1 /* 2 * kernel/lockdep.c 3 * 4 * Runtime locking correctness validator 5 * 6 * Started by Ingo Molnar: 7 * 8 * Copyright (C) 2006,2007 Red Hat, Inc., Ingo Molnar <mingo@redhat.com> 9 * Copyright (C) 2007 Red Hat, Inc., Peter Zijlstra 10 * 11 * this code maps all the lock dependencies as they occur in a live kernel 12 * and will warn about the following classes of locking bugs: 13 * 14 * - lock inversion scenarios 15 * - circular lock dependencies 16 * - hardirq/softirq safe/unsafe locking bugs 17 * 18 * Bugs are reported even if the current locking scenario does not cause 19 * any deadlock at this point. 20 * 21 * I.e. if anytime in the past two locks were taken in a different order, 22 * even if it happened for another task, even if those were different 23 * locks (but of the same class as this lock), this code will detect it. 24 * 25 * Thanks to Arjan van de Ven for coming up with the initial idea of 26 * mapping lock dependencies runtime. 27 */ 28 #define DISABLE_BRANCH_PROFILING 29 #include <linux/mutex.h> 30 #include <linux/sched.h> 31 #include <linux/sched/clock.h> 32 #include <linux/sched/task.h> 33 #include <linux/sched/mm.h> 34 #include <linux/delay.h> 35 #include <linux/module.h> 36 #include <linux/proc_fs.h> 37 #include <linux/seq_file.h> 38 #include <linux/spinlock.h> 39 #include <linux/kallsyms.h> 40 #include <linux/interrupt.h> 41 #include <linux/stacktrace.h> 42 #include <linux/debug_locks.h> 43 #include <linux/irqflags.h> 44 #include <linux/utsname.h> 45 #include <linux/hash.h> 46 #include <linux/ftrace.h> 47 #include <linux/stringify.h> 48 #include <linux/bitmap.h> 49 #include <linux/bitops.h> 50 #include <linux/gfp.h> 51 #include <linux/random.h> 52 #include <linux/jhash.h> 53 #include <linux/nmi.h> 54 #include <linux/rcupdate.h> 55 #include <linux/kprobes.h> 56 57 #include <asm/sections.h> 58 59 #include "lockdep_internals.h" 60 61 #define CREATE_TRACE_POINTS 62 #include <trace/events/lock.h> 63 64 #ifdef CONFIG_PROVE_LOCKING 65 int prove_locking = 1; 66 module_param(prove_locking, int, 0644); 67 #else 68 #define prove_locking 0 69 #endif 70 71 #ifdef CONFIG_LOCK_STAT 72 int lock_stat = 1; 73 module_param(lock_stat, int, 0644); 74 #else 75 #define lock_stat 0 76 #endif 77 78 /* 79 * lockdep_lock: protects the lockdep graph, the hashes and the 80 * class/list/hash allocators. 81 * 82 * This is one of the rare exceptions where it's justified 83 * to use a raw spinlock - we really dont want the spinlock 84 * code to recurse back into the lockdep code... 85 */ 86 static arch_spinlock_t lockdep_lock = (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED; 87 static struct task_struct *lockdep_selftest_task_struct; 88 89 static int graph_lock(void) 90 { 91 arch_spin_lock(&lockdep_lock); 92 /* 93 * Make sure that if another CPU detected a bug while 94 * walking the graph we dont change it (while the other 95 * CPU is busy printing out stuff with the graph lock 96 * dropped already) 97 */ 98 if (!debug_locks) { 99 arch_spin_unlock(&lockdep_lock); 100 return 0; 101 } 102 /* prevent any recursions within lockdep from causing deadlocks */ 103 current->lockdep_recursion++; 104 return 1; 105 } 106 107 static inline int graph_unlock(void) 108 { 109 if (debug_locks && !arch_spin_is_locked(&lockdep_lock)) { 110 /* 111 * The lockdep graph lock isn't locked while we expect it to 112 * be, we're confused now, bye! 113 */ 114 return DEBUG_LOCKS_WARN_ON(1); 115 } 116 117 current->lockdep_recursion--; 118 arch_spin_unlock(&lockdep_lock); 119 return 0; 120 } 121 122 /* 123 * Turn lock debugging off and return with 0 if it was off already, 124 * and also release the graph lock: 125 */ 126 static inline int debug_locks_off_graph_unlock(void) 127 { 128 int ret = debug_locks_off(); 129 130 arch_spin_unlock(&lockdep_lock); 131 132 return ret; 133 } 134 135 unsigned long nr_list_entries; 136 static struct lock_list list_entries[MAX_LOCKDEP_ENTRIES]; 137 static DECLARE_BITMAP(list_entries_in_use, MAX_LOCKDEP_ENTRIES); 138 139 /* 140 * All data structures here are protected by the global debug_lock. 141 * 142 * nr_lock_classes is the number of elements of lock_classes[] that is 143 * in use. 144 */ 145 #define KEYHASH_BITS (MAX_LOCKDEP_KEYS_BITS - 1) 146 #define KEYHASH_SIZE (1UL << KEYHASH_BITS) 147 static struct hlist_head lock_keys_hash[KEYHASH_SIZE]; 148 unsigned long nr_lock_classes; 149 #ifndef CONFIG_DEBUG_LOCKDEP 150 static 151 #endif 152 struct lock_class lock_classes[MAX_LOCKDEP_KEYS]; 153 154 static inline struct lock_class *hlock_class(struct held_lock *hlock) 155 { 156 if (!hlock->class_idx) { 157 /* 158 * Someone passed in garbage, we give up. 159 */ 160 DEBUG_LOCKS_WARN_ON(1); 161 return NULL; 162 } 163 return lock_classes + hlock->class_idx - 1; 164 } 165 166 #ifdef CONFIG_LOCK_STAT 167 static DEFINE_PER_CPU(struct lock_class_stats[MAX_LOCKDEP_KEYS], cpu_lock_stats); 168 169 static inline u64 lockstat_clock(void) 170 { 171 return local_clock(); 172 } 173 174 static int lock_point(unsigned long points[], unsigned long ip) 175 { 176 int i; 177 178 for (i = 0; i < LOCKSTAT_POINTS; i++) { 179 if (points[i] == 0) { 180 points[i] = ip; 181 break; 182 } 183 if (points[i] == ip) 184 break; 185 } 186 187 return i; 188 } 189 190 static void lock_time_inc(struct lock_time *lt, u64 time) 191 { 192 if (time > lt->max) 193 lt->max = time; 194 195 if (time < lt->min || !lt->nr) 196 lt->min = time; 197 198 lt->total += time; 199 lt->nr++; 200 } 201 202 static inline void lock_time_add(struct lock_time *src, struct lock_time *dst) 203 { 204 if (!src->nr) 205 return; 206 207 if (src->max > dst->max) 208 dst->max = src->max; 209 210 if (src->min < dst->min || !dst->nr) 211 dst->min = src->min; 212 213 dst->total += src->total; 214 dst->nr += src->nr; 215 } 216 217 struct lock_class_stats lock_stats(struct lock_class *class) 218 { 219 struct lock_class_stats stats; 220 int cpu, i; 221 222 memset(&stats, 0, sizeof(struct lock_class_stats)); 223 for_each_possible_cpu(cpu) { 224 struct lock_class_stats *pcs = 225 &per_cpu(cpu_lock_stats, cpu)[class - lock_classes]; 226 227 for (i = 0; i < ARRAY_SIZE(stats.contention_point); i++) 228 stats.contention_point[i] += pcs->contention_point[i]; 229 230 for (i = 0; i < ARRAY_SIZE(stats.contending_point); i++) 231 stats.contending_point[i] += pcs->contending_point[i]; 232 233 lock_time_add(&pcs->read_waittime, &stats.read_waittime); 234 lock_time_add(&pcs->write_waittime, &stats.write_waittime); 235 236 lock_time_add(&pcs->read_holdtime, &stats.read_holdtime); 237 lock_time_add(&pcs->write_holdtime, &stats.write_holdtime); 238 239 for (i = 0; i < ARRAY_SIZE(stats.bounces); i++) 240 stats.bounces[i] += pcs->bounces[i]; 241 } 242 243 return stats; 244 } 245 246 void clear_lock_stats(struct lock_class *class) 247 { 248 int cpu; 249 250 for_each_possible_cpu(cpu) { 251 struct lock_class_stats *cpu_stats = 252 &per_cpu(cpu_lock_stats, cpu)[class - lock_classes]; 253 254 memset(cpu_stats, 0, sizeof(struct lock_class_stats)); 255 } 256 memset(class->contention_point, 0, sizeof(class->contention_point)); 257 memset(class->contending_point, 0, sizeof(class->contending_point)); 258 } 259 260 static struct lock_class_stats *get_lock_stats(struct lock_class *class) 261 { 262 return &this_cpu_ptr(cpu_lock_stats)[class - lock_classes]; 263 } 264 265 static void lock_release_holdtime(struct held_lock *hlock) 266 { 267 struct lock_class_stats *stats; 268 u64 holdtime; 269 270 if (!lock_stat) 271 return; 272 273 holdtime = lockstat_clock() - hlock->holdtime_stamp; 274 275 stats = get_lock_stats(hlock_class(hlock)); 276 if (hlock->read) 277 lock_time_inc(&stats->read_holdtime, holdtime); 278 else 279 lock_time_inc(&stats->write_holdtime, holdtime); 280 } 281 #else 282 static inline void lock_release_holdtime(struct held_lock *hlock) 283 { 284 } 285 #endif 286 287 /* 288 * We keep a global list of all lock classes. The list is only accessed with 289 * the lockdep spinlock lock held. free_lock_classes is a list with free 290 * elements. These elements are linked together by the lock_entry member in 291 * struct lock_class. 292 */ 293 LIST_HEAD(all_lock_classes); 294 static LIST_HEAD(free_lock_classes); 295 296 /** 297 * struct pending_free - information about data structures about to be freed 298 * @zapped: Head of a list with struct lock_class elements. 299 * @lock_chains_being_freed: Bitmap that indicates which lock_chains[] elements 300 * are about to be freed. 301 */ 302 struct pending_free { 303 struct list_head zapped; 304 DECLARE_BITMAP(lock_chains_being_freed, MAX_LOCKDEP_CHAINS); 305 }; 306 307 /** 308 * struct delayed_free - data structures used for delayed freeing 309 * 310 * A data structure for delayed freeing of data structures that may be 311 * accessed by RCU readers at the time these were freed. 312 * 313 * @rcu_head: Used to schedule an RCU callback for freeing data structures. 314 * @index: Index of @pf to which freed data structures are added. 315 * @scheduled: Whether or not an RCU callback has been scheduled. 316 * @pf: Array with information about data structures about to be freed. 317 */ 318 static struct delayed_free { 319 struct rcu_head rcu_head; 320 int index; 321 int scheduled; 322 struct pending_free pf[2]; 323 } delayed_free; 324 325 /* 326 * The lockdep classes are in a hash-table as well, for fast lookup: 327 */ 328 #define CLASSHASH_BITS (MAX_LOCKDEP_KEYS_BITS - 1) 329 #define CLASSHASH_SIZE (1UL << CLASSHASH_BITS) 330 #define __classhashfn(key) hash_long((unsigned long)key, CLASSHASH_BITS) 331 #define classhashentry(key) (classhash_table + __classhashfn((key))) 332 333 static struct hlist_head classhash_table[CLASSHASH_SIZE]; 334 335 /* 336 * We put the lock dependency chains into a hash-table as well, to cache 337 * their existence: 338 */ 339 #define CHAINHASH_BITS (MAX_LOCKDEP_CHAINS_BITS-1) 340 #define CHAINHASH_SIZE (1UL << CHAINHASH_BITS) 341 #define __chainhashfn(chain) hash_long(chain, CHAINHASH_BITS) 342 #define chainhashentry(chain) (chainhash_table + __chainhashfn((chain))) 343 344 static struct hlist_head chainhash_table[CHAINHASH_SIZE]; 345 346 /* 347 * The hash key of the lock dependency chains is a hash itself too: 348 * it's a hash of all locks taken up to that lock, including that lock. 349 * It's a 64-bit hash, because it's important for the keys to be 350 * unique. 351 */ 352 static inline u64 iterate_chain_key(u64 key, u32 idx) 353 { 354 u32 k0 = key, k1 = key >> 32; 355 356 __jhash_mix(idx, k0, k1); /* Macro that modifies arguments! */ 357 358 return k0 | (u64)k1 << 32; 359 } 360 361 void lockdep_off(void) 362 { 363 current->lockdep_recursion++; 364 } 365 EXPORT_SYMBOL(lockdep_off); 366 367 void lockdep_on(void) 368 { 369 current->lockdep_recursion--; 370 } 371 EXPORT_SYMBOL(lockdep_on); 372 373 void lockdep_set_selftest_task(struct task_struct *task) 374 { 375 lockdep_selftest_task_struct = task; 376 } 377 378 /* 379 * Debugging switches: 380 */ 381 382 #define VERBOSE 0 383 #define VERY_VERBOSE 0 384 385 #if VERBOSE 386 # define HARDIRQ_VERBOSE 1 387 # define SOFTIRQ_VERBOSE 1 388 #else 389 # define HARDIRQ_VERBOSE 0 390 # define SOFTIRQ_VERBOSE 0 391 #endif 392 393 #if VERBOSE || HARDIRQ_VERBOSE || SOFTIRQ_VERBOSE 394 /* 395 * Quick filtering for interesting events: 396 */ 397 static int class_filter(struct lock_class *class) 398 { 399 #if 0 400 /* Example */ 401 if (class->name_version == 1 && 402 !strcmp(class->name, "lockname")) 403 return 1; 404 if (class->name_version == 1 && 405 !strcmp(class->name, "&struct->lockfield")) 406 return 1; 407 #endif 408 /* Filter everything else. 1 would be to allow everything else */ 409 return 0; 410 } 411 #endif 412 413 static int verbose(struct lock_class *class) 414 { 415 #if VERBOSE 416 return class_filter(class); 417 #endif 418 return 0; 419 } 420 421 /* 422 * Stack-trace: tightly packed array of stack backtrace 423 * addresses. Protected by the graph_lock. 424 */ 425 unsigned long nr_stack_trace_entries; 426 static unsigned long stack_trace[MAX_STACK_TRACE_ENTRIES]; 427 428 static void print_lockdep_off(const char *bug_msg) 429 { 430 printk(KERN_DEBUG "%s\n", bug_msg); 431 printk(KERN_DEBUG "turning off the locking correctness validator.\n"); 432 #ifdef CONFIG_LOCK_STAT 433 printk(KERN_DEBUG "Please attach the output of /proc/lock_stat to the bug report\n"); 434 #endif 435 } 436 437 static int save_trace(struct stack_trace *trace) 438 { 439 trace->nr_entries = 0; 440 trace->max_entries = MAX_STACK_TRACE_ENTRIES - nr_stack_trace_entries; 441 trace->entries = stack_trace + nr_stack_trace_entries; 442 443 trace->skip = 3; 444 445 save_stack_trace(trace); 446 447 /* 448 * Some daft arches put -1 at the end to indicate its a full trace. 449 * 450 * <rant> this is buggy anyway, since it takes a whole extra entry so a 451 * complete trace that maxes out the entries provided will be reported 452 * as incomplete, friggin useless </rant> 453 */ 454 if (trace->nr_entries != 0 && 455 trace->entries[trace->nr_entries-1] == ULONG_MAX) 456 trace->nr_entries--; 457 458 trace->max_entries = trace->nr_entries; 459 460 nr_stack_trace_entries += trace->nr_entries; 461 462 if (nr_stack_trace_entries >= MAX_STACK_TRACE_ENTRIES-1) { 463 if (!debug_locks_off_graph_unlock()) 464 return 0; 465 466 print_lockdep_off("BUG: MAX_STACK_TRACE_ENTRIES too low!"); 467 dump_stack(); 468 469 return 0; 470 } 471 472 return 1; 473 } 474 475 unsigned int nr_hardirq_chains; 476 unsigned int nr_softirq_chains; 477 unsigned int nr_process_chains; 478 unsigned int max_lockdep_depth; 479 480 #ifdef CONFIG_DEBUG_LOCKDEP 481 /* 482 * Various lockdep statistics: 483 */ 484 DEFINE_PER_CPU(struct lockdep_stats, lockdep_stats); 485 #endif 486 487 /* 488 * Locking printouts: 489 */ 490 491 #define __USAGE(__STATE) \ 492 [LOCK_USED_IN_##__STATE] = "IN-"__stringify(__STATE)"-W", \ 493 [LOCK_ENABLED_##__STATE] = __stringify(__STATE)"-ON-W", \ 494 [LOCK_USED_IN_##__STATE##_READ] = "IN-"__stringify(__STATE)"-R",\ 495 [LOCK_ENABLED_##__STATE##_READ] = __stringify(__STATE)"-ON-R", 496 497 static const char *usage_str[] = 498 { 499 #define LOCKDEP_STATE(__STATE) __USAGE(__STATE) 500 #include "lockdep_states.h" 501 #undef LOCKDEP_STATE 502 [LOCK_USED] = "INITIAL USE", 503 }; 504 505 const char * __get_key_name(struct lockdep_subclass_key *key, char *str) 506 { 507 return kallsyms_lookup((unsigned long)key, NULL, NULL, NULL, str); 508 } 509 510 static inline unsigned long lock_flag(enum lock_usage_bit bit) 511 { 512 return 1UL << bit; 513 } 514 515 static char get_usage_char(struct lock_class *class, enum lock_usage_bit bit) 516 { 517 char c = '.'; 518 519 if (class->usage_mask & lock_flag(bit + 2)) 520 c = '+'; 521 if (class->usage_mask & lock_flag(bit)) { 522 c = '-'; 523 if (class->usage_mask & lock_flag(bit + 2)) 524 c = '?'; 525 } 526 527 return c; 528 } 529 530 void get_usage_chars(struct lock_class *class, char usage[LOCK_USAGE_CHARS]) 531 { 532 int i = 0; 533 534 #define LOCKDEP_STATE(__STATE) \ 535 usage[i++] = get_usage_char(class, LOCK_USED_IN_##__STATE); \ 536 usage[i++] = get_usage_char(class, LOCK_USED_IN_##__STATE##_READ); 537 #include "lockdep_states.h" 538 #undef LOCKDEP_STATE 539 540 usage[i] = '\0'; 541 } 542 543 static void __print_lock_name(struct lock_class *class) 544 { 545 char str[KSYM_NAME_LEN]; 546 const char *name; 547 548 name = class->name; 549 if (!name) { 550 name = __get_key_name(class->key, str); 551 printk(KERN_CONT "%s", name); 552 } else { 553 printk(KERN_CONT "%s", name); 554 if (class->name_version > 1) 555 printk(KERN_CONT "#%d", class->name_version); 556 if (class->subclass) 557 printk(KERN_CONT "/%d", class->subclass); 558 } 559 } 560 561 static void print_lock_name(struct lock_class *class) 562 { 563 char usage[LOCK_USAGE_CHARS]; 564 565 get_usage_chars(class, usage); 566 567 printk(KERN_CONT " ("); 568 __print_lock_name(class); 569 printk(KERN_CONT "){%s}", usage); 570 } 571 572 static void print_lockdep_cache(struct lockdep_map *lock) 573 { 574 const char *name; 575 char str[KSYM_NAME_LEN]; 576 577 name = lock->name; 578 if (!name) 579 name = __get_key_name(lock->key->subkeys, str); 580 581 printk(KERN_CONT "%s", name); 582 } 583 584 static void print_lock(struct held_lock *hlock) 585 { 586 /* 587 * We can be called locklessly through debug_show_all_locks() so be 588 * extra careful, the hlock might have been released and cleared. 589 */ 590 unsigned int class_idx = hlock->class_idx; 591 592 /* Don't re-read hlock->class_idx, can't use READ_ONCE() on bitfields: */ 593 barrier(); 594 595 if (!class_idx || (class_idx - 1) >= MAX_LOCKDEP_KEYS) { 596 printk(KERN_CONT "<RELEASED>\n"); 597 return; 598 } 599 600 printk(KERN_CONT "%p", hlock->instance); 601 print_lock_name(lock_classes + class_idx - 1); 602 printk(KERN_CONT ", at: %pS\n", (void *)hlock->acquire_ip); 603 } 604 605 static void lockdep_print_held_locks(struct task_struct *p) 606 { 607 int i, depth = READ_ONCE(p->lockdep_depth); 608 609 if (!depth) 610 printk("no locks held by %s/%d.\n", p->comm, task_pid_nr(p)); 611 else 612 printk("%d lock%s held by %s/%d:\n", depth, 613 depth > 1 ? "s" : "", p->comm, task_pid_nr(p)); 614 /* 615 * It's not reliable to print a task's held locks if it's not sleeping 616 * and it's not the current task. 617 */ 618 if (p->state == TASK_RUNNING && p != current) 619 return; 620 for (i = 0; i < depth; i++) { 621 printk(" #%d: ", i); 622 print_lock(p->held_locks + i); 623 } 624 } 625 626 static void print_kernel_ident(void) 627 { 628 printk("%s %.*s %s\n", init_utsname()->release, 629 (int)strcspn(init_utsname()->version, " "), 630 init_utsname()->version, 631 print_tainted()); 632 } 633 634 static int very_verbose(struct lock_class *class) 635 { 636 #if VERY_VERBOSE 637 return class_filter(class); 638 #endif 639 return 0; 640 } 641 642 /* 643 * Is this the address of a static object: 644 */ 645 #ifdef __KERNEL__ 646 static int static_obj(const void *obj) 647 { 648 unsigned long start = (unsigned long) &_stext, 649 end = (unsigned long) &_end, 650 addr = (unsigned long) obj; 651 652 /* 653 * static variable? 654 */ 655 if ((addr >= start) && (addr < end)) 656 return 1; 657 658 if (arch_is_kernel_data(addr)) 659 return 1; 660 661 /* 662 * in-kernel percpu var? 663 */ 664 if (is_kernel_percpu_address(addr)) 665 return 1; 666 667 /* 668 * module static or percpu var? 669 */ 670 return is_module_address(addr) || is_module_percpu_address(addr); 671 } 672 #endif 673 674 /* 675 * To make lock name printouts unique, we calculate a unique 676 * class->name_version generation counter. The caller must hold the graph 677 * lock. 678 */ 679 static int count_matching_names(struct lock_class *new_class) 680 { 681 struct lock_class *class; 682 int count = 0; 683 684 if (!new_class->name) 685 return 0; 686 687 list_for_each_entry(class, &all_lock_classes, lock_entry) { 688 if (new_class->key - new_class->subclass == class->key) 689 return class->name_version; 690 if (class->name && !strcmp(class->name, new_class->name)) 691 count = max(count, class->name_version); 692 } 693 694 return count + 1; 695 } 696 697 static inline struct lock_class * 698 look_up_lock_class(const struct lockdep_map *lock, unsigned int subclass) 699 { 700 struct lockdep_subclass_key *key; 701 struct hlist_head *hash_head; 702 struct lock_class *class; 703 704 if (unlikely(subclass >= MAX_LOCKDEP_SUBCLASSES)) { 705 debug_locks_off(); 706 printk(KERN_ERR 707 "BUG: looking up invalid subclass: %u\n", subclass); 708 printk(KERN_ERR 709 "turning off the locking correctness validator.\n"); 710 dump_stack(); 711 return NULL; 712 } 713 714 /* 715 * If it is not initialised then it has never been locked, 716 * so it won't be present in the hash table. 717 */ 718 if (unlikely(!lock->key)) 719 return NULL; 720 721 /* 722 * NOTE: the class-key must be unique. For dynamic locks, a static 723 * lock_class_key variable is passed in through the mutex_init() 724 * (or spin_lock_init()) call - which acts as the key. For static 725 * locks we use the lock object itself as the key. 726 */ 727 BUILD_BUG_ON(sizeof(struct lock_class_key) > 728 sizeof(struct lockdep_map)); 729 730 key = lock->key->subkeys + subclass; 731 732 hash_head = classhashentry(key); 733 734 /* 735 * We do an RCU walk of the hash, see lockdep_free_key_range(). 736 */ 737 if (DEBUG_LOCKS_WARN_ON(!irqs_disabled())) 738 return NULL; 739 740 hlist_for_each_entry_rcu(class, hash_head, hash_entry) { 741 if (class->key == key) { 742 /* 743 * Huh! same key, different name? Did someone trample 744 * on some memory? We're most confused. 745 */ 746 WARN_ON_ONCE(class->name != lock->name); 747 return class; 748 } 749 } 750 751 return NULL; 752 } 753 754 /* 755 * Static locks do not have their class-keys yet - for them the key is 756 * the lock object itself. If the lock is in the per cpu area, the 757 * canonical address of the lock (per cpu offset removed) is used. 758 */ 759 static bool assign_lock_key(struct lockdep_map *lock) 760 { 761 unsigned long can_addr, addr = (unsigned long)lock; 762 763 #ifdef __KERNEL__ 764 /* 765 * lockdep_free_key_range() assumes that struct lock_class_key 766 * objects do not overlap. Since we use the address of lock 767 * objects as class key for static objects, check whether the 768 * size of lock_class_key objects does not exceed the size of 769 * the smallest lock object. 770 */ 771 BUILD_BUG_ON(sizeof(struct lock_class_key) > sizeof(raw_spinlock_t)); 772 #endif 773 774 if (__is_kernel_percpu_address(addr, &can_addr)) 775 lock->key = (void *)can_addr; 776 else if (__is_module_percpu_address(addr, &can_addr)) 777 lock->key = (void *)can_addr; 778 else if (static_obj(lock)) 779 lock->key = (void *)lock; 780 else { 781 /* Debug-check: all keys must be persistent! */ 782 debug_locks_off(); 783 pr_err("INFO: trying to register non-static key.\n"); 784 pr_err("the code is fine but needs lockdep annotation.\n"); 785 pr_err("turning off the locking correctness validator.\n"); 786 dump_stack(); 787 return false; 788 } 789 790 return true; 791 } 792 793 #ifdef CONFIG_DEBUG_LOCKDEP 794 795 /* Check whether element @e occurs in list @h */ 796 static bool in_list(struct list_head *e, struct list_head *h) 797 { 798 struct list_head *f; 799 800 list_for_each(f, h) { 801 if (e == f) 802 return true; 803 } 804 805 return false; 806 } 807 808 /* 809 * Check whether entry @e occurs in any of the locks_after or locks_before 810 * lists. 811 */ 812 static bool in_any_class_list(struct list_head *e) 813 { 814 struct lock_class *class; 815 int i; 816 817 for (i = 0; i < ARRAY_SIZE(lock_classes); i++) { 818 class = &lock_classes[i]; 819 if (in_list(e, &class->locks_after) || 820 in_list(e, &class->locks_before)) 821 return true; 822 } 823 return false; 824 } 825 826 static bool class_lock_list_valid(struct lock_class *c, struct list_head *h) 827 { 828 struct lock_list *e; 829 830 list_for_each_entry(e, h, entry) { 831 if (e->links_to != c) { 832 printk(KERN_INFO "class %s: mismatch for lock entry %ld; class %s <> %s", 833 c->name ? : "(?)", 834 (unsigned long)(e - list_entries), 835 e->links_to && e->links_to->name ? 836 e->links_to->name : "(?)", 837 e->class && e->class->name ? e->class->name : 838 "(?)"); 839 return false; 840 } 841 } 842 return true; 843 } 844 845 #ifdef CONFIG_PROVE_LOCKING 846 static u16 chain_hlocks[MAX_LOCKDEP_CHAIN_HLOCKS]; 847 #endif 848 849 static bool check_lock_chain_key(struct lock_chain *chain) 850 { 851 #ifdef CONFIG_PROVE_LOCKING 852 u64 chain_key = 0; 853 int i; 854 855 for (i = chain->base; i < chain->base + chain->depth; i++) 856 chain_key = iterate_chain_key(chain_key, chain_hlocks[i] + 1); 857 /* 858 * The 'unsigned long long' casts avoid that a compiler warning 859 * is reported when building tools/lib/lockdep. 860 */ 861 if (chain->chain_key != chain_key) { 862 printk(KERN_INFO "chain %lld: key %#llx <> %#llx\n", 863 (unsigned long long)(chain - lock_chains), 864 (unsigned long long)chain->chain_key, 865 (unsigned long long)chain_key); 866 return false; 867 } 868 #endif 869 return true; 870 } 871 872 static bool in_any_zapped_class_list(struct lock_class *class) 873 { 874 struct pending_free *pf; 875 int i; 876 877 for (i = 0, pf = delayed_free.pf; i < ARRAY_SIZE(delayed_free.pf); i++, pf++) { 878 if (in_list(&class->lock_entry, &pf->zapped)) 879 return true; 880 } 881 882 return false; 883 } 884 885 static bool __check_data_structures(void) 886 { 887 struct lock_class *class; 888 struct lock_chain *chain; 889 struct hlist_head *head; 890 struct lock_list *e; 891 int i; 892 893 /* Check whether all classes occur in a lock list. */ 894 for (i = 0; i < ARRAY_SIZE(lock_classes); i++) { 895 class = &lock_classes[i]; 896 if (!in_list(&class->lock_entry, &all_lock_classes) && 897 !in_list(&class->lock_entry, &free_lock_classes) && 898 !in_any_zapped_class_list(class)) { 899 printk(KERN_INFO "class %px/%s is not in any class list\n", 900 class, class->name ? : "(?)"); 901 return false; 902 } 903 } 904 905 /* Check whether all classes have valid lock lists. */ 906 for (i = 0; i < ARRAY_SIZE(lock_classes); i++) { 907 class = &lock_classes[i]; 908 if (!class_lock_list_valid(class, &class->locks_before)) 909 return false; 910 if (!class_lock_list_valid(class, &class->locks_after)) 911 return false; 912 } 913 914 /* Check the chain_key of all lock chains. */ 915 for (i = 0; i < ARRAY_SIZE(chainhash_table); i++) { 916 head = chainhash_table + i; 917 hlist_for_each_entry_rcu(chain, head, entry) { 918 if (!check_lock_chain_key(chain)) 919 return false; 920 } 921 } 922 923 /* 924 * Check whether all list entries that are in use occur in a class 925 * lock list. 926 */ 927 for_each_set_bit(i, list_entries_in_use, ARRAY_SIZE(list_entries)) { 928 e = list_entries + i; 929 if (!in_any_class_list(&e->entry)) { 930 printk(KERN_INFO "list entry %d is not in any class list; class %s <> %s\n", 931 (unsigned int)(e - list_entries), 932 e->class->name ? : "(?)", 933 e->links_to->name ? : "(?)"); 934 return false; 935 } 936 } 937 938 /* 939 * Check whether all list entries that are not in use do not occur in 940 * a class lock list. 941 */ 942 for_each_clear_bit(i, list_entries_in_use, ARRAY_SIZE(list_entries)) { 943 e = list_entries + i; 944 if (in_any_class_list(&e->entry)) { 945 printk(KERN_INFO "list entry %d occurs in a class list; class %s <> %s\n", 946 (unsigned int)(e - list_entries), 947 e->class && e->class->name ? e->class->name : 948 "(?)", 949 e->links_to && e->links_to->name ? 950 e->links_to->name : "(?)"); 951 return false; 952 } 953 } 954 955 return true; 956 } 957 958 int check_consistency = 0; 959 module_param(check_consistency, int, 0644); 960 961 static void check_data_structures(void) 962 { 963 static bool once = false; 964 965 if (check_consistency && !once) { 966 if (!__check_data_structures()) { 967 once = true; 968 WARN_ON(once); 969 } 970 } 971 } 972 973 #else /* CONFIG_DEBUG_LOCKDEP */ 974 975 static inline void check_data_structures(void) { } 976 977 #endif /* CONFIG_DEBUG_LOCKDEP */ 978 979 /* 980 * Initialize the lock_classes[] array elements, the free_lock_classes list 981 * and also the delayed_free structure. 982 */ 983 static void init_data_structures_once(void) 984 { 985 static bool ds_initialized, rcu_head_initialized; 986 int i; 987 988 if (likely(rcu_head_initialized)) 989 return; 990 991 if (system_state >= SYSTEM_SCHEDULING) { 992 init_rcu_head(&delayed_free.rcu_head); 993 rcu_head_initialized = true; 994 } 995 996 if (ds_initialized) 997 return; 998 999 ds_initialized = true; 1000 1001 INIT_LIST_HEAD(&delayed_free.pf[0].zapped); 1002 INIT_LIST_HEAD(&delayed_free.pf[1].zapped); 1003 1004 for (i = 0; i < ARRAY_SIZE(lock_classes); i++) { 1005 list_add_tail(&lock_classes[i].lock_entry, &free_lock_classes); 1006 INIT_LIST_HEAD(&lock_classes[i].locks_after); 1007 INIT_LIST_HEAD(&lock_classes[i].locks_before); 1008 } 1009 } 1010 1011 static inline struct hlist_head *keyhashentry(const struct lock_class_key *key) 1012 { 1013 unsigned long hash = hash_long((uintptr_t)key, KEYHASH_BITS); 1014 1015 return lock_keys_hash + hash; 1016 } 1017 1018 /* Register a dynamically allocated key. */ 1019 void lockdep_register_key(struct lock_class_key *key) 1020 { 1021 struct hlist_head *hash_head; 1022 struct lock_class_key *k; 1023 unsigned long flags; 1024 1025 if (WARN_ON_ONCE(static_obj(key))) 1026 return; 1027 hash_head = keyhashentry(key); 1028 1029 raw_local_irq_save(flags); 1030 if (!graph_lock()) 1031 goto restore_irqs; 1032 hlist_for_each_entry_rcu(k, hash_head, hash_entry) { 1033 if (WARN_ON_ONCE(k == key)) 1034 goto out_unlock; 1035 } 1036 hlist_add_head_rcu(&key->hash_entry, hash_head); 1037 out_unlock: 1038 graph_unlock(); 1039 restore_irqs: 1040 raw_local_irq_restore(flags); 1041 } 1042 EXPORT_SYMBOL_GPL(lockdep_register_key); 1043 1044 /* Check whether a key has been registered as a dynamic key. */ 1045 static bool is_dynamic_key(const struct lock_class_key *key) 1046 { 1047 struct hlist_head *hash_head; 1048 struct lock_class_key *k; 1049 bool found = false; 1050 1051 if (WARN_ON_ONCE(static_obj(key))) 1052 return false; 1053 1054 /* 1055 * If lock debugging is disabled lock_keys_hash[] may contain 1056 * pointers to memory that has already been freed. Avoid triggering 1057 * a use-after-free in that case by returning early. 1058 */ 1059 if (!debug_locks) 1060 return true; 1061 1062 hash_head = keyhashentry(key); 1063 1064 rcu_read_lock(); 1065 hlist_for_each_entry_rcu(k, hash_head, hash_entry) { 1066 if (k == key) { 1067 found = true; 1068 break; 1069 } 1070 } 1071 rcu_read_unlock(); 1072 1073 return found; 1074 } 1075 1076 /* 1077 * Register a lock's class in the hash-table, if the class is not present 1078 * yet. Otherwise we look it up. We cache the result in the lock object 1079 * itself, so actual lookup of the hash should be once per lock object. 1080 */ 1081 static struct lock_class * 1082 register_lock_class(struct lockdep_map *lock, unsigned int subclass, int force) 1083 { 1084 struct lockdep_subclass_key *key; 1085 struct hlist_head *hash_head; 1086 struct lock_class *class; 1087 1088 DEBUG_LOCKS_WARN_ON(!irqs_disabled()); 1089 1090 class = look_up_lock_class(lock, subclass); 1091 if (likely(class)) 1092 goto out_set_class_cache; 1093 1094 if (!lock->key) { 1095 if (!assign_lock_key(lock)) 1096 return NULL; 1097 } else if (!static_obj(lock->key) && !is_dynamic_key(lock->key)) { 1098 return NULL; 1099 } 1100 1101 key = lock->key->subkeys + subclass; 1102 hash_head = classhashentry(key); 1103 1104 if (!graph_lock()) { 1105 return NULL; 1106 } 1107 /* 1108 * We have to do the hash-walk again, to avoid races 1109 * with another CPU: 1110 */ 1111 hlist_for_each_entry_rcu(class, hash_head, hash_entry) { 1112 if (class->key == key) 1113 goto out_unlock_set; 1114 } 1115 1116 init_data_structures_once(); 1117 1118 /* Allocate a new lock class and add it to the hash. */ 1119 class = list_first_entry_or_null(&free_lock_classes, typeof(*class), 1120 lock_entry); 1121 if (!class) { 1122 if (!debug_locks_off_graph_unlock()) { 1123 return NULL; 1124 } 1125 1126 print_lockdep_off("BUG: MAX_LOCKDEP_KEYS too low!"); 1127 dump_stack(); 1128 return NULL; 1129 } 1130 nr_lock_classes++; 1131 debug_atomic_inc(nr_unused_locks); 1132 class->key = key; 1133 class->name = lock->name; 1134 class->subclass = subclass; 1135 WARN_ON_ONCE(!list_empty(&class->locks_before)); 1136 WARN_ON_ONCE(!list_empty(&class->locks_after)); 1137 class->name_version = count_matching_names(class); 1138 /* 1139 * We use RCU's safe list-add method to make 1140 * parallel walking of the hash-list safe: 1141 */ 1142 hlist_add_head_rcu(&class->hash_entry, hash_head); 1143 /* 1144 * Remove the class from the free list and add it to the global list 1145 * of classes. 1146 */ 1147 list_move_tail(&class->lock_entry, &all_lock_classes); 1148 1149 if (verbose(class)) { 1150 graph_unlock(); 1151 1152 printk("\nnew class %px: %s", class->key, class->name); 1153 if (class->name_version > 1) 1154 printk(KERN_CONT "#%d", class->name_version); 1155 printk(KERN_CONT "\n"); 1156 dump_stack(); 1157 1158 if (!graph_lock()) { 1159 return NULL; 1160 } 1161 } 1162 out_unlock_set: 1163 graph_unlock(); 1164 1165 out_set_class_cache: 1166 if (!subclass || force) 1167 lock->class_cache[0] = class; 1168 else if (subclass < NR_LOCKDEP_CACHING_CLASSES) 1169 lock->class_cache[subclass] = class; 1170 1171 /* 1172 * Hash collision, did we smoke some? We found a class with a matching 1173 * hash but the subclass -- which is hashed in -- didn't match. 1174 */ 1175 if (DEBUG_LOCKS_WARN_ON(class->subclass != subclass)) 1176 return NULL; 1177 1178 return class; 1179 } 1180 1181 #ifdef CONFIG_PROVE_LOCKING 1182 /* 1183 * Allocate a lockdep entry. (assumes the graph_lock held, returns 1184 * with NULL on failure) 1185 */ 1186 static struct lock_list *alloc_list_entry(void) 1187 { 1188 int idx = find_first_zero_bit(list_entries_in_use, 1189 ARRAY_SIZE(list_entries)); 1190 1191 if (idx >= ARRAY_SIZE(list_entries)) { 1192 if (!debug_locks_off_graph_unlock()) 1193 return NULL; 1194 1195 print_lockdep_off("BUG: MAX_LOCKDEP_ENTRIES too low!"); 1196 dump_stack(); 1197 return NULL; 1198 } 1199 nr_list_entries++; 1200 __set_bit(idx, list_entries_in_use); 1201 return list_entries + idx; 1202 } 1203 1204 /* 1205 * Add a new dependency to the head of the list: 1206 */ 1207 static int add_lock_to_list(struct lock_class *this, 1208 struct lock_class *links_to, struct list_head *head, 1209 unsigned long ip, int distance, 1210 struct stack_trace *trace) 1211 { 1212 struct lock_list *entry; 1213 /* 1214 * Lock not present yet - get a new dependency struct and 1215 * add it to the list: 1216 */ 1217 entry = alloc_list_entry(); 1218 if (!entry) 1219 return 0; 1220 1221 entry->class = this; 1222 entry->links_to = links_to; 1223 entry->distance = distance; 1224 entry->trace = *trace; 1225 /* 1226 * Both allocation and removal are done under the graph lock; but 1227 * iteration is under RCU-sched; see look_up_lock_class() and 1228 * lockdep_free_key_range(). 1229 */ 1230 list_add_tail_rcu(&entry->entry, head); 1231 1232 return 1; 1233 } 1234 1235 /* 1236 * For good efficiency of modular, we use power of 2 1237 */ 1238 #define MAX_CIRCULAR_QUEUE_SIZE 4096UL 1239 #define CQ_MASK (MAX_CIRCULAR_QUEUE_SIZE-1) 1240 1241 /* 1242 * The circular_queue and helpers is used to implement the 1243 * breadth-first search(BFS)algorithem, by which we can build 1244 * the shortest path from the next lock to be acquired to the 1245 * previous held lock if there is a circular between them. 1246 */ 1247 struct circular_queue { 1248 unsigned long element[MAX_CIRCULAR_QUEUE_SIZE]; 1249 unsigned int front, rear; 1250 }; 1251 1252 static struct circular_queue lock_cq; 1253 1254 unsigned int max_bfs_queue_depth; 1255 1256 static unsigned int lockdep_dependency_gen_id; 1257 1258 static inline void __cq_init(struct circular_queue *cq) 1259 { 1260 cq->front = cq->rear = 0; 1261 lockdep_dependency_gen_id++; 1262 } 1263 1264 static inline int __cq_empty(struct circular_queue *cq) 1265 { 1266 return (cq->front == cq->rear); 1267 } 1268 1269 static inline int __cq_full(struct circular_queue *cq) 1270 { 1271 return ((cq->rear + 1) & CQ_MASK) == cq->front; 1272 } 1273 1274 static inline int __cq_enqueue(struct circular_queue *cq, unsigned long elem) 1275 { 1276 if (__cq_full(cq)) 1277 return -1; 1278 1279 cq->element[cq->rear] = elem; 1280 cq->rear = (cq->rear + 1) & CQ_MASK; 1281 return 0; 1282 } 1283 1284 static inline int __cq_dequeue(struct circular_queue *cq, unsigned long *elem) 1285 { 1286 if (__cq_empty(cq)) 1287 return -1; 1288 1289 *elem = cq->element[cq->front]; 1290 cq->front = (cq->front + 1) & CQ_MASK; 1291 return 0; 1292 } 1293 1294 static inline unsigned int __cq_get_elem_count(struct circular_queue *cq) 1295 { 1296 return (cq->rear - cq->front) & CQ_MASK; 1297 } 1298 1299 static inline void mark_lock_accessed(struct lock_list *lock, 1300 struct lock_list *parent) 1301 { 1302 unsigned long nr; 1303 1304 nr = lock - list_entries; 1305 WARN_ON(nr >= ARRAY_SIZE(list_entries)); /* Out-of-bounds, input fail */ 1306 lock->parent = parent; 1307 lock->class->dep_gen_id = lockdep_dependency_gen_id; 1308 } 1309 1310 static inline unsigned long lock_accessed(struct lock_list *lock) 1311 { 1312 unsigned long nr; 1313 1314 nr = lock - list_entries; 1315 WARN_ON(nr >= ARRAY_SIZE(list_entries)); /* Out-of-bounds, input fail */ 1316 return lock->class->dep_gen_id == lockdep_dependency_gen_id; 1317 } 1318 1319 static inline struct lock_list *get_lock_parent(struct lock_list *child) 1320 { 1321 return child->parent; 1322 } 1323 1324 static inline int get_lock_depth(struct lock_list *child) 1325 { 1326 int depth = 0; 1327 struct lock_list *parent; 1328 1329 while ((parent = get_lock_parent(child))) { 1330 child = parent; 1331 depth++; 1332 } 1333 return depth; 1334 } 1335 1336 static int __bfs(struct lock_list *source_entry, 1337 void *data, 1338 int (*match)(struct lock_list *entry, void *data), 1339 struct lock_list **target_entry, 1340 int forward) 1341 { 1342 struct lock_list *entry; 1343 struct list_head *head; 1344 struct circular_queue *cq = &lock_cq; 1345 int ret = 1; 1346 1347 if (match(source_entry, data)) { 1348 *target_entry = source_entry; 1349 ret = 0; 1350 goto exit; 1351 } 1352 1353 if (forward) 1354 head = &source_entry->class->locks_after; 1355 else 1356 head = &source_entry->class->locks_before; 1357 1358 if (list_empty(head)) 1359 goto exit; 1360 1361 __cq_init(cq); 1362 __cq_enqueue(cq, (unsigned long)source_entry); 1363 1364 while (!__cq_empty(cq)) { 1365 struct lock_list *lock; 1366 1367 __cq_dequeue(cq, (unsigned long *)&lock); 1368 1369 if (!lock->class) { 1370 ret = -2; 1371 goto exit; 1372 } 1373 1374 if (forward) 1375 head = &lock->class->locks_after; 1376 else 1377 head = &lock->class->locks_before; 1378 1379 DEBUG_LOCKS_WARN_ON(!irqs_disabled()); 1380 1381 list_for_each_entry_rcu(entry, head, entry) { 1382 if (!lock_accessed(entry)) { 1383 unsigned int cq_depth; 1384 mark_lock_accessed(entry, lock); 1385 if (match(entry, data)) { 1386 *target_entry = entry; 1387 ret = 0; 1388 goto exit; 1389 } 1390 1391 if (__cq_enqueue(cq, (unsigned long)entry)) { 1392 ret = -1; 1393 goto exit; 1394 } 1395 cq_depth = __cq_get_elem_count(cq); 1396 if (max_bfs_queue_depth < cq_depth) 1397 max_bfs_queue_depth = cq_depth; 1398 } 1399 } 1400 } 1401 exit: 1402 return ret; 1403 } 1404 1405 static inline int __bfs_forwards(struct lock_list *src_entry, 1406 void *data, 1407 int (*match)(struct lock_list *entry, void *data), 1408 struct lock_list **target_entry) 1409 { 1410 return __bfs(src_entry, data, match, target_entry, 1); 1411 1412 } 1413 1414 static inline int __bfs_backwards(struct lock_list *src_entry, 1415 void *data, 1416 int (*match)(struct lock_list *entry, void *data), 1417 struct lock_list **target_entry) 1418 { 1419 return __bfs(src_entry, data, match, target_entry, 0); 1420 1421 } 1422 1423 /* 1424 * Recursive, forwards-direction lock-dependency checking, used for 1425 * both noncyclic checking and for hardirq-unsafe/softirq-unsafe 1426 * checking. 1427 */ 1428 1429 /* 1430 * Print a dependency chain entry (this is only done when a deadlock 1431 * has been detected): 1432 */ 1433 static noinline int 1434 print_circular_bug_entry(struct lock_list *target, int depth) 1435 { 1436 if (debug_locks_silent) 1437 return 0; 1438 printk("\n-> #%u", depth); 1439 print_lock_name(target->class); 1440 printk(KERN_CONT ":\n"); 1441 print_stack_trace(&target->trace, 6); 1442 1443 return 0; 1444 } 1445 1446 static void 1447 print_circular_lock_scenario(struct held_lock *src, 1448 struct held_lock *tgt, 1449 struct lock_list *prt) 1450 { 1451 struct lock_class *source = hlock_class(src); 1452 struct lock_class *target = hlock_class(tgt); 1453 struct lock_class *parent = prt->class; 1454 1455 /* 1456 * A direct locking problem where unsafe_class lock is taken 1457 * directly by safe_class lock, then all we need to show 1458 * is the deadlock scenario, as it is obvious that the 1459 * unsafe lock is taken under the safe lock. 1460 * 1461 * But if there is a chain instead, where the safe lock takes 1462 * an intermediate lock (middle_class) where this lock is 1463 * not the same as the safe lock, then the lock chain is 1464 * used to describe the problem. Otherwise we would need 1465 * to show a different CPU case for each link in the chain 1466 * from the safe_class lock to the unsafe_class lock. 1467 */ 1468 if (parent != source) { 1469 printk("Chain exists of:\n "); 1470 __print_lock_name(source); 1471 printk(KERN_CONT " --> "); 1472 __print_lock_name(parent); 1473 printk(KERN_CONT " --> "); 1474 __print_lock_name(target); 1475 printk(KERN_CONT "\n\n"); 1476 } 1477 1478 printk(" Possible unsafe locking scenario:\n\n"); 1479 printk(" CPU0 CPU1\n"); 1480 printk(" ---- ----\n"); 1481 printk(" lock("); 1482 __print_lock_name(target); 1483 printk(KERN_CONT ");\n"); 1484 printk(" lock("); 1485 __print_lock_name(parent); 1486 printk(KERN_CONT ");\n"); 1487 printk(" lock("); 1488 __print_lock_name(target); 1489 printk(KERN_CONT ");\n"); 1490 printk(" lock("); 1491 __print_lock_name(source); 1492 printk(KERN_CONT ");\n"); 1493 printk("\n *** DEADLOCK ***\n\n"); 1494 } 1495 1496 /* 1497 * When a circular dependency is detected, print the 1498 * header first: 1499 */ 1500 static noinline int 1501 print_circular_bug_header(struct lock_list *entry, unsigned int depth, 1502 struct held_lock *check_src, 1503 struct held_lock *check_tgt) 1504 { 1505 struct task_struct *curr = current; 1506 1507 if (debug_locks_silent) 1508 return 0; 1509 1510 pr_warn("\n"); 1511 pr_warn("======================================================\n"); 1512 pr_warn("WARNING: possible circular locking dependency detected\n"); 1513 print_kernel_ident(); 1514 pr_warn("------------------------------------------------------\n"); 1515 pr_warn("%s/%d is trying to acquire lock:\n", 1516 curr->comm, task_pid_nr(curr)); 1517 print_lock(check_src); 1518 1519 pr_warn("\nbut task is already holding lock:\n"); 1520 1521 print_lock(check_tgt); 1522 pr_warn("\nwhich lock already depends on the new lock.\n\n"); 1523 pr_warn("\nthe existing dependency chain (in reverse order) is:\n"); 1524 1525 print_circular_bug_entry(entry, depth); 1526 1527 return 0; 1528 } 1529 1530 static inline int class_equal(struct lock_list *entry, void *data) 1531 { 1532 return entry->class == data; 1533 } 1534 1535 static noinline int print_circular_bug(struct lock_list *this, 1536 struct lock_list *target, 1537 struct held_lock *check_src, 1538 struct held_lock *check_tgt, 1539 struct stack_trace *trace) 1540 { 1541 struct task_struct *curr = current; 1542 struct lock_list *parent; 1543 struct lock_list *first_parent; 1544 int depth; 1545 1546 if (!debug_locks_off_graph_unlock() || debug_locks_silent) 1547 return 0; 1548 1549 if (!save_trace(&this->trace)) 1550 return 0; 1551 1552 depth = get_lock_depth(target); 1553 1554 print_circular_bug_header(target, depth, check_src, check_tgt); 1555 1556 parent = get_lock_parent(target); 1557 first_parent = parent; 1558 1559 while (parent) { 1560 print_circular_bug_entry(parent, --depth); 1561 parent = get_lock_parent(parent); 1562 } 1563 1564 printk("\nother info that might help us debug this:\n\n"); 1565 print_circular_lock_scenario(check_src, check_tgt, 1566 first_parent); 1567 1568 lockdep_print_held_locks(curr); 1569 1570 printk("\nstack backtrace:\n"); 1571 dump_stack(); 1572 1573 return 0; 1574 } 1575 1576 static noinline int print_bfs_bug(int ret) 1577 { 1578 if (!debug_locks_off_graph_unlock()) 1579 return 0; 1580 1581 /* 1582 * Breadth-first-search failed, graph got corrupted? 1583 */ 1584 WARN(1, "lockdep bfs error:%d\n", ret); 1585 1586 return 0; 1587 } 1588 1589 static int noop_count(struct lock_list *entry, void *data) 1590 { 1591 (*(unsigned long *)data)++; 1592 return 0; 1593 } 1594 1595 static unsigned long __lockdep_count_forward_deps(struct lock_list *this) 1596 { 1597 unsigned long count = 0; 1598 struct lock_list *uninitialized_var(target_entry); 1599 1600 __bfs_forwards(this, (void *)&count, noop_count, &target_entry); 1601 1602 return count; 1603 } 1604 unsigned long lockdep_count_forward_deps(struct lock_class *class) 1605 { 1606 unsigned long ret, flags; 1607 struct lock_list this; 1608 1609 this.parent = NULL; 1610 this.class = class; 1611 1612 raw_local_irq_save(flags); 1613 arch_spin_lock(&lockdep_lock); 1614 ret = __lockdep_count_forward_deps(&this); 1615 arch_spin_unlock(&lockdep_lock); 1616 raw_local_irq_restore(flags); 1617 1618 return ret; 1619 } 1620 1621 static unsigned long __lockdep_count_backward_deps(struct lock_list *this) 1622 { 1623 unsigned long count = 0; 1624 struct lock_list *uninitialized_var(target_entry); 1625 1626 __bfs_backwards(this, (void *)&count, noop_count, &target_entry); 1627 1628 return count; 1629 } 1630 1631 unsigned long lockdep_count_backward_deps(struct lock_class *class) 1632 { 1633 unsigned long ret, flags; 1634 struct lock_list this; 1635 1636 this.parent = NULL; 1637 this.class = class; 1638 1639 raw_local_irq_save(flags); 1640 arch_spin_lock(&lockdep_lock); 1641 ret = __lockdep_count_backward_deps(&this); 1642 arch_spin_unlock(&lockdep_lock); 1643 raw_local_irq_restore(flags); 1644 1645 return ret; 1646 } 1647 1648 /* 1649 * Prove that the dependency graph starting at <entry> can not 1650 * lead to <target>. Print an error and return 0 if it does. 1651 */ 1652 static noinline int 1653 check_noncircular(struct lock_list *root, struct lock_class *target, 1654 struct lock_list **target_entry) 1655 { 1656 int result; 1657 1658 debug_atomic_inc(nr_cyclic_checks); 1659 1660 result = __bfs_forwards(root, target, class_equal, target_entry); 1661 1662 return result; 1663 } 1664 1665 static noinline int 1666 check_redundant(struct lock_list *root, struct lock_class *target, 1667 struct lock_list **target_entry) 1668 { 1669 int result; 1670 1671 debug_atomic_inc(nr_redundant_checks); 1672 1673 result = __bfs_forwards(root, target, class_equal, target_entry); 1674 1675 return result; 1676 } 1677 1678 #if defined(CONFIG_TRACE_IRQFLAGS) && defined(CONFIG_PROVE_LOCKING) 1679 /* 1680 * Forwards and backwards subgraph searching, for the purposes of 1681 * proving that two subgraphs can be connected by a new dependency 1682 * without creating any illegal irq-safe -> irq-unsafe lock dependency. 1683 */ 1684 1685 static inline int usage_match(struct lock_list *entry, void *bit) 1686 { 1687 return entry->class->usage_mask & (1 << (enum lock_usage_bit)bit); 1688 } 1689 1690 1691 1692 /* 1693 * Find a node in the forwards-direction dependency sub-graph starting 1694 * at @root->class that matches @bit. 1695 * 1696 * Return 0 if such a node exists in the subgraph, and put that node 1697 * into *@target_entry. 1698 * 1699 * Return 1 otherwise and keep *@target_entry unchanged. 1700 * Return <0 on error. 1701 */ 1702 static int 1703 find_usage_forwards(struct lock_list *root, enum lock_usage_bit bit, 1704 struct lock_list **target_entry) 1705 { 1706 int result; 1707 1708 debug_atomic_inc(nr_find_usage_forwards_checks); 1709 1710 result = __bfs_forwards(root, (void *)bit, usage_match, target_entry); 1711 1712 return result; 1713 } 1714 1715 /* 1716 * Find a node in the backwards-direction dependency sub-graph starting 1717 * at @root->class that matches @bit. 1718 * 1719 * Return 0 if such a node exists in the subgraph, and put that node 1720 * into *@target_entry. 1721 * 1722 * Return 1 otherwise and keep *@target_entry unchanged. 1723 * Return <0 on error. 1724 */ 1725 static int 1726 find_usage_backwards(struct lock_list *root, enum lock_usage_bit bit, 1727 struct lock_list **target_entry) 1728 { 1729 int result; 1730 1731 debug_atomic_inc(nr_find_usage_backwards_checks); 1732 1733 result = __bfs_backwards(root, (void *)bit, usage_match, target_entry); 1734 1735 return result; 1736 } 1737 1738 static void print_lock_class_header(struct lock_class *class, int depth) 1739 { 1740 int bit; 1741 1742 printk("%*s->", depth, ""); 1743 print_lock_name(class); 1744 #ifdef CONFIG_DEBUG_LOCKDEP 1745 printk(KERN_CONT " ops: %lu", debug_class_ops_read(class)); 1746 #endif 1747 printk(KERN_CONT " {\n"); 1748 1749 for (bit = 0; bit < LOCK_USAGE_STATES; bit++) { 1750 if (class->usage_mask & (1 << bit)) { 1751 int len = depth; 1752 1753 len += printk("%*s %s", depth, "", usage_str[bit]); 1754 len += printk(KERN_CONT " at:\n"); 1755 print_stack_trace(class->usage_traces + bit, len); 1756 } 1757 } 1758 printk("%*s }\n", depth, ""); 1759 1760 printk("%*s ... key at: [<%px>] %pS\n", 1761 depth, "", class->key, class->key); 1762 } 1763 1764 /* 1765 * printk the shortest lock dependencies from @start to @end in reverse order: 1766 */ 1767 static void __used 1768 print_shortest_lock_dependencies(struct lock_list *leaf, 1769 struct lock_list *root) 1770 { 1771 struct lock_list *entry = leaf; 1772 int depth; 1773 1774 /*compute depth from generated tree by BFS*/ 1775 depth = get_lock_depth(leaf); 1776 1777 do { 1778 print_lock_class_header(entry->class, depth); 1779 printk("%*s ... acquired at:\n", depth, ""); 1780 print_stack_trace(&entry->trace, 2); 1781 printk("\n"); 1782 1783 if (depth == 0 && (entry != root)) { 1784 printk("lockdep:%s bad path found in chain graph\n", __func__); 1785 break; 1786 } 1787 1788 entry = get_lock_parent(entry); 1789 depth--; 1790 } while (entry && (depth >= 0)); 1791 1792 return; 1793 } 1794 1795 static void 1796 print_irq_lock_scenario(struct lock_list *safe_entry, 1797 struct lock_list *unsafe_entry, 1798 struct lock_class *prev_class, 1799 struct lock_class *next_class) 1800 { 1801 struct lock_class *safe_class = safe_entry->class; 1802 struct lock_class *unsafe_class = unsafe_entry->class; 1803 struct lock_class *middle_class = prev_class; 1804 1805 if (middle_class == safe_class) 1806 middle_class = next_class; 1807 1808 /* 1809 * A direct locking problem where unsafe_class lock is taken 1810 * directly by safe_class lock, then all we need to show 1811 * is the deadlock scenario, as it is obvious that the 1812 * unsafe lock is taken under the safe lock. 1813 * 1814 * But if there is a chain instead, where the safe lock takes 1815 * an intermediate lock (middle_class) where this lock is 1816 * not the same as the safe lock, then the lock chain is 1817 * used to describe the problem. Otherwise we would need 1818 * to show a different CPU case for each link in the chain 1819 * from the safe_class lock to the unsafe_class lock. 1820 */ 1821 if (middle_class != unsafe_class) { 1822 printk("Chain exists of:\n "); 1823 __print_lock_name(safe_class); 1824 printk(KERN_CONT " --> "); 1825 __print_lock_name(middle_class); 1826 printk(KERN_CONT " --> "); 1827 __print_lock_name(unsafe_class); 1828 printk(KERN_CONT "\n\n"); 1829 } 1830 1831 printk(" Possible interrupt unsafe locking scenario:\n\n"); 1832 printk(" CPU0 CPU1\n"); 1833 printk(" ---- ----\n"); 1834 printk(" lock("); 1835 __print_lock_name(unsafe_class); 1836 printk(KERN_CONT ");\n"); 1837 printk(" local_irq_disable();\n"); 1838 printk(" lock("); 1839 __print_lock_name(safe_class); 1840 printk(KERN_CONT ");\n"); 1841 printk(" lock("); 1842 __print_lock_name(middle_class); 1843 printk(KERN_CONT ");\n"); 1844 printk(" <Interrupt>\n"); 1845 printk(" lock("); 1846 __print_lock_name(safe_class); 1847 printk(KERN_CONT ");\n"); 1848 printk("\n *** DEADLOCK ***\n\n"); 1849 } 1850 1851 static int 1852 print_bad_irq_dependency(struct task_struct *curr, 1853 struct lock_list *prev_root, 1854 struct lock_list *next_root, 1855 struct lock_list *backwards_entry, 1856 struct lock_list *forwards_entry, 1857 struct held_lock *prev, 1858 struct held_lock *next, 1859 enum lock_usage_bit bit1, 1860 enum lock_usage_bit bit2, 1861 const char *irqclass) 1862 { 1863 if (!debug_locks_off_graph_unlock() || debug_locks_silent) 1864 return 0; 1865 1866 pr_warn("\n"); 1867 pr_warn("=====================================================\n"); 1868 pr_warn("WARNING: %s-safe -> %s-unsafe lock order detected\n", 1869 irqclass, irqclass); 1870 print_kernel_ident(); 1871 pr_warn("-----------------------------------------------------\n"); 1872 pr_warn("%s/%d [HC%u[%lu]:SC%u[%lu]:HE%u:SE%u] is trying to acquire:\n", 1873 curr->comm, task_pid_nr(curr), 1874 curr->hardirq_context, hardirq_count() >> HARDIRQ_SHIFT, 1875 curr->softirq_context, softirq_count() >> SOFTIRQ_SHIFT, 1876 curr->hardirqs_enabled, 1877 curr->softirqs_enabled); 1878 print_lock(next); 1879 1880 pr_warn("\nand this task is already holding:\n"); 1881 print_lock(prev); 1882 pr_warn("which would create a new lock dependency:\n"); 1883 print_lock_name(hlock_class(prev)); 1884 pr_cont(" ->"); 1885 print_lock_name(hlock_class(next)); 1886 pr_cont("\n"); 1887 1888 pr_warn("\nbut this new dependency connects a %s-irq-safe lock:\n", 1889 irqclass); 1890 print_lock_name(backwards_entry->class); 1891 pr_warn("\n... which became %s-irq-safe at:\n", irqclass); 1892 1893 print_stack_trace(backwards_entry->class->usage_traces + bit1, 1); 1894 1895 pr_warn("\nto a %s-irq-unsafe lock:\n", irqclass); 1896 print_lock_name(forwards_entry->class); 1897 pr_warn("\n... which became %s-irq-unsafe at:\n", irqclass); 1898 pr_warn("..."); 1899 1900 print_stack_trace(forwards_entry->class->usage_traces + bit2, 1); 1901 1902 pr_warn("\nother info that might help us debug this:\n\n"); 1903 print_irq_lock_scenario(backwards_entry, forwards_entry, 1904 hlock_class(prev), hlock_class(next)); 1905 1906 lockdep_print_held_locks(curr); 1907 1908 pr_warn("\nthe dependencies between %s-irq-safe lock and the holding lock:\n", irqclass); 1909 if (!save_trace(&prev_root->trace)) 1910 return 0; 1911 print_shortest_lock_dependencies(backwards_entry, prev_root); 1912 1913 pr_warn("\nthe dependencies between the lock to be acquired"); 1914 pr_warn(" and %s-irq-unsafe lock:\n", irqclass); 1915 if (!save_trace(&next_root->trace)) 1916 return 0; 1917 print_shortest_lock_dependencies(forwards_entry, next_root); 1918 1919 pr_warn("\nstack backtrace:\n"); 1920 dump_stack(); 1921 1922 return 0; 1923 } 1924 1925 static int 1926 check_usage(struct task_struct *curr, struct held_lock *prev, 1927 struct held_lock *next, enum lock_usage_bit bit_backwards, 1928 enum lock_usage_bit bit_forwards, const char *irqclass) 1929 { 1930 int ret; 1931 struct lock_list this, that; 1932 struct lock_list *uninitialized_var(target_entry); 1933 struct lock_list *uninitialized_var(target_entry1); 1934 1935 this.parent = NULL; 1936 1937 this.class = hlock_class(prev); 1938 ret = find_usage_backwards(&this, bit_backwards, &target_entry); 1939 if (ret < 0) 1940 return print_bfs_bug(ret); 1941 if (ret == 1) 1942 return ret; 1943 1944 that.parent = NULL; 1945 that.class = hlock_class(next); 1946 ret = find_usage_forwards(&that, bit_forwards, &target_entry1); 1947 if (ret < 0) 1948 return print_bfs_bug(ret); 1949 if (ret == 1) 1950 return ret; 1951 1952 return print_bad_irq_dependency(curr, &this, &that, 1953 target_entry, target_entry1, 1954 prev, next, 1955 bit_backwards, bit_forwards, irqclass); 1956 } 1957 1958 static const char *state_names[] = { 1959 #define LOCKDEP_STATE(__STATE) \ 1960 __stringify(__STATE), 1961 #include "lockdep_states.h" 1962 #undef LOCKDEP_STATE 1963 }; 1964 1965 static const char *state_rnames[] = { 1966 #define LOCKDEP_STATE(__STATE) \ 1967 __stringify(__STATE)"-READ", 1968 #include "lockdep_states.h" 1969 #undef LOCKDEP_STATE 1970 }; 1971 1972 static inline const char *state_name(enum lock_usage_bit bit) 1973 { 1974 return (bit & LOCK_USAGE_READ_MASK) ? state_rnames[bit >> 2] : state_names[bit >> 2]; 1975 } 1976 1977 static int exclusive_bit(int new_bit) 1978 { 1979 int state = new_bit & LOCK_USAGE_STATE_MASK; 1980 int dir = new_bit & LOCK_USAGE_DIR_MASK; 1981 1982 /* 1983 * keep state, bit flip the direction and strip read. 1984 */ 1985 return state | (dir ^ LOCK_USAGE_DIR_MASK); 1986 } 1987 1988 static int check_irq_usage(struct task_struct *curr, struct held_lock *prev, 1989 struct held_lock *next, enum lock_usage_bit bit) 1990 { 1991 /* 1992 * Prove that the new dependency does not connect a hardirq-safe 1993 * lock with a hardirq-unsafe lock - to achieve this we search 1994 * the backwards-subgraph starting at <prev>, and the 1995 * forwards-subgraph starting at <next>: 1996 */ 1997 if (!check_usage(curr, prev, next, bit, 1998 exclusive_bit(bit), state_name(bit))) 1999 return 0; 2000 2001 bit++; /* _READ */ 2002 2003 /* 2004 * Prove that the new dependency does not connect a hardirq-safe-read 2005 * lock with a hardirq-unsafe lock - to achieve this we search 2006 * the backwards-subgraph starting at <prev>, and the 2007 * forwards-subgraph starting at <next>: 2008 */ 2009 if (!check_usage(curr, prev, next, bit, 2010 exclusive_bit(bit), state_name(bit))) 2011 return 0; 2012 2013 return 1; 2014 } 2015 2016 static int 2017 check_prev_add_irq(struct task_struct *curr, struct held_lock *prev, 2018 struct held_lock *next) 2019 { 2020 #define LOCKDEP_STATE(__STATE) \ 2021 if (!check_irq_usage(curr, prev, next, LOCK_USED_IN_##__STATE)) \ 2022 return 0; 2023 #include "lockdep_states.h" 2024 #undef LOCKDEP_STATE 2025 2026 return 1; 2027 } 2028 2029 static void inc_chains(void) 2030 { 2031 if (current->hardirq_context) 2032 nr_hardirq_chains++; 2033 else { 2034 if (current->softirq_context) 2035 nr_softirq_chains++; 2036 else 2037 nr_process_chains++; 2038 } 2039 } 2040 2041 #else 2042 2043 static inline int 2044 check_prev_add_irq(struct task_struct *curr, struct held_lock *prev, 2045 struct held_lock *next) 2046 { 2047 return 1; 2048 } 2049 2050 static inline void inc_chains(void) 2051 { 2052 nr_process_chains++; 2053 } 2054 2055 #endif 2056 2057 static void 2058 print_deadlock_scenario(struct held_lock *nxt, 2059 struct held_lock *prv) 2060 { 2061 struct lock_class *next = hlock_class(nxt); 2062 struct lock_class *prev = hlock_class(prv); 2063 2064 printk(" Possible unsafe locking scenario:\n\n"); 2065 printk(" CPU0\n"); 2066 printk(" ----\n"); 2067 printk(" lock("); 2068 __print_lock_name(prev); 2069 printk(KERN_CONT ");\n"); 2070 printk(" lock("); 2071 __print_lock_name(next); 2072 printk(KERN_CONT ");\n"); 2073 printk("\n *** DEADLOCK ***\n\n"); 2074 printk(" May be due to missing lock nesting notation\n\n"); 2075 } 2076 2077 static int 2078 print_deadlock_bug(struct task_struct *curr, struct held_lock *prev, 2079 struct held_lock *next) 2080 { 2081 if (!debug_locks_off_graph_unlock() || debug_locks_silent) 2082 return 0; 2083 2084 pr_warn("\n"); 2085 pr_warn("============================================\n"); 2086 pr_warn("WARNING: possible recursive locking detected\n"); 2087 print_kernel_ident(); 2088 pr_warn("--------------------------------------------\n"); 2089 pr_warn("%s/%d is trying to acquire lock:\n", 2090 curr->comm, task_pid_nr(curr)); 2091 print_lock(next); 2092 pr_warn("\nbut task is already holding lock:\n"); 2093 print_lock(prev); 2094 2095 pr_warn("\nother info that might help us debug this:\n"); 2096 print_deadlock_scenario(next, prev); 2097 lockdep_print_held_locks(curr); 2098 2099 pr_warn("\nstack backtrace:\n"); 2100 dump_stack(); 2101 2102 return 0; 2103 } 2104 2105 /* 2106 * Check whether we are holding such a class already. 2107 * 2108 * (Note that this has to be done separately, because the graph cannot 2109 * detect such classes of deadlocks.) 2110 * 2111 * Returns: 0 on deadlock detected, 1 on OK, 2 on recursive read 2112 */ 2113 static int 2114 check_deadlock(struct task_struct *curr, struct held_lock *next, 2115 struct lockdep_map *next_instance, int read) 2116 { 2117 struct held_lock *prev; 2118 struct held_lock *nest = NULL; 2119 int i; 2120 2121 for (i = 0; i < curr->lockdep_depth; i++) { 2122 prev = curr->held_locks + i; 2123 2124 if (prev->instance == next->nest_lock) 2125 nest = prev; 2126 2127 if (hlock_class(prev) != hlock_class(next)) 2128 continue; 2129 2130 /* 2131 * Allow read-after-read recursion of the same 2132 * lock class (i.e. read_lock(lock)+read_lock(lock)): 2133 */ 2134 if ((read == 2) && prev->read) 2135 return 2; 2136 2137 /* 2138 * We're holding the nest_lock, which serializes this lock's 2139 * nesting behaviour. 2140 */ 2141 if (nest) 2142 return 2; 2143 2144 return print_deadlock_bug(curr, prev, next); 2145 } 2146 return 1; 2147 } 2148 2149 /* 2150 * There was a chain-cache miss, and we are about to add a new dependency 2151 * to a previous lock. We recursively validate the following rules: 2152 * 2153 * - would the adding of the <prev> -> <next> dependency create a 2154 * circular dependency in the graph? [== circular deadlock] 2155 * 2156 * - does the new prev->next dependency connect any hardirq-safe lock 2157 * (in the full backwards-subgraph starting at <prev>) with any 2158 * hardirq-unsafe lock (in the full forwards-subgraph starting at 2159 * <next>)? [== illegal lock inversion with hardirq contexts] 2160 * 2161 * - does the new prev->next dependency connect any softirq-safe lock 2162 * (in the full backwards-subgraph starting at <prev>) with any 2163 * softirq-unsafe lock (in the full forwards-subgraph starting at 2164 * <next>)? [== illegal lock inversion with softirq contexts] 2165 * 2166 * any of these scenarios could lead to a deadlock. 2167 * 2168 * Then if all the validations pass, we add the forwards and backwards 2169 * dependency. 2170 */ 2171 static int 2172 check_prev_add(struct task_struct *curr, struct held_lock *prev, 2173 struct held_lock *next, int distance, struct stack_trace *trace, 2174 int (*save)(struct stack_trace *trace)) 2175 { 2176 struct lock_list *uninitialized_var(target_entry); 2177 struct lock_list *entry; 2178 struct lock_list this; 2179 int ret; 2180 2181 if (!hlock_class(prev)->key || !hlock_class(next)->key) { 2182 /* 2183 * The warning statements below may trigger a use-after-free 2184 * of the class name. It is better to trigger a use-after free 2185 * and to have the class name most of the time instead of not 2186 * having the class name available. 2187 */ 2188 WARN_ONCE(!debug_locks_silent && !hlock_class(prev)->key, 2189 "Detected use-after-free of lock class %px/%s\n", 2190 hlock_class(prev), 2191 hlock_class(prev)->name); 2192 WARN_ONCE(!debug_locks_silent && !hlock_class(next)->key, 2193 "Detected use-after-free of lock class %px/%s\n", 2194 hlock_class(next), 2195 hlock_class(next)->name); 2196 return 2; 2197 } 2198 2199 /* 2200 * Prove that the new <prev> -> <next> dependency would not 2201 * create a circular dependency in the graph. (We do this by 2202 * forward-recursing into the graph starting at <next>, and 2203 * checking whether we can reach <prev>.) 2204 * 2205 * We are using global variables to control the recursion, to 2206 * keep the stackframe size of the recursive functions low: 2207 */ 2208 this.class = hlock_class(next); 2209 this.parent = NULL; 2210 ret = check_noncircular(&this, hlock_class(prev), &target_entry); 2211 if (unlikely(!ret)) { 2212 if (!trace->entries) { 2213 /* 2214 * If @save fails here, the printing might trigger 2215 * a WARN but because of the !nr_entries it should 2216 * not do bad things. 2217 */ 2218 save(trace); 2219 } 2220 return print_circular_bug(&this, target_entry, next, prev, trace); 2221 } 2222 else if (unlikely(ret < 0)) 2223 return print_bfs_bug(ret); 2224 2225 if (!check_prev_add_irq(curr, prev, next)) 2226 return 0; 2227 2228 /* 2229 * For recursive read-locks we do all the dependency checks, 2230 * but we dont store read-triggered dependencies (only 2231 * write-triggered dependencies). This ensures that only the 2232 * write-side dependencies matter, and that if for example a 2233 * write-lock never takes any other locks, then the reads are 2234 * equivalent to a NOP. 2235 */ 2236 if (next->read == 2 || prev->read == 2) 2237 return 1; 2238 /* 2239 * Is the <prev> -> <next> dependency already present? 2240 * 2241 * (this may occur even though this is a new chain: consider 2242 * e.g. the L1 -> L2 -> L3 -> L4 and the L5 -> L1 -> L2 -> L3 2243 * chains - the second one will be new, but L1 already has 2244 * L2 added to its dependency list, due to the first chain.) 2245 */ 2246 list_for_each_entry(entry, &hlock_class(prev)->locks_after, entry) { 2247 if (entry->class == hlock_class(next)) { 2248 if (distance == 1) 2249 entry->distance = 1; 2250 return 1; 2251 } 2252 } 2253 2254 /* 2255 * Is the <prev> -> <next> link redundant? 2256 */ 2257 this.class = hlock_class(prev); 2258 this.parent = NULL; 2259 ret = check_redundant(&this, hlock_class(next), &target_entry); 2260 if (!ret) { 2261 debug_atomic_inc(nr_redundant); 2262 return 2; 2263 } 2264 if (ret < 0) 2265 return print_bfs_bug(ret); 2266 2267 2268 if (!trace->entries && !save(trace)) 2269 return 0; 2270 2271 /* 2272 * Ok, all validations passed, add the new lock 2273 * to the previous lock's dependency list: 2274 */ 2275 ret = add_lock_to_list(hlock_class(next), hlock_class(prev), 2276 &hlock_class(prev)->locks_after, 2277 next->acquire_ip, distance, trace); 2278 2279 if (!ret) 2280 return 0; 2281 2282 ret = add_lock_to_list(hlock_class(prev), hlock_class(next), 2283 &hlock_class(next)->locks_before, 2284 next->acquire_ip, distance, trace); 2285 if (!ret) 2286 return 0; 2287 2288 return 2; 2289 } 2290 2291 /* 2292 * Add the dependency to all directly-previous locks that are 'relevant'. 2293 * The ones that are relevant are (in increasing distance from curr): 2294 * all consecutive trylock entries and the final non-trylock entry - or 2295 * the end of this context's lock-chain - whichever comes first. 2296 */ 2297 static int 2298 check_prevs_add(struct task_struct *curr, struct held_lock *next) 2299 { 2300 int depth = curr->lockdep_depth; 2301 struct held_lock *hlock; 2302 struct stack_trace trace = { 2303 .nr_entries = 0, 2304 .max_entries = 0, 2305 .entries = NULL, 2306 .skip = 0, 2307 }; 2308 2309 /* 2310 * Debugging checks. 2311 * 2312 * Depth must not be zero for a non-head lock: 2313 */ 2314 if (!depth) 2315 goto out_bug; 2316 /* 2317 * At least two relevant locks must exist for this 2318 * to be a head: 2319 */ 2320 if (curr->held_locks[depth].irq_context != 2321 curr->held_locks[depth-1].irq_context) 2322 goto out_bug; 2323 2324 for (;;) { 2325 int distance = curr->lockdep_depth - depth + 1; 2326 hlock = curr->held_locks + depth - 1; 2327 2328 /* 2329 * Only non-recursive-read entries get new dependencies 2330 * added: 2331 */ 2332 if (hlock->read != 2 && hlock->check) { 2333 int ret = check_prev_add(curr, hlock, next, distance, &trace, save_trace); 2334 if (!ret) 2335 return 0; 2336 2337 /* 2338 * Stop after the first non-trylock entry, 2339 * as non-trylock entries have added their 2340 * own direct dependencies already, so this 2341 * lock is connected to them indirectly: 2342 */ 2343 if (!hlock->trylock) 2344 break; 2345 } 2346 2347 depth--; 2348 /* 2349 * End of lock-stack? 2350 */ 2351 if (!depth) 2352 break; 2353 /* 2354 * Stop the search if we cross into another context: 2355 */ 2356 if (curr->held_locks[depth].irq_context != 2357 curr->held_locks[depth-1].irq_context) 2358 break; 2359 } 2360 return 1; 2361 out_bug: 2362 if (!debug_locks_off_graph_unlock()) 2363 return 0; 2364 2365 /* 2366 * Clearly we all shouldn't be here, but since we made it we 2367 * can reliable say we messed up our state. See the above two 2368 * gotos for reasons why we could possibly end up here. 2369 */ 2370 WARN_ON(1); 2371 2372 return 0; 2373 } 2374 2375 struct lock_chain lock_chains[MAX_LOCKDEP_CHAINS]; 2376 static DECLARE_BITMAP(lock_chains_in_use, MAX_LOCKDEP_CHAINS); 2377 int nr_chain_hlocks; 2378 static u16 chain_hlocks[MAX_LOCKDEP_CHAIN_HLOCKS]; 2379 2380 struct lock_class *lock_chain_get_class(struct lock_chain *chain, int i) 2381 { 2382 return lock_classes + chain_hlocks[chain->base + i]; 2383 } 2384 2385 /* 2386 * Returns the index of the first held_lock of the current chain 2387 */ 2388 static inline int get_first_held_lock(struct task_struct *curr, 2389 struct held_lock *hlock) 2390 { 2391 int i; 2392 struct held_lock *hlock_curr; 2393 2394 for (i = curr->lockdep_depth - 1; i >= 0; i--) { 2395 hlock_curr = curr->held_locks + i; 2396 if (hlock_curr->irq_context != hlock->irq_context) 2397 break; 2398 2399 } 2400 2401 return ++i; 2402 } 2403 2404 #ifdef CONFIG_DEBUG_LOCKDEP 2405 /* 2406 * Returns the next chain_key iteration 2407 */ 2408 static u64 print_chain_key_iteration(int class_idx, u64 chain_key) 2409 { 2410 u64 new_chain_key = iterate_chain_key(chain_key, class_idx); 2411 2412 printk(" class_idx:%d -> chain_key:%016Lx", 2413 class_idx, 2414 (unsigned long long)new_chain_key); 2415 return new_chain_key; 2416 } 2417 2418 static void 2419 print_chain_keys_held_locks(struct task_struct *curr, struct held_lock *hlock_next) 2420 { 2421 struct held_lock *hlock; 2422 u64 chain_key = 0; 2423 int depth = curr->lockdep_depth; 2424 int i; 2425 2426 printk("depth: %u\n", depth + 1); 2427 for (i = get_first_held_lock(curr, hlock_next); i < depth; i++) { 2428 hlock = curr->held_locks + i; 2429 chain_key = print_chain_key_iteration(hlock->class_idx, chain_key); 2430 2431 print_lock(hlock); 2432 } 2433 2434 print_chain_key_iteration(hlock_next->class_idx, chain_key); 2435 print_lock(hlock_next); 2436 } 2437 2438 static void print_chain_keys_chain(struct lock_chain *chain) 2439 { 2440 int i; 2441 u64 chain_key = 0; 2442 int class_id; 2443 2444 printk("depth: %u\n", chain->depth); 2445 for (i = 0; i < chain->depth; i++) { 2446 class_id = chain_hlocks[chain->base + i]; 2447 chain_key = print_chain_key_iteration(class_id + 1, chain_key); 2448 2449 print_lock_name(lock_classes + class_id); 2450 printk("\n"); 2451 } 2452 } 2453 2454 static void print_collision(struct task_struct *curr, 2455 struct held_lock *hlock_next, 2456 struct lock_chain *chain) 2457 { 2458 pr_warn("\n"); 2459 pr_warn("============================\n"); 2460 pr_warn("WARNING: chain_key collision\n"); 2461 print_kernel_ident(); 2462 pr_warn("----------------------------\n"); 2463 pr_warn("%s/%d: ", current->comm, task_pid_nr(current)); 2464 pr_warn("Hash chain already cached but the contents don't match!\n"); 2465 2466 pr_warn("Held locks:"); 2467 print_chain_keys_held_locks(curr, hlock_next); 2468 2469 pr_warn("Locks in cached chain:"); 2470 print_chain_keys_chain(chain); 2471 2472 pr_warn("\nstack backtrace:\n"); 2473 dump_stack(); 2474 } 2475 #endif 2476 2477 /* 2478 * Checks whether the chain and the current held locks are consistent 2479 * in depth and also in content. If they are not it most likely means 2480 * that there was a collision during the calculation of the chain_key. 2481 * Returns: 0 not passed, 1 passed 2482 */ 2483 static int check_no_collision(struct task_struct *curr, 2484 struct held_lock *hlock, 2485 struct lock_chain *chain) 2486 { 2487 #ifdef CONFIG_DEBUG_LOCKDEP 2488 int i, j, id; 2489 2490 i = get_first_held_lock(curr, hlock); 2491 2492 if (DEBUG_LOCKS_WARN_ON(chain->depth != curr->lockdep_depth - (i - 1))) { 2493 print_collision(curr, hlock, chain); 2494 return 0; 2495 } 2496 2497 for (j = 0; j < chain->depth - 1; j++, i++) { 2498 id = curr->held_locks[i].class_idx - 1; 2499 2500 if (DEBUG_LOCKS_WARN_ON(chain_hlocks[chain->base + j] != id)) { 2501 print_collision(curr, hlock, chain); 2502 return 0; 2503 } 2504 } 2505 #endif 2506 return 1; 2507 } 2508 2509 /* 2510 * Given an index that is >= -1, return the index of the next lock chain. 2511 * Return -2 if there is no next lock chain. 2512 */ 2513 long lockdep_next_lockchain(long i) 2514 { 2515 i = find_next_bit(lock_chains_in_use, ARRAY_SIZE(lock_chains), i + 1); 2516 return i < ARRAY_SIZE(lock_chains) ? i : -2; 2517 } 2518 2519 unsigned long lock_chain_count(void) 2520 { 2521 return bitmap_weight(lock_chains_in_use, ARRAY_SIZE(lock_chains)); 2522 } 2523 2524 /* Must be called with the graph lock held. */ 2525 static struct lock_chain *alloc_lock_chain(void) 2526 { 2527 int idx = find_first_zero_bit(lock_chains_in_use, 2528 ARRAY_SIZE(lock_chains)); 2529 2530 if (unlikely(idx >= ARRAY_SIZE(lock_chains))) 2531 return NULL; 2532 __set_bit(idx, lock_chains_in_use); 2533 return lock_chains + idx; 2534 } 2535 2536 /* 2537 * Adds a dependency chain into chain hashtable. And must be called with 2538 * graph_lock held. 2539 * 2540 * Return 0 if fail, and graph_lock is released. 2541 * Return 1 if succeed, with graph_lock held. 2542 */ 2543 static inline int add_chain_cache(struct task_struct *curr, 2544 struct held_lock *hlock, 2545 u64 chain_key) 2546 { 2547 struct lock_class *class = hlock_class(hlock); 2548 struct hlist_head *hash_head = chainhashentry(chain_key); 2549 struct lock_chain *chain; 2550 int i, j; 2551 2552 /* 2553 * The caller must hold the graph lock, ensure we've got IRQs 2554 * disabled to make this an IRQ-safe lock.. for recursion reasons 2555 * lockdep won't complain about its own locking errors. 2556 */ 2557 if (DEBUG_LOCKS_WARN_ON(!irqs_disabled())) 2558 return 0; 2559 2560 chain = alloc_lock_chain(); 2561 if (!chain) { 2562 if (!debug_locks_off_graph_unlock()) 2563 return 0; 2564 2565 print_lockdep_off("BUG: MAX_LOCKDEP_CHAINS too low!"); 2566 dump_stack(); 2567 return 0; 2568 } 2569 chain->chain_key = chain_key; 2570 chain->irq_context = hlock->irq_context; 2571 i = get_first_held_lock(curr, hlock); 2572 chain->depth = curr->lockdep_depth + 1 - i; 2573 2574 BUILD_BUG_ON((1UL << 24) <= ARRAY_SIZE(chain_hlocks)); 2575 BUILD_BUG_ON((1UL << 6) <= ARRAY_SIZE(curr->held_locks)); 2576 BUILD_BUG_ON((1UL << 8*sizeof(chain_hlocks[0])) <= ARRAY_SIZE(lock_classes)); 2577 2578 if (likely(nr_chain_hlocks + chain->depth <= MAX_LOCKDEP_CHAIN_HLOCKS)) { 2579 chain->base = nr_chain_hlocks; 2580 for (j = 0; j < chain->depth - 1; j++, i++) { 2581 int lock_id = curr->held_locks[i].class_idx - 1; 2582 chain_hlocks[chain->base + j] = lock_id; 2583 } 2584 chain_hlocks[chain->base + j] = class - lock_classes; 2585 nr_chain_hlocks += chain->depth; 2586 } else { 2587 if (!debug_locks_off_graph_unlock()) 2588 return 0; 2589 2590 print_lockdep_off("BUG: MAX_LOCKDEP_CHAIN_HLOCKS too low!"); 2591 dump_stack(); 2592 return 0; 2593 } 2594 2595 hlist_add_head_rcu(&chain->entry, hash_head); 2596 debug_atomic_inc(chain_lookup_misses); 2597 inc_chains(); 2598 2599 return 1; 2600 } 2601 2602 /* 2603 * Look up a dependency chain. Must be called with either the graph lock or 2604 * the RCU read lock held. 2605 */ 2606 static inline struct lock_chain *lookup_chain_cache(u64 chain_key) 2607 { 2608 struct hlist_head *hash_head = chainhashentry(chain_key); 2609 struct lock_chain *chain; 2610 2611 hlist_for_each_entry_rcu(chain, hash_head, entry) { 2612 if (READ_ONCE(chain->chain_key) == chain_key) { 2613 debug_atomic_inc(chain_lookup_hits); 2614 return chain; 2615 } 2616 } 2617 return NULL; 2618 } 2619 2620 /* 2621 * If the key is not present yet in dependency chain cache then 2622 * add it and return 1 - in this case the new dependency chain is 2623 * validated. If the key is already hashed, return 0. 2624 * (On return with 1 graph_lock is held.) 2625 */ 2626 static inline int lookup_chain_cache_add(struct task_struct *curr, 2627 struct held_lock *hlock, 2628 u64 chain_key) 2629 { 2630 struct lock_class *class = hlock_class(hlock); 2631 struct lock_chain *chain = lookup_chain_cache(chain_key); 2632 2633 if (chain) { 2634 cache_hit: 2635 if (!check_no_collision(curr, hlock, chain)) 2636 return 0; 2637 2638 if (very_verbose(class)) { 2639 printk("\nhash chain already cached, key: " 2640 "%016Lx tail class: [%px] %s\n", 2641 (unsigned long long)chain_key, 2642 class->key, class->name); 2643 } 2644 2645 return 0; 2646 } 2647 2648 if (very_verbose(class)) { 2649 printk("\nnew hash chain, key: %016Lx tail class: [%px] %s\n", 2650 (unsigned long long)chain_key, class->key, class->name); 2651 } 2652 2653 if (!graph_lock()) 2654 return 0; 2655 2656 /* 2657 * We have to walk the chain again locked - to avoid duplicates: 2658 */ 2659 chain = lookup_chain_cache(chain_key); 2660 if (chain) { 2661 graph_unlock(); 2662 goto cache_hit; 2663 } 2664 2665 if (!add_chain_cache(curr, hlock, chain_key)) 2666 return 0; 2667 2668 return 1; 2669 } 2670 2671 static int validate_chain(struct task_struct *curr, struct lockdep_map *lock, 2672 struct held_lock *hlock, int chain_head, u64 chain_key) 2673 { 2674 /* 2675 * Trylock needs to maintain the stack of held locks, but it 2676 * does not add new dependencies, because trylock can be done 2677 * in any order. 2678 * 2679 * We look up the chain_key and do the O(N^2) check and update of 2680 * the dependencies only if this is a new dependency chain. 2681 * (If lookup_chain_cache_add() return with 1 it acquires 2682 * graph_lock for us) 2683 */ 2684 if (!hlock->trylock && hlock->check && 2685 lookup_chain_cache_add(curr, hlock, chain_key)) { 2686 /* 2687 * Check whether last held lock: 2688 * 2689 * - is irq-safe, if this lock is irq-unsafe 2690 * - is softirq-safe, if this lock is hardirq-unsafe 2691 * 2692 * And check whether the new lock's dependency graph 2693 * could lead back to the previous lock. 2694 * 2695 * any of these scenarios could lead to a deadlock. If 2696 * All validations 2697 */ 2698 int ret = check_deadlock(curr, hlock, lock, hlock->read); 2699 2700 if (!ret) 2701 return 0; 2702 /* 2703 * Mark recursive read, as we jump over it when 2704 * building dependencies (just like we jump over 2705 * trylock entries): 2706 */ 2707 if (ret == 2) 2708 hlock->read = 2; 2709 /* 2710 * Add dependency only if this lock is not the head 2711 * of the chain, and if it's not a secondary read-lock: 2712 */ 2713 if (!chain_head && ret != 2) { 2714 if (!check_prevs_add(curr, hlock)) 2715 return 0; 2716 } 2717 2718 graph_unlock(); 2719 } else { 2720 /* after lookup_chain_cache_add(): */ 2721 if (unlikely(!debug_locks)) 2722 return 0; 2723 } 2724 2725 return 1; 2726 } 2727 #else 2728 static inline int validate_chain(struct task_struct *curr, 2729 struct lockdep_map *lock, struct held_lock *hlock, 2730 int chain_head, u64 chain_key) 2731 { 2732 return 1; 2733 } 2734 #endif 2735 2736 /* 2737 * We are building curr_chain_key incrementally, so double-check 2738 * it from scratch, to make sure that it's done correctly: 2739 */ 2740 static void check_chain_key(struct task_struct *curr) 2741 { 2742 #ifdef CONFIG_DEBUG_LOCKDEP 2743 struct held_lock *hlock, *prev_hlock = NULL; 2744 unsigned int i; 2745 u64 chain_key = 0; 2746 2747 for (i = 0; i < curr->lockdep_depth; i++) { 2748 hlock = curr->held_locks + i; 2749 if (chain_key != hlock->prev_chain_key) { 2750 debug_locks_off(); 2751 /* 2752 * We got mighty confused, our chain keys don't match 2753 * with what we expect, someone trample on our task state? 2754 */ 2755 WARN(1, "hm#1, depth: %u [%u], %016Lx != %016Lx\n", 2756 curr->lockdep_depth, i, 2757 (unsigned long long)chain_key, 2758 (unsigned long long)hlock->prev_chain_key); 2759 return; 2760 } 2761 /* 2762 * Whoops ran out of static storage again? 2763 */ 2764 if (DEBUG_LOCKS_WARN_ON(hlock->class_idx > MAX_LOCKDEP_KEYS)) 2765 return; 2766 2767 if (prev_hlock && (prev_hlock->irq_context != 2768 hlock->irq_context)) 2769 chain_key = 0; 2770 chain_key = iterate_chain_key(chain_key, hlock->class_idx); 2771 prev_hlock = hlock; 2772 } 2773 if (chain_key != curr->curr_chain_key) { 2774 debug_locks_off(); 2775 /* 2776 * More smoking hash instead of calculating it, damn see these 2777 * numbers float.. I bet that a pink elephant stepped on my memory. 2778 */ 2779 WARN(1, "hm#2, depth: %u [%u], %016Lx != %016Lx\n", 2780 curr->lockdep_depth, i, 2781 (unsigned long long)chain_key, 2782 (unsigned long long)curr->curr_chain_key); 2783 } 2784 #endif 2785 } 2786 2787 static void 2788 print_usage_bug_scenario(struct held_lock *lock) 2789 { 2790 struct lock_class *class = hlock_class(lock); 2791 2792 printk(" Possible unsafe locking scenario:\n\n"); 2793 printk(" CPU0\n"); 2794 printk(" ----\n"); 2795 printk(" lock("); 2796 __print_lock_name(class); 2797 printk(KERN_CONT ");\n"); 2798 printk(" <Interrupt>\n"); 2799 printk(" lock("); 2800 __print_lock_name(class); 2801 printk(KERN_CONT ");\n"); 2802 printk("\n *** DEADLOCK ***\n\n"); 2803 } 2804 2805 static int 2806 print_usage_bug(struct task_struct *curr, struct held_lock *this, 2807 enum lock_usage_bit prev_bit, enum lock_usage_bit new_bit) 2808 { 2809 if (!debug_locks_off_graph_unlock() || debug_locks_silent) 2810 return 0; 2811 2812 pr_warn("\n"); 2813 pr_warn("================================\n"); 2814 pr_warn("WARNING: inconsistent lock state\n"); 2815 print_kernel_ident(); 2816 pr_warn("--------------------------------\n"); 2817 2818 pr_warn("inconsistent {%s} -> {%s} usage.\n", 2819 usage_str[prev_bit], usage_str[new_bit]); 2820 2821 pr_warn("%s/%d [HC%u[%lu]:SC%u[%lu]:HE%u:SE%u] takes:\n", 2822 curr->comm, task_pid_nr(curr), 2823 trace_hardirq_context(curr), hardirq_count() >> HARDIRQ_SHIFT, 2824 trace_softirq_context(curr), softirq_count() >> SOFTIRQ_SHIFT, 2825 trace_hardirqs_enabled(curr), 2826 trace_softirqs_enabled(curr)); 2827 print_lock(this); 2828 2829 pr_warn("{%s} state was registered at:\n", usage_str[prev_bit]); 2830 print_stack_trace(hlock_class(this)->usage_traces + prev_bit, 1); 2831 2832 print_irqtrace_events(curr); 2833 pr_warn("\nother info that might help us debug this:\n"); 2834 print_usage_bug_scenario(this); 2835 2836 lockdep_print_held_locks(curr); 2837 2838 pr_warn("\nstack backtrace:\n"); 2839 dump_stack(); 2840 2841 return 0; 2842 } 2843 2844 /* 2845 * Print out an error if an invalid bit is set: 2846 */ 2847 static inline int 2848 valid_state(struct task_struct *curr, struct held_lock *this, 2849 enum lock_usage_bit new_bit, enum lock_usage_bit bad_bit) 2850 { 2851 if (unlikely(hlock_class(this)->usage_mask & (1 << bad_bit))) 2852 return print_usage_bug(curr, this, bad_bit, new_bit); 2853 return 1; 2854 } 2855 2856 static int mark_lock(struct task_struct *curr, struct held_lock *this, 2857 enum lock_usage_bit new_bit); 2858 2859 #if defined(CONFIG_TRACE_IRQFLAGS) && defined(CONFIG_PROVE_LOCKING) 2860 2861 /* 2862 * print irq inversion bug: 2863 */ 2864 static int 2865 print_irq_inversion_bug(struct task_struct *curr, 2866 struct lock_list *root, struct lock_list *other, 2867 struct held_lock *this, int forwards, 2868 const char *irqclass) 2869 { 2870 struct lock_list *entry = other; 2871 struct lock_list *middle = NULL; 2872 int depth; 2873 2874 if (!debug_locks_off_graph_unlock() || debug_locks_silent) 2875 return 0; 2876 2877 pr_warn("\n"); 2878 pr_warn("========================================================\n"); 2879 pr_warn("WARNING: possible irq lock inversion dependency detected\n"); 2880 print_kernel_ident(); 2881 pr_warn("--------------------------------------------------------\n"); 2882 pr_warn("%s/%d just changed the state of lock:\n", 2883 curr->comm, task_pid_nr(curr)); 2884 print_lock(this); 2885 if (forwards) 2886 pr_warn("but this lock took another, %s-unsafe lock in the past:\n", irqclass); 2887 else 2888 pr_warn("but this lock was taken by another, %s-safe lock in the past:\n", irqclass); 2889 print_lock_name(other->class); 2890 pr_warn("\n\nand interrupts could create inverse lock ordering between them.\n\n"); 2891 2892 pr_warn("\nother info that might help us debug this:\n"); 2893 2894 /* Find a middle lock (if one exists) */ 2895 depth = get_lock_depth(other); 2896 do { 2897 if (depth == 0 && (entry != root)) { 2898 pr_warn("lockdep:%s bad path found in chain graph\n", __func__); 2899 break; 2900 } 2901 middle = entry; 2902 entry = get_lock_parent(entry); 2903 depth--; 2904 } while (entry && entry != root && (depth >= 0)); 2905 if (forwards) 2906 print_irq_lock_scenario(root, other, 2907 middle ? middle->class : root->class, other->class); 2908 else 2909 print_irq_lock_scenario(other, root, 2910 middle ? middle->class : other->class, root->class); 2911 2912 lockdep_print_held_locks(curr); 2913 2914 pr_warn("\nthe shortest dependencies between 2nd lock and 1st lock:\n"); 2915 if (!save_trace(&root->trace)) 2916 return 0; 2917 print_shortest_lock_dependencies(other, root); 2918 2919 pr_warn("\nstack backtrace:\n"); 2920 dump_stack(); 2921 2922 return 0; 2923 } 2924 2925 /* 2926 * Prove that in the forwards-direction subgraph starting at <this> 2927 * there is no lock matching <mask>: 2928 */ 2929 static int 2930 check_usage_forwards(struct task_struct *curr, struct held_lock *this, 2931 enum lock_usage_bit bit, const char *irqclass) 2932 { 2933 int ret; 2934 struct lock_list root; 2935 struct lock_list *uninitialized_var(target_entry); 2936 2937 root.parent = NULL; 2938 root.class = hlock_class(this); 2939 ret = find_usage_forwards(&root, bit, &target_entry); 2940 if (ret < 0) 2941 return print_bfs_bug(ret); 2942 if (ret == 1) 2943 return ret; 2944 2945 return print_irq_inversion_bug(curr, &root, target_entry, 2946 this, 1, irqclass); 2947 } 2948 2949 /* 2950 * Prove that in the backwards-direction subgraph starting at <this> 2951 * there is no lock matching <mask>: 2952 */ 2953 static int 2954 check_usage_backwards(struct task_struct *curr, struct held_lock *this, 2955 enum lock_usage_bit bit, const char *irqclass) 2956 { 2957 int ret; 2958 struct lock_list root; 2959 struct lock_list *uninitialized_var(target_entry); 2960 2961 root.parent = NULL; 2962 root.class = hlock_class(this); 2963 ret = find_usage_backwards(&root, bit, &target_entry); 2964 if (ret < 0) 2965 return print_bfs_bug(ret); 2966 if (ret == 1) 2967 return ret; 2968 2969 return print_irq_inversion_bug(curr, &root, target_entry, 2970 this, 0, irqclass); 2971 } 2972 2973 void print_irqtrace_events(struct task_struct *curr) 2974 { 2975 printk("irq event stamp: %u\n", curr->irq_events); 2976 printk("hardirqs last enabled at (%u): [<%px>] %pS\n", 2977 curr->hardirq_enable_event, (void *)curr->hardirq_enable_ip, 2978 (void *)curr->hardirq_enable_ip); 2979 printk("hardirqs last disabled at (%u): [<%px>] %pS\n", 2980 curr->hardirq_disable_event, (void *)curr->hardirq_disable_ip, 2981 (void *)curr->hardirq_disable_ip); 2982 printk("softirqs last enabled at (%u): [<%px>] %pS\n", 2983 curr->softirq_enable_event, (void *)curr->softirq_enable_ip, 2984 (void *)curr->softirq_enable_ip); 2985 printk("softirqs last disabled at (%u): [<%px>] %pS\n", 2986 curr->softirq_disable_event, (void *)curr->softirq_disable_ip, 2987 (void *)curr->softirq_disable_ip); 2988 } 2989 2990 static int HARDIRQ_verbose(struct lock_class *class) 2991 { 2992 #if HARDIRQ_VERBOSE 2993 return class_filter(class); 2994 #endif 2995 return 0; 2996 } 2997 2998 static int SOFTIRQ_verbose(struct lock_class *class) 2999 { 3000 #if SOFTIRQ_VERBOSE 3001 return class_filter(class); 3002 #endif 3003 return 0; 3004 } 3005 3006 #define STRICT_READ_CHECKS 1 3007 3008 static int (*state_verbose_f[])(struct lock_class *class) = { 3009 #define LOCKDEP_STATE(__STATE) \ 3010 __STATE##_verbose, 3011 #include "lockdep_states.h" 3012 #undef LOCKDEP_STATE 3013 }; 3014 3015 static inline int state_verbose(enum lock_usage_bit bit, 3016 struct lock_class *class) 3017 { 3018 return state_verbose_f[bit >> 2](class); 3019 } 3020 3021 typedef int (*check_usage_f)(struct task_struct *, struct held_lock *, 3022 enum lock_usage_bit bit, const char *name); 3023 3024 static int 3025 mark_lock_irq(struct task_struct *curr, struct held_lock *this, 3026 enum lock_usage_bit new_bit) 3027 { 3028 int excl_bit = exclusive_bit(new_bit); 3029 int read = new_bit & LOCK_USAGE_READ_MASK; 3030 int dir = new_bit & LOCK_USAGE_DIR_MASK; 3031 3032 /* 3033 * mark USED_IN has to look forwards -- to ensure no dependency 3034 * has ENABLED state, which would allow recursion deadlocks. 3035 * 3036 * mark ENABLED has to look backwards -- to ensure no dependee 3037 * has USED_IN state, which, again, would allow recursion deadlocks. 3038 */ 3039 check_usage_f usage = dir ? 3040 check_usage_backwards : check_usage_forwards; 3041 3042 /* 3043 * Validate that this particular lock does not have conflicting 3044 * usage states. 3045 */ 3046 if (!valid_state(curr, this, new_bit, excl_bit)) 3047 return 0; 3048 3049 /* 3050 * Validate that the lock dependencies don't have conflicting usage 3051 * states. 3052 */ 3053 if ((!read || !dir || STRICT_READ_CHECKS) && 3054 !usage(curr, this, excl_bit, state_name(new_bit & ~LOCK_USAGE_READ_MASK))) 3055 return 0; 3056 3057 /* 3058 * Check for read in write conflicts 3059 */ 3060 if (!read) { 3061 if (!valid_state(curr, this, new_bit, excl_bit + LOCK_USAGE_READ_MASK)) 3062 return 0; 3063 3064 if (STRICT_READ_CHECKS && 3065 !usage(curr, this, excl_bit + LOCK_USAGE_READ_MASK, 3066 state_name(new_bit + LOCK_USAGE_READ_MASK))) 3067 return 0; 3068 } 3069 3070 if (state_verbose(new_bit, hlock_class(this))) 3071 return 2; 3072 3073 return 1; 3074 } 3075 3076 /* 3077 * Mark all held locks with a usage bit: 3078 */ 3079 static int 3080 mark_held_locks(struct task_struct *curr, enum lock_usage_bit base_bit) 3081 { 3082 struct held_lock *hlock; 3083 int i; 3084 3085 for (i = 0; i < curr->lockdep_depth; i++) { 3086 enum lock_usage_bit hlock_bit = base_bit; 3087 hlock = curr->held_locks + i; 3088 3089 if (hlock->read) 3090 hlock_bit += LOCK_USAGE_READ_MASK; 3091 3092 BUG_ON(hlock_bit >= LOCK_USAGE_STATES); 3093 3094 if (!hlock->check) 3095 continue; 3096 3097 if (!mark_lock(curr, hlock, hlock_bit)) 3098 return 0; 3099 } 3100 3101 return 1; 3102 } 3103 3104 /* 3105 * Hardirqs will be enabled: 3106 */ 3107 static void __trace_hardirqs_on_caller(unsigned long ip) 3108 { 3109 struct task_struct *curr = current; 3110 3111 /* we'll do an OFF -> ON transition: */ 3112 curr->hardirqs_enabled = 1; 3113 3114 /* 3115 * We are going to turn hardirqs on, so set the 3116 * usage bit for all held locks: 3117 */ 3118 if (!mark_held_locks(curr, LOCK_ENABLED_HARDIRQ)) 3119 return; 3120 /* 3121 * If we have softirqs enabled, then set the usage 3122 * bit for all held locks. (disabled hardirqs prevented 3123 * this bit from being set before) 3124 */ 3125 if (curr->softirqs_enabled) 3126 if (!mark_held_locks(curr, LOCK_ENABLED_SOFTIRQ)) 3127 return; 3128 3129 curr->hardirq_enable_ip = ip; 3130 curr->hardirq_enable_event = ++curr->irq_events; 3131 debug_atomic_inc(hardirqs_on_events); 3132 } 3133 3134 void lockdep_hardirqs_on(unsigned long ip) 3135 { 3136 if (unlikely(!debug_locks || current->lockdep_recursion)) 3137 return; 3138 3139 if (unlikely(current->hardirqs_enabled)) { 3140 /* 3141 * Neither irq nor preemption are disabled here 3142 * so this is racy by nature but losing one hit 3143 * in a stat is not a big deal. 3144 */ 3145 __debug_atomic_inc(redundant_hardirqs_on); 3146 return; 3147 } 3148 3149 /* 3150 * We're enabling irqs and according to our state above irqs weren't 3151 * already enabled, yet we find the hardware thinks they are in fact 3152 * enabled.. someone messed up their IRQ state tracing. 3153 */ 3154 if (DEBUG_LOCKS_WARN_ON(!irqs_disabled())) 3155 return; 3156 3157 /* 3158 * See the fine text that goes along with this variable definition. 3159 */ 3160 if (DEBUG_LOCKS_WARN_ON(unlikely(early_boot_irqs_disabled))) 3161 return; 3162 3163 /* 3164 * Can't allow enabling interrupts while in an interrupt handler, 3165 * that's general bad form and such. Recursion, limited stack etc.. 3166 */ 3167 if (DEBUG_LOCKS_WARN_ON(current->hardirq_context)) 3168 return; 3169 3170 current->lockdep_recursion = 1; 3171 __trace_hardirqs_on_caller(ip); 3172 current->lockdep_recursion = 0; 3173 } 3174 NOKPROBE_SYMBOL(lockdep_hardirqs_on); 3175 3176 /* 3177 * Hardirqs were disabled: 3178 */ 3179 void lockdep_hardirqs_off(unsigned long ip) 3180 { 3181 struct task_struct *curr = current; 3182 3183 if (unlikely(!debug_locks || current->lockdep_recursion)) 3184 return; 3185 3186 /* 3187 * So we're supposed to get called after you mask local IRQs, but for 3188 * some reason the hardware doesn't quite think you did a proper job. 3189 */ 3190 if (DEBUG_LOCKS_WARN_ON(!irqs_disabled())) 3191 return; 3192 3193 if (curr->hardirqs_enabled) { 3194 /* 3195 * We have done an ON -> OFF transition: 3196 */ 3197 curr->hardirqs_enabled = 0; 3198 curr->hardirq_disable_ip = ip; 3199 curr->hardirq_disable_event = ++curr->irq_events; 3200 debug_atomic_inc(hardirqs_off_events); 3201 } else 3202 debug_atomic_inc(redundant_hardirqs_off); 3203 } 3204 NOKPROBE_SYMBOL(lockdep_hardirqs_off); 3205 3206 /* 3207 * Softirqs will be enabled: 3208 */ 3209 void trace_softirqs_on(unsigned long ip) 3210 { 3211 struct task_struct *curr = current; 3212 3213 if (unlikely(!debug_locks || current->lockdep_recursion)) 3214 return; 3215 3216 /* 3217 * We fancy IRQs being disabled here, see softirq.c, avoids 3218 * funny state and nesting things. 3219 */ 3220 if (DEBUG_LOCKS_WARN_ON(!irqs_disabled())) 3221 return; 3222 3223 if (curr->softirqs_enabled) { 3224 debug_atomic_inc(redundant_softirqs_on); 3225 return; 3226 } 3227 3228 current->lockdep_recursion = 1; 3229 /* 3230 * We'll do an OFF -> ON transition: 3231 */ 3232 curr->softirqs_enabled = 1; 3233 curr->softirq_enable_ip = ip; 3234 curr->softirq_enable_event = ++curr->irq_events; 3235 debug_atomic_inc(softirqs_on_events); 3236 /* 3237 * We are going to turn softirqs on, so set the 3238 * usage bit for all held locks, if hardirqs are 3239 * enabled too: 3240 */ 3241 if (curr->hardirqs_enabled) 3242 mark_held_locks(curr, LOCK_ENABLED_SOFTIRQ); 3243 current->lockdep_recursion = 0; 3244 } 3245 3246 /* 3247 * Softirqs were disabled: 3248 */ 3249 void trace_softirqs_off(unsigned long ip) 3250 { 3251 struct task_struct *curr = current; 3252 3253 if (unlikely(!debug_locks || current->lockdep_recursion)) 3254 return; 3255 3256 /* 3257 * We fancy IRQs being disabled here, see softirq.c 3258 */ 3259 if (DEBUG_LOCKS_WARN_ON(!irqs_disabled())) 3260 return; 3261 3262 if (curr->softirqs_enabled) { 3263 /* 3264 * We have done an ON -> OFF transition: 3265 */ 3266 curr->softirqs_enabled = 0; 3267 curr->softirq_disable_ip = ip; 3268 curr->softirq_disable_event = ++curr->irq_events; 3269 debug_atomic_inc(softirqs_off_events); 3270 /* 3271 * Whoops, we wanted softirqs off, so why aren't they? 3272 */ 3273 DEBUG_LOCKS_WARN_ON(!softirq_count()); 3274 } else 3275 debug_atomic_inc(redundant_softirqs_off); 3276 } 3277 3278 static int mark_irqflags(struct task_struct *curr, struct held_lock *hlock) 3279 { 3280 /* 3281 * If non-trylock use in a hardirq or softirq context, then 3282 * mark the lock as used in these contexts: 3283 */ 3284 if (!hlock->trylock) { 3285 if (hlock->read) { 3286 if (curr->hardirq_context) 3287 if (!mark_lock(curr, hlock, 3288 LOCK_USED_IN_HARDIRQ_READ)) 3289 return 0; 3290 if (curr->softirq_context) 3291 if (!mark_lock(curr, hlock, 3292 LOCK_USED_IN_SOFTIRQ_READ)) 3293 return 0; 3294 } else { 3295 if (curr->hardirq_context) 3296 if (!mark_lock(curr, hlock, LOCK_USED_IN_HARDIRQ)) 3297 return 0; 3298 if (curr->softirq_context) 3299 if (!mark_lock(curr, hlock, LOCK_USED_IN_SOFTIRQ)) 3300 return 0; 3301 } 3302 } 3303 if (!hlock->hardirqs_off) { 3304 if (hlock->read) { 3305 if (!mark_lock(curr, hlock, 3306 LOCK_ENABLED_HARDIRQ_READ)) 3307 return 0; 3308 if (curr->softirqs_enabled) 3309 if (!mark_lock(curr, hlock, 3310 LOCK_ENABLED_SOFTIRQ_READ)) 3311 return 0; 3312 } else { 3313 if (!mark_lock(curr, hlock, 3314 LOCK_ENABLED_HARDIRQ)) 3315 return 0; 3316 if (curr->softirqs_enabled) 3317 if (!mark_lock(curr, hlock, 3318 LOCK_ENABLED_SOFTIRQ)) 3319 return 0; 3320 } 3321 } 3322 3323 return 1; 3324 } 3325 3326 static inline unsigned int task_irq_context(struct task_struct *task) 3327 { 3328 return 2 * !!task->hardirq_context + !!task->softirq_context; 3329 } 3330 3331 static int separate_irq_context(struct task_struct *curr, 3332 struct held_lock *hlock) 3333 { 3334 unsigned int depth = curr->lockdep_depth; 3335 3336 /* 3337 * Keep track of points where we cross into an interrupt context: 3338 */ 3339 if (depth) { 3340 struct held_lock *prev_hlock; 3341 3342 prev_hlock = curr->held_locks + depth-1; 3343 /* 3344 * If we cross into another context, reset the 3345 * hash key (this also prevents the checking and the 3346 * adding of the dependency to 'prev'): 3347 */ 3348 if (prev_hlock->irq_context != hlock->irq_context) 3349 return 1; 3350 } 3351 return 0; 3352 } 3353 3354 #else /* defined(CONFIG_TRACE_IRQFLAGS) && defined(CONFIG_PROVE_LOCKING) */ 3355 3356 static inline 3357 int mark_lock_irq(struct task_struct *curr, struct held_lock *this, 3358 enum lock_usage_bit new_bit) 3359 { 3360 WARN_ON(1); /* Impossible innit? when we don't have TRACE_IRQFLAG */ 3361 return 1; 3362 } 3363 3364 static inline int mark_irqflags(struct task_struct *curr, 3365 struct held_lock *hlock) 3366 { 3367 return 1; 3368 } 3369 3370 static inline unsigned int task_irq_context(struct task_struct *task) 3371 { 3372 return 0; 3373 } 3374 3375 static inline int separate_irq_context(struct task_struct *curr, 3376 struct held_lock *hlock) 3377 { 3378 return 0; 3379 } 3380 3381 #endif /* defined(CONFIG_TRACE_IRQFLAGS) && defined(CONFIG_PROVE_LOCKING) */ 3382 3383 /* 3384 * Mark a lock with a usage bit, and validate the state transition: 3385 */ 3386 static int mark_lock(struct task_struct *curr, struct held_lock *this, 3387 enum lock_usage_bit new_bit) 3388 { 3389 unsigned int new_mask = 1 << new_bit, ret = 1; 3390 3391 /* 3392 * If already set then do not dirty the cacheline, 3393 * nor do any checks: 3394 */ 3395 if (likely(hlock_class(this)->usage_mask & new_mask)) 3396 return 1; 3397 3398 if (!graph_lock()) 3399 return 0; 3400 /* 3401 * Make sure we didn't race: 3402 */ 3403 if (unlikely(hlock_class(this)->usage_mask & new_mask)) { 3404 graph_unlock(); 3405 return 1; 3406 } 3407 3408 hlock_class(this)->usage_mask |= new_mask; 3409 3410 if (!save_trace(hlock_class(this)->usage_traces + new_bit)) 3411 return 0; 3412 3413 switch (new_bit) { 3414 #define LOCKDEP_STATE(__STATE) \ 3415 case LOCK_USED_IN_##__STATE: \ 3416 case LOCK_USED_IN_##__STATE##_READ: \ 3417 case LOCK_ENABLED_##__STATE: \ 3418 case LOCK_ENABLED_##__STATE##_READ: 3419 #include "lockdep_states.h" 3420 #undef LOCKDEP_STATE 3421 ret = mark_lock_irq(curr, this, new_bit); 3422 if (!ret) 3423 return 0; 3424 break; 3425 case LOCK_USED: 3426 debug_atomic_dec(nr_unused_locks); 3427 break; 3428 default: 3429 if (!debug_locks_off_graph_unlock()) 3430 return 0; 3431 WARN_ON(1); 3432 return 0; 3433 } 3434 3435 graph_unlock(); 3436 3437 /* 3438 * We must printk outside of the graph_lock: 3439 */ 3440 if (ret == 2) { 3441 printk("\nmarked lock as {%s}:\n", usage_str[new_bit]); 3442 print_lock(this); 3443 print_irqtrace_events(curr); 3444 dump_stack(); 3445 } 3446 3447 return ret; 3448 } 3449 3450 /* 3451 * Initialize a lock instance's lock-class mapping info: 3452 */ 3453 void lockdep_init_map(struct lockdep_map *lock, const char *name, 3454 struct lock_class_key *key, int subclass) 3455 { 3456 int i; 3457 3458 for (i = 0; i < NR_LOCKDEP_CACHING_CLASSES; i++) 3459 lock->class_cache[i] = NULL; 3460 3461 #ifdef CONFIG_LOCK_STAT 3462 lock->cpu = raw_smp_processor_id(); 3463 #endif 3464 3465 /* 3466 * Can't be having no nameless bastards around this place! 3467 */ 3468 if (DEBUG_LOCKS_WARN_ON(!name)) { 3469 lock->name = "NULL"; 3470 return; 3471 } 3472 3473 lock->name = name; 3474 3475 /* 3476 * No key, no joy, we need to hash something. 3477 */ 3478 if (DEBUG_LOCKS_WARN_ON(!key)) 3479 return; 3480 /* 3481 * Sanity check, the lock-class key must either have been allocated 3482 * statically or must have been registered as a dynamic key. 3483 */ 3484 if (!static_obj(key) && !is_dynamic_key(key)) { 3485 if (debug_locks) 3486 printk(KERN_ERR "BUG: key %px has not been registered!\n", key); 3487 DEBUG_LOCKS_WARN_ON(1); 3488 return; 3489 } 3490 lock->key = key; 3491 3492 if (unlikely(!debug_locks)) 3493 return; 3494 3495 if (subclass) { 3496 unsigned long flags; 3497 3498 if (DEBUG_LOCKS_WARN_ON(current->lockdep_recursion)) 3499 return; 3500 3501 raw_local_irq_save(flags); 3502 current->lockdep_recursion = 1; 3503 register_lock_class(lock, subclass, 1); 3504 current->lockdep_recursion = 0; 3505 raw_local_irq_restore(flags); 3506 } 3507 } 3508 EXPORT_SYMBOL_GPL(lockdep_init_map); 3509 3510 struct lock_class_key __lockdep_no_validate__; 3511 EXPORT_SYMBOL_GPL(__lockdep_no_validate__); 3512 3513 static int 3514 print_lock_nested_lock_not_held(struct task_struct *curr, 3515 struct held_lock *hlock, 3516 unsigned long ip) 3517 { 3518 if (!debug_locks_off()) 3519 return 0; 3520 if (debug_locks_silent) 3521 return 0; 3522 3523 pr_warn("\n"); 3524 pr_warn("==================================\n"); 3525 pr_warn("WARNING: Nested lock was not taken\n"); 3526 print_kernel_ident(); 3527 pr_warn("----------------------------------\n"); 3528 3529 pr_warn("%s/%d is trying to lock:\n", curr->comm, task_pid_nr(curr)); 3530 print_lock(hlock); 3531 3532 pr_warn("\nbut this task is not holding:\n"); 3533 pr_warn("%s\n", hlock->nest_lock->name); 3534 3535 pr_warn("\nstack backtrace:\n"); 3536 dump_stack(); 3537 3538 pr_warn("\nother info that might help us debug this:\n"); 3539 lockdep_print_held_locks(curr); 3540 3541 pr_warn("\nstack backtrace:\n"); 3542 dump_stack(); 3543 3544 return 0; 3545 } 3546 3547 static int __lock_is_held(const struct lockdep_map *lock, int read); 3548 3549 /* 3550 * This gets called for every mutex_lock*()/spin_lock*() operation. 3551 * We maintain the dependency maps and validate the locking attempt: 3552 * 3553 * The callers must make sure that IRQs are disabled before calling it, 3554 * otherwise we could get an interrupt which would want to take locks, 3555 * which would end up in lockdep again. 3556 */ 3557 static int __lock_acquire(struct lockdep_map *lock, unsigned int subclass, 3558 int trylock, int read, int check, int hardirqs_off, 3559 struct lockdep_map *nest_lock, unsigned long ip, 3560 int references, int pin_count) 3561 { 3562 struct task_struct *curr = current; 3563 struct lock_class *class = NULL; 3564 struct held_lock *hlock; 3565 unsigned int depth; 3566 int chain_head = 0; 3567 int class_idx; 3568 u64 chain_key; 3569 3570 if (unlikely(!debug_locks)) 3571 return 0; 3572 3573 if (!prove_locking || lock->key == &__lockdep_no_validate__) 3574 check = 0; 3575 3576 if (subclass < NR_LOCKDEP_CACHING_CLASSES) 3577 class = lock->class_cache[subclass]; 3578 /* 3579 * Not cached? 3580 */ 3581 if (unlikely(!class)) { 3582 class = register_lock_class(lock, subclass, 0); 3583 if (!class) 3584 return 0; 3585 } 3586 3587 debug_class_ops_inc(class); 3588 3589 if (very_verbose(class)) { 3590 printk("\nacquire class [%px] %s", class->key, class->name); 3591 if (class->name_version > 1) 3592 printk(KERN_CONT "#%d", class->name_version); 3593 printk(KERN_CONT "\n"); 3594 dump_stack(); 3595 } 3596 3597 /* 3598 * Add the lock to the list of currently held locks. 3599 * (we dont increase the depth just yet, up until the 3600 * dependency checks are done) 3601 */ 3602 depth = curr->lockdep_depth; 3603 /* 3604 * Ran out of static storage for our per-task lock stack again have we? 3605 */ 3606 if (DEBUG_LOCKS_WARN_ON(depth >= MAX_LOCK_DEPTH)) 3607 return 0; 3608 3609 class_idx = class - lock_classes + 1; 3610 3611 if (depth) { 3612 hlock = curr->held_locks + depth - 1; 3613 if (hlock->class_idx == class_idx && nest_lock) { 3614 if (hlock->references) { 3615 /* 3616 * Check: unsigned int references:12, overflow. 3617 */ 3618 if (DEBUG_LOCKS_WARN_ON(hlock->references == (1 << 12)-1)) 3619 return 0; 3620 3621 hlock->references++; 3622 } else { 3623 hlock->references = 2; 3624 } 3625 3626 return 1; 3627 } 3628 } 3629 3630 hlock = curr->held_locks + depth; 3631 /* 3632 * Plain impossible, we just registered it and checked it weren't no 3633 * NULL like.. I bet this mushroom I ate was good! 3634 */ 3635 if (DEBUG_LOCKS_WARN_ON(!class)) 3636 return 0; 3637 hlock->class_idx = class_idx; 3638 hlock->acquire_ip = ip; 3639 hlock->instance = lock; 3640 hlock->nest_lock = nest_lock; 3641 hlock->irq_context = task_irq_context(curr); 3642 hlock->trylock = trylock; 3643 hlock->read = read; 3644 hlock->check = check; 3645 hlock->hardirqs_off = !!hardirqs_off; 3646 hlock->references = references; 3647 #ifdef CONFIG_LOCK_STAT 3648 hlock->waittime_stamp = 0; 3649 hlock->holdtime_stamp = lockstat_clock(); 3650 #endif 3651 hlock->pin_count = pin_count; 3652 3653 if (check && !mark_irqflags(curr, hlock)) 3654 return 0; 3655 3656 /* mark it as used: */ 3657 if (!mark_lock(curr, hlock, LOCK_USED)) 3658 return 0; 3659 3660 /* 3661 * Calculate the chain hash: it's the combined hash of all the 3662 * lock keys along the dependency chain. We save the hash value 3663 * at every step so that we can get the current hash easily 3664 * after unlock. The chain hash is then used to cache dependency 3665 * results. 3666 * 3667 * The 'key ID' is what is the most compact key value to drive 3668 * the hash, not class->key. 3669 */ 3670 /* 3671 * Whoops, we did it again.. ran straight out of our static allocation. 3672 */ 3673 if (DEBUG_LOCKS_WARN_ON(class_idx > MAX_LOCKDEP_KEYS)) 3674 return 0; 3675 3676 chain_key = curr->curr_chain_key; 3677 if (!depth) { 3678 /* 3679 * How can we have a chain hash when we ain't got no keys?! 3680 */ 3681 if (DEBUG_LOCKS_WARN_ON(chain_key != 0)) 3682 return 0; 3683 chain_head = 1; 3684 } 3685 3686 hlock->prev_chain_key = chain_key; 3687 if (separate_irq_context(curr, hlock)) { 3688 chain_key = 0; 3689 chain_head = 1; 3690 } 3691 chain_key = iterate_chain_key(chain_key, class_idx); 3692 3693 if (nest_lock && !__lock_is_held(nest_lock, -1)) 3694 return print_lock_nested_lock_not_held(curr, hlock, ip); 3695 3696 if (!debug_locks_silent) { 3697 WARN_ON_ONCE(depth && !hlock_class(hlock - 1)->key); 3698 WARN_ON_ONCE(!hlock_class(hlock)->key); 3699 } 3700 3701 if (!validate_chain(curr, lock, hlock, chain_head, chain_key)) 3702 return 0; 3703 3704 curr->curr_chain_key = chain_key; 3705 curr->lockdep_depth++; 3706 check_chain_key(curr); 3707 #ifdef CONFIG_DEBUG_LOCKDEP 3708 if (unlikely(!debug_locks)) 3709 return 0; 3710 #endif 3711 if (unlikely(curr->lockdep_depth >= MAX_LOCK_DEPTH)) { 3712 debug_locks_off(); 3713 print_lockdep_off("BUG: MAX_LOCK_DEPTH too low!"); 3714 printk(KERN_DEBUG "depth: %i max: %lu!\n", 3715 curr->lockdep_depth, MAX_LOCK_DEPTH); 3716 3717 lockdep_print_held_locks(current); 3718 debug_show_all_locks(); 3719 dump_stack(); 3720 3721 return 0; 3722 } 3723 3724 if (unlikely(curr->lockdep_depth > max_lockdep_depth)) 3725 max_lockdep_depth = curr->lockdep_depth; 3726 3727 return 1; 3728 } 3729 3730 static int 3731 print_unlock_imbalance_bug(struct task_struct *curr, struct lockdep_map *lock, 3732 unsigned long ip) 3733 { 3734 if (!debug_locks_off()) 3735 return 0; 3736 if (debug_locks_silent) 3737 return 0; 3738 3739 pr_warn("\n"); 3740 pr_warn("=====================================\n"); 3741 pr_warn("WARNING: bad unlock balance detected!\n"); 3742 print_kernel_ident(); 3743 pr_warn("-------------------------------------\n"); 3744 pr_warn("%s/%d is trying to release lock (", 3745 curr->comm, task_pid_nr(curr)); 3746 print_lockdep_cache(lock); 3747 pr_cont(") at:\n"); 3748 print_ip_sym(ip); 3749 pr_warn("but there are no more locks to release!\n"); 3750 pr_warn("\nother info that might help us debug this:\n"); 3751 lockdep_print_held_locks(curr); 3752 3753 pr_warn("\nstack backtrace:\n"); 3754 dump_stack(); 3755 3756 return 0; 3757 } 3758 3759 static int match_held_lock(const struct held_lock *hlock, 3760 const struct lockdep_map *lock) 3761 { 3762 if (hlock->instance == lock) 3763 return 1; 3764 3765 if (hlock->references) { 3766 const struct lock_class *class = lock->class_cache[0]; 3767 3768 if (!class) 3769 class = look_up_lock_class(lock, 0); 3770 3771 /* 3772 * If look_up_lock_class() failed to find a class, we're trying 3773 * to test if we hold a lock that has never yet been acquired. 3774 * Clearly if the lock hasn't been acquired _ever_, we're not 3775 * holding it either, so report failure. 3776 */ 3777 if (!class) 3778 return 0; 3779 3780 /* 3781 * References, but not a lock we're actually ref-counting? 3782 * State got messed up, follow the sites that change ->references 3783 * and try to make sense of it. 3784 */ 3785 if (DEBUG_LOCKS_WARN_ON(!hlock->nest_lock)) 3786 return 0; 3787 3788 if (hlock->class_idx == class - lock_classes + 1) 3789 return 1; 3790 } 3791 3792 return 0; 3793 } 3794 3795 /* @depth must not be zero */ 3796 static struct held_lock *find_held_lock(struct task_struct *curr, 3797 struct lockdep_map *lock, 3798 unsigned int depth, int *idx) 3799 { 3800 struct held_lock *ret, *hlock, *prev_hlock; 3801 int i; 3802 3803 i = depth - 1; 3804 hlock = curr->held_locks + i; 3805 ret = hlock; 3806 if (match_held_lock(hlock, lock)) 3807 goto out; 3808 3809 ret = NULL; 3810 for (i--, prev_hlock = hlock--; 3811 i >= 0; 3812 i--, prev_hlock = hlock--) { 3813 /* 3814 * We must not cross into another context: 3815 */ 3816 if (prev_hlock->irq_context != hlock->irq_context) { 3817 ret = NULL; 3818 break; 3819 } 3820 if (match_held_lock(hlock, lock)) { 3821 ret = hlock; 3822 break; 3823 } 3824 } 3825 3826 out: 3827 *idx = i; 3828 return ret; 3829 } 3830 3831 static int reacquire_held_locks(struct task_struct *curr, unsigned int depth, 3832 int idx) 3833 { 3834 struct held_lock *hlock; 3835 3836 if (DEBUG_LOCKS_WARN_ON(!irqs_disabled())) 3837 return 0; 3838 3839 for (hlock = curr->held_locks + idx; idx < depth; idx++, hlock++) { 3840 if (!__lock_acquire(hlock->instance, 3841 hlock_class(hlock)->subclass, 3842 hlock->trylock, 3843 hlock->read, hlock->check, 3844 hlock->hardirqs_off, 3845 hlock->nest_lock, hlock->acquire_ip, 3846 hlock->references, hlock->pin_count)) 3847 return 1; 3848 } 3849 return 0; 3850 } 3851 3852 static int 3853 __lock_set_class(struct lockdep_map *lock, const char *name, 3854 struct lock_class_key *key, unsigned int subclass, 3855 unsigned long ip) 3856 { 3857 struct task_struct *curr = current; 3858 struct held_lock *hlock; 3859 struct lock_class *class; 3860 unsigned int depth; 3861 int i; 3862 3863 if (unlikely(!debug_locks)) 3864 return 0; 3865 3866 depth = curr->lockdep_depth; 3867 /* 3868 * This function is about (re)setting the class of a held lock, 3869 * yet we're not actually holding any locks. Naughty user! 3870 */ 3871 if (DEBUG_LOCKS_WARN_ON(!depth)) 3872 return 0; 3873 3874 hlock = find_held_lock(curr, lock, depth, &i); 3875 if (!hlock) 3876 return print_unlock_imbalance_bug(curr, lock, ip); 3877 3878 lockdep_init_map(lock, name, key, 0); 3879 class = register_lock_class(lock, subclass, 0); 3880 hlock->class_idx = class - lock_classes + 1; 3881 3882 curr->lockdep_depth = i; 3883 curr->curr_chain_key = hlock->prev_chain_key; 3884 3885 if (reacquire_held_locks(curr, depth, i)) 3886 return 0; 3887 3888 /* 3889 * I took it apart and put it back together again, except now I have 3890 * these 'spare' parts.. where shall I put them. 3891 */ 3892 if (DEBUG_LOCKS_WARN_ON(curr->lockdep_depth != depth)) 3893 return 0; 3894 return 1; 3895 } 3896 3897 static int __lock_downgrade(struct lockdep_map *lock, unsigned long ip) 3898 { 3899 struct task_struct *curr = current; 3900 struct held_lock *hlock; 3901 unsigned int depth; 3902 int i; 3903 3904 if (unlikely(!debug_locks)) 3905 return 0; 3906 3907 depth = curr->lockdep_depth; 3908 /* 3909 * This function is about (re)setting the class of a held lock, 3910 * yet we're not actually holding any locks. Naughty user! 3911 */ 3912 if (DEBUG_LOCKS_WARN_ON(!depth)) 3913 return 0; 3914 3915 hlock = find_held_lock(curr, lock, depth, &i); 3916 if (!hlock) 3917 return print_unlock_imbalance_bug(curr, lock, ip); 3918 3919 curr->lockdep_depth = i; 3920 curr->curr_chain_key = hlock->prev_chain_key; 3921 3922 WARN(hlock->read, "downgrading a read lock"); 3923 hlock->read = 1; 3924 hlock->acquire_ip = ip; 3925 3926 if (reacquire_held_locks(curr, depth, i)) 3927 return 0; 3928 3929 /* 3930 * I took it apart and put it back together again, except now I have 3931 * these 'spare' parts.. where shall I put them. 3932 */ 3933 if (DEBUG_LOCKS_WARN_ON(curr->lockdep_depth != depth)) 3934 return 0; 3935 return 1; 3936 } 3937 3938 /* 3939 * Remove the lock to the list of currently held locks - this gets 3940 * called on mutex_unlock()/spin_unlock*() (or on a failed 3941 * mutex_lock_interruptible()). 3942 * 3943 * @nested is an hysterical artifact, needs a tree wide cleanup. 3944 */ 3945 static int 3946 __lock_release(struct lockdep_map *lock, int nested, unsigned long ip) 3947 { 3948 struct task_struct *curr = current; 3949 struct held_lock *hlock; 3950 unsigned int depth; 3951 int i; 3952 3953 if (unlikely(!debug_locks)) 3954 return 0; 3955 3956 depth = curr->lockdep_depth; 3957 /* 3958 * So we're all set to release this lock.. wait what lock? We don't 3959 * own any locks, you've been drinking again? 3960 */ 3961 if (DEBUG_LOCKS_WARN_ON(depth <= 0)) 3962 return print_unlock_imbalance_bug(curr, lock, ip); 3963 3964 /* 3965 * Check whether the lock exists in the current stack 3966 * of held locks: 3967 */ 3968 hlock = find_held_lock(curr, lock, depth, &i); 3969 if (!hlock) 3970 return print_unlock_imbalance_bug(curr, lock, ip); 3971 3972 if (hlock->instance == lock) 3973 lock_release_holdtime(hlock); 3974 3975 WARN(hlock->pin_count, "releasing a pinned lock\n"); 3976 3977 if (hlock->references) { 3978 hlock->references--; 3979 if (hlock->references) { 3980 /* 3981 * We had, and after removing one, still have 3982 * references, the current lock stack is still 3983 * valid. We're done! 3984 */ 3985 return 1; 3986 } 3987 } 3988 3989 /* 3990 * We have the right lock to unlock, 'hlock' points to it. 3991 * Now we remove it from the stack, and add back the other 3992 * entries (if any), recalculating the hash along the way: 3993 */ 3994 3995 curr->lockdep_depth = i; 3996 curr->curr_chain_key = hlock->prev_chain_key; 3997 3998 /* 3999 * The most likely case is when the unlock is on the innermost 4000 * lock. In this case, we are done! 4001 */ 4002 if (i == depth-1) 4003 return 1; 4004 4005 if (reacquire_held_locks(curr, depth, i + 1)) 4006 return 0; 4007 4008 /* 4009 * We had N bottles of beer on the wall, we drank one, but now 4010 * there's not N-1 bottles of beer left on the wall... 4011 */ 4012 DEBUG_LOCKS_WARN_ON(curr->lockdep_depth != depth-1); 4013 4014 /* 4015 * Since reacquire_held_locks() would have called check_chain_key() 4016 * indirectly via __lock_acquire(), we don't need to do it again 4017 * on return. 4018 */ 4019 return 0; 4020 } 4021 4022 static nokprobe_inline 4023 int __lock_is_held(const struct lockdep_map *lock, int read) 4024 { 4025 struct task_struct *curr = current; 4026 int i; 4027 4028 for (i = 0; i < curr->lockdep_depth; i++) { 4029 struct held_lock *hlock = curr->held_locks + i; 4030 4031 if (match_held_lock(hlock, lock)) { 4032 if (read == -1 || hlock->read == read) 4033 return 1; 4034 4035 return 0; 4036 } 4037 } 4038 4039 return 0; 4040 } 4041 4042 static struct pin_cookie __lock_pin_lock(struct lockdep_map *lock) 4043 { 4044 struct pin_cookie cookie = NIL_COOKIE; 4045 struct task_struct *curr = current; 4046 int i; 4047 4048 if (unlikely(!debug_locks)) 4049 return cookie; 4050 4051 for (i = 0; i < curr->lockdep_depth; i++) { 4052 struct held_lock *hlock = curr->held_locks + i; 4053 4054 if (match_held_lock(hlock, lock)) { 4055 /* 4056 * Grab 16bits of randomness; this is sufficient to not 4057 * be guessable and still allows some pin nesting in 4058 * our u32 pin_count. 4059 */ 4060 cookie.val = 1 + (prandom_u32() >> 16); 4061 hlock->pin_count += cookie.val; 4062 return cookie; 4063 } 4064 } 4065 4066 WARN(1, "pinning an unheld lock\n"); 4067 return cookie; 4068 } 4069 4070 static void __lock_repin_lock(struct lockdep_map *lock, struct pin_cookie cookie) 4071 { 4072 struct task_struct *curr = current; 4073 int i; 4074 4075 if (unlikely(!debug_locks)) 4076 return; 4077 4078 for (i = 0; i < curr->lockdep_depth; i++) { 4079 struct held_lock *hlock = curr->held_locks + i; 4080 4081 if (match_held_lock(hlock, lock)) { 4082 hlock->pin_count += cookie.val; 4083 return; 4084 } 4085 } 4086 4087 WARN(1, "pinning an unheld lock\n"); 4088 } 4089 4090 static void __lock_unpin_lock(struct lockdep_map *lock, struct pin_cookie cookie) 4091 { 4092 struct task_struct *curr = current; 4093 int i; 4094 4095 if (unlikely(!debug_locks)) 4096 return; 4097 4098 for (i = 0; i < curr->lockdep_depth; i++) { 4099 struct held_lock *hlock = curr->held_locks + i; 4100 4101 if (match_held_lock(hlock, lock)) { 4102 if (WARN(!hlock->pin_count, "unpinning an unpinned lock\n")) 4103 return; 4104 4105 hlock->pin_count -= cookie.val; 4106 4107 if (WARN((int)hlock->pin_count < 0, "pin count corrupted\n")) 4108 hlock->pin_count = 0; 4109 4110 return; 4111 } 4112 } 4113 4114 WARN(1, "unpinning an unheld lock\n"); 4115 } 4116 4117 /* 4118 * Check whether we follow the irq-flags state precisely: 4119 */ 4120 static void check_flags(unsigned long flags) 4121 { 4122 #if defined(CONFIG_PROVE_LOCKING) && defined(CONFIG_DEBUG_LOCKDEP) && \ 4123 defined(CONFIG_TRACE_IRQFLAGS) 4124 if (!debug_locks) 4125 return; 4126 4127 if (irqs_disabled_flags(flags)) { 4128 if (DEBUG_LOCKS_WARN_ON(current->hardirqs_enabled)) { 4129 printk("possible reason: unannotated irqs-off.\n"); 4130 } 4131 } else { 4132 if (DEBUG_LOCKS_WARN_ON(!current->hardirqs_enabled)) { 4133 printk("possible reason: unannotated irqs-on.\n"); 4134 } 4135 } 4136 4137 /* 4138 * We dont accurately track softirq state in e.g. 4139 * hardirq contexts (such as on 4KSTACKS), so only 4140 * check if not in hardirq contexts: 4141 */ 4142 if (!hardirq_count()) { 4143 if (softirq_count()) { 4144 /* like the above, but with softirqs */ 4145 DEBUG_LOCKS_WARN_ON(current->softirqs_enabled); 4146 } else { 4147 /* lick the above, does it taste good? */ 4148 DEBUG_LOCKS_WARN_ON(!current->softirqs_enabled); 4149 } 4150 } 4151 4152 if (!debug_locks) 4153 print_irqtrace_events(current); 4154 #endif 4155 } 4156 4157 void lock_set_class(struct lockdep_map *lock, const char *name, 4158 struct lock_class_key *key, unsigned int subclass, 4159 unsigned long ip) 4160 { 4161 unsigned long flags; 4162 4163 if (unlikely(current->lockdep_recursion)) 4164 return; 4165 4166 raw_local_irq_save(flags); 4167 current->lockdep_recursion = 1; 4168 check_flags(flags); 4169 if (__lock_set_class(lock, name, key, subclass, ip)) 4170 check_chain_key(current); 4171 current->lockdep_recursion = 0; 4172 raw_local_irq_restore(flags); 4173 } 4174 EXPORT_SYMBOL_GPL(lock_set_class); 4175 4176 void lock_downgrade(struct lockdep_map *lock, unsigned long ip) 4177 { 4178 unsigned long flags; 4179 4180 if (unlikely(current->lockdep_recursion)) 4181 return; 4182 4183 raw_local_irq_save(flags); 4184 current->lockdep_recursion = 1; 4185 check_flags(flags); 4186 if (__lock_downgrade(lock, ip)) 4187 check_chain_key(current); 4188 current->lockdep_recursion = 0; 4189 raw_local_irq_restore(flags); 4190 } 4191 EXPORT_SYMBOL_GPL(lock_downgrade); 4192 4193 /* 4194 * We are not always called with irqs disabled - do that here, 4195 * and also avoid lockdep recursion: 4196 */ 4197 void lock_acquire(struct lockdep_map *lock, unsigned int subclass, 4198 int trylock, int read, int check, 4199 struct lockdep_map *nest_lock, unsigned long ip) 4200 { 4201 unsigned long flags; 4202 4203 if (unlikely(current->lockdep_recursion)) 4204 return; 4205 4206 raw_local_irq_save(flags); 4207 check_flags(flags); 4208 4209 current->lockdep_recursion = 1; 4210 trace_lock_acquire(lock, subclass, trylock, read, check, nest_lock, ip); 4211 __lock_acquire(lock, subclass, trylock, read, check, 4212 irqs_disabled_flags(flags), nest_lock, ip, 0, 0); 4213 current->lockdep_recursion = 0; 4214 raw_local_irq_restore(flags); 4215 } 4216 EXPORT_SYMBOL_GPL(lock_acquire); 4217 4218 void lock_release(struct lockdep_map *lock, int nested, 4219 unsigned long ip) 4220 { 4221 unsigned long flags; 4222 4223 if (unlikely(current->lockdep_recursion)) 4224 return; 4225 4226 raw_local_irq_save(flags); 4227 check_flags(flags); 4228 current->lockdep_recursion = 1; 4229 trace_lock_release(lock, ip); 4230 if (__lock_release(lock, nested, ip)) 4231 check_chain_key(current); 4232 current->lockdep_recursion = 0; 4233 raw_local_irq_restore(flags); 4234 } 4235 EXPORT_SYMBOL_GPL(lock_release); 4236 4237 int lock_is_held_type(const struct lockdep_map *lock, int read) 4238 { 4239 unsigned long flags; 4240 int ret = 0; 4241 4242 if (unlikely(current->lockdep_recursion)) 4243 return 1; /* avoid false negative lockdep_assert_held() */ 4244 4245 raw_local_irq_save(flags); 4246 check_flags(flags); 4247 4248 current->lockdep_recursion = 1; 4249 ret = __lock_is_held(lock, read); 4250 current->lockdep_recursion = 0; 4251 raw_local_irq_restore(flags); 4252 4253 return ret; 4254 } 4255 EXPORT_SYMBOL_GPL(lock_is_held_type); 4256 NOKPROBE_SYMBOL(lock_is_held_type); 4257 4258 struct pin_cookie lock_pin_lock(struct lockdep_map *lock) 4259 { 4260 struct pin_cookie cookie = NIL_COOKIE; 4261 unsigned long flags; 4262 4263 if (unlikely(current->lockdep_recursion)) 4264 return cookie; 4265 4266 raw_local_irq_save(flags); 4267 check_flags(flags); 4268 4269 current->lockdep_recursion = 1; 4270 cookie = __lock_pin_lock(lock); 4271 current->lockdep_recursion = 0; 4272 raw_local_irq_restore(flags); 4273 4274 return cookie; 4275 } 4276 EXPORT_SYMBOL_GPL(lock_pin_lock); 4277 4278 void lock_repin_lock(struct lockdep_map *lock, struct pin_cookie cookie) 4279 { 4280 unsigned long flags; 4281 4282 if (unlikely(current->lockdep_recursion)) 4283 return; 4284 4285 raw_local_irq_save(flags); 4286 check_flags(flags); 4287 4288 current->lockdep_recursion = 1; 4289 __lock_repin_lock(lock, cookie); 4290 current->lockdep_recursion = 0; 4291 raw_local_irq_restore(flags); 4292 } 4293 EXPORT_SYMBOL_GPL(lock_repin_lock); 4294 4295 void lock_unpin_lock(struct lockdep_map *lock, struct pin_cookie cookie) 4296 { 4297 unsigned long flags; 4298 4299 if (unlikely(current->lockdep_recursion)) 4300 return; 4301 4302 raw_local_irq_save(flags); 4303 check_flags(flags); 4304 4305 current->lockdep_recursion = 1; 4306 __lock_unpin_lock(lock, cookie); 4307 current->lockdep_recursion = 0; 4308 raw_local_irq_restore(flags); 4309 } 4310 EXPORT_SYMBOL_GPL(lock_unpin_lock); 4311 4312 #ifdef CONFIG_LOCK_STAT 4313 static int 4314 print_lock_contention_bug(struct task_struct *curr, struct lockdep_map *lock, 4315 unsigned long ip) 4316 { 4317 if (!debug_locks_off()) 4318 return 0; 4319 if (debug_locks_silent) 4320 return 0; 4321 4322 pr_warn("\n"); 4323 pr_warn("=================================\n"); 4324 pr_warn("WARNING: bad contention detected!\n"); 4325 print_kernel_ident(); 4326 pr_warn("---------------------------------\n"); 4327 pr_warn("%s/%d is trying to contend lock (", 4328 curr->comm, task_pid_nr(curr)); 4329 print_lockdep_cache(lock); 4330 pr_cont(") at:\n"); 4331 print_ip_sym(ip); 4332 pr_warn("but there are no locks held!\n"); 4333 pr_warn("\nother info that might help us debug this:\n"); 4334 lockdep_print_held_locks(curr); 4335 4336 pr_warn("\nstack backtrace:\n"); 4337 dump_stack(); 4338 4339 return 0; 4340 } 4341 4342 static void 4343 __lock_contended(struct lockdep_map *lock, unsigned long ip) 4344 { 4345 struct task_struct *curr = current; 4346 struct held_lock *hlock; 4347 struct lock_class_stats *stats; 4348 unsigned int depth; 4349 int i, contention_point, contending_point; 4350 4351 depth = curr->lockdep_depth; 4352 /* 4353 * Whee, we contended on this lock, except it seems we're not 4354 * actually trying to acquire anything much at all.. 4355 */ 4356 if (DEBUG_LOCKS_WARN_ON(!depth)) 4357 return; 4358 4359 hlock = find_held_lock(curr, lock, depth, &i); 4360 if (!hlock) { 4361 print_lock_contention_bug(curr, lock, ip); 4362 return; 4363 } 4364 4365 if (hlock->instance != lock) 4366 return; 4367 4368 hlock->waittime_stamp = lockstat_clock(); 4369 4370 contention_point = lock_point(hlock_class(hlock)->contention_point, ip); 4371 contending_point = lock_point(hlock_class(hlock)->contending_point, 4372 lock->ip); 4373 4374 stats = get_lock_stats(hlock_class(hlock)); 4375 if (contention_point < LOCKSTAT_POINTS) 4376 stats->contention_point[contention_point]++; 4377 if (contending_point < LOCKSTAT_POINTS) 4378 stats->contending_point[contending_point]++; 4379 if (lock->cpu != smp_processor_id()) 4380 stats->bounces[bounce_contended + !!hlock->read]++; 4381 } 4382 4383 static void 4384 __lock_acquired(struct lockdep_map *lock, unsigned long ip) 4385 { 4386 struct task_struct *curr = current; 4387 struct held_lock *hlock; 4388 struct lock_class_stats *stats; 4389 unsigned int depth; 4390 u64 now, waittime = 0; 4391 int i, cpu; 4392 4393 depth = curr->lockdep_depth; 4394 /* 4395 * Yay, we acquired ownership of this lock we didn't try to 4396 * acquire, how the heck did that happen? 4397 */ 4398 if (DEBUG_LOCKS_WARN_ON(!depth)) 4399 return; 4400 4401 hlock = find_held_lock(curr, lock, depth, &i); 4402 if (!hlock) { 4403 print_lock_contention_bug(curr, lock, _RET_IP_); 4404 return; 4405 } 4406 4407 if (hlock->instance != lock) 4408 return; 4409 4410 cpu = smp_processor_id(); 4411 if (hlock->waittime_stamp) { 4412 now = lockstat_clock(); 4413 waittime = now - hlock->waittime_stamp; 4414 hlock->holdtime_stamp = now; 4415 } 4416 4417 trace_lock_acquired(lock, ip); 4418 4419 stats = get_lock_stats(hlock_class(hlock)); 4420 if (waittime) { 4421 if (hlock->read) 4422 lock_time_inc(&stats->read_waittime, waittime); 4423 else 4424 lock_time_inc(&stats->write_waittime, waittime); 4425 } 4426 if (lock->cpu != cpu) 4427 stats->bounces[bounce_acquired + !!hlock->read]++; 4428 4429 lock->cpu = cpu; 4430 lock->ip = ip; 4431 } 4432 4433 void lock_contended(struct lockdep_map *lock, unsigned long ip) 4434 { 4435 unsigned long flags; 4436 4437 if (unlikely(!lock_stat || !debug_locks)) 4438 return; 4439 4440 if (unlikely(current->lockdep_recursion)) 4441 return; 4442 4443 raw_local_irq_save(flags); 4444 check_flags(flags); 4445 current->lockdep_recursion = 1; 4446 trace_lock_contended(lock, ip); 4447 __lock_contended(lock, ip); 4448 current->lockdep_recursion = 0; 4449 raw_local_irq_restore(flags); 4450 } 4451 EXPORT_SYMBOL_GPL(lock_contended); 4452 4453 void lock_acquired(struct lockdep_map *lock, unsigned long ip) 4454 { 4455 unsigned long flags; 4456 4457 if (unlikely(!lock_stat || !debug_locks)) 4458 return; 4459 4460 if (unlikely(current->lockdep_recursion)) 4461 return; 4462 4463 raw_local_irq_save(flags); 4464 check_flags(flags); 4465 current->lockdep_recursion = 1; 4466 __lock_acquired(lock, ip); 4467 current->lockdep_recursion = 0; 4468 raw_local_irq_restore(flags); 4469 } 4470 EXPORT_SYMBOL_GPL(lock_acquired); 4471 #endif 4472 4473 /* 4474 * Used by the testsuite, sanitize the validator state 4475 * after a simulated failure: 4476 */ 4477 4478 void lockdep_reset(void) 4479 { 4480 unsigned long flags; 4481 int i; 4482 4483 raw_local_irq_save(flags); 4484 current->curr_chain_key = 0; 4485 current->lockdep_depth = 0; 4486 current->lockdep_recursion = 0; 4487 memset(current->held_locks, 0, MAX_LOCK_DEPTH*sizeof(struct held_lock)); 4488 nr_hardirq_chains = 0; 4489 nr_softirq_chains = 0; 4490 nr_process_chains = 0; 4491 debug_locks = 1; 4492 for (i = 0; i < CHAINHASH_SIZE; i++) 4493 INIT_HLIST_HEAD(chainhash_table + i); 4494 raw_local_irq_restore(flags); 4495 } 4496 4497 /* Remove a class from a lock chain. Must be called with the graph lock held. */ 4498 static void remove_class_from_lock_chain(struct pending_free *pf, 4499 struct lock_chain *chain, 4500 struct lock_class *class) 4501 { 4502 #ifdef CONFIG_PROVE_LOCKING 4503 struct lock_chain *new_chain; 4504 u64 chain_key; 4505 int i; 4506 4507 for (i = chain->base; i < chain->base + chain->depth; i++) { 4508 if (chain_hlocks[i] != class - lock_classes) 4509 continue; 4510 /* The code below leaks one chain_hlock[] entry. */ 4511 if (--chain->depth > 0) { 4512 memmove(&chain_hlocks[i], &chain_hlocks[i + 1], 4513 (chain->base + chain->depth - i) * 4514 sizeof(chain_hlocks[0])); 4515 } 4516 /* 4517 * Each lock class occurs at most once in a lock chain so once 4518 * we found a match we can break out of this loop. 4519 */ 4520 goto recalc; 4521 } 4522 /* Since the chain has not been modified, return. */ 4523 return; 4524 4525 recalc: 4526 chain_key = 0; 4527 for (i = chain->base; i < chain->base + chain->depth; i++) 4528 chain_key = iterate_chain_key(chain_key, chain_hlocks[i] + 1); 4529 if (chain->depth && chain->chain_key == chain_key) 4530 return; 4531 /* Overwrite the chain key for concurrent RCU readers. */ 4532 WRITE_ONCE(chain->chain_key, chain_key); 4533 /* 4534 * Note: calling hlist_del_rcu() from inside a 4535 * hlist_for_each_entry_rcu() loop is safe. 4536 */ 4537 hlist_del_rcu(&chain->entry); 4538 __set_bit(chain - lock_chains, pf->lock_chains_being_freed); 4539 if (chain->depth == 0) 4540 return; 4541 /* 4542 * If the modified lock chain matches an existing lock chain, drop 4543 * the modified lock chain. 4544 */ 4545 if (lookup_chain_cache(chain_key)) 4546 return; 4547 new_chain = alloc_lock_chain(); 4548 if (WARN_ON_ONCE(!new_chain)) { 4549 debug_locks_off(); 4550 return; 4551 } 4552 *new_chain = *chain; 4553 hlist_add_head_rcu(&new_chain->entry, chainhashentry(chain_key)); 4554 #endif 4555 } 4556 4557 /* Must be called with the graph lock held. */ 4558 static void remove_class_from_lock_chains(struct pending_free *pf, 4559 struct lock_class *class) 4560 { 4561 struct lock_chain *chain; 4562 struct hlist_head *head; 4563 int i; 4564 4565 for (i = 0; i < ARRAY_SIZE(chainhash_table); i++) { 4566 head = chainhash_table + i; 4567 hlist_for_each_entry_rcu(chain, head, entry) { 4568 remove_class_from_lock_chain(pf, chain, class); 4569 } 4570 } 4571 } 4572 4573 /* 4574 * Remove all references to a lock class. The caller must hold the graph lock. 4575 */ 4576 static void zap_class(struct pending_free *pf, struct lock_class *class) 4577 { 4578 struct lock_list *entry; 4579 int i; 4580 4581 WARN_ON_ONCE(!class->key); 4582 4583 /* 4584 * Remove all dependencies this lock is 4585 * involved in: 4586 */ 4587 for_each_set_bit(i, list_entries_in_use, ARRAY_SIZE(list_entries)) { 4588 entry = list_entries + i; 4589 if (entry->class != class && entry->links_to != class) 4590 continue; 4591 __clear_bit(i, list_entries_in_use); 4592 nr_list_entries--; 4593 list_del_rcu(&entry->entry); 4594 } 4595 if (list_empty(&class->locks_after) && 4596 list_empty(&class->locks_before)) { 4597 list_move_tail(&class->lock_entry, &pf->zapped); 4598 hlist_del_rcu(&class->hash_entry); 4599 WRITE_ONCE(class->key, NULL); 4600 WRITE_ONCE(class->name, NULL); 4601 nr_lock_classes--; 4602 } else { 4603 WARN_ONCE(true, "%s() failed for class %s\n", __func__, 4604 class->name); 4605 } 4606 4607 remove_class_from_lock_chains(pf, class); 4608 } 4609 4610 static void reinit_class(struct lock_class *class) 4611 { 4612 void *const p = class; 4613 const unsigned int offset = offsetof(struct lock_class, key); 4614 4615 WARN_ON_ONCE(!class->lock_entry.next); 4616 WARN_ON_ONCE(!list_empty(&class->locks_after)); 4617 WARN_ON_ONCE(!list_empty(&class->locks_before)); 4618 memset(p + offset, 0, sizeof(*class) - offset); 4619 WARN_ON_ONCE(!class->lock_entry.next); 4620 WARN_ON_ONCE(!list_empty(&class->locks_after)); 4621 WARN_ON_ONCE(!list_empty(&class->locks_before)); 4622 } 4623 4624 static inline int within(const void *addr, void *start, unsigned long size) 4625 { 4626 return addr >= start && addr < start + size; 4627 } 4628 4629 static bool inside_selftest(void) 4630 { 4631 return current == lockdep_selftest_task_struct; 4632 } 4633 4634 /* The caller must hold the graph lock. */ 4635 static struct pending_free *get_pending_free(void) 4636 { 4637 return delayed_free.pf + delayed_free.index; 4638 } 4639 4640 static void free_zapped_rcu(struct rcu_head *cb); 4641 4642 /* 4643 * Schedule an RCU callback if no RCU callback is pending. Must be called with 4644 * the graph lock held. 4645 */ 4646 static void call_rcu_zapped(struct pending_free *pf) 4647 { 4648 WARN_ON_ONCE(inside_selftest()); 4649 4650 if (list_empty(&pf->zapped)) 4651 return; 4652 4653 if (delayed_free.scheduled) 4654 return; 4655 4656 delayed_free.scheduled = true; 4657 4658 WARN_ON_ONCE(delayed_free.pf + delayed_free.index != pf); 4659 delayed_free.index ^= 1; 4660 4661 call_rcu(&delayed_free.rcu_head, free_zapped_rcu); 4662 } 4663 4664 /* The caller must hold the graph lock. May be called from RCU context. */ 4665 static void __free_zapped_classes(struct pending_free *pf) 4666 { 4667 struct lock_class *class; 4668 4669 check_data_structures(); 4670 4671 list_for_each_entry(class, &pf->zapped, lock_entry) 4672 reinit_class(class); 4673 4674 list_splice_init(&pf->zapped, &free_lock_classes); 4675 4676 #ifdef CONFIG_PROVE_LOCKING 4677 bitmap_andnot(lock_chains_in_use, lock_chains_in_use, 4678 pf->lock_chains_being_freed, ARRAY_SIZE(lock_chains)); 4679 bitmap_clear(pf->lock_chains_being_freed, 0, ARRAY_SIZE(lock_chains)); 4680 #endif 4681 } 4682 4683 static void free_zapped_rcu(struct rcu_head *ch) 4684 { 4685 struct pending_free *pf; 4686 unsigned long flags; 4687 4688 if (WARN_ON_ONCE(ch != &delayed_free.rcu_head)) 4689 return; 4690 4691 raw_local_irq_save(flags); 4692 if (!graph_lock()) 4693 goto out_irq; 4694 4695 /* closed head */ 4696 pf = delayed_free.pf + (delayed_free.index ^ 1); 4697 __free_zapped_classes(pf); 4698 delayed_free.scheduled = false; 4699 4700 /* 4701 * If there's anything on the open list, close and start a new callback. 4702 */ 4703 call_rcu_zapped(delayed_free.pf + delayed_free.index); 4704 4705 graph_unlock(); 4706 out_irq: 4707 raw_local_irq_restore(flags); 4708 } 4709 4710 /* 4711 * Remove all lock classes from the class hash table and from the 4712 * all_lock_classes list whose key or name is in the address range [start, 4713 * start + size). Move these lock classes to the zapped_classes list. Must 4714 * be called with the graph lock held. 4715 */ 4716 static void __lockdep_free_key_range(struct pending_free *pf, void *start, 4717 unsigned long size) 4718 { 4719 struct lock_class *class; 4720 struct hlist_head *head; 4721 int i; 4722 4723 /* Unhash all classes that were created by a module. */ 4724 for (i = 0; i < CLASSHASH_SIZE; i++) { 4725 head = classhash_table + i; 4726 hlist_for_each_entry_rcu(class, head, hash_entry) { 4727 if (!within(class->key, start, size) && 4728 !within(class->name, start, size)) 4729 continue; 4730 zap_class(pf, class); 4731 } 4732 } 4733 } 4734 4735 /* 4736 * Used in module.c to remove lock classes from memory that is going to be 4737 * freed; and possibly re-used by other modules. 4738 * 4739 * We will have had one synchronize_rcu() before getting here, so we're 4740 * guaranteed nobody will look up these exact classes -- they're properly dead 4741 * but still allocated. 4742 */ 4743 static void lockdep_free_key_range_reg(void *start, unsigned long size) 4744 { 4745 struct pending_free *pf; 4746 unsigned long flags; 4747 int locked; 4748 4749 init_data_structures_once(); 4750 4751 raw_local_irq_save(flags); 4752 locked = graph_lock(); 4753 if (!locked) 4754 goto out_irq; 4755 4756 pf = get_pending_free(); 4757 __lockdep_free_key_range(pf, start, size); 4758 call_rcu_zapped(pf); 4759 4760 graph_unlock(); 4761 out_irq: 4762 raw_local_irq_restore(flags); 4763 4764 /* 4765 * Wait for any possible iterators from look_up_lock_class() to pass 4766 * before continuing to free the memory they refer to. 4767 */ 4768 synchronize_rcu(); 4769 } 4770 4771 /* 4772 * Free all lockdep keys in the range [start, start+size). Does not sleep. 4773 * Ignores debug_locks. Must only be used by the lockdep selftests. 4774 */ 4775 static void lockdep_free_key_range_imm(void *start, unsigned long size) 4776 { 4777 struct pending_free *pf = delayed_free.pf; 4778 unsigned long flags; 4779 4780 init_data_structures_once(); 4781 4782 raw_local_irq_save(flags); 4783 arch_spin_lock(&lockdep_lock); 4784 __lockdep_free_key_range(pf, start, size); 4785 __free_zapped_classes(pf); 4786 arch_spin_unlock(&lockdep_lock); 4787 raw_local_irq_restore(flags); 4788 } 4789 4790 void lockdep_free_key_range(void *start, unsigned long size) 4791 { 4792 init_data_structures_once(); 4793 4794 if (inside_selftest()) 4795 lockdep_free_key_range_imm(start, size); 4796 else 4797 lockdep_free_key_range_reg(start, size); 4798 } 4799 4800 /* 4801 * Check whether any element of the @lock->class_cache[] array refers to a 4802 * registered lock class. The caller must hold either the graph lock or the 4803 * RCU read lock. 4804 */ 4805 static bool lock_class_cache_is_registered(struct lockdep_map *lock) 4806 { 4807 struct lock_class *class; 4808 struct hlist_head *head; 4809 int i, j; 4810 4811 for (i = 0; i < CLASSHASH_SIZE; i++) { 4812 head = classhash_table + i; 4813 hlist_for_each_entry_rcu(class, head, hash_entry) { 4814 for (j = 0; j < NR_LOCKDEP_CACHING_CLASSES; j++) 4815 if (lock->class_cache[j] == class) 4816 return true; 4817 } 4818 } 4819 return false; 4820 } 4821 4822 /* The caller must hold the graph lock. Does not sleep. */ 4823 static void __lockdep_reset_lock(struct pending_free *pf, 4824 struct lockdep_map *lock) 4825 { 4826 struct lock_class *class; 4827 int j; 4828 4829 /* 4830 * Remove all classes this lock might have: 4831 */ 4832 for (j = 0; j < MAX_LOCKDEP_SUBCLASSES; j++) { 4833 /* 4834 * If the class exists we look it up and zap it: 4835 */ 4836 class = look_up_lock_class(lock, j); 4837 if (class) 4838 zap_class(pf, class); 4839 } 4840 /* 4841 * Debug check: in the end all mapped classes should 4842 * be gone. 4843 */ 4844 if (WARN_ON_ONCE(lock_class_cache_is_registered(lock))) 4845 debug_locks_off(); 4846 } 4847 4848 /* 4849 * Remove all information lockdep has about a lock if debug_locks == 1. Free 4850 * released data structures from RCU context. 4851 */ 4852 static void lockdep_reset_lock_reg(struct lockdep_map *lock) 4853 { 4854 struct pending_free *pf; 4855 unsigned long flags; 4856 int locked; 4857 4858 raw_local_irq_save(flags); 4859 locked = graph_lock(); 4860 if (!locked) 4861 goto out_irq; 4862 4863 pf = get_pending_free(); 4864 __lockdep_reset_lock(pf, lock); 4865 call_rcu_zapped(pf); 4866 4867 graph_unlock(); 4868 out_irq: 4869 raw_local_irq_restore(flags); 4870 } 4871 4872 /* 4873 * Reset a lock. Does not sleep. Ignores debug_locks. Must only be used by the 4874 * lockdep selftests. 4875 */ 4876 static void lockdep_reset_lock_imm(struct lockdep_map *lock) 4877 { 4878 struct pending_free *pf = delayed_free.pf; 4879 unsigned long flags; 4880 4881 raw_local_irq_save(flags); 4882 arch_spin_lock(&lockdep_lock); 4883 __lockdep_reset_lock(pf, lock); 4884 __free_zapped_classes(pf); 4885 arch_spin_unlock(&lockdep_lock); 4886 raw_local_irq_restore(flags); 4887 } 4888 4889 void lockdep_reset_lock(struct lockdep_map *lock) 4890 { 4891 init_data_structures_once(); 4892 4893 if (inside_selftest()) 4894 lockdep_reset_lock_imm(lock); 4895 else 4896 lockdep_reset_lock_reg(lock); 4897 } 4898 4899 /* Unregister a dynamically allocated key. */ 4900 void lockdep_unregister_key(struct lock_class_key *key) 4901 { 4902 struct hlist_head *hash_head = keyhashentry(key); 4903 struct lock_class_key *k; 4904 struct pending_free *pf; 4905 unsigned long flags; 4906 bool found = false; 4907 4908 might_sleep(); 4909 4910 if (WARN_ON_ONCE(static_obj(key))) 4911 return; 4912 4913 raw_local_irq_save(flags); 4914 if (!graph_lock()) 4915 goto out_irq; 4916 4917 pf = get_pending_free(); 4918 hlist_for_each_entry_rcu(k, hash_head, hash_entry) { 4919 if (k == key) { 4920 hlist_del_rcu(&k->hash_entry); 4921 found = true; 4922 break; 4923 } 4924 } 4925 WARN_ON_ONCE(!found); 4926 __lockdep_free_key_range(pf, key, 1); 4927 call_rcu_zapped(pf); 4928 graph_unlock(); 4929 out_irq: 4930 raw_local_irq_restore(flags); 4931 4932 /* Wait until is_dynamic_key() has finished accessing k->hash_entry. */ 4933 synchronize_rcu(); 4934 } 4935 EXPORT_SYMBOL_GPL(lockdep_unregister_key); 4936 4937 void __init lockdep_init(void) 4938 { 4939 printk("Lock dependency validator: Copyright (c) 2006 Red Hat, Inc., Ingo Molnar\n"); 4940 4941 printk("... MAX_LOCKDEP_SUBCLASSES: %lu\n", MAX_LOCKDEP_SUBCLASSES); 4942 printk("... MAX_LOCK_DEPTH: %lu\n", MAX_LOCK_DEPTH); 4943 printk("... MAX_LOCKDEP_KEYS: %lu\n", MAX_LOCKDEP_KEYS); 4944 printk("... CLASSHASH_SIZE: %lu\n", CLASSHASH_SIZE); 4945 printk("... MAX_LOCKDEP_ENTRIES: %lu\n", MAX_LOCKDEP_ENTRIES); 4946 printk("... MAX_LOCKDEP_CHAINS: %lu\n", MAX_LOCKDEP_CHAINS); 4947 printk("... CHAINHASH_SIZE: %lu\n", CHAINHASH_SIZE); 4948 4949 printk(" memory used by lock dependency info: %zu kB\n", 4950 (sizeof(lock_classes) + 4951 sizeof(classhash_table) + 4952 sizeof(list_entries) + 4953 sizeof(list_entries_in_use) + 4954 sizeof(chainhash_table) + 4955 sizeof(delayed_free) 4956 #ifdef CONFIG_PROVE_LOCKING 4957 + sizeof(lock_cq) 4958 + sizeof(lock_chains) 4959 + sizeof(lock_chains_in_use) 4960 + sizeof(chain_hlocks) 4961 #endif 4962 ) / 1024 4963 ); 4964 4965 printk(" per task-struct memory footprint: %zu bytes\n", 4966 sizeof(((struct task_struct *)NULL)->held_locks)); 4967 } 4968 4969 static void 4970 print_freed_lock_bug(struct task_struct *curr, const void *mem_from, 4971 const void *mem_to, struct held_lock *hlock) 4972 { 4973 if (!debug_locks_off()) 4974 return; 4975 if (debug_locks_silent) 4976 return; 4977 4978 pr_warn("\n"); 4979 pr_warn("=========================\n"); 4980 pr_warn("WARNING: held lock freed!\n"); 4981 print_kernel_ident(); 4982 pr_warn("-------------------------\n"); 4983 pr_warn("%s/%d is freeing memory %px-%px, with a lock still held there!\n", 4984 curr->comm, task_pid_nr(curr), mem_from, mem_to-1); 4985 print_lock(hlock); 4986 lockdep_print_held_locks(curr); 4987 4988 pr_warn("\nstack backtrace:\n"); 4989 dump_stack(); 4990 } 4991 4992 static inline int not_in_range(const void* mem_from, unsigned long mem_len, 4993 const void* lock_from, unsigned long lock_len) 4994 { 4995 return lock_from + lock_len <= mem_from || 4996 mem_from + mem_len <= lock_from; 4997 } 4998 4999 /* 5000 * Called when kernel memory is freed (or unmapped), or if a lock 5001 * is destroyed or reinitialized - this code checks whether there is 5002 * any held lock in the memory range of <from> to <to>: 5003 */ 5004 void debug_check_no_locks_freed(const void *mem_from, unsigned long mem_len) 5005 { 5006 struct task_struct *curr = current; 5007 struct held_lock *hlock; 5008 unsigned long flags; 5009 int i; 5010 5011 if (unlikely(!debug_locks)) 5012 return; 5013 5014 raw_local_irq_save(flags); 5015 for (i = 0; i < curr->lockdep_depth; i++) { 5016 hlock = curr->held_locks + i; 5017 5018 if (not_in_range(mem_from, mem_len, hlock->instance, 5019 sizeof(*hlock->instance))) 5020 continue; 5021 5022 print_freed_lock_bug(curr, mem_from, mem_from + mem_len, hlock); 5023 break; 5024 } 5025 raw_local_irq_restore(flags); 5026 } 5027 EXPORT_SYMBOL_GPL(debug_check_no_locks_freed); 5028 5029 static void print_held_locks_bug(void) 5030 { 5031 if (!debug_locks_off()) 5032 return; 5033 if (debug_locks_silent) 5034 return; 5035 5036 pr_warn("\n"); 5037 pr_warn("====================================\n"); 5038 pr_warn("WARNING: %s/%d still has locks held!\n", 5039 current->comm, task_pid_nr(current)); 5040 print_kernel_ident(); 5041 pr_warn("------------------------------------\n"); 5042 lockdep_print_held_locks(current); 5043 pr_warn("\nstack backtrace:\n"); 5044 dump_stack(); 5045 } 5046 5047 void debug_check_no_locks_held(void) 5048 { 5049 if (unlikely(current->lockdep_depth > 0)) 5050 print_held_locks_bug(); 5051 } 5052 EXPORT_SYMBOL_GPL(debug_check_no_locks_held); 5053 5054 #ifdef __KERNEL__ 5055 void debug_show_all_locks(void) 5056 { 5057 struct task_struct *g, *p; 5058 5059 if (unlikely(!debug_locks)) { 5060 pr_warn("INFO: lockdep is turned off.\n"); 5061 return; 5062 } 5063 pr_warn("\nShowing all locks held in the system:\n"); 5064 5065 rcu_read_lock(); 5066 for_each_process_thread(g, p) { 5067 if (!p->lockdep_depth) 5068 continue; 5069 lockdep_print_held_locks(p); 5070 touch_nmi_watchdog(); 5071 touch_all_softlockup_watchdogs(); 5072 } 5073 rcu_read_unlock(); 5074 5075 pr_warn("\n"); 5076 pr_warn("=============================================\n\n"); 5077 } 5078 EXPORT_SYMBOL_GPL(debug_show_all_locks); 5079 #endif 5080 5081 /* 5082 * Careful: only use this function if you are sure that 5083 * the task cannot run in parallel! 5084 */ 5085 void debug_show_held_locks(struct task_struct *task) 5086 { 5087 if (unlikely(!debug_locks)) { 5088 printk("INFO: lockdep is turned off.\n"); 5089 return; 5090 } 5091 lockdep_print_held_locks(task); 5092 } 5093 EXPORT_SYMBOL_GPL(debug_show_held_locks); 5094 5095 asmlinkage __visible void lockdep_sys_exit(void) 5096 { 5097 struct task_struct *curr = current; 5098 5099 if (unlikely(curr->lockdep_depth)) { 5100 if (!debug_locks_off()) 5101 return; 5102 pr_warn("\n"); 5103 pr_warn("================================================\n"); 5104 pr_warn("WARNING: lock held when returning to user space!\n"); 5105 print_kernel_ident(); 5106 pr_warn("------------------------------------------------\n"); 5107 pr_warn("%s/%d is leaving the kernel with locks still held!\n", 5108 curr->comm, curr->pid); 5109 lockdep_print_held_locks(curr); 5110 } 5111 5112 /* 5113 * The lock history for each syscall should be independent. So wipe the 5114 * slate clean on return to userspace. 5115 */ 5116 lockdep_invariant_state(false); 5117 } 5118 5119 void lockdep_rcu_suspicious(const char *file, const int line, const char *s) 5120 { 5121 struct task_struct *curr = current; 5122 5123 /* Note: the following can be executed concurrently, so be careful. */ 5124 pr_warn("\n"); 5125 pr_warn("=============================\n"); 5126 pr_warn("WARNING: suspicious RCU usage\n"); 5127 print_kernel_ident(); 5128 pr_warn("-----------------------------\n"); 5129 pr_warn("%s:%d %s!\n", file, line, s); 5130 pr_warn("\nother info that might help us debug this:\n\n"); 5131 pr_warn("\n%srcu_scheduler_active = %d, debug_locks = %d\n", 5132 !rcu_lockdep_current_cpu_online() 5133 ? "RCU used illegally from offline CPU!\n" 5134 : !rcu_is_watching() 5135 ? "RCU used illegally from idle CPU!\n" 5136 : "", 5137 rcu_scheduler_active, debug_locks); 5138 5139 /* 5140 * If a CPU is in the RCU-free window in idle (ie: in the section 5141 * between rcu_idle_enter() and rcu_idle_exit(), then RCU 5142 * considers that CPU to be in an "extended quiescent state", 5143 * which means that RCU will be completely ignoring that CPU. 5144 * Therefore, rcu_read_lock() and friends have absolutely no 5145 * effect on a CPU running in that state. In other words, even if 5146 * such an RCU-idle CPU has called rcu_read_lock(), RCU might well 5147 * delete data structures out from under it. RCU really has no 5148 * choice here: we need to keep an RCU-free window in idle where 5149 * the CPU may possibly enter into low power mode. This way we can 5150 * notice an extended quiescent state to other CPUs that started a grace 5151 * period. Otherwise we would delay any grace period as long as we run 5152 * in the idle task. 5153 * 5154 * So complain bitterly if someone does call rcu_read_lock(), 5155 * rcu_read_lock_bh() and so on from extended quiescent states. 5156 */ 5157 if (!rcu_is_watching()) 5158 pr_warn("RCU used illegally from extended quiescent state!\n"); 5159 5160 lockdep_print_held_locks(curr); 5161 pr_warn("\nstack backtrace:\n"); 5162 dump_stack(); 5163 } 5164 EXPORT_SYMBOL_GPL(lockdep_rcu_suspicious); 5165