1702 (instance->li_flags & LI_RECURSEMASK) == 0) 1703 panic("Lock (%s) %s not recursed @ %s:%d.", 1704 lock->lo_class->lc_name, lock->lo_name, file, line); 1705 if ((flags & LA_NOTRECURSED) != 0 && 1706 (instance->li_flags & LI_RECURSEMASK) != 0) 1707 panic("Lock (%s) %s recursed @ %s:%d.", 1708 lock->lo_class->lc_name, lock->lo_name, file, line); 1709 break; 1710 default: 1711 panic("Invalid lock assertion at %s:%d.", file, line); 1712 1713 } 1714 #endif /* INVARIANT_SUPPORT */ 1715 } 1716 1717 #ifdef DDB 1718 static void 1719 witness_list(struct thread *td) 1720 { 1721 1722 KASSERT(!witness_cold, ("%s: witness_cold", __func__)); 1723 KASSERT(db_active, ("%s: not in the debugger", __func__)); 1724 1725 if (witness_watch == 0) 1726 return; 1727 1728 witness_list_locks(&td->td_sleeplocks); 1729 1730 /* 1731 * We only handle spinlocks if td == curthread. This is somewhat broken 1732 * if td is currently executing on some other CPU and holds spin locks 1733 * as we won't display those locks. If we had a MI way of getting 1734 * the per-cpu data for a given cpu then we could use 1735 * td->td_oncpu to get the list of spinlocks for this thread 1736 * and "fix" this. 1737 * 1738 * That still wouldn't really fix this unless we locked sched_lock 1739 * or stopped the other CPU to make sure it wasn't changing the list 1740 * out from under us. It is probably best to just not try to handle 1741 * threads on other CPU's for now. 1742 */ 1743 if (td == curthread && PCPU_GET(spinlocks) != NULL) 1744 witness_list_locks(PCPU_PTR(spinlocks)); 1745 } 1746 1747 DB_SHOW_COMMAND(locks, db_witness_list) 1748 { 1749 struct thread *td; 1750 pid_t pid; 1751 struct proc *p; 1752 1753 if (have_addr) { 1754 pid = (addr % 16) + ((addr >> 4) % 16) * 10 + 1755 ((addr >> 8) % 16) * 100 + ((addr >> 12) % 16) * 1000 + 1756 ((addr >> 16) % 16) * 10000; 1757 /* sx_slock(&allproc_lock); */ 1758 FOREACH_PROC_IN_SYSTEM(p) { 1759 if (p->p_pid == pid) 1760 break; 1761 } 1762 /* sx_sunlock(&allproc_lock); */ 1763 if (p == NULL) { 1764 db_printf("pid %d not found\n", pid); 1765 return; 1766 } 1767 FOREACH_THREAD_IN_PROC(p, td) { 1768 witness_list(td); 1769 } 1770 } else { 1771 td = curthread; 1772 witness_list(td); 1773 } 1774 } 1775 1776 DB_SHOW_COMMAND(witness, db_witness_display) 1777 { 1778 1779 witness_display(db_printf); 1780 } 1781 #endif 1782