1 // SPDX-License-Identifier: GPL-2.0-only 2 /* 3 * kernel/lockdep.c 4 * 5 * Runtime locking correctness validator 6 * 7 * Started by Ingo Molnar: 8 * 9 * Copyright (C) 2006,2007 Red Hat, Inc., Ingo Molnar <mingo@redhat.com> 10 * Copyright (C) 2007 Red Hat, Inc., Peter Zijlstra 11 * 12 * this code maps all the lock dependencies as they occur in a live kernel 13 * and will warn about the following classes of locking bugs: 14 * 15 * - lock inversion scenarios 16 * - circular lock dependencies 17 * - hardirq/softirq safe/unsafe locking bugs 18 * 19 * Bugs are reported even if the current locking scenario does not cause 20 * any deadlock at this point. 21 * 22 * I.e. if anytime in the past two locks were taken in a different order, 23 * even if it happened for another task, even if those were different 24 * locks (but of the same class as this lock), this code will detect it. 25 * 26 * Thanks to Arjan van de Ven for coming up with the initial idea of 27 * mapping lock dependencies runtime. 28 */ 29 #define DISABLE_BRANCH_PROFILING 30 #include <linux/mutex.h> 31 #include <linux/sched.h> 32 #include <linux/sched/clock.h> 33 #include <linux/sched/task.h> 34 #include <linux/sched/mm.h> 35 #include <linux/delay.h> 36 #include <linux/module.h> 37 #include <linux/proc_fs.h> 38 #include <linux/seq_file.h> 39 #include <linux/spinlock.h> 40 #include <linux/kallsyms.h> 41 #include <linux/interrupt.h> 42 #include <linux/stacktrace.h> 43 #include <linux/debug_locks.h> 44 #include <linux/irqflags.h> 45 #include <linux/utsname.h> 46 #include <linux/hash.h> 47 #include <linux/ftrace.h> 48 #include <linux/stringify.h> 49 #include <linux/bitmap.h> 50 #include <linux/bitops.h> 51 #include <linux/gfp.h> 52 #include <linux/random.h> 53 #include <linux/jhash.h> 54 #include <linux/nmi.h> 55 #include <linux/rcupdate.h> 56 #include <linux/kprobes.h> 57 #include <linux/lockdep.h> 58 59 #include <asm/sections.h> 60 61 #include "lockdep_internals.h" 62 63 #include <trace/events/lock.h> 64 65 #ifdef CONFIG_PROVE_LOCKING 66 static int prove_locking = 1; 67 module_param(prove_locking, int, 0644); 68 #else 69 #define prove_locking 0 70 #endif 71 72 #ifdef CONFIG_LOCK_STAT 73 static int lock_stat = 1; 74 module_param(lock_stat, int, 0644); 75 #else 76 #define lock_stat 0 77 #endif 78 79 #ifdef CONFIG_SYSCTL 80 static struct ctl_table kern_lockdep_table[] = { 81 #ifdef CONFIG_PROVE_LOCKING 82 { 83 .procname = "prove_locking", 84 .data = &prove_locking, 85 .maxlen = sizeof(int), 86 .mode = 0644, 87 .proc_handler = proc_dointvec, 88 }, 89 #endif /* CONFIG_PROVE_LOCKING */ 90 #ifdef CONFIG_LOCK_STAT 91 { 92 .procname = "lock_stat", 93 .data = &lock_stat, 94 .maxlen = sizeof(int), 95 .mode = 0644, 96 .proc_handler = proc_dointvec, 97 }, 98 #endif /* CONFIG_LOCK_STAT */ 99 { } 100 }; 101 102 static __init int kernel_lockdep_sysctls_init(void) 103 { 104 register_sysctl_init("kernel", kern_lockdep_table); 105 return 0; 106 } 107 late_initcall(kernel_lockdep_sysctls_init); 108 #endif /* CONFIG_SYSCTL */ 109 110 DEFINE_PER_CPU(unsigned int, lockdep_recursion); 111 EXPORT_PER_CPU_SYMBOL_GPL(lockdep_recursion); 112 113 static __always_inline bool lockdep_enabled(void) 114 { 115 if (!debug_locks) 116 return false; 117 118 if (this_cpu_read(lockdep_recursion)) 119 return false; 120 121 if (current->lockdep_recursion) 122 return false; 123 124 return true; 125 } 126 127 /* 128 * lockdep_lock: protects the lockdep graph, the hashes and the 129 * class/list/hash allocators. 130 * 131 * This is one of the rare exceptions where it's justified 132 * to use a raw spinlock - we really dont want the spinlock 133 * code to recurse back into the lockdep code... 134 */ 135 static arch_spinlock_t __lock = (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED; 136 static struct task_struct *__owner; 137 138 static inline void lockdep_lock(void) 139 { 140 DEBUG_LOCKS_WARN_ON(!irqs_disabled()); 141 142 __this_cpu_inc(lockdep_recursion); 143 arch_spin_lock(&__lock); 144 __owner = current; 145 } 146 147 static inline void lockdep_unlock(void) 148 { 149 DEBUG_LOCKS_WARN_ON(!irqs_disabled()); 150 151 if (debug_locks && DEBUG_LOCKS_WARN_ON(__owner != current)) 152 return; 153 154 __owner = NULL; 155 arch_spin_unlock(&__lock); 156 __this_cpu_dec(lockdep_recursion); 157 } 158 159 static inline bool lockdep_assert_locked(void) 160 { 161 return DEBUG_LOCKS_WARN_ON(__owner != current); 162 } 163 164 static struct task_struct *lockdep_selftest_task_struct; 165 166 167 static int graph_lock(void) 168 { 169 lockdep_lock(); 170 /* 171 * Make sure that if another CPU detected a bug while 172 * walking the graph we dont change it (while the other 173 * CPU is busy printing out stuff with the graph lock 174 * dropped already) 175 */ 176 if (!debug_locks) { 177 lockdep_unlock(); 178 return 0; 179 } 180 return 1; 181 } 182 183 static inline void graph_unlock(void) 184 { 185 lockdep_unlock(); 186 } 187 188 /* 189 * Turn lock debugging off and return with 0 if it was off already, 190 * and also release the graph lock: 191 */ 192 static inline int debug_locks_off_graph_unlock(void) 193 { 194 int ret = debug_locks_off(); 195 196 lockdep_unlock(); 197 198 return ret; 199 } 200 201 unsigned long nr_list_entries; 202 static struct lock_list list_entries[MAX_LOCKDEP_ENTRIES]; 203 static DECLARE_BITMAP(list_entries_in_use, MAX_LOCKDEP_ENTRIES); 204 205 /* 206 * All data structures here are protected by the global debug_lock. 207 * 208 * nr_lock_classes is the number of elements of lock_classes[] that is 209 * in use. 210 */ 211 #define KEYHASH_BITS (MAX_LOCKDEP_KEYS_BITS - 1) 212 #define KEYHASH_SIZE (1UL << KEYHASH_BITS) 213 static struct hlist_head lock_keys_hash[KEYHASH_SIZE]; 214 unsigned long nr_lock_classes; 215 unsigned long nr_zapped_classes; 216 unsigned long max_lock_class_idx; 217 struct lock_class lock_classes[MAX_LOCKDEP_KEYS]; 218 DECLARE_BITMAP(lock_classes_in_use, MAX_LOCKDEP_KEYS); 219 220 static inline struct lock_class *hlock_class(struct held_lock *hlock) 221 { 222 unsigned int class_idx = hlock->class_idx; 223 224 /* Don't re-read hlock->class_idx, can't use READ_ONCE() on bitfield */ 225 barrier(); 226 227 if (!test_bit(class_idx, lock_classes_in_use)) { 228 /* 229 * Someone passed in garbage, we give up. 230 */ 231 DEBUG_LOCKS_WARN_ON(1); 232 return NULL; 233 } 234 235 /* 236 * At this point, if the passed hlock->class_idx is still garbage, 237 * we just have to live with it 238 */ 239 return lock_classes + class_idx; 240 } 241 242 #ifdef CONFIG_LOCK_STAT 243 static DEFINE_PER_CPU(struct lock_class_stats[MAX_LOCKDEP_KEYS], cpu_lock_stats); 244 245 static inline u64 lockstat_clock(void) 246 { 247 return local_clock(); 248 } 249 250 static int lock_point(unsigned long points[], unsigned long ip) 251 { 252 int i; 253 254 for (i = 0; i < LOCKSTAT_POINTS; i++) { 255 if (points[i] == 0) { 256 points[i] = ip; 257 break; 258 } 259 if (points[i] == ip) 260 break; 261 } 262 263 return i; 264 } 265 266 static void lock_time_inc(struct lock_time *lt, u64 time) 267 { 268 if (time > lt->max) 269 lt->max = time; 270 271 if (time < lt->min || !lt->nr) 272 lt->min = time; 273 274 lt->total += time; 275 lt->nr++; 276 } 277 278 static inline void lock_time_add(struct lock_time *src, struct lock_time *dst) 279 { 280 if (!src->nr) 281 return; 282 283 if (src->max > dst->max) 284 dst->max = src->max; 285 286 if (src->min < dst->min || !dst->nr) 287 dst->min = src->min; 288 289 dst->total += src->total; 290 dst->nr += src->nr; 291 } 292 293 struct lock_class_stats lock_stats(struct lock_class *class) 294 { 295 struct lock_class_stats stats; 296 int cpu, i; 297 298 memset(&stats, 0, sizeof(struct lock_class_stats)); 299 for_each_possible_cpu(cpu) { 300 struct lock_class_stats *pcs = 301 &per_cpu(cpu_lock_stats, cpu)[class - lock_classes]; 302 303 for (i = 0; i < ARRAY_SIZE(stats.contention_point); i++) 304 stats.contention_point[i] += pcs->contention_point[i]; 305 306 for (i = 0; i < ARRAY_SIZE(stats.contending_point); i++) 307 stats.contending_point[i] += pcs->contending_point[i]; 308 309 lock_time_add(&pcs->read_waittime, &stats.read_waittime); 310 lock_time_add(&pcs->write_waittime, &stats.write_waittime); 311 312 lock_time_add(&pcs->read_holdtime, &stats.read_holdtime); 313 lock_time_add(&pcs->write_holdtime, &stats.write_holdtime); 314 315 for (i = 0; i < ARRAY_SIZE(stats.bounces); i++) 316 stats.bounces[i] += pcs->bounces[i]; 317 } 318 319 return stats; 320 } 321 322 void clear_lock_stats(struct lock_class *class) 323 { 324 int cpu; 325 326 for_each_possible_cpu(cpu) { 327 struct lock_class_stats *cpu_stats = 328 &per_cpu(cpu_lock_stats, cpu)[class - lock_classes]; 329 330 memset(cpu_stats, 0, sizeof(struct lock_class_stats)); 331 } 332 memset(class->contention_point, 0, sizeof(class->contention_point)); 333 memset(class->contending_point, 0, sizeof(class->contending_point)); 334 } 335 336 static struct lock_class_stats *get_lock_stats(struct lock_class *class) 337 { 338 return &this_cpu_ptr(cpu_lock_stats)[class - lock_classes]; 339 } 340 341 static void lock_release_holdtime(struct held_lock *hlock) 342 { 343 struct lock_class_stats *stats; 344 u64 holdtime; 345 346 if (!lock_stat) 347 return; 348 349 holdtime = lockstat_clock() - hlock->holdtime_stamp; 350 351 stats = get_lock_stats(hlock_class(hlock)); 352 if (hlock->read) 353 lock_time_inc(&stats->read_holdtime, holdtime); 354 else 355 lock_time_inc(&stats->write_holdtime, holdtime); 356 } 357 #else 358 static inline void lock_release_holdtime(struct held_lock *hlock) 359 { 360 } 361 #endif 362 363 /* 364 * We keep a global list of all lock classes. The list is only accessed with 365 * the lockdep spinlock lock held. free_lock_classes is a list with free 366 * elements. These elements are linked together by the lock_entry member in 367 * struct lock_class. 368 */ 369 static LIST_HEAD(all_lock_classes); 370 static LIST_HEAD(free_lock_classes); 371 372 /** 373 * struct pending_free - information about data structures about to be freed 374 * @zapped: Head of a list with struct lock_class elements. 375 * @lock_chains_being_freed: Bitmap that indicates which lock_chains[] elements 376 * are about to be freed. 377 */ 378 struct pending_free { 379 struct list_head zapped; 380 DECLARE_BITMAP(lock_chains_being_freed, MAX_LOCKDEP_CHAINS); 381 }; 382 383 /** 384 * struct delayed_free - data structures used for delayed freeing 385 * 386 * A data structure for delayed freeing of data structures that may be 387 * accessed by RCU readers at the time these were freed. 388 * 389 * @rcu_head: Used to schedule an RCU callback for freeing data structures. 390 * @index: Index of @pf to which freed data structures are added. 391 * @scheduled: Whether or not an RCU callback has been scheduled. 392 * @pf: Array with information about data structures about to be freed. 393 */ 394 static struct delayed_free { 395 struct rcu_head rcu_head; 396 int index; 397 int scheduled; 398 struct pending_free pf[2]; 399 } delayed_free; 400 401 /* 402 * The lockdep classes are in a hash-table as well, for fast lookup: 403 */ 404 #define CLASSHASH_BITS (MAX_LOCKDEP_KEYS_BITS - 1) 405 #define CLASSHASH_SIZE (1UL << CLASSHASH_BITS) 406 #define __classhashfn(key) hash_long((unsigned long)key, CLASSHASH_BITS) 407 #define classhashentry(key) (classhash_table + __classhashfn((key))) 408 409 static struct hlist_head classhash_table[CLASSHASH_SIZE]; 410 411 /* 412 * We put the lock dependency chains into a hash-table as well, to cache 413 * their existence: 414 */ 415 #define CHAINHASH_BITS (MAX_LOCKDEP_CHAINS_BITS-1) 416 #define CHAINHASH_SIZE (1UL << CHAINHASH_BITS) 417 #define __chainhashfn(chain) hash_long(chain, CHAINHASH_BITS) 418 #define chainhashentry(chain) (chainhash_table + __chainhashfn((chain))) 419 420 static struct hlist_head chainhash_table[CHAINHASH_SIZE]; 421 422 /* 423 * the id of held_lock 424 */ 425 static inline u16 hlock_id(struct held_lock *hlock) 426 { 427 BUILD_BUG_ON(MAX_LOCKDEP_KEYS_BITS + 2 > 16); 428 429 return (hlock->class_idx | (hlock->read << MAX_LOCKDEP_KEYS_BITS)); 430 } 431 432 static inline unsigned int chain_hlock_class_idx(u16 hlock_id) 433 { 434 return hlock_id & (MAX_LOCKDEP_KEYS - 1); 435 } 436 437 /* 438 * The hash key of the lock dependency chains is a hash itself too: 439 * it's a hash of all locks taken up to that lock, including that lock. 440 * It's a 64-bit hash, because it's important for the keys to be 441 * unique. 442 */ 443 static inline u64 iterate_chain_key(u64 key, u32 idx) 444 { 445 u32 k0 = key, k1 = key >> 32; 446 447 __jhash_mix(idx, k0, k1); /* Macro that modifies arguments! */ 448 449 return k0 | (u64)k1 << 32; 450 } 451 452 void lockdep_init_task(struct task_struct *task) 453 { 454 task->lockdep_depth = 0; /* no locks held yet */ 455 task->curr_chain_key = INITIAL_CHAIN_KEY; 456 task->lockdep_recursion = 0; 457 } 458 459 static __always_inline void lockdep_recursion_inc(void) 460 { 461 __this_cpu_inc(lockdep_recursion); 462 } 463 464 static __always_inline void lockdep_recursion_finish(void) 465 { 466 if (WARN_ON_ONCE(__this_cpu_dec_return(lockdep_recursion))) 467 __this_cpu_write(lockdep_recursion, 0); 468 } 469 470 void lockdep_set_selftest_task(struct task_struct *task) 471 { 472 lockdep_selftest_task_struct = task; 473 } 474 475 /* 476 * Debugging switches: 477 */ 478 479 #define VERBOSE 0 480 #define VERY_VERBOSE 0 481 482 #if VERBOSE 483 # define HARDIRQ_VERBOSE 1 484 # define SOFTIRQ_VERBOSE 1 485 #else 486 # define HARDIRQ_VERBOSE 0 487 # define SOFTIRQ_VERBOSE 0 488 #endif 489 490 #if VERBOSE || HARDIRQ_VERBOSE || SOFTIRQ_VERBOSE 491 /* 492 * Quick filtering for interesting events: 493 */ 494 static int class_filter(struct lock_class *class) 495 { 496 #if 0 497 /* Example */ 498 if (class->name_version == 1 && 499 !strcmp(class->name, "lockname")) 500 return 1; 501 if (class->name_version == 1 && 502 !strcmp(class->name, "&struct->lockfield")) 503 return 1; 504 #endif 505 /* Filter everything else. 1 would be to allow everything else */ 506 return 0; 507 } 508 #endif 509 510 static int verbose(struct lock_class *class) 511 { 512 #if VERBOSE 513 return class_filter(class); 514 #endif 515 return 0; 516 } 517 518 static void print_lockdep_off(const char *bug_msg) 519 { 520 printk(KERN_DEBUG "%s\n", bug_msg); 521 printk(KERN_DEBUG "turning off the locking correctness validator.\n"); 522 #ifdef CONFIG_LOCK_STAT 523 printk(KERN_DEBUG "Please attach the output of /proc/lock_stat to the bug report\n"); 524 #endif 525 } 526 527 unsigned long nr_stack_trace_entries; 528 529 #ifdef CONFIG_PROVE_LOCKING 530 /** 531 * struct lock_trace - single stack backtrace 532 * @hash_entry: Entry in a stack_trace_hash[] list. 533 * @hash: jhash() of @entries. 534 * @nr_entries: Number of entries in @entries. 535 * @entries: Actual stack backtrace. 536 */ 537 struct lock_trace { 538 struct hlist_node hash_entry; 539 u32 hash; 540 u32 nr_entries; 541 unsigned long entries[] __aligned(sizeof(unsigned long)); 542 }; 543 #define LOCK_TRACE_SIZE_IN_LONGS \ 544 (sizeof(struct lock_trace) / sizeof(unsigned long)) 545 /* 546 * Stack-trace: sequence of lock_trace structures. Protected by the graph_lock. 547 */ 548 static unsigned long stack_trace[MAX_STACK_TRACE_ENTRIES]; 549 static struct hlist_head stack_trace_hash[STACK_TRACE_HASH_SIZE]; 550 551 static bool traces_identical(struct lock_trace *t1, struct lock_trace *t2) 552 { 553 return t1->hash == t2->hash && t1->nr_entries == t2->nr_entries && 554 memcmp(t1->entries, t2->entries, 555 t1->nr_entries * sizeof(t1->entries[0])) == 0; 556 } 557 558 static struct lock_trace *save_trace(void) 559 { 560 struct lock_trace *trace, *t2; 561 struct hlist_head *hash_head; 562 u32 hash; 563 int max_entries; 564 565 BUILD_BUG_ON_NOT_POWER_OF_2(STACK_TRACE_HASH_SIZE); 566 BUILD_BUG_ON(LOCK_TRACE_SIZE_IN_LONGS >= MAX_STACK_TRACE_ENTRIES); 567 568 trace = (struct lock_trace *)(stack_trace + nr_stack_trace_entries); 569 max_entries = MAX_STACK_TRACE_ENTRIES - nr_stack_trace_entries - 570 LOCK_TRACE_SIZE_IN_LONGS; 571 572 if (max_entries <= 0) { 573 if (!debug_locks_off_graph_unlock()) 574 return NULL; 575 576 print_lockdep_off("BUG: MAX_STACK_TRACE_ENTRIES too low!"); 577 dump_stack(); 578 579 return NULL; 580 } 581 trace->nr_entries = stack_trace_save(trace->entries, max_entries, 3); 582 583 hash = jhash(trace->entries, trace->nr_entries * 584 sizeof(trace->entries[0]), 0); 585 trace->hash = hash; 586 hash_head = stack_trace_hash + (hash & (STACK_TRACE_HASH_SIZE - 1)); 587 hlist_for_each_entry(t2, hash_head, hash_entry) { 588 if (traces_identical(trace, t2)) 589 return t2; 590 } 591 nr_stack_trace_entries += LOCK_TRACE_SIZE_IN_LONGS + trace->nr_entries; 592 hlist_add_head(&trace->hash_entry, hash_head); 593 594 return trace; 595 } 596 597 /* Return the number of stack traces in the stack_trace[] array. */ 598 u64 lockdep_stack_trace_count(void) 599 { 600 struct lock_trace *trace; 601 u64 c = 0; 602 int i; 603 604 for (i = 0; i < ARRAY_SIZE(stack_trace_hash); i++) { 605 hlist_for_each_entry(trace, &stack_trace_hash[i], hash_entry) { 606 c++; 607 } 608 } 609 610 return c; 611 } 612 613 /* Return the number of stack hash chains that have at least one stack trace. */ 614 u64 lockdep_stack_hash_count(void) 615 { 616 u64 c = 0; 617 int i; 618 619 for (i = 0; i < ARRAY_SIZE(stack_trace_hash); i++) 620 if (!hlist_empty(&stack_trace_hash[i])) 621 c++; 622 623 return c; 624 } 625 #endif 626 627 unsigned int nr_hardirq_chains; 628 unsigned int nr_softirq_chains; 629 unsigned int nr_process_chains; 630 unsigned int max_lockdep_depth; 631 632 #ifdef CONFIG_DEBUG_LOCKDEP 633 /* 634 * Various lockdep statistics: 635 */ 636 DEFINE_PER_CPU(struct lockdep_stats, lockdep_stats); 637 #endif 638 639 #ifdef CONFIG_PROVE_LOCKING 640 /* 641 * Locking printouts: 642 */ 643 644 #define __USAGE(__STATE) \ 645 [LOCK_USED_IN_##__STATE] = "IN-"__stringify(__STATE)"-W", \ 646 [LOCK_ENABLED_##__STATE] = __stringify(__STATE)"-ON-W", \ 647 [LOCK_USED_IN_##__STATE##_READ] = "IN-"__stringify(__STATE)"-R",\ 648 [LOCK_ENABLED_##__STATE##_READ] = __stringify(__STATE)"-ON-R", 649 650 static const char *usage_str[] = 651 { 652 #define LOCKDEP_STATE(__STATE) __USAGE(__STATE) 653 #include "lockdep_states.h" 654 #undef LOCKDEP_STATE 655 [LOCK_USED] = "INITIAL USE", 656 [LOCK_USED_READ] = "INITIAL READ USE", 657 /* abused as string storage for verify_lock_unused() */ 658 [LOCK_USAGE_STATES] = "IN-NMI", 659 }; 660 #endif 661 662 const char *__get_key_name(const struct lockdep_subclass_key *key, char *str) 663 { 664 return kallsyms_lookup((unsigned long)key, NULL, NULL, NULL, str); 665 } 666 667 static inline unsigned long lock_flag(enum lock_usage_bit bit) 668 { 669 return 1UL << bit; 670 } 671 672 static char get_usage_char(struct lock_class *class, enum lock_usage_bit bit) 673 { 674 /* 675 * The usage character defaults to '.' (i.e., irqs disabled and not in 676 * irq context), which is the safest usage category. 677 */ 678 char c = '.'; 679 680 /* 681 * The order of the following usage checks matters, which will 682 * result in the outcome character as follows: 683 * 684 * - '+': irq is enabled and not in irq context 685 * - '-': in irq context and irq is disabled 686 * - '?': in irq context and irq is enabled 687 */ 688 if (class->usage_mask & lock_flag(bit + LOCK_USAGE_DIR_MASK)) { 689 c = '+'; 690 if (class->usage_mask & lock_flag(bit)) 691 c = '?'; 692 } else if (class->usage_mask & lock_flag(bit)) 693 c = '-'; 694 695 return c; 696 } 697 698 void get_usage_chars(struct lock_class *class, char usage[LOCK_USAGE_CHARS]) 699 { 700 int i = 0; 701 702 #define LOCKDEP_STATE(__STATE) \ 703 usage[i++] = get_usage_char(class, LOCK_USED_IN_##__STATE); \ 704 usage[i++] = get_usage_char(class, LOCK_USED_IN_##__STATE##_READ); 705 #include "lockdep_states.h" 706 #undef LOCKDEP_STATE 707 708 usage[i] = '\0'; 709 } 710 711 static void __print_lock_name(struct lock_class *class) 712 { 713 char str[KSYM_NAME_LEN]; 714 const char *name; 715 716 name = class->name; 717 if (!name) { 718 name = __get_key_name(class->key, str); 719 printk(KERN_CONT "%s", name); 720 } else { 721 printk(KERN_CONT "%s", name); 722 if (class->name_version > 1) 723 printk(KERN_CONT "#%d", class->name_version); 724 if (class->subclass) 725 printk(KERN_CONT "/%d", class->subclass); 726 } 727 } 728 729 static void print_lock_name(struct lock_class *class) 730 { 731 char usage[LOCK_USAGE_CHARS]; 732 733 get_usage_chars(class, usage); 734 735 printk(KERN_CONT " ("); 736 __print_lock_name(class); 737 printk(KERN_CONT "){%s}-{%d:%d}", usage, 738 class->wait_type_outer ?: class->wait_type_inner, 739 class->wait_type_inner); 740 } 741 742 static void print_lockdep_cache(struct lockdep_map *lock) 743 { 744 const char *name; 745 char str[KSYM_NAME_LEN]; 746 747 name = lock->name; 748 if (!name) 749 name = __get_key_name(lock->key->subkeys, str); 750 751 printk(KERN_CONT "%s", name); 752 } 753 754 static void print_lock(struct held_lock *hlock) 755 { 756 /* 757 * We can be called locklessly through debug_show_all_locks() so be 758 * extra careful, the hlock might have been released and cleared. 759 * 760 * If this indeed happens, lets pretend it does not hurt to continue 761 * to print the lock unless the hlock class_idx does not point to a 762 * registered class. The rationale here is: since we don't attempt 763 * to distinguish whether we are in this situation, if it just 764 * happened we can't count on class_idx to tell either. 765 */ 766 struct lock_class *lock = hlock_class(hlock); 767 768 if (!lock) { 769 printk(KERN_CONT "<RELEASED>\n"); 770 return; 771 } 772 773 printk(KERN_CONT "%px", hlock->instance); 774 print_lock_name(lock); 775 printk(KERN_CONT ", at: %pS\n", (void *)hlock->acquire_ip); 776 } 777 778 static void lockdep_print_held_locks(struct task_struct *p) 779 { 780 int i, depth = READ_ONCE(p->lockdep_depth); 781 782 if (!depth) 783 printk("no locks held by %s/%d.\n", p->comm, task_pid_nr(p)); 784 else 785 printk("%d lock%s held by %s/%d:\n", depth, 786 depth > 1 ? "s" : "", p->comm, task_pid_nr(p)); 787 /* 788 * It's not reliable to print a task's held locks if it's not sleeping 789 * and it's not the current task. 790 */ 791 if (p != current && task_is_running(p)) 792 return; 793 for (i = 0; i < depth; i++) { 794 printk(" #%d: ", i); 795 print_lock(p->held_locks + i); 796 } 797 } 798 799 static void print_kernel_ident(void) 800 { 801 printk("%s %.*s %s\n", init_utsname()->release, 802 (int)strcspn(init_utsname()->version, " "), 803 init_utsname()->version, 804 print_tainted()); 805 } 806 807 static int very_verbose(struct lock_class *class) 808 { 809 #if VERY_VERBOSE 810 return class_filter(class); 811 #endif 812 return 0; 813 } 814 815 /* 816 * Is this the address of a static object: 817 */ 818 #ifdef __KERNEL__ 819 /* 820 * Check if an address is part of freed initmem. After initmem is freed, 821 * memory can be allocated from it, and such allocations would then have 822 * addresses within the range [_stext, _end]. 823 */ 824 #ifndef arch_is_kernel_initmem_freed 825 static int arch_is_kernel_initmem_freed(unsigned long addr) 826 { 827 if (system_state < SYSTEM_FREEING_INITMEM) 828 return 0; 829 830 return init_section_contains((void *)addr, 1); 831 } 832 #endif 833 834 static int static_obj(const void *obj) 835 { 836 unsigned long start = (unsigned long) &_stext, 837 end = (unsigned long) &_end, 838 addr = (unsigned long) obj; 839 840 if (arch_is_kernel_initmem_freed(addr)) 841 return 0; 842 843 /* 844 * static variable? 845 */ 846 if ((addr >= start) && (addr < end)) 847 return 1; 848 849 /* 850 * in-kernel percpu var? 851 */ 852 if (is_kernel_percpu_address(addr)) 853 return 1; 854 855 /* 856 * module static or percpu var? 857 */ 858 return is_module_address(addr) || is_module_percpu_address(addr); 859 } 860 #endif 861 862 /* 863 * To make lock name printouts unique, we calculate a unique 864 * class->name_version generation counter. The caller must hold the graph 865 * lock. 866 */ 867 static int count_matching_names(struct lock_class *new_class) 868 { 869 struct lock_class *class; 870 int count = 0; 871 872 if (!new_class->name) 873 return 0; 874 875 list_for_each_entry(class, &all_lock_classes, lock_entry) { 876 if (new_class->key - new_class->subclass == class->key) 877 return class->name_version; 878 if (class->name && !strcmp(class->name, new_class->name)) 879 count = max(count, class->name_version); 880 } 881 882 return count + 1; 883 } 884 885 /* used from NMI context -- must be lockless */ 886 static noinstr struct lock_class * 887 look_up_lock_class(const struct lockdep_map *lock, unsigned int subclass) 888 { 889 struct lockdep_subclass_key *key; 890 struct hlist_head *hash_head; 891 struct lock_class *class; 892 893 if (unlikely(subclass >= MAX_LOCKDEP_SUBCLASSES)) { 894 instrumentation_begin(); 895 debug_locks_off(); 896 printk(KERN_ERR 897 "BUG: looking up invalid subclass: %u\n", subclass); 898 printk(KERN_ERR 899 "turning off the locking correctness validator.\n"); 900 dump_stack(); 901 instrumentation_end(); 902 return NULL; 903 } 904 905 /* 906 * If it is not initialised then it has never been locked, 907 * so it won't be present in the hash table. 908 */ 909 if (unlikely(!lock->key)) 910 return NULL; 911 912 /* 913 * NOTE: the class-key must be unique. For dynamic locks, a static 914 * lock_class_key variable is passed in through the mutex_init() 915 * (or spin_lock_init()) call - which acts as the key. For static 916 * locks we use the lock object itself as the key. 917 */ 918 BUILD_BUG_ON(sizeof(struct lock_class_key) > 919 sizeof(struct lockdep_map)); 920 921 key = lock->key->subkeys + subclass; 922 923 hash_head = classhashentry(key); 924 925 /* 926 * We do an RCU walk of the hash, see lockdep_free_key_range(). 927 */ 928 if (DEBUG_LOCKS_WARN_ON(!irqs_disabled())) 929 return NULL; 930 931 hlist_for_each_entry_rcu_notrace(class, hash_head, hash_entry) { 932 if (class->key == key) { 933 /* 934 * Huh! same key, different name? Did someone trample 935 * on some memory? We're most confused. 936 */ 937 WARN_ON_ONCE(class->name != lock->name && 938 lock->key != &__lockdep_no_validate__); 939 return class; 940 } 941 } 942 943 return NULL; 944 } 945 946 /* 947 * Static locks do not have their class-keys yet - for them the key is 948 * the lock object itself. If the lock is in the per cpu area, the 949 * canonical address of the lock (per cpu offset removed) is used. 950 */ 951 static bool assign_lock_key(struct lockdep_map *lock) 952 { 953 unsigned long can_addr, addr = (unsigned long)lock; 954 955 #ifdef __KERNEL__ 956 /* 957 * lockdep_free_key_range() assumes that struct lock_class_key 958 * objects do not overlap. Since we use the address of lock 959 * objects as class key for static objects, check whether the 960 * size of lock_class_key objects does not exceed the size of 961 * the smallest lock object. 962 */ 963 BUILD_BUG_ON(sizeof(struct lock_class_key) > sizeof(raw_spinlock_t)); 964 #endif 965 966 if (__is_kernel_percpu_address(addr, &can_addr)) 967 lock->key = (void *)can_addr; 968 else if (__is_module_percpu_address(addr, &can_addr)) 969 lock->key = (void *)can_addr; 970 else if (static_obj(lock)) 971 lock->key = (void *)lock; 972 else { 973 /* Debug-check: all keys must be persistent! */ 974 debug_locks_off(); 975 pr_err("INFO: trying to register non-static key.\n"); 976 pr_err("The code is fine but needs lockdep annotation, or maybe\n"); 977 pr_err("you didn't initialize this object before use?\n"); 978 pr_err("turning off the locking correctness validator.\n"); 979 dump_stack(); 980 return false; 981 } 982 983 return true; 984 } 985 986 #ifdef CONFIG_DEBUG_LOCKDEP 987 988 /* Check whether element @e occurs in list @h */ 989 static bool in_list(struct list_head *e, struct list_head *h) 990 { 991 struct list_head *f; 992 993 list_for_each(f, h) { 994 if (e == f) 995 return true; 996 } 997 998 return false; 999 } 1000 1001 /* 1002 * Check whether entry @e occurs in any of the locks_after or locks_before 1003 * lists. 1004 */ 1005 static bool in_any_class_list(struct list_head *e) 1006 { 1007 struct lock_class *class; 1008 int i; 1009 1010 for (i = 0; i < ARRAY_SIZE(lock_classes); i++) { 1011 class = &lock_classes[i]; 1012 if (in_list(e, &class->locks_after) || 1013 in_list(e, &class->locks_before)) 1014 return true; 1015 } 1016 return false; 1017 } 1018 1019 static bool class_lock_list_valid(struct lock_class *c, struct list_head *h) 1020 { 1021 struct lock_list *e; 1022 1023 list_for_each_entry(e, h, entry) { 1024 if (e->links_to != c) { 1025 printk(KERN_INFO "class %s: mismatch for lock entry %ld; class %s <> %s", 1026 c->name ? : "(?)", 1027 (unsigned long)(e - list_entries), 1028 e->links_to && e->links_to->name ? 1029 e->links_to->name : "(?)", 1030 e->class && e->class->name ? e->class->name : 1031 "(?)"); 1032 return false; 1033 } 1034 } 1035 return true; 1036 } 1037 1038 #ifdef CONFIG_PROVE_LOCKING 1039 static u16 chain_hlocks[MAX_LOCKDEP_CHAIN_HLOCKS]; 1040 #endif 1041 1042 static bool check_lock_chain_key(struct lock_chain *chain) 1043 { 1044 #ifdef CONFIG_PROVE_LOCKING 1045 u64 chain_key = INITIAL_CHAIN_KEY; 1046 int i; 1047 1048 for (i = chain->base; i < chain->base + chain->depth; i++) 1049 chain_key = iterate_chain_key(chain_key, chain_hlocks[i]); 1050 /* 1051 * The 'unsigned long long' casts avoid that a compiler warning 1052 * is reported when building tools/lib/lockdep. 1053 */ 1054 if (chain->chain_key != chain_key) { 1055 printk(KERN_INFO "chain %lld: key %#llx <> %#llx\n", 1056 (unsigned long long)(chain - lock_chains), 1057 (unsigned long long)chain->chain_key, 1058 (unsigned long long)chain_key); 1059 return false; 1060 } 1061 #endif 1062 return true; 1063 } 1064 1065 static bool in_any_zapped_class_list(struct lock_class *class) 1066 { 1067 struct pending_free *pf; 1068 int i; 1069 1070 for (i = 0, pf = delayed_free.pf; i < ARRAY_SIZE(delayed_free.pf); i++, pf++) { 1071 if (in_list(&class->lock_entry, &pf->zapped)) 1072 return true; 1073 } 1074 1075 return false; 1076 } 1077 1078 static bool __check_data_structures(void) 1079 { 1080 struct lock_class *class; 1081 struct lock_chain *chain; 1082 struct hlist_head *head; 1083 struct lock_list *e; 1084 int i; 1085 1086 /* Check whether all classes occur in a lock list. */ 1087 for (i = 0; i < ARRAY_SIZE(lock_classes); i++) { 1088 class = &lock_classes[i]; 1089 if (!in_list(&class->lock_entry, &all_lock_classes) && 1090 !in_list(&class->lock_entry, &free_lock_classes) && 1091 !in_any_zapped_class_list(class)) { 1092 printk(KERN_INFO "class %px/%s is not in any class list\n", 1093 class, class->name ? : "(?)"); 1094 return false; 1095 } 1096 } 1097 1098 /* Check whether all classes have valid lock lists. */ 1099 for (i = 0; i < ARRAY_SIZE(lock_classes); i++) { 1100 class = &lock_classes[i]; 1101 if (!class_lock_list_valid(class, &class->locks_before)) 1102 return false; 1103 if (!class_lock_list_valid(class, &class->locks_after)) 1104 return false; 1105 } 1106 1107 /* Check the chain_key of all lock chains. */ 1108 for (i = 0; i < ARRAY_SIZE(chainhash_table); i++) { 1109 head = chainhash_table + i; 1110 hlist_for_each_entry_rcu(chain, head, entry) { 1111 if (!check_lock_chain_key(chain)) 1112 return false; 1113 } 1114 } 1115 1116 /* 1117 * Check whether all list entries that are in use occur in a class 1118 * lock list. 1119 */ 1120 for_each_set_bit(i, list_entries_in_use, ARRAY_SIZE(list_entries)) { 1121 e = list_entries + i; 1122 if (!in_any_class_list(&e->entry)) { 1123 printk(KERN_INFO "list entry %d is not in any class list; class %s <> %s\n", 1124 (unsigned int)(e - list_entries), 1125 e->class->name ? : "(?)", 1126 e->links_to->name ? : "(?)"); 1127 return false; 1128 } 1129 } 1130 1131 /* 1132 * Check whether all list entries that are not in use do not occur in 1133 * a class lock list. 1134 */ 1135 for_each_clear_bit(i, list_entries_in_use, ARRAY_SIZE(list_entries)) { 1136 e = list_entries + i; 1137 if (in_any_class_list(&e->entry)) { 1138 printk(KERN_INFO "list entry %d occurs in a class list; class %s <> %s\n", 1139 (unsigned int)(e - list_entries), 1140 e->class && e->class->name ? e->class->name : 1141 "(?)", 1142 e->links_to && e->links_to->name ? 1143 e->links_to->name : "(?)"); 1144 return false; 1145 } 1146 } 1147 1148 return true; 1149 } 1150 1151 int check_consistency = 0; 1152 module_param(check_consistency, int, 0644); 1153 1154 static void check_data_structures(void) 1155 { 1156 static bool once = false; 1157 1158 if (check_consistency && !once) { 1159 if (!__check_data_structures()) { 1160 once = true; 1161 WARN_ON(once); 1162 } 1163 } 1164 } 1165 1166 #else /* CONFIG_DEBUG_LOCKDEP */ 1167 1168 static inline void check_data_structures(void) { } 1169 1170 #endif /* CONFIG_DEBUG_LOCKDEP */ 1171 1172 static void init_chain_block_buckets(void); 1173 1174 /* 1175 * Initialize the lock_classes[] array elements, the free_lock_classes list 1176 * and also the delayed_free structure. 1177 */ 1178 static void init_data_structures_once(void) 1179 { 1180 static bool __read_mostly ds_initialized, rcu_head_initialized; 1181 int i; 1182 1183 if (likely(rcu_head_initialized)) 1184 return; 1185 1186 if (system_state >= SYSTEM_SCHEDULING) { 1187 init_rcu_head(&delayed_free.rcu_head); 1188 rcu_head_initialized = true; 1189 } 1190 1191 if (ds_initialized) 1192 return; 1193 1194 ds_initialized = true; 1195 1196 INIT_LIST_HEAD(&delayed_free.pf[0].zapped); 1197 INIT_LIST_HEAD(&delayed_free.pf[1].zapped); 1198 1199 for (i = 0; i < ARRAY_SIZE(lock_classes); i++) { 1200 list_add_tail(&lock_classes[i].lock_entry, &free_lock_classes); 1201 INIT_LIST_HEAD(&lock_classes[i].locks_after); 1202 INIT_LIST_HEAD(&lock_classes[i].locks_before); 1203 } 1204 init_chain_block_buckets(); 1205 } 1206 1207 static inline struct hlist_head *keyhashentry(const struct lock_class_key *key) 1208 { 1209 unsigned long hash = hash_long((uintptr_t)key, KEYHASH_BITS); 1210 1211 return lock_keys_hash + hash; 1212 } 1213 1214 /* Register a dynamically allocated key. */ 1215 void lockdep_register_key(struct lock_class_key *key) 1216 { 1217 struct hlist_head *hash_head; 1218 struct lock_class_key *k; 1219 unsigned long flags; 1220 1221 if (WARN_ON_ONCE(static_obj(key))) 1222 return; 1223 hash_head = keyhashentry(key); 1224 1225 raw_local_irq_save(flags); 1226 if (!graph_lock()) 1227 goto restore_irqs; 1228 hlist_for_each_entry_rcu(k, hash_head, hash_entry) { 1229 if (WARN_ON_ONCE(k == key)) 1230 goto out_unlock; 1231 } 1232 hlist_add_head_rcu(&key->hash_entry, hash_head); 1233 out_unlock: 1234 graph_unlock(); 1235 restore_irqs: 1236 raw_local_irq_restore(flags); 1237 } 1238 EXPORT_SYMBOL_GPL(lockdep_register_key); 1239 1240 /* Check whether a key has been registered as a dynamic key. */ 1241 static bool is_dynamic_key(const struct lock_class_key *key) 1242 { 1243 struct hlist_head *hash_head; 1244 struct lock_class_key *k; 1245 bool found = false; 1246 1247 if (WARN_ON_ONCE(static_obj(key))) 1248 return false; 1249 1250 /* 1251 * If lock debugging is disabled lock_keys_hash[] may contain 1252 * pointers to memory that has already been freed. Avoid triggering 1253 * a use-after-free in that case by returning early. 1254 */ 1255 if (!debug_locks) 1256 return true; 1257 1258 hash_head = keyhashentry(key); 1259 1260 rcu_read_lock(); 1261 hlist_for_each_entry_rcu(k, hash_head, hash_entry) { 1262 if (k == key) { 1263 found = true; 1264 break; 1265 } 1266 } 1267 rcu_read_unlock(); 1268 1269 return found; 1270 } 1271 1272 /* 1273 * Register a lock's class in the hash-table, if the class is not present 1274 * yet. Otherwise we look it up. We cache the result in the lock object 1275 * itself, so actual lookup of the hash should be once per lock object. 1276 */ 1277 static struct lock_class * 1278 register_lock_class(struct lockdep_map *lock, unsigned int subclass, int force) 1279 { 1280 struct lockdep_subclass_key *key; 1281 struct hlist_head *hash_head; 1282 struct lock_class *class; 1283 int idx; 1284 1285 DEBUG_LOCKS_WARN_ON(!irqs_disabled()); 1286 1287 class = look_up_lock_class(lock, subclass); 1288 if (likely(class)) 1289 goto out_set_class_cache; 1290 1291 if (!lock->key) { 1292 if (!assign_lock_key(lock)) 1293 return NULL; 1294 } else if (!static_obj(lock->key) && !is_dynamic_key(lock->key)) { 1295 return NULL; 1296 } 1297 1298 key = lock->key->subkeys + subclass; 1299 hash_head = classhashentry(key); 1300 1301 if (!graph_lock()) { 1302 return NULL; 1303 } 1304 /* 1305 * We have to do the hash-walk again, to avoid races 1306 * with another CPU: 1307 */ 1308 hlist_for_each_entry_rcu(class, hash_head, hash_entry) { 1309 if (class->key == key) 1310 goto out_unlock_set; 1311 } 1312 1313 init_data_structures_once(); 1314 1315 /* Allocate a new lock class and add it to the hash. */ 1316 class = list_first_entry_or_null(&free_lock_classes, typeof(*class), 1317 lock_entry); 1318 if (!class) { 1319 if (!debug_locks_off_graph_unlock()) { 1320 return NULL; 1321 } 1322 1323 print_lockdep_off("BUG: MAX_LOCKDEP_KEYS too low!"); 1324 dump_stack(); 1325 return NULL; 1326 } 1327 nr_lock_classes++; 1328 __set_bit(class - lock_classes, lock_classes_in_use); 1329 debug_atomic_inc(nr_unused_locks); 1330 class->key = key; 1331 class->name = lock->name; 1332 class->subclass = subclass; 1333 WARN_ON_ONCE(!list_empty(&class->locks_before)); 1334 WARN_ON_ONCE(!list_empty(&class->locks_after)); 1335 class->name_version = count_matching_names(class); 1336 class->wait_type_inner = lock->wait_type_inner; 1337 class->wait_type_outer = lock->wait_type_outer; 1338 class->lock_type = lock->lock_type; 1339 /* 1340 * We use RCU's safe list-add method to make 1341 * parallel walking of the hash-list safe: 1342 */ 1343 hlist_add_head_rcu(&class->hash_entry, hash_head); 1344 /* 1345 * Remove the class from the free list and add it to the global list 1346 * of classes. 1347 */ 1348 list_move_tail(&class->lock_entry, &all_lock_classes); 1349 idx = class - lock_classes; 1350 if (idx > max_lock_class_idx) 1351 max_lock_class_idx = idx; 1352 1353 if (verbose(class)) { 1354 graph_unlock(); 1355 1356 printk("\nnew class %px: %s", class->key, class->name); 1357 if (class->name_version > 1) 1358 printk(KERN_CONT "#%d", class->name_version); 1359 printk(KERN_CONT "\n"); 1360 dump_stack(); 1361 1362 if (!graph_lock()) { 1363 return NULL; 1364 } 1365 } 1366 out_unlock_set: 1367 graph_unlock(); 1368 1369 out_set_class_cache: 1370 if (!subclass || force) 1371 lock->class_cache[0] = class; 1372 else if (subclass < NR_LOCKDEP_CACHING_CLASSES) 1373 lock->class_cache[subclass] = class; 1374 1375 /* 1376 * Hash collision, did we smoke some? We found a class with a matching 1377 * hash but the subclass -- which is hashed in -- didn't match. 1378 */ 1379 if (DEBUG_LOCKS_WARN_ON(class->subclass != subclass)) 1380 return NULL; 1381 1382 return class; 1383 } 1384 1385 #ifdef CONFIG_PROVE_LOCKING 1386 /* 1387 * Allocate a lockdep entry. (assumes the graph_lock held, returns 1388 * with NULL on failure) 1389 */ 1390 static struct lock_list *alloc_list_entry(void) 1391 { 1392 int idx = find_first_zero_bit(list_entries_in_use, 1393 ARRAY_SIZE(list_entries)); 1394 1395 if (idx >= ARRAY_SIZE(list_entries)) { 1396 if (!debug_locks_off_graph_unlock()) 1397 return NULL; 1398 1399 print_lockdep_off("BUG: MAX_LOCKDEP_ENTRIES too low!"); 1400 dump_stack(); 1401 return NULL; 1402 } 1403 nr_list_entries++; 1404 __set_bit(idx, list_entries_in_use); 1405 return list_entries + idx; 1406 } 1407 1408 /* 1409 * Add a new dependency to the head of the list: 1410 */ 1411 static int add_lock_to_list(struct lock_class *this, 1412 struct lock_class *links_to, struct list_head *head, 1413 u16 distance, u8 dep, 1414 const struct lock_trace *trace) 1415 { 1416 struct lock_list *entry; 1417 /* 1418 * Lock not present yet - get a new dependency struct and 1419 * add it to the list: 1420 */ 1421 entry = alloc_list_entry(); 1422 if (!entry) 1423 return 0; 1424 1425 entry->class = this; 1426 entry->links_to = links_to; 1427 entry->dep = dep; 1428 entry->distance = distance; 1429 entry->trace = trace; 1430 /* 1431 * Both allocation and removal are done under the graph lock; but 1432 * iteration is under RCU-sched; see look_up_lock_class() and 1433 * lockdep_free_key_range(). 1434 */ 1435 list_add_tail_rcu(&entry->entry, head); 1436 1437 return 1; 1438 } 1439 1440 /* 1441 * For good efficiency of modular, we use power of 2 1442 */ 1443 #define MAX_CIRCULAR_QUEUE_SIZE (1UL << CONFIG_LOCKDEP_CIRCULAR_QUEUE_BITS) 1444 #define CQ_MASK (MAX_CIRCULAR_QUEUE_SIZE-1) 1445 1446 /* 1447 * The circular_queue and helpers are used to implement graph 1448 * breadth-first search (BFS) algorithm, by which we can determine 1449 * whether there is a path from a lock to another. In deadlock checks, 1450 * a path from the next lock to be acquired to a previous held lock 1451 * indicates that adding the <prev> -> <next> lock dependency will 1452 * produce a circle in the graph. Breadth-first search instead of 1453 * depth-first search is used in order to find the shortest (circular) 1454 * path. 1455 */ 1456 struct circular_queue { 1457 struct lock_list *element[MAX_CIRCULAR_QUEUE_SIZE]; 1458 unsigned int front, rear; 1459 }; 1460 1461 static struct circular_queue lock_cq; 1462 1463 unsigned int max_bfs_queue_depth; 1464 1465 static unsigned int lockdep_dependency_gen_id; 1466 1467 static inline void __cq_init(struct circular_queue *cq) 1468 { 1469 cq->front = cq->rear = 0; 1470 lockdep_dependency_gen_id++; 1471 } 1472 1473 static inline int __cq_empty(struct circular_queue *cq) 1474 { 1475 return (cq->front == cq->rear); 1476 } 1477 1478 static inline int __cq_full(struct circular_queue *cq) 1479 { 1480 return ((cq->rear + 1) & CQ_MASK) == cq->front; 1481 } 1482 1483 static inline int __cq_enqueue(struct circular_queue *cq, struct lock_list *elem) 1484 { 1485 if (__cq_full(cq)) 1486 return -1; 1487 1488 cq->element[cq->rear] = elem; 1489 cq->rear = (cq->rear + 1) & CQ_MASK; 1490 return 0; 1491 } 1492 1493 /* 1494 * Dequeue an element from the circular_queue, return a lock_list if 1495 * the queue is not empty, or NULL if otherwise. 1496 */ 1497 static inline struct lock_list * __cq_dequeue(struct circular_queue *cq) 1498 { 1499 struct lock_list * lock; 1500 1501 if (__cq_empty(cq)) 1502 return NULL; 1503 1504 lock = cq->element[cq->front]; 1505 cq->front = (cq->front + 1) & CQ_MASK; 1506 1507 return lock; 1508 } 1509 1510 static inline unsigned int __cq_get_elem_count(struct circular_queue *cq) 1511 { 1512 return (cq->rear - cq->front) & CQ_MASK; 1513 } 1514 1515 static inline void mark_lock_accessed(struct lock_list *lock) 1516 { 1517 lock->class->dep_gen_id = lockdep_dependency_gen_id; 1518 } 1519 1520 static inline void visit_lock_entry(struct lock_list *lock, 1521 struct lock_list *parent) 1522 { 1523 lock->parent = parent; 1524 } 1525 1526 static inline unsigned long lock_accessed(struct lock_list *lock) 1527 { 1528 return lock->class->dep_gen_id == lockdep_dependency_gen_id; 1529 } 1530 1531 static inline struct lock_list *get_lock_parent(struct lock_list *child) 1532 { 1533 return child->parent; 1534 } 1535 1536 static inline int get_lock_depth(struct lock_list *child) 1537 { 1538 int depth = 0; 1539 struct lock_list *parent; 1540 1541 while ((parent = get_lock_parent(child))) { 1542 child = parent; 1543 depth++; 1544 } 1545 return depth; 1546 } 1547 1548 /* 1549 * Return the forward or backward dependency list. 1550 * 1551 * @lock: the lock_list to get its class's dependency list 1552 * @offset: the offset to struct lock_class to determine whether it is 1553 * locks_after or locks_before 1554 */ 1555 static inline struct list_head *get_dep_list(struct lock_list *lock, int offset) 1556 { 1557 void *lock_class = lock->class; 1558 1559 return lock_class + offset; 1560 } 1561 /* 1562 * Return values of a bfs search: 1563 * 1564 * BFS_E* indicates an error 1565 * BFS_R* indicates a result (match or not) 1566 * 1567 * BFS_EINVALIDNODE: Find a invalid node in the graph. 1568 * 1569 * BFS_EQUEUEFULL: The queue is full while doing the bfs. 1570 * 1571 * BFS_RMATCH: Find the matched node in the graph, and put that node into 1572 * *@target_entry. 1573 * 1574 * BFS_RNOMATCH: Haven't found the matched node and keep *@target_entry 1575 * _unchanged_. 1576 */ 1577 enum bfs_result { 1578 BFS_EINVALIDNODE = -2, 1579 BFS_EQUEUEFULL = -1, 1580 BFS_RMATCH = 0, 1581 BFS_RNOMATCH = 1, 1582 }; 1583 1584 /* 1585 * bfs_result < 0 means error 1586 */ 1587 static inline bool bfs_error(enum bfs_result res) 1588 { 1589 return res < 0; 1590 } 1591 1592 /* 1593 * DEP_*_BIT in lock_list::dep 1594 * 1595 * For dependency @prev -> @next: 1596 * 1597 * SR: @prev is shared reader (->read != 0) and @next is recursive reader 1598 * (->read == 2) 1599 * ER: @prev is exclusive locker (->read == 0) and @next is recursive reader 1600 * SN: @prev is shared reader and @next is non-recursive locker (->read != 2) 1601 * EN: @prev is exclusive locker and @next is non-recursive locker 1602 * 1603 * Note that we define the value of DEP_*_BITs so that: 1604 * bit0 is prev->read == 0 1605 * bit1 is next->read != 2 1606 */ 1607 #define DEP_SR_BIT (0 + (0 << 1)) /* 0 */ 1608 #define DEP_ER_BIT (1 + (0 << 1)) /* 1 */ 1609 #define DEP_SN_BIT (0 + (1 << 1)) /* 2 */ 1610 #define DEP_EN_BIT (1 + (1 << 1)) /* 3 */ 1611 1612 #define DEP_SR_MASK (1U << (DEP_SR_BIT)) 1613 #define DEP_ER_MASK (1U << (DEP_ER_BIT)) 1614 #define DEP_SN_MASK (1U << (DEP_SN_BIT)) 1615 #define DEP_EN_MASK (1U << (DEP_EN_BIT)) 1616 1617 static inline unsigned int 1618 __calc_dep_bit(struct held_lock *prev, struct held_lock *next) 1619 { 1620 return (prev->read == 0) + ((next->read != 2) << 1); 1621 } 1622 1623 static inline u8 calc_dep(struct held_lock *prev, struct held_lock *next) 1624 { 1625 return 1U << __calc_dep_bit(prev, next); 1626 } 1627 1628 /* 1629 * calculate the dep_bit for backwards edges. We care about whether @prev is 1630 * shared and whether @next is recursive. 1631 */ 1632 static inline unsigned int 1633 __calc_dep_bitb(struct held_lock *prev, struct held_lock *next) 1634 { 1635 return (next->read != 2) + ((prev->read == 0) << 1); 1636 } 1637 1638 static inline u8 calc_depb(struct held_lock *prev, struct held_lock *next) 1639 { 1640 return 1U << __calc_dep_bitb(prev, next); 1641 } 1642 1643 /* 1644 * Initialize a lock_list entry @lock belonging to @class as the root for a BFS 1645 * search. 1646 */ 1647 static inline void __bfs_init_root(struct lock_list *lock, 1648 struct lock_class *class) 1649 { 1650 lock->class = class; 1651 lock->parent = NULL; 1652 lock->only_xr = 0; 1653 } 1654 1655 /* 1656 * Initialize a lock_list entry @lock based on a lock acquisition @hlock as the 1657 * root for a BFS search. 1658 * 1659 * ->only_xr of the initial lock node is set to @hlock->read == 2, to make sure 1660 * that <prev> -> @hlock and @hlock -> <whatever __bfs() found> is not -(*R)-> 1661 * and -(S*)->. 1662 */ 1663 static inline void bfs_init_root(struct lock_list *lock, 1664 struct held_lock *hlock) 1665 { 1666 __bfs_init_root(lock, hlock_class(hlock)); 1667 lock->only_xr = (hlock->read == 2); 1668 } 1669 1670 /* 1671 * Similar to bfs_init_root() but initialize the root for backwards BFS. 1672 * 1673 * ->only_xr of the initial lock node is set to @hlock->read != 0, to make sure 1674 * that <next> -> @hlock and @hlock -> <whatever backwards BFS found> is not 1675 * -(*S)-> and -(R*)-> (reverse order of -(*R)-> and -(S*)->). 1676 */ 1677 static inline void bfs_init_rootb(struct lock_list *lock, 1678 struct held_lock *hlock) 1679 { 1680 __bfs_init_root(lock, hlock_class(hlock)); 1681 lock->only_xr = (hlock->read != 0); 1682 } 1683 1684 static inline struct lock_list *__bfs_next(struct lock_list *lock, int offset) 1685 { 1686 if (!lock || !lock->parent) 1687 return NULL; 1688 1689 return list_next_or_null_rcu(get_dep_list(lock->parent, offset), 1690 &lock->entry, struct lock_list, entry); 1691 } 1692 1693 /* 1694 * Breadth-First Search to find a strong path in the dependency graph. 1695 * 1696 * @source_entry: the source of the path we are searching for. 1697 * @data: data used for the second parameter of @match function 1698 * @match: match function for the search 1699 * @target_entry: pointer to the target of a matched path 1700 * @offset: the offset to struct lock_class to determine whether it is 1701 * locks_after or locks_before 1702 * 1703 * We may have multiple edges (considering different kinds of dependencies, 1704 * e.g. ER and SN) between two nodes in the dependency graph. But 1705 * only the strong dependency path in the graph is relevant to deadlocks. A 1706 * strong dependency path is a dependency path that doesn't have two adjacent 1707 * dependencies as -(*R)-> -(S*)->, please see: 1708 * 1709 * Documentation/locking/lockdep-design.rst 1710 * 1711 * for more explanation of the definition of strong dependency paths 1712 * 1713 * In __bfs(), we only traverse in the strong dependency path: 1714 * 1715 * In lock_list::only_xr, we record whether the previous dependency only 1716 * has -(*R)-> in the search, and if it does (prev only has -(*R)->), we 1717 * filter out any -(S*)-> in the current dependency and after that, the 1718 * ->only_xr is set according to whether we only have -(*R)-> left. 1719 */ 1720 static enum bfs_result __bfs(struct lock_list *source_entry, 1721 void *data, 1722 bool (*match)(struct lock_list *entry, void *data), 1723 bool (*skip)(struct lock_list *entry, void *data), 1724 struct lock_list **target_entry, 1725 int offset) 1726 { 1727 struct circular_queue *cq = &lock_cq; 1728 struct lock_list *lock = NULL; 1729 struct lock_list *entry; 1730 struct list_head *head; 1731 unsigned int cq_depth; 1732 bool first; 1733 1734 lockdep_assert_locked(); 1735 1736 __cq_init(cq); 1737 __cq_enqueue(cq, source_entry); 1738 1739 while ((lock = __bfs_next(lock, offset)) || (lock = __cq_dequeue(cq))) { 1740 if (!lock->class) 1741 return BFS_EINVALIDNODE; 1742 1743 /* 1744 * Step 1: check whether we already finish on this one. 1745 * 1746 * If we have visited all the dependencies from this @lock to 1747 * others (iow, if we have visited all lock_list entries in 1748 * @lock->class->locks_{after,before}) we skip, otherwise go 1749 * and visit all the dependencies in the list and mark this 1750 * list accessed. 1751 */ 1752 if (lock_accessed(lock)) 1753 continue; 1754 else 1755 mark_lock_accessed(lock); 1756 1757 /* 1758 * Step 2: check whether prev dependency and this form a strong 1759 * dependency path. 1760 */ 1761 if (lock->parent) { /* Parent exists, check prev dependency */ 1762 u8 dep = lock->dep; 1763 bool prev_only_xr = lock->parent->only_xr; 1764 1765 /* 1766 * Mask out all -(S*)-> if we only have *R in previous 1767 * step, because -(*R)-> -(S*)-> don't make up a strong 1768 * dependency. 1769 */ 1770 if (prev_only_xr) 1771 dep &= ~(DEP_SR_MASK | DEP_SN_MASK); 1772 1773 /* If nothing left, we skip */ 1774 if (!dep) 1775 continue; 1776 1777 /* If there are only -(*R)-> left, set that for the next step */ 1778 lock->only_xr = !(dep & (DEP_SN_MASK | DEP_EN_MASK)); 1779 } 1780 1781 /* 1782 * Step 3: we haven't visited this and there is a strong 1783 * dependency path to this, so check with @match. 1784 * If @skip is provide and returns true, we skip this 1785 * lock (and any path this lock is in). 1786 */ 1787 if (skip && skip(lock, data)) 1788 continue; 1789 1790 if (match(lock, data)) { 1791 *target_entry = lock; 1792 return BFS_RMATCH; 1793 } 1794 1795 /* 1796 * Step 4: if not match, expand the path by adding the 1797 * forward or backwards dependencies in the search 1798 * 1799 */ 1800 first = true; 1801 head = get_dep_list(lock, offset); 1802 list_for_each_entry_rcu(entry, head, entry) { 1803 visit_lock_entry(entry, lock); 1804 1805 /* 1806 * Note we only enqueue the first of the list into the 1807 * queue, because we can always find a sibling 1808 * dependency from one (see __bfs_next()), as a result 1809 * the space of queue is saved. 1810 */ 1811 if (!first) 1812 continue; 1813 1814 first = false; 1815 1816 if (__cq_enqueue(cq, entry)) 1817 return BFS_EQUEUEFULL; 1818 1819 cq_depth = __cq_get_elem_count(cq); 1820 if (max_bfs_queue_depth < cq_depth) 1821 max_bfs_queue_depth = cq_depth; 1822 } 1823 } 1824 1825 return BFS_RNOMATCH; 1826 } 1827 1828 static inline enum bfs_result 1829 __bfs_forwards(struct lock_list *src_entry, 1830 void *data, 1831 bool (*match)(struct lock_list *entry, void *data), 1832 bool (*skip)(struct lock_list *entry, void *data), 1833 struct lock_list **target_entry) 1834 { 1835 return __bfs(src_entry, data, match, skip, target_entry, 1836 offsetof(struct lock_class, locks_after)); 1837 1838 } 1839 1840 static inline enum bfs_result 1841 __bfs_backwards(struct lock_list *src_entry, 1842 void *data, 1843 bool (*match)(struct lock_list *entry, void *data), 1844 bool (*skip)(struct lock_list *entry, void *data), 1845 struct lock_list **target_entry) 1846 { 1847 return __bfs(src_entry, data, match, skip, target_entry, 1848 offsetof(struct lock_class, locks_before)); 1849 1850 } 1851 1852 static void print_lock_trace(const struct lock_trace *trace, 1853 unsigned int spaces) 1854 { 1855 stack_trace_print(trace->entries, trace->nr_entries, spaces); 1856 } 1857 1858 /* 1859 * Print a dependency chain entry (this is only done when a deadlock 1860 * has been detected): 1861 */ 1862 static noinline void 1863 print_circular_bug_entry(struct lock_list *target, int depth) 1864 { 1865 if (debug_locks_silent) 1866 return; 1867 printk("\n-> #%u", depth); 1868 print_lock_name(target->class); 1869 printk(KERN_CONT ":\n"); 1870 print_lock_trace(target->trace, 6); 1871 } 1872 1873 static void 1874 print_circular_lock_scenario(struct held_lock *src, 1875 struct held_lock *tgt, 1876 struct lock_list *prt) 1877 { 1878 struct lock_class *source = hlock_class(src); 1879 struct lock_class *target = hlock_class(tgt); 1880 struct lock_class *parent = prt->class; 1881 1882 /* 1883 * A direct locking problem where unsafe_class lock is taken 1884 * directly by safe_class lock, then all we need to show 1885 * is the deadlock scenario, as it is obvious that the 1886 * unsafe lock is taken under the safe lock. 1887 * 1888 * But if there is a chain instead, where the safe lock takes 1889 * an intermediate lock (middle_class) where this lock is 1890 * not the same as the safe lock, then the lock chain is 1891 * used to describe the problem. Otherwise we would need 1892 * to show a different CPU case for each link in the chain 1893 * from the safe_class lock to the unsafe_class lock. 1894 */ 1895 if (parent != source) { 1896 printk("Chain exists of:\n "); 1897 __print_lock_name(source); 1898 printk(KERN_CONT " --> "); 1899 __print_lock_name(parent); 1900 printk(KERN_CONT " --> "); 1901 __print_lock_name(target); 1902 printk(KERN_CONT "\n\n"); 1903 } 1904 1905 printk(" Possible unsafe locking scenario:\n\n"); 1906 printk(" CPU0 CPU1\n"); 1907 printk(" ---- ----\n"); 1908 printk(" lock("); 1909 __print_lock_name(target); 1910 printk(KERN_CONT ");\n"); 1911 printk(" lock("); 1912 __print_lock_name(parent); 1913 printk(KERN_CONT ");\n"); 1914 printk(" lock("); 1915 __print_lock_name(target); 1916 printk(KERN_CONT ");\n"); 1917 printk(" lock("); 1918 __print_lock_name(source); 1919 printk(KERN_CONT ");\n"); 1920 printk("\n *** DEADLOCK ***\n\n"); 1921 } 1922 1923 /* 1924 * When a circular dependency is detected, print the 1925 * header first: 1926 */ 1927 static noinline void 1928 print_circular_bug_header(struct lock_list *entry, unsigned int depth, 1929 struct held_lock *check_src, 1930 struct held_lock *check_tgt) 1931 { 1932 struct task_struct *curr = current; 1933 1934 if (debug_locks_silent) 1935 return; 1936 1937 pr_warn("\n"); 1938 pr_warn("======================================================\n"); 1939 pr_warn("WARNING: possible circular locking dependency detected\n"); 1940 print_kernel_ident(); 1941 pr_warn("------------------------------------------------------\n"); 1942 pr_warn("%s/%d is trying to acquire lock:\n", 1943 curr->comm, task_pid_nr(curr)); 1944 print_lock(check_src); 1945 1946 pr_warn("\nbut task is already holding lock:\n"); 1947 1948 print_lock(check_tgt); 1949 pr_warn("\nwhich lock already depends on the new lock.\n\n"); 1950 pr_warn("\nthe existing dependency chain (in reverse order) is:\n"); 1951 1952 print_circular_bug_entry(entry, depth); 1953 } 1954 1955 /* 1956 * We are about to add A -> B into the dependency graph, and in __bfs() a 1957 * strong dependency path A -> .. -> B is found: hlock_class equals 1958 * entry->class. 1959 * 1960 * If A -> .. -> B can replace A -> B in any __bfs() search (means the former 1961 * is _stronger_ than or equal to the latter), we consider A -> B as redundant. 1962 * For example if A -> .. -> B is -(EN)-> (i.e. A -(E*)-> .. -(*N)-> B), and A 1963 * -> B is -(ER)-> or -(EN)->, then we don't need to add A -> B into the 1964 * dependency graph, as any strong path ..-> A -> B ->.. we can get with 1965 * having dependency A -> B, we could already get a equivalent path ..-> A -> 1966 * .. -> B -> .. with A -> .. -> B. Therefore A -> B is redundant. 1967 * 1968 * We need to make sure both the start and the end of A -> .. -> B is not 1969 * weaker than A -> B. For the start part, please see the comment in 1970 * check_redundant(). For the end part, we need: 1971 * 1972 * Either 1973 * 1974 * a) A -> B is -(*R)-> (everything is not weaker than that) 1975 * 1976 * or 1977 * 1978 * b) A -> .. -> B is -(*N)-> (nothing is stronger than this) 1979 * 1980 */ 1981 static inline bool hlock_equal(struct lock_list *entry, void *data) 1982 { 1983 struct held_lock *hlock = (struct held_lock *)data; 1984 1985 return hlock_class(hlock) == entry->class && /* Found A -> .. -> B */ 1986 (hlock->read == 2 || /* A -> B is -(*R)-> */ 1987 !entry->only_xr); /* A -> .. -> B is -(*N)-> */ 1988 } 1989 1990 /* 1991 * We are about to add B -> A into the dependency graph, and in __bfs() a 1992 * strong dependency path A -> .. -> B is found: hlock_class equals 1993 * entry->class. 1994 * 1995 * We will have a deadlock case (conflict) if A -> .. -> B -> A is a strong 1996 * dependency cycle, that means: 1997 * 1998 * Either 1999 * 2000 * a) B -> A is -(E*)-> 2001 * 2002 * or 2003 * 2004 * b) A -> .. -> B is -(*N)-> (i.e. A -> .. -(*N)-> B) 2005 * 2006 * as then we don't have -(*R)-> -(S*)-> in the cycle. 2007 */ 2008 static inline bool hlock_conflict(struct lock_list *entry, void *data) 2009 { 2010 struct held_lock *hlock = (struct held_lock *)data; 2011 2012 return hlock_class(hlock) == entry->class && /* Found A -> .. -> B */ 2013 (hlock->read == 0 || /* B -> A is -(E*)-> */ 2014 !entry->only_xr); /* A -> .. -> B is -(*N)-> */ 2015 } 2016 2017 static noinline void print_circular_bug(struct lock_list *this, 2018 struct lock_list *target, 2019 struct held_lock *check_src, 2020 struct held_lock *check_tgt) 2021 { 2022 struct task_struct *curr = current; 2023 struct lock_list *parent; 2024 struct lock_list *first_parent; 2025 int depth; 2026 2027 if (!debug_locks_off_graph_unlock() || debug_locks_silent) 2028 return; 2029 2030 this->trace = save_trace(); 2031 if (!this->trace) 2032 return; 2033 2034 depth = get_lock_depth(target); 2035 2036 print_circular_bug_header(target, depth, check_src, check_tgt); 2037 2038 parent = get_lock_parent(target); 2039 first_parent = parent; 2040 2041 while (parent) { 2042 print_circular_bug_entry(parent, --depth); 2043 parent = get_lock_parent(parent); 2044 } 2045 2046 printk("\nother info that might help us debug this:\n\n"); 2047 print_circular_lock_scenario(check_src, check_tgt, 2048 first_parent); 2049 2050 lockdep_print_held_locks(curr); 2051 2052 printk("\nstack backtrace:\n"); 2053 dump_stack(); 2054 } 2055 2056 static noinline void print_bfs_bug(int ret) 2057 { 2058 if (!debug_locks_off_graph_unlock()) 2059 return; 2060 2061 /* 2062 * Breadth-first-search failed, graph got corrupted? 2063 */ 2064 WARN(1, "lockdep bfs error:%d\n", ret); 2065 } 2066 2067 static bool noop_count(struct lock_list *entry, void *data) 2068 { 2069 (*(unsigned long *)data)++; 2070 return false; 2071 } 2072 2073 static unsigned long __lockdep_count_forward_deps(struct lock_list *this) 2074 { 2075 unsigned long count = 0; 2076 struct lock_list *target_entry; 2077 2078 __bfs_forwards(this, (void *)&count, noop_count, NULL, &target_entry); 2079 2080 return count; 2081 } 2082 unsigned long lockdep_count_forward_deps(struct lock_class *class) 2083 { 2084 unsigned long ret, flags; 2085 struct lock_list this; 2086 2087 __bfs_init_root(&this, class); 2088 2089 raw_local_irq_save(flags); 2090 lockdep_lock(); 2091 ret = __lockdep_count_forward_deps(&this); 2092 lockdep_unlock(); 2093 raw_local_irq_restore(flags); 2094 2095 return ret; 2096 } 2097 2098 static unsigned long __lockdep_count_backward_deps(struct lock_list *this) 2099 { 2100 unsigned long count = 0; 2101 struct lock_list *target_entry; 2102 2103 __bfs_backwards(this, (void *)&count, noop_count, NULL, &target_entry); 2104 2105 return count; 2106 } 2107 2108 unsigned long lockdep_count_backward_deps(struct lock_class *class) 2109 { 2110 unsigned long ret, flags; 2111 struct lock_list this; 2112 2113 __bfs_init_root(&this, class); 2114 2115 raw_local_irq_save(flags); 2116 lockdep_lock(); 2117 ret = __lockdep_count_backward_deps(&this); 2118 lockdep_unlock(); 2119 raw_local_irq_restore(flags); 2120 2121 return ret; 2122 } 2123 2124 /* 2125 * Check that the dependency graph starting at <src> can lead to 2126 * <target> or not. 2127 */ 2128 static noinline enum bfs_result 2129 check_path(struct held_lock *target, struct lock_list *src_entry, 2130 bool (*match)(struct lock_list *entry, void *data), 2131 bool (*skip)(struct lock_list *entry, void *data), 2132 struct lock_list **target_entry) 2133 { 2134 enum bfs_result ret; 2135 2136 ret = __bfs_forwards(src_entry, target, match, skip, target_entry); 2137 2138 if (unlikely(bfs_error(ret))) 2139 print_bfs_bug(ret); 2140 2141 return ret; 2142 } 2143 2144 /* 2145 * Prove that the dependency graph starting at <src> can not 2146 * lead to <target>. If it can, there is a circle when adding 2147 * <target> -> <src> dependency. 2148 * 2149 * Print an error and return BFS_RMATCH if it does. 2150 */ 2151 static noinline enum bfs_result 2152 check_noncircular(struct held_lock *src, struct held_lock *target, 2153 struct lock_trace **const trace) 2154 { 2155 enum bfs_result ret; 2156 struct lock_list *target_entry; 2157 struct lock_list src_entry; 2158 2159 bfs_init_root(&src_entry, src); 2160 2161 debug_atomic_inc(nr_cyclic_checks); 2162 2163 ret = check_path(target, &src_entry, hlock_conflict, NULL, &target_entry); 2164 2165 if (unlikely(ret == BFS_RMATCH)) { 2166 if (!*trace) { 2167 /* 2168 * If save_trace fails here, the printing might 2169 * trigger a WARN but because of the !nr_entries it 2170 * should not do bad things. 2171 */ 2172 *trace = save_trace(); 2173 } 2174 2175 print_circular_bug(&src_entry, target_entry, src, target); 2176 } 2177 2178 return ret; 2179 } 2180 2181 #ifdef CONFIG_TRACE_IRQFLAGS 2182 2183 /* 2184 * Forwards and backwards subgraph searching, for the purposes of 2185 * proving that two subgraphs can be connected by a new dependency 2186 * without creating any illegal irq-safe -> irq-unsafe lock dependency. 2187 * 2188 * A irq safe->unsafe deadlock happens with the following conditions: 2189 * 2190 * 1) We have a strong dependency path A -> ... -> B 2191 * 2192 * 2) and we have ENABLED_IRQ usage of B and USED_IN_IRQ usage of A, therefore 2193 * irq can create a new dependency B -> A (consider the case that a holder 2194 * of B gets interrupted by an irq whose handler will try to acquire A). 2195 * 2196 * 3) the dependency circle A -> ... -> B -> A we get from 1) and 2) is a 2197 * strong circle: 2198 * 2199 * For the usage bits of B: 2200 * a) if A -> B is -(*N)->, then B -> A could be any type, so any 2201 * ENABLED_IRQ usage suffices. 2202 * b) if A -> B is -(*R)->, then B -> A must be -(E*)->, so only 2203 * ENABLED_IRQ_*_READ usage suffices. 2204 * 2205 * For the usage bits of A: 2206 * c) if A -> B is -(E*)->, then B -> A could be any type, so any 2207 * USED_IN_IRQ usage suffices. 2208 * d) if A -> B is -(S*)->, then B -> A must be -(*N)->, so only 2209 * USED_IN_IRQ_*_READ usage suffices. 2210 */ 2211 2212 /* 2213 * There is a strong dependency path in the dependency graph: A -> B, and now 2214 * we need to decide which usage bit of A should be accumulated to detect 2215 * safe->unsafe bugs. 2216 * 2217 * Note that usage_accumulate() is used in backwards search, so ->only_xr 2218 * stands for whether A -> B only has -(S*)-> (in this case ->only_xr is true). 2219 * 2220 * As above, if only_xr is false, which means A -> B has -(E*)-> dependency 2221 * path, any usage of A should be considered. Otherwise, we should only 2222 * consider _READ usage. 2223 */ 2224 static inline bool usage_accumulate(struct lock_list *entry, void *mask) 2225 { 2226 if (!entry->only_xr) 2227 *(unsigned long *)mask |= entry->class->usage_mask; 2228 else /* Mask out _READ usage bits */ 2229 *(unsigned long *)mask |= (entry->class->usage_mask & LOCKF_IRQ); 2230 2231 return false; 2232 } 2233 2234 /* 2235 * There is a strong dependency path in the dependency graph: A -> B, and now 2236 * we need to decide which usage bit of B conflicts with the usage bits of A, 2237 * i.e. which usage bit of B may introduce safe->unsafe deadlocks. 2238 * 2239 * As above, if only_xr is false, which means A -> B has -(*N)-> dependency 2240 * path, any usage of B should be considered. Otherwise, we should only 2241 * consider _READ usage. 2242 */ 2243 static inline bool usage_match(struct lock_list *entry, void *mask) 2244 { 2245 if (!entry->only_xr) 2246 return !!(entry->class->usage_mask & *(unsigned long *)mask); 2247 else /* Mask out _READ usage bits */ 2248 return !!((entry->class->usage_mask & LOCKF_IRQ) & *(unsigned long *)mask); 2249 } 2250 2251 static inline bool usage_skip(struct lock_list *entry, void *mask) 2252 { 2253 /* 2254 * Skip local_lock() for irq inversion detection. 2255 * 2256 * For !RT, local_lock() is not a real lock, so it won't carry any 2257 * dependency. 2258 * 2259 * For RT, an irq inversion happens when we have lock A and B, and on 2260 * some CPU we can have: 2261 * 2262 * lock(A); 2263 * <interrupted> 2264 * lock(B); 2265 * 2266 * where lock(B) cannot sleep, and we have a dependency B -> ... -> A. 2267 * 2268 * Now we prove local_lock() cannot exist in that dependency. First we 2269 * have the observation for any lock chain L1 -> ... -> Ln, for any 2270 * 1 <= i <= n, Li.inner_wait_type <= L1.inner_wait_type, otherwise 2271 * wait context check will complain. And since B is not a sleep lock, 2272 * therefore B.inner_wait_type >= 2, and since the inner_wait_type of 2273 * local_lock() is 3, which is greater than 2, therefore there is no 2274 * way the local_lock() exists in the dependency B -> ... -> A. 2275 * 2276 * As a result, we will skip local_lock(), when we search for irq 2277 * inversion bugs. 2278 */ 2279 if (entry->class->lock_type == LD_LOCK_PERCPU) { 2280 if (DEBUG_LOCKS_WARN_ON(entry->class->wait_type_inner < LD_WAIT_CONFIG)) 2281 return false; 2282 2283 return true; 2284 } 2285 2286 return false; 2287 } 2288 2289 /* 2290 * Find a node in the forwards-direction dependency sub-graph starting 2291 * at @root->class that matches @bit. 2292 * 2293 * Return BFS_MATCH if such a node exists in the subgraph, and put that node 2294 * into *@target_entry. 2295 */ 2296 static enum bfs_result 2297 find_usage_forwards(struct lock_list *root, unsigned long usage_mask, 2298 struct lock_list **target_entry) 2299 { 2300 enum bfs_result result; 2301 2302 debug_atomic_inc(nr_find_usage_forwards_checks); 2303 2304 result = __bfs_forwards(root, &usage_mask, usage_match, usage_skip, target_entry); 2305 2306 return result; 2307 } 2308 2309 /* 2310 * Find a node in the backwards-direction dependency sub-graph starting 2311 * at @root->class that matches @bit. 2312 */ 2313 static enum bfs_result 2314 find_usage_backwards(struct lock_list *root, unsigned long usage_mask, 2315 struct lock_list **target_entry) 2316 { 2317 enum bfs_result result; 2318 2319 debug_atomic_inc(nr_find_usage_backwards_checks); 2320 2321 result = __bfs_backwards(root, &usage_mask, usage_match, usage_skip, target_entry); 2322 2323 return result; 2324 } 2325 2326 static void print_lock_class_header(struct lock_class *class, int depth) 2327 { 2328 int bit; 2329 2330 printk("%*s->", depth, ""); 2331 print_lock_name(class); 2332 #ifdef CONFIG_DEBUG_LOCKDEP 2333 printk(KERN_CONT " ops: %lu", debug_class_ops_read(class)); 2334 #endif 2335 printk(KERN_CONT " {\n"); 2336 2337 for (bit = 0; bit < LOCK_TRACE_STATES; bit++) { 2338 if (class->usage_mask & (1 << bit)) { 2339 int len = depth; 2340 2341 len += printk("%*s %s", depth, "", usage_str[bit]); 2342 len += printk(KERN_CONT " at:\n"); 2343 print_lock_trace(class->usage_traces[bit], len); 2344 } 2345 } 2346 printk("%*s }\n", depth, ""); 2347 2348 printk("%*s ... key at: [<%px>] %pS\n", 2349 depth, "", class->key, class->key); 2350 } 2351 2352 /* 2353 * Dependency path printing: 2354 * 2355 * After BFS we get a lock dependency path (linked via ->parent of lock_list), 2356 * printing out each lock in the dependency path will help on understanding how 2357 * the deadlock could happen. Here are some details about dependency path 2358 * printing: 2359 * 2360 * 1) A lock_list can be either forwards or backwards for a lock dependency, 2361 * for a lock dependency A -> B, there are two lock_lists: 2362 * 2363 * a) lock_list in the ->locks_after list of A, whose ->class is B and 2364 * ->links_to is A. In this case, we can say the lock_list is 2365 * "A -> B" (forwards case). 2366 * 2367 * b) lock_list in the ->locks_before list of B, whose ->class is A 2368 * and ->links_to is B. In this case, we can say the lock_list is 2369 * "B <- A" (bacwards case). 2370 * 2371 * The ->trace of both a) and b) point to the call trace where B was 2372 * acquired with A held. 2373 * 2374 * 2) A "helper" lock_list is introduced during BFS, this lock_list doesn't 2375 * represent a certain lock dependency, it only provides an initial entry 2376 * for BFS. For example, BFS may introduce a "helper" lock_list whose 2377 * ->class is A, as a result BFS will search all dependencies starting with 2378 * A, e.g. A -> B or A -> C. 2379 * 2380 * The notation of a forwards helper lock_list is like "-> A", which means 2381 * we should search the forwards dependencies starting with "A", e.g A -> B 2382 * or A -> C. 2383 * 2384 * The notation of a bacwards helper lock_list is like "<- B", which means 2385 * we should search the backwards dependencies ending with "B", e.g. 2386 * B <- A or B <- C. 2387 */ 2388 2389 /* 2390 * printk the shortest lock dependencies from @root to @leaf in reverse order. 2391 * 2392 * We have a lock dependency path as follow: 2393 * 2394 * @root @leaf 2395 * | | 2396 * V V 2397 * ->parent ->parent 2398 * | lock_list | <--------- | lock_list | ... | lock_list | <--------- | lock_list | 2399 * | -> L1 | | L1 -> L2 | ... |Ln-2 -> Ln-1| | Ln-1 -> Ln| 2400 * 2401 * , so it's natural that we start from @leaf and print every ->class and 2402 * ->trace until we reach the @root. 2403 */ 2404 static void __used 2405 print_shortest_lock_dependencies(struct lock_list *leaf, 2406 struct lock_list *root) 2407 { 2408 struct lock_list *entry = leaf; 2409 int depth; 2410 2411 /*compute depth from generated tree by BFS*/ 2412 depth = get_lock_depth(leaf); 2413 2414 do { 2415 print_lock_class_header(entry->class, depth); 2416 printk("%*s ... acquired at:\n", depth, ""); 2417 print_lock_trace(entry->trace, 2); 2418 printk("\n"); 2419 2420 if (depth == 0 && (entry != root)) { 2421 printk("lockdep:%s bad path found in chain graph\n", __func__); 2422 break; 2423 } 2424 2425 entry = get_lock_parent(entry); 2426 depth--; 2427 } while (entry && (depth >= 0)); 2428 } 2429 2430 /* 2431 * printk the shortest lock dependencies from @leaf to @root. 2432 * 2433 * We have a lock dependency path (from a backwards search) as follow: 2434 * 2435 * @leaf @root 2436 * | | 2437 * V V 2438 * ->parent ->parent 2439 * | lock_list | ---------> | lock_list | ... | lock_list | ---------> | lock_list | 2440 * | L2 <- L1 | | L3 <- L2 | ... | Ln <- Ln-1 | | <- Ln | 2441 * 2442 * , so when we iterate from @leaf to @root, we actually print the lock 2443 * dependency path L1 -> L2 -> .. -> Ln in the non-reverse order. 2444 * 2445 * Another thing to notice here is that ->class of L2 <- L1 is L1, while the 2446 * ->trace of L2 <- L1 is the call trace of L2, in fact we don't have the call 2447 * trace of L1 in the dependency path, which is alright, because most of the 2448 * time we can figure out where L1 is held from the call trace of L2. 2449 */ 2450 static void __used 2451 print_shortest_lock_dependencies_backwards(struct lock_list *leaf, 2452 struct lock_list *root) 2453 { 2454 struct lock_list *entry = leaf; 2455 const struct lock_trace *trace = NULL; 2456 int depth; 2457 2458 /*compute depth from generated tree by BFS*/ 2459 depth = get_lock_depth(leaf); 2460 2461 do { 2462 print_lock_class_header(entry->class, depth); 2463 if (trace) { 2464 printk("%*s ... acquired at:\n", depth, ""); 2465 print_lock_trace(trace, 2); 2466 printk("\n"); 2467 } 2468 2469 /* 2470 * Record the pointer to the trace for the next lock_list 2471 * entry, see the comments for the function. 2472 */ 2473 trace = entry->trace; 2474 2475 if (depth == 0 && (entry != root)) { 2476 printk("lockdep:%s bad path found in chain graph\n", __func__); 2477 break; 2478 } 2479 2480 entry = get_lock_parent(entry); 2481 depth--; 2482 } while (entry && (depth >= 0)); 2483 } 2484 2485 static void 2486 print_irq_lock_scenario(struct lock_list *safe_entry, 2487 struct lock_list *unsafe_entry, 2488 struct lock_class *prev_class, 2489 struct lock_class *next_class) 2490 { 2491 struct lock_class *safe_class = safe_entry->class; 2492 struct lock_class *unsafe_class = unsafe_entry->class; 2493 struct lock_class *middle_class = prev_class; 2494 2495 if (middle_class == safe_class) 2496 middle_class = next_class; 2497 2498 /* 2499 * A direct locking problem where unsafe_class lock is taken 2500 * directly by safe_class lock, then all we need to show 2501 * is the deadlock scenario, as it is obvious that the 2502 * unsafe lock is taken under the safe lock. 2503 * 2504 * But if there is a chain instead, where the safe lock takes 2505 * an intermediate lock (middle_class) where this lock is 2506 * not the same as the safe lock, then the lock chain is 2507 * used to describe the problem. Otherwise we would need 2508 * to show a different CPU case for each link in the chain 2509 * from the safe_class lock to the unsafe_class lock. 2510 */ 2511 if (middle_class != unsafe_class) { 2512 printk("Chain exists of:\n "); 2513 __print_lock_name(safe_class); 2514 printk(KERN_CONT " --> "); 2515 __print_lock_name(middle_class); 2516 printk(KERN_CONT " --> "); 2517 __print_lock_name(unsafe_class); 2518 printk(KERN_CONT "\n\n"); 2519 } 2520 2521 printk(" Possible interrupt unsafe locking scenario:\n\n"); 2522 printk(" CPU0 CPU1\n"); 2523 printk(" ---- ----\n"); 2524 printk(" lock("); 2525 __print_lock_name(unsafe_class); 2526 printk(KERN_CONT ");\n"); 2527 printk(" local_irq_disable();\n"); 2528 printk(" lock("); 2529 __print_lock_name(safe_class); 2530 printk(KERN_CONT ");\n"); 2531 printk(" lock("); 2532 __print_lock_name(middle_class); 2533 printk(KERN_CONT ");\n"); 2534 printk(" <Interrupt>\n"); 2535 printk(" lock("); 2536 __print_lock_name(safe_class); 2537 printk(KERN_CONT ");\n"); 2538 printk("\n *** DEADLOCK ***\n\n"); 2539 } 2540 2541 static void 2542 print_bad_irq_dependency(struct task_struct *curr, 2543 struct lock_list *prev_root, 2544 struct lock_list *next_root, 2545 struct lock_list *backwards_entry, 2546 struct lock_list *forwards_entry, 2547 struct held_lock *prev, 2548 struct held_lock *next, 2549 enum lock_usage_bit bit1, 2550 enum lock_usage_bit bit2, 2551 const char *irqclass) 2552 { 2553 if (!debug_locks_off_graph_unlock() || debug_locks_silent) 2554 return; 2555 2556 pr_warn("\n"); 2557 pr_warn("=====================================================\n"); 2558 pr_warn("WARNING: %s-safe -> %s-unsafe lock order detected\n", 2559 irqclass, irqclass); 2560 print_kernel_ident(); 2561 pr_warn("-----------------------------------------------------\n"); 2562 pr_warn("%s/%d [HC%u[%lu]:SC%u[%lu]:HE%u:SE%u] is trying to acquire:\n", 2563 curr->comm, task_pid_nr(curr), 2564 lockdep_hardirq_context(), hardirq_count() >> HARDIRQ_SHIFT, 2565 curr->softirq_context, softirq_count() >> SOFTIRQ_SHIFT, 2566 lockdep_hardirqs_enabled(), 2567 curr->softirqs_enabled); 2568 print_lock(next); 2569 2570 pr_warn("\nand this task is already holding:\n"); 2571 print_lock(prev); 2572 pr_warn("which would create a new lock dependency:\n"); 2573 print_lock_name(hlock_class(prev)); 2574 pr_cont(" ->"); 2575 print_lock_name(hlock_class(next)); 2576 pr_cont("\n"); 2577 2578 pr_warn("\nbut this new dependency connects a %s-irq-safe lock:\n", 2579 irqclass); 2580 print_lock_name(backwards_entry->class); 2581 pr_warn("\n... which became %s-irq-safe at:\n", irqclass); 2582 2583 print_lock_trace(backwards_entry->class->usage_traces[bit1], 1); 2584 2585 pr_warn("\nto a %s-irq-unsafe lock:\n", irqclass); 2586 print_lock_name(forwards_entry->class); 2587 pr_warn("\n... which became %s-irq-unsafe at:\n", irqclass); 2588 pr_warn("..."); 2589 2590 print_lock_trace(forwards_entry->class->usage_traces[bit2], 1); 2591 2592 pr_warn("\nother info that might help us debug this:\n\n"); 2593 print_irq_lock_scenario(backwards_entry, forwards_entry, 2594 hlock_class(prev), hlock_class(next)); 2595 2596 lockdep_print_held_locks(curr); 2597 2598 pr_warn("\nthe dependencies between %s-irq-safe lock and the holding lock:\n", irqclass); 2599 print_shortest_lock_dependencies_backwards(backwards_entry, prev_root); 2600 2601 pr_warn("\nthe dependencies between the lock to be acquired"); 2602 pr_warn(" and %s-irq-unsafe lock:\n", irqclass); 2603 next_root->trace = save_trace(); 2604 if (!next_root->trace) 2605 return; 2606 print_shortest_lock_dependencies(forwards_entry, next_root); 2607 2608 pr_warn("\nstack backtrace:\n"); 2609 dump_stack(); 2610 } 2611 2612 static const char *state_names[] = { 2613 #define LOCKDEP_STATE(__STATE) \ 2614 __stringify(__STATE), 2615 #include "lockdep_states.h" 2616 #undef LOCKDEP_STATE 2617 }; 2618 2619 static const char *state_rnames[] = { 2620 #define LOCKDEP_STATE(__STATE) \ 2621 __stringify(__STATE)"-READ", 2622 #include "lockdep_states.h" 2623 #undef LOCKDEP_STATE 2624 }; 2625 2626 static inline const char *state_name(enum lock_usage_bit bit) 2627 { 2628 if (bit & LOCK_USAGE_READ_MASK) 2629 return state_rnames[bit >> LOCK_USAGE_DIR_MASK]; 2630 else 2631 return state_names[bit >> LOCK_USAGE_DIR_MASK]; 2632 } 2633 2634 /* 2635 * The bit number is encoded like: 2636 * 2637 * bit0: 0 exclusive, 1 read lock 2638 * bit1: 0 used in irq, 1 irq enabled 2639 * bit2-n: state 2640 */ 2641 static int exclusive_bit(int new_bit) 2642 { 2643 int state = new_bit & LOCK_USAGE_STATE_MASK; 2644 int dir = new_bit & LOCK_USAGE_DIR_MASK; 2645 2646 /* 2647 * keep state, bit flip the direction and strip read. 2648 */ 2649 return state | (dir ^ LOCK_USAGE_DIR_MASK); 2650 } 2651 2652 /* 2653 * Observe that when given a bitmask where each bitnr is encoded as above, a 2654 * right shift of the mask transforms the individual bitnrs as -1 and 2655 * conversely, a left shift transforms into +1 for the individual bitnrs. 2656 * 2657 * So for all bits whose number have LOCK_ENABLED_* set (bitnr1 == 1), we can 2658 * create the mask with those bit numbers using LOCK_USED_IN_* (bitnr1 == 0) 2659 * instead by subtracting the bit number by 2, or shifting the mask right by 2. 2660 * 2661 * Similarly, bitnr1 == 0 becomes bitnr1 == 1 by adding 2, or shifting left 2. 2662 * 2663 * So split the mask (note that LOCKF_ENABLED_IRQ_ALL|LOCKF_USED_IN_IRQ_ALL is 2664 * all bits set) and recompose with bitnr1 flipped. 2665 */ 2666 static unsigned long invert_dir_mask(unsigned long mask) 2667 { 2668 unsigned long excl = 0; 2669 2670 /* Invert dir */ 2671 excl |= (mask & LOCKF_ENABLED_IRQ_ALL) >> LOCK_USAGE_DIR_MASK; 2672 excl |= (mask & LOCKF_USED_IN_IRQ_ALL) << LOCK_USAGE_DIR_MASK; 2673 2674 return excl; 2675 } 2676 2677 /* 2678 * Note that a LOCK_ENABLED_IRQ_*_READ usage and a LOCK_USED_IN_IRQ_*_READ 2679 * usage may cause deadlock too, for example: 2680 * 2681 * P1 P2 2682 * <irq disabled> 2683 * write_lock(l1); <irq enabled> 2684 * read_lock(l2); 2685 * write_lock(l2); 2686 * <in irq> 2687 * read_lock(l1); 2688 * 2689 * , in above case, l1 will be marked as LOCK_USED_IN_IRQ_HARDIRQ_READ and l2 2690 * will marked as LOCK_ENABLE_IRQ_HARDIRQ_READ, and this is a possible 2691 * deadlock. 2692 * 2693 * In fact, all of the following cases may cause deadlocks: 2694 * 2695 * LOCK_USED_IN_IRQ_* -> LOCK_ENABLED_IRQ_* 2696 * LOCK_USED_IN_IRQ_*_READ -> LOCK_ENABLED_IRQ_* 2697 * LOCK_USED_IN_IRQ_* -> LOCK_ENABLED_IRQ_*_READ 2698 * LOCK_USED_IN_IRQ_*_READ -> LOCK_ENABLED_IRQ_*_READ 2699 * 2700 * As a result, to calculate the "exclusive mask", first we invert the 2701 * direction (USED_IN/ENABLED) of the original mask, and 1) for all bits with 2702 * bitnr0 set (LOCK_*_READ), add those with bitnr0 cleared (LOCK_*). 2) for all 2703 * bits with bitnr0 cleared (LOCK_*_READ), add those with bitnr0 set (LOCK_*). 2704 */ 2705 static unsigned long exclusive_mask(unsigned long mask) 2706 { 2707 unsigned long excl = invert_dir_mask(mask); 2708 2709 excl |= (excl & LOCKF_IRQ_READ) >> LOCK_USAGE_READ_MASK; 2710 excl |= (excl & LOCKF_IRQ) << LOCK_USAGE_READ_MASK; 2711 2712 return excl; 2713 } 2714 2715 /* 2716 * Retrieve the _possible_ original mask to which @mask is 2717 * exclusive. Ie: this is the opposite of exclusive_mask(). 2718 * Note that 2 possible original bits can match an exclusive 2719 * bit: one has LOCK_USAGE_READ_MASK set, the other has it 2720 * cleared. So both are returned for each exclusive bit. 2721 */ 2722 static unsigned long original_mask(unsigned long mask) 2723 { 2724 unsigned long excl = invert_dir_mask(mask); 2725 2726 /* Include read in existing usages */ 2727 excl |= (excl & LOCKF_IRQ_READ) >> LOCK_USAGE_READ_MASK; 2728 excl |= (excl & LOCKF_IRQ) << LOCK_USAGE_READ_MASK; 2729 2730 return excl; 2731 } 2732 2733 /* 2734 * Find the first pair of bit match between an original 2735 * usage mask and an exclusive usage mask. 2736 */ 2737 static int find_exclusive_match(unsigned long mask, 2738 unsigned long excl_mask, 2739 enum lock_usage_bit *bitp, 2740 enum lock_usage_bit *excl_bitp) 2741 { 2742 int bit, excl, excl_read; 2743 2744 for_each_set_bit(bit, &mask, LOCK_USED) { 2745 /* 2746 * exclusive_bit() strips the read bit, however, 2747 * LOCK_ENABLED_IRQ_*_READ may cause deadlocks too, so we need 2748 * to search excl | LOCK_USAGE_READ_MASK as well. 2749 */ 2750 excl = exclusive_bit(bit); 2751 excl_read = excl | LOCK_USAGE_READ_MASK; 2752 if (excl_mask & lock_flag(excl)) { 2753 *bitp = bit; 2754 *excl_bitp = excl; 2755 return 0; 2756 } else if (excl_mask & lock_flag(excl_read)) { 2757 *bitp = bit; 2758 *excl_bitp = excl_read; 2759 return 0; 2760 } 2761 } 2762 return -1; 2763 } 2764 2765 /* 2766 * Prove that the new dependency does not connect a hardirq-safe(-read) 2767 * lock with a hardirq-unsafe lock - to achieve this we search 2768 * the backwards-subgraph starting at <prev>, and the 2769 * forwards-subgraph starting at <next>: 2770 */ 2771 static int check_irq_usage(struct task_struct *curr, struct held_lock *prev, 2772 struct held_lock *next) 2773 { 2774 unsigned long usage_mask = 0, forward_mask, backward_mask; 2775 enum lock_usage_bit forward_bit = 0, backward_bit = 0; 2776 struct lock_list *target_entry1; 2777 struct lock_list *target_entry; 2778 struct lock_list this, that; 2779 enum bfs_result ret; 2780 2781 /* 2782 * Step 1: gather all hard/soft IRQs usages backward in an 2783 * accumulated usage mask. 2784 */ 2785 bfs_init_rootb(&this, prev); 2786 2787 ret = __bfs_backwards(&this, &usage_mask, usage_accumulate, usage_skip, NULL); 2788 if (bfs_error(ret)) { 2789 print_bfs_bug(ret); 2790 return 0; 2791 } 2792 2793 usage_mask &= LOCKF_USED_IN_IRQ_ALL; 2794 if (!usage_mask) 2795 return 1; 2796 2797 /* 2798 * Step 2: find exclusive uses forward that match the previous 2799 * backward accumulated mask. 2800 */ 2801 forward_mask = exclusive_mask(usage_mask); 2802 2803 bfs_init_root(&that, next); 2804 2805 ret = find_usage_forwards(&that, forward_mask, &target_entry1); 2806 if (bfs_error(ret)) { 2807 print_bfs_bug(ret); 2808 return 0; 2809 } 2810 if (ret == BFS_RNOMATCH) 2811 return 1; 2812 2813 /* 2814 * Step 3: we found a bad match! Now retrieve a lock from the backward 2815 * list whose usage mask matches the exclusive usage mask from the 2816 * lock found on the forward list. 2817 * 2818 * Note, we should only keep the LOCKF_ENABLED_IRQ_ALL bits, considering 2819 * the follow case: 2820 * 2821 * When trying to add A -> B to the graph, we find that there is a 2822 * hardirq-safe L, that L -> ... -> A, and another hardirq-unsafe M, 2823 * that B -> ... -> M. However M is **softirq-safe**, if we use exact 2824 * invert bits of M's usage_mask, we will find another lock N that is 2825 * **softirq-unsafe** and N -> ... -> A, however N -> .. -> M will not 2826 * cause a inversion deadlock. 2827 */ 2828 backward_mask = original_mask(target_entry1->class->usage_mask & LOCKF_ENABLED_IRQ_ALL); 2829 2830 ret = find_usage_backwards(&this, backward_mask, &target_entry); 2831 if (bfs_error(ret)) { 2832 print_bfs_bug(ret); 2833 return 0; 2834 } 2835 if (DEBUG_LOCKS_WARN_ON(ret == BFS_RNOMATCH)) 2836 return 1; 2837 2838 /* 2839 * Step 4: narrow down to a pair of incompatible usage bits 2840 * and report it. 2841 */ 2842 ret = find_exclusive_match(target_entry->class->usage_mask, 2843 target_entry1->class->usage_mask, 2844 &backward_bit, &forward_bit); 2845 if (DEBUG_LOCKS_WARN_ON(ret == -1)) 2846 return 1; 2847 2848 print_bad_irq_dependency(curr, &this, &that, 2849 target_entry, target_entry1, 2850 prev, next, 2851 backward_bit, forward_bit, 2852 state_name(backward_bit)); 2853 2854 return 0; 2855 } 2856 2857 #else 2858 2859 static inline int check_irq_usage(struct task_struct *curr, 2860 struct held_lock *prev, struct held_lock *next) 2861 { 2862 return 1; 2863 } 2864 2865 static inline bool usage_skip(struct lock_list *entry, void *mask) 2866 { 2867 return false; 2868 } 2869 2870 #endif /* CONFIG_TRACE_IRQFLAGS */ 2871 2872 #ifdef CONFIG_LOCKDEP_SMALL 2873 /* 2874 * Check that the dependency graph starting at <src> can lead to 2875 * <target> or not. If it can, <src> -> <target> dependency is already 2876 * in the graph. 2877 * 2878 * Return BFS_RMATCH if it does, or BFS_RNOMATCH if it does not, return BFS_E* if 2879 * any error appears in the bfs search. 2880 */ 2881 static noinline enum bfs_result 2882 check_redundant(struct held_lock *src, struct held_lock *target) 2883 { 2884 enum bfs_result ret; 2885 struct lock_list *target_entry; 2886 struct lock_list src_entry; 2887 2888 bfs_init_root(&src_entry, src); 2889 /* 2890 * Special setup for check_redundant(). 2891 * 2892 * To report redundant, we need to find a strong dependency path that 2893 * is equal to or stronger than <src> -> <target>. So if <src> is E, 2894 * we need to let __bfs() only search for a path starting at a -(E*)->, 2895 * we achieve this by setting the initial node's ->only_xr to true in 2896 * that case. And if <prev> is S, we set initial ->only_xr to false 2897 * because both -(S*)-> (equal) and -(E*)-> (stronger) are redundant. 2898 */ 2899 src_entry.only_xr = src->read == 0; 2900 2901 debug_atomic_inc(nr_redundant_checks); 2902 2903 /* 2904 * Note: we skip local_lock() for redundant check, because as the 2905 * comment in usage_skip(), A -> local_lock() -> B and A -> B are not 2906 * the same. 2907 */ 2908 ret = check_path(target, &src_entry, hlock_equal, usage_skip, &target_entry); 2909 2910 if (ret == BFS_RMATCH) 2911 debug_atomic_inc(nr_redundant); 2912 2913 return ret; 2914 } 2915 2916 #else 2917 2918 static inline enum bfs_result 2919 check_redundant(struct held_lock *src, struct held_lock *target) 2920 { 2921 return BFS_RNOMATCH; 2922 } 2923 2924 #endif 2925 2926 static void inc_chains(int irq_context) 2927 { 2928 if (irq_context & LOCK_CHAIN_HARDIRQ_CONTEXT) 2929 nr_hardirq_chains++; 2930 else if (irq_context & LOCK_CHAIN_SOFTIRQ_CONTEXT) 2931 nr_softirq_chains++; 2932 else 2933 nr_process_chains++; 2934 } 2935 2936 static void dec_chains(int irq_context) 2937 { 2938 if (irq_context & LOCK_CHAIN_HARDIRQ_CONTEXT) 2939 nr_hardirq_chains--; 2940 else if (irq_context & LOCK_CHAIN_SOFTIRQ_CONTEXT) 2941 nr_softirq_chains--; 2942 else 2943 nr_process_chains--; 2944 } 2945 2946 static void 2947 print_deadlock_scenario(struct held_lock *nxt, struct held_lock *prv) 2948 { 2949 struct lock_class *next = hlock_class(nxt); 2950 struct lock_class *prev = hlock_class(prv); 2951 2952 printk(" Possible unsafe locking scenario:\n\n"); 2953 printk(" CPU0\n"); 2954 printk(" ----\n"); 2955 printk(" lock("); 2956 __print_lock_name(prev); 2957 printk(KERN_CONT ");\n"); 2958 printk(" lock("); 2959 __print_lock_name(next); 2960 printk(KERN_CONT ");\n"); 2961 printk("\n *** DEADLOCK ***\n\n"); 2962 printk(" May be due to missing lock nesting notation\n\n"); 2963 } 2964 2965 static void 2966 print_deadlock_bug(struct task_struct *curr, struct held_lock *prev, 2967 struct held_lock *next) 2968 { 2969 if (!debug_locks_off_graph_unlock() || debug_locks_silent) 2970 return; 2971 2972 pr_warn("\n"); 2973 pr_warn("============================================\n"); 2974 pr_warn("WARNING: possible recursive locking detected\n"); 2975 print_kernel_ident(); 2976 pr_warn("--------------------------------------------\n"); 2977 pr_warn("%s/%d is trying to acquire lock:\n", 2978 curr->comm, task_pid_nr(curr)); 2979 print_lock(next); 2980 pr_warn("\nbut task is already holding lock:\n"); 2981 print_lock(prev); 2982 2983 pr_warn("\nother info that might help us debug this:\n"); 2984 print_deadlock_scenario(next, prev); 2985 lockdep_print_held_locks(curr); 2986 2987 pr_warn("\nstack backtrace:\n"); 2988 dump_stack(); 2989 } 2990 2991 /* 2992 * Check whether we are holding such a class already. 2993 * 2994 * (Note that this has to be done separately, because the graph cannot 2995 * detect such classes of deadlocks.) 2996 * 2997 * Returns: 0 on deadlock detected, 1 on OK, 2 if another lock with the same 2998 * lock class is held but nest_lock is also held, i.e. we rely on the 2999 * nest_lock to avoid the deadlock. 3000 */ 3001 static int 3002 check_deadlock(struct task_struct *curr, struct held_lock *next) 3003 { 3004 struct held_lock *prev; 3005 struct held_lock *nest = NULL; 3006 int i; 3007 3008 for (i = 0; i < curr->lockdep_depth; i++) { 3009 prev = curr->held_locks + i; 3010 3011 if (prev->instance == next->nest_lock) 3012 nest = prev; 3013 3014 if (hlock_class(prev) != hlock_class(next)) 3015 continue; 3016 3017 /* 3018 * Allow read-after-read recursion of the same 3019 * lock class (i.e. read_lock(lock)+read_lock(lock)): 3020 */ 3021 if ((next->read == 2) && prev->read) 3022 continue; 3023 3024 /* 3025 * We're holding the nest_lock, which serializes this lock's 3026 * nesting behaviour. 3027 */ 3028 if (nest) 3029 return 2; 3030 3031 print_deadlock_bug(curr, prev, next); 3032 return 0; 3033 } 3034 return 1; 3035 } 3036 3037 /* 3038 * There was a chain-cache miss, and we are about to add a new dependency 3039 * to a previous lock. We validate the following rules: 3040 * 3041 * - would the adding of the <prev> -> <next> dependency create a 3042 * circular dependency in the graph? [== circular deadlock] 3043 * 3044 * - does the new prev->next dependency connect any hardirq-safe lock 3045 * (in the full backwards-subgraph starting at <prev>) with any 3046 * hardirq-unsafe lock (in the full forwards-subgraph starting at 3047 * <next>)? [== illegal lock inversion with hardirq contexts] 3048 * 3049 * - does the new prev->next dependency connect any softirq-safe lock 3050 * (in the full backwards-subgraph starting at <prev>) with any 3051 * softirq-unsafe lock (in the full forwards-subgraph starting at 3052 * <next>)? [== illegal lock inversion with softirq contexts] 3053 * 3054 * any of these scenarios could lead to a deadlock. 3055 * 3056 * Then if all the validations pass, we add the forwards and backwards 3057 * dependency. 3058 */ 3059 static int 3060 check_prev_add(struct task_struct *curr, struct held_lock *prev, 3061 struct held_lock *next, u16 distance, 3062 struct lock_trace **const trace) 3063 { 3064 struct lock_list *entry; 3065 enum bfs_result ret; 3066 3067 if (!hlock_class(prev)->key || !hlock_class(next)->key) { 3068 /* 3069 * The warning statements below may trigger a use-after-free 3070 * of the class name. It is better to trigger a use-after free 3071 * and to have the class name most of the time instead of not 3072 * having the class name available. 3073 */ 3074 WARN_ONCE(!debug_locks_silent && !hlock_class(prev)->key, 3075 "Detected use-after-free of lock class %px/%s\n", 3076 hlock_class(prev), 3077 hlock_class(prev)->name); 3078 WARN_ONCE(!debug_locks_silent && !hlock_class(next)->key, 3079 "Detected use-after-free of lock class %px/%s\n", 3080 hlock_class(next), 3081 hlock_class(next)->name); 3082 return 2; 3083 } 3084 3085 /* 3086 * Prove that the new <prev> -> <next> dependency would not 3087 * create a circular dependency in the graph. (We do this by 3088 * a breadth-first search into the graph starting at <next>, 3089 * and check whether we can reach <prev>.) 3090 * 3091 * The search is limited by the size of the circular queue (i.e., 3092 * MAX_CIRCULAR_QUEUE_SIZE) which keeps track of a breadth of nodes 3093 * in the graph whose neighbours are to be checked. 3094 */ 3095 ret = check_noncircular(next, prev, trace); 3096 if (unlikely(bfs_error(ret) || ret == BFS_RMATCH)) 3097 return 0; 3098 3099 if (!check_irq_usage(curr, prev, next)) 3100 return 0; 3101 3102 /* 3103 * Is the <prev> -> <next> dependency already present? 3104 * 3105 * (this may occur even though this is a new chain: consider 3106 * e.g. the L1 -> L2 -> L3 -> L4 and the L5 -> L1 -> L2 -> L3 3107 * chains - the second one will be new, but L1 already has 3108 * L2 added to its dependency list, due to the first chain.) 3109 */ 3110 list_for_each_entry(entry, &hlock_class(prev)->locks_after, entry) { 3111 if (entry->class == hlock_class(next)) { 3112 if (distance == 1) 3113 entry->distance = 1; 3114 entry->dep |= calc_dep(prev, next); 3115 3116 /* 3117 * Also, update the reverse dependency in @next's 3118 * ->locks_before list. 3119 * 3120 * Here we reuse @entry as the cursor, which is fine 3121 * because we won't go to the next iteration of the 3122 * outer loop: 3123 * 3124 * For normal cases, we return in the inner loop. 3125 * 3126 * If we fail to return, we have inconsistency, i.e. 3127 * <prev>::locks_after contains <next> while 3128 * <next>::locks_before doesn't contain <prev>. In 3129 * that case, we return after the inner and indicate 3130 * something is wrong. 3131 */ 3132 list_for_each_entry(entry, &hlock_class(next)->locks_before, entry) { 3133 if (entry->class == hlock_class(prev)) { 3134 if (distance == 1) 3135 entry->distance = 1; 3136 entry->dep |= calc_depb(prev, next); 3137 return 1; 3138 } 3139 } 3140 3141 /* <prev> is not found in <next>::locks_before */ 3142 return 0; 3143 } 3144 } 3145 3146 /* 3147 * Is the <prev> -> <next> link redundant? 3148 */ 3149 ret = check_redundant(prev, next); 3150 if (bfs_error(ret)) 3151 return 0; 3152 else if (ret == BFS_RMATCH) 3153 return 2; 3154 3155 if (!*trace) { 3156 *trace = save_trace(); 3157 if (!*trace) 3158 return 0; 3159 } 3160 3161 /* 3162 * Ok, all validations passed, add the new lock 3163 * to the previous lock's dependency list: 3164 */ 3165 ret = add_lock_to_list(hlock_class(next), hlock_class(prev), 3166 &hlock_class(prev)->locks_after, distance, 3167 calc_dep(prev, next), *trace); 3168 3169 if (!ret) 3170 return 0; 3171 3172 ret = add_lock_to_list(hlock_class(prev), hlock_class(next), 3173 &hlock_class(next)->locks_before, distance, 3174 calc_depb(prev, next), *trace); 3175 if (!ret) 3176 return 0; 3177 3178 return 2; 3179 } 3180 3181 /* 3182 * Add the dependency to all directly-previous locks that are 'relevant'. 3183 * The ones that are relevant are (in increasing distance from curr): 3184 * all consecutive trylock entries and the final non-trylock entry - or 3185 * the end of this context's lock-chain - whichever comes first. 3186 */ 3187 static int 3188 check_prevs_add(struct task_struct *curr, struct held_lock *next) 3189 { 3190 struct lock_trace *trace = NULL; 3191 int depth = curr->lockdep_depth; 3192 struct held_lock *hlock; 3193 3194 /* 3195 * Debugging checks. 3196 * 3197 * Depth must not be zero for a non-head lock: 3198 */ 3199 if (!depth) 3200 goto out_bug; 3201 /* 3202 * At least two relevant locks must exist for this 3203 * to be a head: 3204 */ 3205 if (curr->held_locks[depth].irq_context != 3206 curr->held_locks[depth-1].irq_context) 3207 goto out_bug; 3208 3209 for (;;) { 3210 u16 distance = curr->lockdep_depth - depth + 1; 3211 hlock = curr->held_locks + depth - 1; 3212 3213 if (hlock->check) { 3214 int ret = check_prev_add(curr, hlock, next, distance, &trace); 3215 if (!ret) 3216 return 0; 3217 3218 /* 3219 * Stop after the first non-trylock entry, 3220 * as non-trylock entries have added their 3221 * own direct dependencies already, so this 3222 * lock is connected to them indirectly: 3223 */ 3224 if (!hlock->trylock) 3225 break; 3226 } 3227 3228 depth--; 3229 /* 3230 * End of lock-stack? 3231 */ 3232 if (!depth) 3233 break; 3234 /* 3235 * Stop the search if we cross into another context: 3236 */ 3237 if (curr->held_locks[depth].irq_context != 3238 curr->held_locks[depth-1].irq_context) 3239 break; 3240 } 3241 return 1; 3242 out_bug: 3243 if (!debug_locks_off_graph_unlock()) 3244 return 0; 3245 3246 /* 3247 * Clearly we all shouldn't be here, but since we made it we 3248 * can reliable say we messed up our state. See the above two 3249 * gotos for reasons why we could possibly end up here. 3250 */ 3251 WARN_ON(1); 3252 3253 return 0; 3254 } 3255 3256 struct lock_chain lock_chains[MAX_LOCKDEP_CHAINS]; 3257 static DECLARE_BITMAP(lock_chains_in_use, MAX_LOCKDEP_CHAINS); 3258 static u16 chain_hlocks[MAX_LOCKDEP_CHAIN_HLOCKS]; 3259 unsigned long nr_zapped_lock_chains; 3260 unsigned int nr_free_chain_hlocks; /* Free chain_hlocks in buckets */ 3261 unsigned int nr_lost_chain_hlocks; /* Lost chain_hlocks */ 3262 unsigned int nr_large_chain_blocks; /* size > MAX_CHAIN_BUCKETS */ 3263 3264 /* 3265 * The first 2 chain_hlocks entries in the chain block in the bucket 3266 * list contains the following meta data: 3267 * 3268 * entry[0]: 3269 * Bit 15 - always set to 1 (it is not a class index) 3270 * Bits 0-14 - upper 15 bits of the next block index 3271 * entry[1] - lower 16 bits of next block index 3272 * 3273 * A next block index of all 1 bits means it is the end of the list. 3274 * 3275 * On the unsized bucket (bucket-0), the 3rd and 4th entries contain 3276 * the chain block size: 3277 * 3278 * entry[2] - upper 16 bits of the chain block size 3279 * entry[3] - lower 16 bits of the chain block size 3280 */ 3281 #define MAX_CHAIN_BUCKETS 16 3282 #define CHAIN_BLK_FLAG (1U << 15) 3283 #define CHAIN_BLK_LIST_END 0xFFFFU 3284 3285 static int chain_block_buckets[MAX_CHAIN_BUCKETS]; 3286 3287 static inline int size_to_bucket(int size) 3288 { 3289 if (size > MAX_CHAIN_BUCKETS) 3290 return 0; 3291 3292 return size - 1; 3293 } 3294 3295 /* 3296 * Iterate all the chain blocks in a bucket. 3297 */ 3298 #define for_each_chain_block(bucket, prev, curr) \ 3299 for ((prev) = -1, (curr) = chain_block_buckets[bucket]; \ 3300 (curr) >= 0; \ 3301 (prev) = (curr), (curr) = chain_block_next(curr)) 3302 3303 /* 3304 * next block or -1 3305 */ 3306 static inline int chain_block_next(int offset) 3307 { 3308 int next = chain_hlocks[offset]; 3309 3310 WARN_ON_ONCE(!(next & CHAIN_BLK_FLAG)); 3311 3312 if (next == CHAIN_BLK_LIST_END) 3313 return -1; 3314 3315 next &= ~CHAIN_BLK_FLAG; 3316 next <<= 16; 3317 next |= chain_hlocks[offset + 1]; 3318 3319 return next; 3320 } 3321 3322 /* 3323 * bucket-0 only 3324 */ 3325 static inline int chain_block_size(int offset) 3326 { 3327 return (chain_hlocks[offset + 2] << 16) | chain_hlocks[offset + 3]; 3328 } 3329 3330 static inline void init_chain_block(int offset, int next, int bucket, int size) 3331 { 3332 chain_hlocks[offset] = (next >> 16) | CHAIN_BLK_FLAG; 3333 chain_hlocks[offset + 1] = (u16)next; 3334 3335 if (size && !bucket) { 3336 chain_hlocks[offset + 2] = size >> 16; 3337 chain_hlocks[offset + 3] = (u16)size; 3338 } 3339 } 3340 3341 static inline void add_chain_block(int offset, int size) 3342 { 3343 int bucket = size_to_bucket(size); 3344 int next = chain_block_buckets[bucket]; 3345 int prev, curr; 3346 3347 if (unlikely(size < 2)) { 3348 /* 3349 * We can't store single entries on the freelist. Leak them. 3350 * 3351 * One possible way out would be to uniquely mark them, other 3352 * than with CHAIN_BLK_FLAG, such that we can recover them when 3353 * the block before it is re-added. 3354 */ 3355 if (size) 3356 nr_lost_chain_hlocks++; 3357 return; 3358 } 3359 3360 nr_free_chain_hlocks += size; 3361 if (!bucket) { 3362 nr_large_chain_blocks++; 3363 3364 /* 3365 * Variable sized, sort large to small. 3366 */ 3367 for_each_chain_block(0, prev, curr) { 3368 if (size >= chain_block_size(curr)) 3369 break; 3370 } 3371 init_chain_block(offset, curr, 0, size); 3372 if (prev < 0) 3373 chain_block_buckets[0] = offset; 3374 else 3375 init_chain_block(prev, offset, 0, 0); 3376 return; 3377 } 3378 /* 3379 * Fixed size, add to head. 3380 */ 3381 init_chain_block(offset, next, bucket, size); 3382 chain_block_buckets[bucket] = offset; 3383 } 3384 3385 /* 3386 * Only the first block in the list can be deleted. 3387 * 3388 * For the variable size bucket[0], the first block (the largest one) is 3389 * returned, broken up and put back into the pool. So if a chain block of 3390 * length > MAX_CHAIN_BUCKETS is ever used and zapped, it will just be 3391 * queued up after the primordial chain block and never be used until the 3392 * hlock entries in the primordial chain block is almost used up. That 3393 * causes fragmentation and reduce allocation efficiency. That can be 3394 * monitored by looking at the "large chain blocks" number in lockdep_stats. 3395 */ 3396 static inline void del_chain_block(int bucket, int size, int next) 3397 { 3398 nr_free_chain_hlocks -= size; 3399 chain_block_buckets[bucket] = next; 3400 3401 if (!bucket) 3402 nr_large_chain_blocks--; 3403 } 3404 3405 static void init_chain_block_buckets(void) 3406 { 3407 int i; 3408 3409 for (i = 0; i < MAX_CHAIN_BUCKETS; i++) 3410 chain_block_buckets[i] = -1; 3411 3412 add_chain_block(0, ARRAY_SIZE(chain_hlocks)); 3413 } 3414 3415 /* 3416 * Return offset of a chain block of the right size or -1 if not found. 3417 * 3418 * Fairly simple worst-fit allocator with the addition of a number of size 3419 * specific free lists. 3420 */ 3421 static int alloc_chain_hlocks(int req) 3422 { 3423 int bucket, curr, size; 3424 3425 /* 3426 * We rely on the MSB to act as an escape bit to denote freelist 3427 * pointers. Make sure this bit isn't set in 'normal' class_idx usage. 3428 */ 3429 BUILD_BUG_ON((MAX_LOCKDEP_KEYS-1) & CHAIN_BLK_FLAG); 3430 3431 init_data_structures_once(); 3432 3433 if (nr_free_chain_hlocks < req) 3434 return -1; 3435 3436 /* 3437 * We require a minimum of 2 (u16) entries to encode a freelist 3438 * 'pointer'. 3439 */ 3440 req = max(req, 2); 3441 bucket = size_to_bucket(req); 3442 curr = chain_block_buckets[bucket]; 3443 3444 if (bucket) { 3445 if (curr >= 0) { 3446 del_chain_block(bucket, req, chain_block_next(curr)); 3447 return curr; 3448 } 3449 /* Try bucket 0 */ 3450 curr = chain_block_buckets[0]; 3451 } 3452 3453 /* 3454 * The variable sized freelist is sorted by size; the first entry is 3455 * the largest. Use it if it fits. 3456 */ 3457 if (curr >= 0) { 3458 size = chain_block_size(curr); 3459 if (likely(size >= req)) { 3460 del_chain_block(0, size, chain_block_next(curr)); 3461 add_chain_block(curr + req, size - req); 3462 return curr; 3463 } 3464 } 3465 3466 /* 3467 * Last resort, split a block in a larger sized bucket. 3468 */ 3469 for (size = MAX_CHAIN_BUCKETS; size > req; size--) { 3470 bucket = size_to_bucket(size); 3471 curr = chain_block_buckets[bucket]; 3472 if (curr < 0) 3473 continue; 3474 3475 del_chain_block(bucket, size, chain_block_next(curr)); 3476 add_chain_block(curr + req, size - req); 3477 return curr; 3478 } 3479 3480 return -1; 3481 } 3482 3483 static inline void free_chain_hlocks(int base, int size) 3484 { 3485 add_chain_block(base, max(size, 2)); 3486 } 3487 3488 struct lock_class *lock_chain_get_class(struct lock_chain *chain, int i) 3489 { 3490 u16 chain_hlock = chain_hlocks[chain->base + i]; 3491 unsigned int class_idx = chain_hlock_class_idx(chain_hlock); 3492 3493 return lock_classes + class_idx; 3494 } 3495 3496 /* 3497 * Returns the index of the first held_lock of the current chain 3498 */ 3499 static inline int get_first_held_lock(struct task_struct *curr, 3500 struct held_lock *hlock) 3501 { 3502 int i; 3503 struct held_lock *hlock_curr; 3504 3505 for (i = curr->lockdep_depth - 1; i >= 0; i--) { 3506 hlock_curr = curr->held_locks + i; 3507 if (hlock_curr->irq_context != hlock->irq_context) 3508 break; 3509 3510 } 3511 3512 return ++i; 3513 } 3514 3515 #ifdef CONFIG_DEBUG_LOCKDEP 3516 /* 3517 * Returns the next chain_key iteration 3518 */ 3519 static u64 print_chain_key_iteration(u16 hlock_id, u64 chain_key) 3520 { 3521 u64 new_chain_key = iterate_chain_key(chain_key, hlock_id); 3522 3523 printk(" hlock_id:%d -> chain_key:%016Lx", 3524 (unsigned int)hlock_id, 3525 (unsigned long long)new_chain_key); 3526 return new_chain_key; 3527 } 3528 3529 static void 3530 print_chain_keys_held_locks(struct task_struct *curr, struct held_lock *hlock_next) 3531 { 3532 struct held_lock *hlock; 3533 u64 chain_key = INITIAL_CHAIN_KEY; 3534 int depth = curr->lockdep_depth; 3535 int i = get_first_held_lock(curr, hlock_next); 3536 3537 printk("depth: %u (irq_context %u)\n", depth - i + 1, 3538 hlock_next->irq_context); 3539 for (; i < depth; i++) { 3540 hlock = curr->held_locks + i; 3541 chain_key = print_chain_key_iteration(hlock_id(hlock), chain_key); 3542 3543 print_lock(hlock); 3544 } 3545 3546 print_chain_key_iteration(hlock_id(hlock_next), chain_key); 3547 print_lock(hlock_next); 3548 } 3549 3550 static void print_chain_keys_chain(struct lock_chain *chain) 3551 { 3552 int i; 3553 u64 chain_key = INITIAL_CHAIN_KEY; 3554 u16 hlock_id; 3555 3556 printk("depth: %u\n", chain->depth); 3557 for (i = 0; i < chain->depth; i++) { 3558 hlock_id = chain_hlocks[chain->base + i]; 3559 chain_key = print_chain_key_iteration(hlock_id, chain_key); 3560 3561 print_lock_name(lock_classes + chain_hlock_class_idx(hlock_id)); 3562 printk("\n"); 3563 } 3564 } 3565 3566 static void print_collision(struct task_struct *curr, 3567 struct held_lock *hlock_next, 3568 struct lock_chain *chain) 3569 { 3570 pr_warn("\n"); 3571 pr_warn("============================\n"); 3572 pr_warn("WARNING: chain_key collision\n"); 3573 print_kernel_ident(); 3574 pr_warn("----------------------------\n"); 3575 pr_warn("%s/%d: ", current->comm, task_pid_nr(current)); 3576 pr_warn("Hash chain already cached but the contents don't match!\n"); 3577 3578 pr_warn("Held locks:"); 3579 print_chain_keys_held_locks(curr, hlock_next); 3580 3581 pr_warn("Locks in cached chain:"); 3582 print_chain_keys_chain(chain); 3583 3584 pr_warn("\nstack backtrace:\n"); 3585 dump_stack(); 3586 } 3587 #endif 3588 3589 /* 3590 * Checks whether the chain and the current held locks are consistent 3591 * in depth and also in content. If they are not it most likely means 3592 * that there was a collision during the calculation of the chain_key. 3593 * Returns: 0 not passed, 1 passed 3594 */ 3595 static int check_no_collision(struct task_struct *curr, 3596 struct held_lock *hlock, 3597 struct lock_chain *chain) 3598 { 3599 #ifdef CONFIG_DEBUG_LOCKDEP 3600 int i, j, id; 3601 3602 i = get_first_held_lock(curr, hlock); 3603 3604 if (DEBUG_LOCKS_WARN_ON(chain->depth != curr->lockdep_depth - (i - 1))) { 3605 print_collision(curr, hlock, chain); 3606 return 0; 3607 } 3608 3609 for (j = 0; j < chain->depth - 1; j++, i++) { 3610 id = hlock_id(&curr->held_locks[i]); 3611 3612 if (DEBUG_LOCKS_WARN_ON(chain_hlocks[chain->base + j] != id)) { 3613 print_collision(curr, hlock, chain); 3614 return 0; 3615 } 3616 } 3617 #endif 3618 return 1; 3619 } 3620 3621 /* 3622 * Given an index that is >= -1, return the index of the next lock chain. 3623 * Return -2 if there is no next lock chain. 3624 */ 3625 long lockdep_next_lockchain(long i) 3626 { 3627 i = find_next_bit(lock_chains_in_use, ARRAY_SIZE(lock_chains), i + 1); 3628 return i < ARRAY_SIZE(lock_chains) ? i : -2; 3629 } 3630 3631 unsigned long lock_chain_count(void) 3632 { 3633 return bitmap_weight(lock_chains_in_use, ARRAY_SIZE(lock_chains)); 3634 } 3635 3636 /* Must be called with the graph lock held. */ 3637 static struct lock_chain *alloc_lock_chain(void) 3638 { 3639 int idx = find_first_zero_bit(lock_chains_in_use, 3640 ARRAY_SIZE(lock_chains)); 3641 3642 if (unlikely(idx >= ARRAY_SIZE(lock_chains))) 3643 return NULL; 3644 __set_bit(idx, lock_chains_in_use); 3645 return lock_chains + idx; 3646 } 3647 3648 /* 3649 * Adds a dependency chain into chain hashtable. And must be called with 3650 * graph_lock held. 3651 * 3652 * Return 0 if fail, and graph_lock is released. 3653 * Return 1 if succeed, with graph_lock held. 3654 */ 3655 static inline int add_chain_cache(struct task_struct *curr, 3656 struct held_lock *hlock, 3657 u64 chain_key) 3658 { 3659 struct hlist_head *hash_head = chainhashentry(chain_key); 3660 struct lock_chain *chain; 3661 int i, j; 3662 3663 /* 3664 * The caller must hold the graph lock, ensure we've got IRQs 3665 * disabled to make this an IRQ-safe lock.. for recursion reasons 3666 * lockdep won't complain about its own locking errors. 3667 */ 3668 if (lockdep_assert_locked()) 3669 return 0; 3670 3671 chain = alloc_lock_chain(); 3672 if (!chain) { 3673 if (!debug_locks_off_graph_unlock()) 3674 return 0; 3675 3676 print_lockdep_off("BUG: MAX_LOCKDEP_CHAINS too low!"); 3677 dump_stack(); 3678 return 0; 3679 } 3680 chain->chain_key = chain_key; 3681 chain->irq_context = hlock->irq_context; 3682 i = get_first_held_lock(curr, hlock); 3683 chain->depth = curr->lockdep_depth + 1 - i; 3684 3685 BUILD_BUG_ON((1UL << 24) <= ARRAY_SIZE(chain_hlocks)); 3686 BUILD_BUG_ON((1UL << 6) <= ARRAY_SIZE(curr->held_locks)); 3687 BUILD_BUG_ON((1UL << 8*sizeof(chain_hlocks[0])) <= ARRAY_SIZE(lock_classes)); 3688 3689 j = alloc_chain_hlocks(chain->depth); 3690 if (j < 0) { 3691 if (!debug_locks_off_graph_unlock()) 3692 return 0; 3693 3694 print_lockdep_off("BUG: MAX_LOCKDEP_CHAIN_HLOCKS too low!"); 3695 dump_stack(); 3696 return 0; 3697 } 3698 3699 chain->base = j; 3700 for (j = 0; j < chain->depth - 1; j++, i++) { 3701 int lock_id = hlock_id(curr->held_locks + i); 3702 3703 chain_hlocks[chain->base + j] = lock_id; 3704 } 3705 chain_hlocks[chain->base + j] = hlock_id(hlock); 3706 hlist_add_head_rcu(&chain->entry, hash_head); 3707 debug_atomic_inc(chain_lookup_misses); 3708 inc_chains(chain->irq_context); 3709 3710 return 1; 3711 } 3712 3713 /* 3714 * Look up a dependency chain. Must be called with either the graph lock or 3715 * the RCU read lock held. 3716 */ 3717 static inline struct lock_chain *lookup_chain_cache(u64 chain_key) 3718 { 3719 struct hlist_head *hash_head = chainhashentry(chain_key); 3720 struct lock_chain *chain; 3721 3722 hlist_for_each_entry_rcu(chain, hash_head, entry) { 3723 if (READ_ONCE(chain->chain_key) == chain_key) { 3724 debug_atomic_inc(chain_lookup_hits); 3725 return chain; 3726 } 3727 } 3728 return NULL; 3729 } 3730 3731 /* 3732 * If the key is not present yet in dependency chain cache then 3733 * add it and return 1 - in this case the new dependency chain is 3734 * validated. If the key is already hashed, return 0. 3735 * (On return with 1 graph_lock is held.) 3736 */ 3737 static inline int lookup_chain_cache_add(struct task_struct *curr, 3738 struct held_lock *hlock, 3739 u64 chain_key) 3740 { 3741 struct lock_class *class = hlock_class(hlock); 3742 struct lock_chain *chain = lookup_chain_cache(chain_key); 3743 3744 if (chain) { 3745 cache_hit: 3746 if (!check_no_collision(curr, hlock, chain)) 3747 return 0; 3748 3749 if (very_verbose(class)) { 3750 printk("\nhash chain already cached, key: " 3751 "%016Lx tail class: [%px] %s\n", 3752 (unsigned long long)chain_key, 3753 class->key, class->name); 3754 } 3755 3756 return 0; 3757 } 3758 3759 if (very_verbose(class)) { 3760 printk("\nnew hash chain, key: %016Lx tail class: [%px] %s\n", 3761 (unsigned long long)chain_key, class->key, class->name); 3762 } 3763 3764 if (!graph_lock()) 3765 return 0; 3766 3767 /* 3768 * We have to walk the chain again locked - to avoid duplicates: 3769 */ 3770 chain = lookup_chain_cache(chain_key); 3771 if (chain) { 3772 graph_unlock(); 3773 goto cache_hit; 3774 } 3775 3776 if (!add_chain_cache(curr, hlock, chain_key)) 3777 return 0; 3778 3779 return 1; 3780 } 3781 3782 static int validate_chain(struct task_struct *curr, 3783 struct held_lock *hlock, 3784 int chain_head, u64 chain_key) 3785 { 3786 /* 3787 * Trylock needs to maintain the stack of held locks, but it 3788 * does not add new dependencies, because trylock can be done 3789 * in any order. 3790 * 3791 * We look up the chain_key and do the O(N^2) check and update of 3792 * the dependencies only if this is a new dependency chain. 3793 * (If lookup_chain_cache_add() return with 1 it acquires 3794 * graph_lock for us) 3795 */ 3796 if (!hlock->trylock && hlock->check && 3797 lookup_chain_cache_add(curr, hlock, chain_key)) { 3798 /* 3799 * Check whether last held lock: 3800 * 3801 * - is irq-safe, if this lock is irq-unsafe 3802 * - is softirq-safe, if this lock is hardirq-unsafe 3803 * 3804 * And check whether the new lock's dependency graph 3805 * could lead back to the previous lock: 3806 * 3807 * - within the current held-lock stack 3808 * - across our accumulated lock dependency records 3809 * 3810 * any of these scenarios could lead to a deadlock. 3811 */ 3812 /* 3813 * The simple case: does the current hold the same lock 3814 * already? 3815 */ 3816 int ret = check_deadlock(curr, hlock); 3817 3818 if (!ret) 3819 return 0; 3820 /* 3821 * Add dependency only if this lock is not the head 3822 * of the chain, and if the new lock introduces no more 3823 * lock dependency (because we already hold a lock with the 3824 * same lock class) nor deadlock (because the nest_lock 3825 * serializes nesting locks), see the comments for 3826 * check_deadlock(). 3827 */ 3828 if (!chain_head && ret != 2) { 3829 if (!check_prevs_add(curr, hlock)) 3830 return 0; 3831 } 3832 3833 graph_unlock(); 3834 } else { 3835 /* after lookup_chain_cache_add(): */ 3836 if (unlikely(!debug_locks)) 3837 return 0; 3838 } 3839 3840 return 1; 3841 } 3842 #else 3843 static inline int validate_chain(struct task_struct *curr, 3844 struct held_lock *hlock, 3845 int chain_head, u64 chain_key) 3846 { 3847 return 1; 3848 } 3849 3850 static void init_chain_block_buckets(void) { } 3851 #endif /* CONFIG_PROVE_LOCKING */ 3852 3853 /* 3854 * We are building curr_chain_key incrementally, so double-check 3855 * it from scratch, to make sure that it's done correctly: 3856 */ 3857 static void check_chain_key(struct task_struct *curr) 3858 { 3859 #ifdef CONFIG_DEBUG_LOCKDEP 3860 struct held_lock *hlock, *prev_hlock = NULL; 3861 unsigned int i; 3862 u64 chain_key = INITIAL_CHAIN_KEY; 3863 3864 for (i = 0; i < curr->lockdep_depth; i++) { 3865 hlock = curr->held_locks + i; 3866 if (chain_key != hlock->prev_chain_key) { 3867 debug_locks_off(); 3868 /* 3869 * We got mighty confused, our chain keys don't match 3870 * with what we expect, someone trample on our task state? 3871 */ 3872 WARN(1, "hm#1, depth: %u [%u], %016Lx != %016Lx\n", 3873 curr->lockdep_depth, i, 3874 (unsigned long long)chain_key, 3875 (unsigned long long)hlock->prev_chain_key); 3876 return; 3877 } 3878 3879 /* 3880 * hlock->class_idx can't go beyond MAX_LOCKDEP_KEYS, but is 3881 * it registered lock class index? 3882 */ 3883 if (DEBUG_LOCKS_WARN_ON(!test_bit(hlock->class_idx, lock_classes_in_use))) 3884 return; 3885 3886 if (prev_hlock && (prev_hlock->irq_context != 3887 hlock->irq_context)) 3888 chain_key = INITIAL_CHAIN_KEY; 3889 chain_key = iterate_chain_key(chain_key, hlock_id(hlock)); 3890 prev_hlock = hlock; 3891 } 3892 if (chain_key != curr->curr_chain_key) { 3893 debug_locks_off(); 3894 /* 3895 * More smoking hash instead of calculating it, damn see these 3896 * numbers float.. I bet that a pink elephant stepped on my memory. 3897 */ 3898 WARN(1, "hm#2, depth: %u [%u], %016Lx != %016Lx\n", 3899 curr->lockdep_depth, i, 3900 (unsigned long long)chain_key, 3901 (unsigned long long)curr->curr_chain_key); 3902 } 3903 #endif 3904 } 3905 3906 #ifdef CONFIG_PROVE_LOCKING 3907 static int mark_lock(struct task_struct *curr, struct held_lock *this, 3908 enum lock_usage_bit new_bit); 3909 3910 static void print_usage_bug_scenario(struct held_lock *lock) 3911 { 3912 struct lock_class *class = hlock_class(lock); 3913 3914 printk(" Possible unsafe locking scenario:\n\n"); 3915 printk(" CPU0\n"); 3916 printk(" ----\n"); 3917 printk(" lock("); 3918 __print_lock_name(class); 3919 printk(KERN_CONT ");\n"); 3920 printk(" <Interrupt>\n"); 3921 printk(" lock("); 3922 __print_lock_name(class); 3923 printk(KERN_CONT ");\n"); 3924 printk("\n *** DEADLOCK ***\n\n"); 3925 } 3926 3927 static void 3928 print_usage_bug(struct task_struct *curr, struct held_lock *this, 3929 enum lock_usage_bit prev_bit, enum lock_usage_bit new_bit) 3930 { 3931 if (!debug_locks_off() || debug_locks_silent) 3932 return; 3933 3934 pr_warn("\n"); 3935 pr_warn("================================\n"); 3936 pr_warn("WARNING: inconsistent lock state\n"); 3937 print_kernel_ident(); 3938 pr_warn("--------------------------------\n"); 3939 3940 pr_warn("inconsistent {%s} -> {%s} usage.\n", 3941 usage_str[prev_bit], usage_str[new_bit]); 3942 3943 pr_warn("%s/%d [HC%u[%lu]:SC%u[%lu]:HE%u:SE%u] takes:\n", 3944 curr->comm, task_pid_nr(curr), 3945 lockdep_hardirq_context(), hardirq_count() >> HARDIRQ_SHIFT, 3946 lockdep_softirq_context(curr), softirq_count() >> SOFTIRQ_SHIFT, 3947 lockdep_hardirqs_enabled(), 3948 lockdep_softirqs_enabled(curr)); 3949 print_lock(this); 3950 3951 pr_warn("{%s} state was registered at:\n", usage_str[prev_bit]); 3952 print_lock_trace(hlock_class(this)->usage_traces[prev_bit], 1); 3953 3954 print_irqtrace_events(curr); 3955 pr_warn("\nother info that might help us debug this:\n"); 3956 print_usage_bug_scenario(this); 3957 3958 lockdep_print_held_locks(curr); 3959 3960 pr_warn("\nstack backtrace:\n"); 3961 dump_stack(); 3962 } 3963 3964 /* 3965 * Print out an error if an invalid bit is set: 3966 */ 3967 static inline int 3968 valid_state(struct task_struct *curr, struct held_lock *this, 3969 enum lock_usage_bit new_bit, enum lock_usage_bit bad_bit) 3970 { 3971 if (unlikely(hlock_class(this)->usage_mask & (1 << bad_bit))) { 3972 graph_unlock(); 3973 print_usage_bug(curr, this, bad_bit, new_bit); 3974 return 0; 3975 } 3976 return 1; 3977 } 3978 3979 3980 /* 3981 * print irq inversion bug: 3982 */ 3983 static void 3984 print_irq_inversion_bug(struct task_struct *curr, 3985 struct lock_list *root, struct lock_list *other, 3986 struct held_lock *this, int forwards, 3987 const char *irqclass) 3988 { 3989 struct lock_list *entry = other; 3990 struct lock_list *middle = NULL; 3991 int depth; 3992 3993 if (!debug_locks_off_graph_unlock() || debug_locks_silent) 3994 return; 3995 3996 pr_warn("\n"); 3997 pr_warn("========================================================\n"); 3998 pr_warn("WARNING: possible irq lock inversion dependency detected\n"); 3999 print_kernel_ident(); 4000 pr_warn("--------------------------------------------------------\n"); 4001 pr_warn("%s/%d just changed the state of lock:\n", 4002 curr->comm, task_pid_nr(curr)); 4003 print_lock(this); 4004 if (forwards) 4005 pr_warn("but this lock took another, %s-unsafe lock in the past:\n", irqclass); 4006 else 4007 pr_warn("but this lock was taken by another, %s-safe lock in the past:\n", irqclass); 4008 print_lock_name(other->class); 4009 pr_warn("\n\nand interrupts could create inverse lock ordering between them.\n\n"); 4010 4011 pr_warn("\nother info that might help us debug this:\n"); 4012 4013 /* Find a middle lock (if one exists) */ 4014 depth = get_lock_depth(other); 4015 do { 4016 if (depth == 0 && (entry != root)) { 4017 pr_warn("lockdep:%s bad path found in chain graph\n", __func__); 4018 break; 4019 } 4020 middle = entry; 4021 entry = get_lock_parent(entry); 4022 depth--; 4023 } while (entry && entry != root && (depth >= 0)); 4024 if (forwards) 4025 print_irq_lock_scenario(root, other, 4026 middle ? middle->class : root->class, other->class); 4027 else 4028 print_irq_lock_scenario(other, root, 4029 middle ? middle->class : other->class, root->class); 4030 4031 lockdep_print_held_locks(curr); 4032 4033 pr_warn("\nthe shortest dependencies between 2nd lock and 1st lock:\n"); 4034 root->trace = save_trace(); 4035 if (!root->trace) 4036 return; 4037 print_shortest_lock_dependencies(other, root); 4038 4039 pr_warn("\nstack backtrace:\n"); 4040 dump_stack(); 4041 } 4042 4043 /* 4044 * Prove that in the forwards-direction subgraph starting at <this> 4045 * there is no lock matching <mask>: 4046 */ 4047 static int 4048 check_usage_forwards(struct task_struct *curr, struct held_lock *this, 4049 enum lock_usage_bit bit) 4050 { 4051 enum bfs_result ret; 4052 struct lock_list root; 4053 struct lock_list *target_entry; 4054 enum lock_usage_bit read_bit = bit + LOCK_USAGE_READ_MASK; 4055 unsigned usage_mask = lock_flag(bit) | lock_flag(read_bit); 4056 4057 bfs_init_root(&root, this); 4058 ret = find_usage_forwards(&root, usage_mask, &target_entry); 4059 if (bfs_error(ret)) { 4060 print_bfs_bug(ret); 4061 return 0; 4062 } 4063 if (ret == BFS_RNOMATCH) 4064 return 1; 4065 4066 /* Check whether write or read usage is the match */ 4067 if (target_entry->class->usage_mask & lock_flag(bit)) { 4068 print_irq_inversion_bug(curr, &root, target_entry, 4069 this, 1, state_name(bit)); 4070 } else { 4071 print_irq_inversion_bug(curr, &root, target_entry, 4072 this, 1, state_name(read_bit)); 4073 } 4074 4075 return 0; 4076 } 4077 4078 /* 4079 * Prove that in the backwards-direction subgraph starting at <this> 4080 * there is no lock matching <mask>: 4081 */ 4082 static int 4083 check_usage_backwards(struct task_struct *curr, struct held_lock *this, 4084 enum lock_usage_bit bit) 4085 { 4086 enum bfs_result ret; 4087 struct lock_list root; 4088 struct lock_list *target_entry; 4089 enum lock_usage_bit read_bit = bit + LOCK_USAGE_READ_MASK; 4090 unsigned usage_mask = lock_flag(bit) | lock_flag(read_bit); 4091 4092 bfs_init_rootb(&root, this); 4093 ret = find_usage_backwards(&root, usage_mask, &target_entry); 4094 if (bfs_error(ret)) { 4095 print_bfs_bug(ret); 4096 return 0; 4097 } 4098 if (ret == BFS_RNOMATCH) 4099 return 1; 4100 4101 /* Check whether write or read usage is the match */ 4102 if (target_entry->class->usage_mask & lock_flag(bit)) { 4103 print_irq_inversion_bug(curr, &root, target_entry, 4104 this, 0, state_name(bit)); 4105 } else { 4106 print_irq_inversion_bug(curr, &root, target_entry, 4107 this, 0, state_name(read_bit)); 4108 } 4109 4110 return 0; 4111 } 4112 4113 void print_irqtrace_events(struct task_struct *curr) 4114 { 4115 const struct irqtrace_events *trace = &curr->irqtrace; 4116 4117 printk("irq event stamp: %u\n", trace->irq_events); 4118 printk("hardirqs last enabled at (%u): [<%px>] %pS\n", 4119 trace->hardirq_enable_event, (void *)trace->hardirq_enable_ip, 4120 (void *)trace->hardirq_enable_ip); 4121 printk("hardirqs last disabled at (%u): [<%px>] %pS\n", 4122 trace->hardirq_disable_event, (void *)trace->hardirq_disable_ip, 4123 (void *)trace->hardirq_disable_ip); 4124 printk("softirqs last enabled at (%u): [<%px>] %pS\n", 4125 trace->softirq_enable_event, (void *)trace->softirq_enable_ip, 4126 (void *)trace->softirq_enable_ip); 4127 printk("softirqs last disabled at (%u): [<%px>] %pS\n", 4128 trace->softirq_disable_event, (void *)trace->softirq_disable_ip, 4129 (void *)trace->softirq_disable_ip); 4130 } 4131 4132 static int HARDIRQ_verbose(struct lock_class *class) 4133 { 4134 #if HARDIRQ_VERBOSE 4135 return class_filter(class); 4136 #endif 4137 return 0; 4138 } 4139 4140 static int SOFTIRQ_verbose(struct lock_class *class) 4141 { 4142 #if SOFTIRQ_VERBOSE 4143 return class_filter(class); 4144 #endif 4145 return 0; 4146 } 4147 4148 static int (*state_verbose_f[])(struct lock_class *class) = { 4149 #define LOCKDEP_STATE(__STATE) \ 4150 __STATE##_verbose, 4151 #include "lockdep_states.h" 4152 #undef LOCKDEP_STATE 4153 }; 4154 4155 static inline int state_verbose(enum lock_usage_bit bit, 4156 struct lock_class *class) 4157 { 4158 return state_verbose_f[bit >> LOCK_USAGE_DIR_MASK](class); 4159 } 4160 4161 typedef int (*check_usage_f)(struct task_struct *, struct held_lock *, 4162 enum lock_usage_bit bit, const char *name); 4163 4164 static int 4165 mark_lock_irq(struct task_struct *curr, struct held_lock *this, 4166 enum lock_usage_bit new_bit) 4167 { 4168 int excl_bit = exclusive_bit(new_bit); 4169 int read = new_bit & LOCK_USAGE_READ_MASK; 4170 int dir = new_bit & LOCK_USAGE_DIR_MASK; 4171 4172 /* 4173 * Validate that this particular lock does not have conflicting 4174 * usage states. 4175 */ 4176 if (!valid_state(curr, this, new_bit, excl_bit)) 4177 return 0; 4178 4179 /* 4180 * Check for read in write conflicts 4181 */ 4182 if (!read && !valid_state(curr, this, new_bit, 4183 excl_bit + LOCK_USAGE_READ_MASK)) 4184 return 0; 4185 4186 4187 /* 4188 * Validate that the lock dependencies don't have conflicting usage 4189 * states. 4190 */ 4191 if (dir) { 4192 /* 4193 * mark ENABLED has to look backwards -- to ensure no dependee 4194 * has USED_IN state, which, again, would allow recursion deadlocks. 4195 */ 4196 if (!check_usage_backwards(curr, this, excl_bit)) 4197 return 0; 4198 } else { 4199 /* 4200 * mark USED_IN has to look forwards -- to ensure no dependency 4201 * has ENABLED state, which would allow recursion deadlocks. 4202 */ 4203 if (!check_usage_forwards(curr, this, excl_bit)) 4204 return 0; 4205 } 4206 4207 if (state_verbose(new_bit, hlock_class(this))) 4208 return 2; 4209 4210 return 1; 4211 } 4212 4213 /* 4214 * Mark all held locks with a usage bit: 4215 */ 4216 static int 4217 mark_held_locks(struct task_struct *curr, enum lock_usage_bit base_bit) 4218 { 4219 struct held_lock *hlock; 4220 int i; 4221 4222 for (i = 0; i < curr->lockdep_depth; i++) { 4223 enum lock_usage_bit hlock_bit = base_bit; 4224 hlock = curr->held_locks + i; 4225 4226 if (hlock->read) 4227 hlock_bit += LOCK_USAGE_READ_MASK; 4228 4229 BUG_ON(hlock_bit >= LOCK_USAGE_STATES); 4230 4231 if (!hlock->check) 4232 continue; 4233 4234 if (!mark_lock(curr, hlock, hlock_bit)) 4235 return 0; 4236 } 4237 4238 return 1; 4239 } 4240 4241 /* 4242 * Hardirqs will be enabled: 4243 */ 4244 static void __trace_hardirqs_on_caller(void) 4245 { 4246 struct task_struct *curr = current; 4247 4248 /* 4249 * We are going to turn hardirqs on, so set the 4250 * usage bit for all held locks: 4251 */ 4252 if (!mark_held_locks(curr, LOCK_ENABLED_HARDIRQ)) 4253 return; 4254 /* 4255 * If we have softirqs enabled, then set the usage 4256 * bit for all held locks. (disabled hardirqs prevented 4257 * this bit from being set before) 4258 */ 4259 if (curr->softirqs_enabled) 4260 mark_held_locks(curr, LOCK_ENABLED_SOFTIRQ); 4261 } 4262 4263 /** 4264 * lockdep_hardirqs_on_prepare - Prepare for enabling interrupts 4265 * 4266 * Invoked before a possible transition to RCU idle from exit to user or 4267 * guest mode. This ensures that all RCU operations are done before RCU 4268 * stops watching. After the RCU transition lockdep_hardirqs_on() has to be 4269 * invoked to set the final state. 4270 */ 4271 void lockdep_hardirqs_on_prepare(void) 4272 { 4273 if (unlikely(!debug_locks)) 4274 return; 4275 4276 /* 4277 * NMIs do not (and cannot) track lock dependencies, nothing to do. 4278 */ 4279 if (unlikely(in_nmi())) 4280 return; 4281 4282 if (unlikely(this_cpu_read(lockdep_recursion))) 4283 return; 4284 4285 if (unlikely(lockdep_hardirqs_enabled())) { 4286 /* 4287 * Neither irq nor preemption are disabled here 4288 * so this is racy by nature but losing one hit 4289 * in a stat is not a big deal. 4290 */ 4291 __debug_atomic_inc(redundant_hardirqs_on); 4292 return; 4293 } 4294 4295 /* 4296 * We're enabling irqs and according to our state above irqs weren't 4297 * already enabled, yet we find the hardware thinks they are in fact 4298 * enabled.. someone messed up their IRQ state tracing. 4299 */ 4300 if (DEBUG_LOCKS_WARN_ON(!irqs_disabled())) 4301 return; 4302 4303 /* 4304 * See the fine text that goes along with this variable definition. 4305 */ 4306 if (DEBUG_LOCKS_WARN_ON(early_boot_irqs_disabled)) 4307 return; 4308 4309 /* 4310 * Can't allow enabling interrupts while in an interrupt handler, 4311 * that's general bad form and such. Recursion, limited stack etc.. 4312 */ 4313 if (DEBUG_LOCKS_WARN_ON(lockdep_hardirq_context())) 4314 return; 4315 4316 current->hardirq_chain_key = current->curr_chain_key; 4317 4318 lockdep_recursion_inc(); 4319 __trace_hardirqs_on_caller(); 4320 lockdep_recursion_finish(); 4321 } 4322 EXPORT_SYMBOL_GPL(lockdep_hardirqs_on_prepare); 4323 4324 void noinstr lockdep_hardirqs_on(unsigned long ip) 4325 { 4326 struct irqtrace_events *trace = ¤t->irqtrace; 4327 4328 if (unlikely(!debug_locks)) 4329 return; 4330 4331 /* 4332 * NMIs can happen in the middle of local_irq_{en,dis}able() where the 4333 * tracking state and hardware state are out of sync. 4334 * 4335 * NMIs must save lockdep_hardirqs_enabled() to restore IRQ state from, 4336 * and not rely on hardware state like normal interrupts. 4337 */ 4338 if (unlikely(in_nmi())) { 4339 if (!IS_ENABLED(CONFIG_TRACE_IRQFLAGS_NMI)) 4340 return; 4341 4342 /* 4343 * Skip: 4344 * - recursion check, because NMI can hit lockdep; 4345 * - hardware state check, because above; 4346 * - chain_key check, see lockdep_hardirqs_on_prepare(). 4347 */ 4348 goto skip_checks; 4349 } 4350 4351 if (unlikely(this_cpu_read(lockdep_recursion))) 4352 return; 4353 4354 if (lockdep_hardirqs_enabled()) { 4355 /* 4356 * Neither irq nor preemption are disabled here 4357 * so this is racy by nature but losing one hit 4358 * in a stat is not a big deal. 4359 */ 4360 __debug_atomic_inc(redundant_hardirqs_on); 4361 return; 4362 } 4363 4364 /* 4365 * We're enabling irqs and according to our state above irqs weren't 4366 * already enabled, yet we find the hardware thinks they are in fact 4367 * enabled.. someone messed up their IRQ state tracing. 4368 */ 4369 if (DEBUG_LOCKS_WARN_ON(!irqs_disabled())) 4370 return; 4371 4372 /* 4373 * Ensure the lock stack remained unchanged between 4374 * lockdep_hardirqs_on_prepare() and lockdep_hardirqs_on(). 4375 */ 4376 DEBUG_LOCKS_WARN_ON(current->hardirq_chain_key != 4377 current->curr_chain_key); 4378 4379 skip_checks: 4380 /* we'll do an OFF -> ON transition: */ 4381 __this_cpu_write(hardirqs_enabled, 1); 4382 trace->hardirq_enable_ip = ip; 4383 trace->hardirq_enable_event = ++trace->irq_events; 4384 debug_atomic_inc(hardirqs_on_events); 4385 } 4386 EXPORT_SYMBOL_GPL(lockdep_hardirqs_on); 4387 4388 /* 4389 * Hardirqs were disabled: 4390 */ 4391 void noinstr lockdep_hardirqs_off(unsigned long ip) 4392 { 4393 if (unlikely(!debug_locks)) 4394 return; 4395 4396 /* 4397 * Matching lockdep_hardirqs_on(), allow NMIs in the middle of lockdep; 4398 * they will restore the software state. This ensures the software 4399 * state is consistent inside NMIs as well. 4400 */ 4401 if (in_nmi()) { 4402 if (!IS_ENABLED(CONFIG_TRACE_IRQFLAGS_NMI)) 4403 return; 4404 } else if (__this_cpu_read(lockdep_recursion)) 4405 return; 4406 4407 /* 4408 * So we're supposed to get called after you mask local IRQs, but for 4409 * some reason the hardware doesn't quite think you did a proper job. 4410 */ 4411 if (DEBUG_LOCKS_WARN_ON(!irqs_disabled())) 4412 return; 4413 4414 if (lockdep_hardirqs_enabled()) { 4415 struct irqtrace_events *trace = ¤t->irqtrace; 4416 4417 /* 4418 * We have done an ON -> OFF transition: 4419 */ 4420 __this_cpu_write(hardirqs_enabled, 0); 4421 trace->hardirq_disable_ip = ip; 4422 trace->hardirq_disable_event = ++trace->irq_events; 4423 debug_atomic_inc(hardirqs_off_events); 4424 } else { 4425 debug_atomic_inc(redundant_hardirqs_off); 4426 } 4427 } 4428 EXPORT_SYMBOL_GPL(lockdep_hardirqs_off); 4429 4430 /* 4431 * Softirqs will be enabled: 4432 */ 4433 void lockdep_softirqs_on(unsigned long ip) 4434 { 4435 struct irqtrace_events *trace = ¤t->irqtrace; 4436 4437 if (unlikely(!lockdep_enabled())) 4438 return; 4439 4440 /* 4441 * We fancy IRQs being disabled here, see softirq.c, avoids 4442 * funny state and nesting things. 4443 */ 4444 if (DEBUG_LOCKS_WARN_ON(!irqs_disabled())) 4445 return; 4446 4447 if (current->softirqs_enabled) { 4448 debug_atomic_inc(redundant_softirqs_on); 4449 return; 4450 } 4451 4452 lockdep_recursion_inc(); 4453 /* 4454 * We'll do an OFF -> ON transition: 4455 */ 4456 current->softirqs_enabled = 1; 4457 trace->softirq_enable_ip = ip; 4458 trace->softirq_enable_event = ++trace->irq_events; 4459 debug_atomic_inc(softirqs_on_events); 4460 /* 4461 * We are going to turn softirqs on, so set the 4462 * usage bit for all held locks, if hardirqs are 4463 * enabled too: 4464 */ 4465 if (lockdep_hardirqs_enabled()) 4466 mark_held_locks(current, LOCK_ENABLED_SOFTIRQ); 4467 lockdep_recursion_finish(); 4468 } 4469 4470 /* 4471 * Softirqs were disabled: 4472 */ 4473 void lockdep_softirqs_off(unsigned long ip) 4474 { 4475 if (unlikely(!lockdep_enabled())) 4476 return; 4477 4478 /* 4479 * We fancy IRQs being disabled here, see softirq.c 4480 */ 4481 if (DEBUG_LOCKS_WARN_ON(!irqs_disabled())) 4482 return; 4483 4484 if (current->softirqs_enabled) { 4485 struct irqtrace_events *trace = ¤t->irqtrace; 4486 4487 /* 4488 * We have done an ON -> OFF transition: 4489 */ 4490 current->softirqs_enabled = 0; 4491 trace->softirq_disable_ip = ip; 4492 trace->softirq_disable_event = ++trace->irq_events; 4493 debug_atomic_inc(softirqs_off_events); 4494 /* 4495 * Whoops, we wanted softirqs off, so why aren't they? 4496 */ 4497 DEBUG_LOCKS_WARN_ON(!softirq_count()); 4498 } else 4499 debug_atomic_inc(redundant_softirqs_off); 4500 } 4501 4502 static int 4503 mark_usage(struct task_struct *curr, struct held_lock *hlock, int check) 4504 { 4505 if (!check) 4506 goto lock_used; 4507 4508 /* 4509 * If non-trylock use in a hardirq or softirq context, then 4510 * mark the lock as used in these contexts: 4511 */ 4512 if (!hlock->trylock) { 4513 if (hlock->read) { 4514 if (lockdep_hardirq_context()) 4515 if (!mark_lock(curr, hlock, 4516 LOCK_USED_IN_HARDIRQ_READ)) 4517 return 0; 4518 if (curr->softirq_context) 4519 if (!mark_lock(curr, hlock, 4520 LOCK_USED_IN_SOFTIRQ_READ)) 4521 return 0; 4522 } else { 4523 if (lockdep_hardirq_context()) 4524 if (!mark_lock(curr, hlock, LOCK_USED_IN_HARDIRQ)) 4525 return 0; 4526 if (curr->softirq_context) 4527 if (!mark_lock(curr, hlock, LOCK_USED_IN_SOFTIRQ)) 4528 return 0; 4529 } 4530 } 4531 if (!hlock->hardirqs_off) { 4532 if (hlock->read) { 4533 if (!mark_lock(curr, hlock, 4534 LOCK_ENABLED_HARDIRQ_READ)) 4535 return 0; 4536 if (curr->softirqs_enabled) 4537 if (!mark_lock(curr, hlock, 4538 LOCK_ENABLED_SOFTIRQ_READ)) 4539 return 0; 4540 } else { 4541 if (!mark_lock(curr, hlock, 4542 LOCK_ENABLED_HARDIRQ)) 4543 return 0; 4544 if (curr->softirqs_enabled) 4545 if (!mark_lock(curr, hlock, 4546 LOCK_ENABLED_SOFTIRQ)) 4547 return 0; 4548 } 4549 } 4550 4551 lock_used: 4552 /* mark it as used: */ 4553 if (!mark_lock(curr, hlock, LOCK_USED)) 4554 return 0; 4555 4556 return 1; 4557 } 4558 4559 static inline unsigned int task_irq_context(struct task_struct *task) 4560 { 4561 return LOCK_CHAIN_HARDIRQ_CONTEXT * !!lockdep_hardirq_context() + 4562 LOCK_CHAIN_SOFTIRQ_CONTEXT * !!task->softirq_context; 4563 } 4564 4565 static int separate_irq_context(struct task_struct *curr, 4566 struct held_lock *hlock) 4567 { 4568 unsigned int depth = curr->lockdep_depth; 4569 4570 /* 4571 * Keep track of points where we cross into an interrupt context: 4572 */ 4573 if (depth) { 4574 struct held_lock *prev_hlock; 4575 4576 prev_hlock = curr->held_locks + depth-1; 4577 /* 4578 * If we cross into another context, reset the 4579 * hash key (this also prevents the checking and the 4580 * adding of the dependency to 'prev'): 4581 */ 4582 if (prev_hlock->irq_context != hlock->irq_context) 4583 return 1; 4584 } 4585 return 0; 4586 } 4587 4588 /* 4589 * Mark a lock with a usage bit, and validate the state transition: 4590 */ 4591 static int mark_lock(struct task_struct *curr, struct held_lock *this, 4592 enum lock_usage_bit new_bit) 4593 { 4594 unsigned int new_mask, ret = 1; 4595 4596 if (new_bit >= LOCK_USAGE_STATES) { 4597 DEBUG_LOCKS_WARN_ON(1); 4598 return 0; 4599 } 4600 4601 if (new_bit == LOCK_USED && this->read) 4602 new_bit = LOCK_USED_READ; 4603 4604 new_mask = 1 << new_bit; 4605 4606 /* 4607 * If already set then do not dirty the cacheline, 4608 * nor do any checks: 4609 */ 4610 if (likely(hlock_class(this)->usage_mask & new_mask)) 4611 return 1; 4612 4613 if (!graph_lock()) 4614 return 0; 4615 /* 4616 * Make sure we didn't race: 4617 */ 4618 if (unlikely(hlock_class(this)->usage_mask & new_mask)) 4619 goto unlock; 4620 4621 if (!hlock_class(this)->usage_mask) 4622 debug_atomic_dec(nr_unused_locks); 4623 4624 hlock_class(this)->usage_mask |= new_mask; 4625 4626 if (new_bit < LOCK_TRACE_STATES) { 4627 if (!(hlock_class(this)->usage_traces[new_bit] = save_trace())) 4628 return 0; 4629 } 4630 4631 if (new_bit < LOCK_USED) { 4632 ret = mark_lock_irq(curr, this, new_bit); 4633 if (!ret) 4634 return 0; 4635 } 4636 4637 unlock: 4638 graph_unlock(); 4639 4640 /* 4641 * We must printk outside of the graph_lock: 4642 */ 4643 if (ret == 2) { 4644 printk("\nmarked lock as {%s}:\n", usage_str[new_bit]); 4645 print_lock(this); 4646 print_irqtrace_events(curr); 4647 dump_stack(); 4648 } 4649 4650 return ret; 4651 } 4652 4653 static inline short task_wait_context(struct task_struct *curr) 4654 { 4655 /* 4656 * Set appropriate wait type for the context; for IRQs we have to take 4657 * into account force_irqthread as that is implied by PREEMPT_RT. 4658 */ 4659 if (lockdep_hardirq_context()) { 4660 /* 4661 * Check if force_irqthreads will run us threaded. 4662 */ 4663 if (curr->hardirq_threaded || curr->irq_config) 4664 return LD_WAIT_CONFIG; 4665 4666 return LD_WAIT_SPIN; 4667 } else if (curr->softirq_context) { 4668 /* 4669 * Softirqs are always threaded. 4670 */ 4671 return LD_WAIT_CONFIG; 4672 } 4673 4674 return LD_WAIT_MAX; 4675 } 4676 4677 static int 4678 print_lock_invalid_wait_context(struct task_struct *curr, 4679 struct held_lock *hlock) 4680 { 4681 short curr_inner; 4682 4683 if (!debug_locks_off()) 4684 return 0; 4685 if (debug_locks_silent) 4686 return 0; 4687 4688 pr_warn("\n"); 4689 pr_warn("=============================\n"); 4690 pr_warn("[ BUG: Invalid wait context ]\n"); 4691 print_kernel_ident(); 4692 pr_warn("-----------------------------\n"); 4693 4694 pr_warn("%s/%d is trying to lock:\n", curr->comm, task_pid_nr(curr)); 4695 print_lock(hlock); 4696 4697 pr_warn("other info that might help us debug this:\n"); 4698 4699 curr_inner = task_wait_context(curr); 4700 pr_warn("context-{%d:%d}\n", curr_inner, curr_inner); 4701 4702 lockdep_print_held_locks(curr); 4703 4704 pr_warn("stack backtrace:\n"); 4705 dump_stack(); 4706 4707 return 0; 4708 } 4709 4710 /* 4711 * Verify the wait_type context. 4712 * 4713 * This check validates we take locks in the right wait-type order; that is it 4714 * ensures that we do not take mutexes inside spinlocks and do not attempt to 4715 * acquire spinlocks inside raw_spinlocks and the sort. 4716 * 4717 * The entire thing is slightly more complex because of RCU, RCU is a lock that 4718 * can be taken from (pretty much) any context but also has constraints. 4719 * However when taken in a stricter environment the RCU lock does not loosen 4720 * the constraints. 4721 * 4722 * Therefore we must look for the strictest environment in the lock stack and 4723 * compare that to the lock we're trying to acquire. 4724 */ 4725 static int check_wait_context(struct task_struct *curr, struct held_lock *next) 4726 { 4727 u8 next_inner = hlock_class(next)->wait_type_inner; 4728 u8 next_outer = hlock_class(next)->wait_type_outer; 4729 u8 curr_inner; 4730 int depth; 4731 4732 if (!next_inner || next->trylock) 4733 return 0; 4734 4735 if (!next_outer) 4736 next_outer = next_inner; 4737 4738 /* 4739 * Find start of current irq_context.. 4740 */ 4741 for (depth = curr->lockdep_depth - 1; depth >= 0; depth--) { 4742 struct held_lock *prev = curr->held_locks + depth; 4743 if (prev->irq_context != next->irq_context) 4744 break; 4745 } 4746 depth++; 4747 4748 curr_inner = task_wait_context(curr); 4749 4750 for (; depth < curr->lockdep_depth; depth++) { 4751 struct held_lock *prev = curr->held_locks + depth; 4752 u8 prev_inner = hlock_class(prev)->wait_type_inner; 4753 4754 if (prev_inner) { 4755 /* 4756 * We can have a bigger inner than a previous one 4757 * when outer is smaller than inner, as with RCU. 4758 * 4759 * Also due to trylocks. 4760 */ 4761 curr_inner = min(curr_inner, prev_inner); 4762 } 4763 } 4764 4765 if (next_outer > curr_inner) 4766 return print_lock_invalid_wait_context(curr, next); 4767 4768 return 0; 4769 } 4770 4771 #else /* CONFIG_PROVE_LOCKING */ 4772 4773 static inline int 4774 mark_usage(struct task_struct *curr, struct held_lock *hlock, int check) 4775 { 4776 return 1; 4777 } 4778 4779 static inline unsigned int task_irq_context(struct task_struct *task) 4780 { 4781 return 0; 4782 } 4783 4784 static inline int separate_irq_context(struct task_struct *curr, 4785 struct held_lock *hlock) 4786 { 4787 return 0; 4788 } 4789 4790 static inline int check_wait_context(struct task_struct *curr, 4791 struct held_lock *next) 4792 { 4793 return 0; 4794 } 4795 4796 #endif /* CONFIG_PROVE_LOCKING */ 4797 4798 /* 4799 * Initialize a lock instance's lock-class mapping info: 4800 */ 4801 void lockdep_init_map_type(struct lockdep_map *lock, const char *name, 4802 struct lock_class_key *key, int subclass, 4803 u8 inner, u8 outer, u8 lock_type) 4804 { 4805 int i; 4806 4807 for (i = 0; i < NR_LOCKDEP_CACHING_CLASSES; i++) 4808 lock->class_cache[i] = NULL; 4809 4810 #ifdef CONFIG_LOCK_STAT 4811 lock->cpu = raw_smp_processor_id(); 4812 #endif 4813 4814 /* 4815 * Can't be having no nameless bastards around this place! 4816 */ 4817 if (DEBUG_LOCKS_WARN_ON(!name)) { 4818 lock->name = "NULL"; 4819 return; 4820 } 4821 4822 lock->name = name; 4823 4824 lock->wait_type_outer = outer; 4825 lock->wait_type_inner = inner; 4826 lock->lock_type = lock_type; 4827 4828 /* 4829 * No key, no joy, we need to hash something. 4830 */ 4831 if (DEBUG_LOCKS_WARN_ON(!key)) 4832 return; 4833 /* 4834 * Sanity check, the lock-class key must either have been allocated 4835 * statically or must have been registered as a dynamic key. 4836 */ 4837 if (!static_obj(key) && !is_dynamic_key(key)) { 4838 if (debug_locks) 4839 printk(KERN_ERR "BUG: key %px has not been registered!\n", key); 4840 DEBUG_LOCKS_WARN_ON(1); 4841 return; 4842 } 4843 lock->key = key; 4844 4845 if (unlikely(!debug_locks)) 4846 return; 4847 4848 if (subclass) { 4849 unsigned long flags; 4850 4851 if (DEBUG_LOCKS_WARN_ON(!lockdep_enabled())) 4852 return; 4853 4854 raw_local_irq_save(flags); 4855 lockdep_recursion_inc(); 4856 register_lock_class(lock, subclass, 1); 4857 lockdep_recursion_finish(); 4858 raw_local_irq_restore(flags); 4859 } 4860 } 4861 EXPORT_SYMBOL_GPL(lockdep_init_map_type); 4862 4863 struct lock_class_key __lockdep_no_validate__; 4864 EXPORT_SYMBOL_GPL(__lockdep_no_validate__); 4865 4866 static void 4867 print_lock_nested_lock_not_held(struct task_struct *curr, 4868 struct held_lock *hlock) 4869 { 4870 if (!debug_locks_off()) 4871 return; 4872 if (debug_locks_silent) 4873 return; 4874 4875 pr_warn("\n"); 4876 pr_warn("==================================\n"); 4877 pr_warn("WARNING: Nested lock was not taken\n"); 4878 print_kernel_ident(); 4879 pr_warn("----------------------------------\n"); 4880 4881 pr_warn("%s/%d is trying to lock:\n", curr->comm, task_pid_nr(curr)); 4882 print_lock(hlock); 4883 4884 pr_warn("\nbut this task is not holding:\n"); 4885 pr_warn("%s\n", hlock->nest_lock->name); 4886 4887 pr_warn("\nstack backtrace:\n"); 4888 dump_stack(); 4889 4890 pr_warn("\nother info that might help us debug this:\n"); 4891 lockdep_print_held_locks(curr); 4892 4893 pr_warn("\nstack backtrace:\n"); 4894 dump_stack(); 4895 } 4896 4897 static int __lock_is_held(const struct lockdep_map *lock, int read); 4898 4899 /* 4900 * This gets called for every mutex_lock*()/spin_lock*() operation. 4901 * We maintain the dependency maps and validate the locking attempt: 4902 * 4903 * The callers must make sure that IRQs are disabled before calling it, 4904 * otherwise we could get an interrupt which would want to take locks, 4905 * which would end up in lockdep again. 4906 */ 4907 static int __lock_acquire(struct lockdep_map *lock, unsigned int subclass, 4908 int trylock, int read, int check, int hardirqs_off, 4909 struct lockdep_map *nest_lock, unsigned long ip, 4910 int references, int pin_count) 4911 { 4912 struct task_struct *curr = current; 4913 struct lock_class *class = NULL; 4914 struct held_lock *hlock; 4915 unsigned int depth; 4916 int chain_head = 0; 4917 int class_idx; 4918 u64 chain_key; 4919 4920 if (unlikely(!debug_locks)) 4921 return 0; 4922 4923 if (!prove_locking || lock->key == &__lockdep_no_validate__) 4924 check = 0; 4925 4926 if (subclass < NR_LOCKDEP_CACHING_CLASSES) 4927 class = lock->class_cache[subclass]; 4928 /* 4929 * Not cached? 4930 */ 4931 if (unlikely(!class)) { 4932 class = register_lock_class(lock, subclass, 0); 4933 if (!class) 4934 return 0; 4935 } 4936 4937 debug_class_ops_inc(class); 4938 4939 if (very_verbose(class)) { 4940 printk("\nacquire class [%px] %s", class->key, class->name); 4941 if (class->name_version > 1) 4942 printk(KERN_CONT "#%d", class->name_version); 4943 printk(KERN_CONT "\n"); 4944 dump_stack(); 4945 } 4946 4947 /* 4948 * Add the lock to the list of currently held locks. 4949 * (we dont increase the depth just yet, up until the 4950 * dependency checks are done) 4951 */ 4952 depth = curr->lockdep_depth; 4953 /* 4954 * Ran out of static storage for our per-task lock stack again have we? 4955 */ 4956 if (DEBUG_LOCKS_WARN_ON(depth >= MAX_LOCK_DEPTH)) 4957 return 0; 4958 4959 class_idx = class - lock_classes; 4960 4961 if (depth) { /* we're holding locks */ 4962 hlock = curr->held_locks + depth - 1; 4963 if (hlock->class_idx == class_idx && nest_lock) { 4964 if (!references) 4965 references++; 4966 4967 if (!hlock->references) 4968 hlock->references++; 4969 4970 hlock->references += references; 4971 4972 /* Overflow */ 4973 if (DEBUG_LOCKS_WARN_ON(hlock->references < references)) 4974 return 0; 4975 4976 return 2; 4977 } 4978 } 4979 4980 hlock = curr->held_locks + depth; 4981 /* 4982 * Plain impossible, we just registered it and checked it weren't no 4983 * NULL like.. I bet this mushroom I ate was good! 4984 */ 4985 if (DEBUG_LOCKS_WARN_ON(!class)) 4986 return 0; 4987 hlock->class_idx = class_idx; 4988 hlock->acquire_ip = ip; 4989 hlock->instance = lock; 4990 hlock->nest_lock = nest_lock; 4991 hlock->irq_context = task_irq_context(curr); 4992 hlock->trylock = trylock; 4993 hlock->read = read; 4994 hlock->check = check; 4995 hlock->hardirqs_off = !!hardirqs_off; 4996 hlock->references = references; 4997 #ifdef CONFIG_LOCK_STAT 4998 hlock->waittime_stamp = 0; 4999 hlock->holdtime_stamp = lockstat_clock(); 5000 #endif 5001 hlock->pin_count = pin_count; 5002 5003 if (check_wait_context(curr, hlock)) 5004 return 0; 5005 5006 /* Initialize the lock usage bit */ 5007 if (!mark_usage(curr, hlock, check)) 5008 return 0; 5009 5010 /* 5011 * Calculate the chain hash: it's the combined hash of all the 5012 * lock keys along the dependency chain. We save the hash value 5013 * at every step so that we can get the current hash easily 5014 * after unlock. The chain hash is then used to cache dependency 5015 * results. 5016 * 5017 * The 'key ID' is what is the most compact key value to drive 5018 * the hash, not class->key. 5019 */ 5020 /* 5021 * Whoops, we did it again.. class_idx is invalid. 5022 */ 5023 if (DEBUG_LOCKS_WARN_ON(!test_bit(class_idx, lock_classes_in_use))) 5024 return 0; 5025 5026 chain_key = curr->curr_chain_key; 5027 if (!depth) { 5028 /* 5029 * How can we have a chain hash when we ain't got no keys?! 5030 */ 5031 if (DEBUG_LOCKS_WARN_ON(chain_key != INITIAL_CHAIN_KEY)) 5032 return 0; 5033 chain_head = 1; 5034 } 5035 5036 hlock->prev_chain_key = chain_key; 5037 if (separate_irq_context(curr, hlock)) { 5038 chain_key = INITIAL_CHAIN_KEY; 5039 chain_head = 1; 5040 } 5041 chain_key = iterate_chain_key(chain_key, hlock_id(hlock)); 5042 5043 if (nest_lock && !__lock_is_held(nest_lock, -1)) { 5044 print_lock_nested_lock_not_held(curr, hlock); 5045 return 0; 5046 } 5047 5048 if (!debug_locks_silent) { 5049 WARN_ON_ONCE(depth && !hlock_class(hlock - 1)->key); 5050 WARN_ON_ONCE(!hlock_class(hlock)->key); 5051 } 5052 5053 if (!validate_chain(curr, hlock, chain_head, chain_key)) 5054 return 0; 5055 5056 curr->curr_chain_key = chain_key; 5057 curr->lockdep_depth++; 5058 check_chain_key(curr); 5059 #ifdef CONFIG_DEBUG_LOCKDEP 5060 if (unlikely(!debug_locks)) 5061 return 0; 5062 #endif 5063 if (unlikely(curr->lockdep_depth >= MAX_LOCK_DEPTH)) { 5064 debug_locks_off(); 5065 print_lockdep_off("BUG: MAX_LOCK_DEPTH too low!"); 5066 printk(KERN_DEBUG "depth: %i max: %lu!\n", 5067 curr->lockdep_depth, MAX_LOCK_DEPTH); 5068 5069 lockdep_print_held_locks(current); 5070 debug_show_all_locks(); 5071 dump_stack(); 5072 5073 return 0; 5074 } 5075 5076 if (unlikely(curr->lockdep_depth > max_lockdep_depth)) 5077 max_lockdep_depth = curr->lockdep_depth; 5078 5079 return 1; 5080 } 5081 5082 static void print_unlock_imbalance_bug(struct task_struct *curr, 5083 struct lockdep_map *lock, 5084 unsigned long ip) 5085 { 5086 if (!debug_locks_off()) 5087 return; 5088 if (debug_locks_silent) 5089 return; 5090 5091 pr_warn("\n"); 5092 pr_warn("=====================================\n"); 5093 pr_warn("WARNING: bad unlock balance detected!\n"); 5094 print_kernel_ident(); 5095 pr_warn("-------------------------------------\n"); 5096 pr_warn("%s/%d is trying to release lock (", 5097 curr->comm, task_pid_nr(curr)); 5098 print_lockdep_cache(lock); 5099 pr_cont(") at:\n"); 5100 print_ip_sym(KERN_WARNING, ip); 5101 pr_warn("but there are no more locks to release!\n"); 5102 pr_warn("\nother info that might help us debug this:\n"); 5103 lockdep_print_held_locks(curr); 5104 5105 pr_warn("\nstack backtrace:\n"); 5106 dump_stack(); 5107 } 5108 5109 static noinstr int match_held_lock(const struct held_lock *hlock, 5110 const struct lockdep_map *lock) 5111 { 5112 if (hlock->instance == lock) 5113 return 1; 5114 5115 if (hlock->references) { 5116 const struct lock_class *class = lock->class_cache[0]; 5117 5118 if (!class) 5119 class = look_up_lock_class(lock, 0); 5120 5121 /* 5122 * If look_up_lock_class() failed to find a class, we're trying 5123 * to test if we hold a lock that has never yet been acquired. 5124 * Clearly if the lock hasn't been acquired _ever_, we're not 5125 * holding it either, so report failure. 5126 */ 5127 if (!class) 5128 return 0; 5129 5130 /* 5131 * References, but not a lock we're actually ref-counting? 5132 * State got messed up, follow the sites that change ->references 5133 * and try to make sense of it. 5134 */ 5135 if (DEBUG_LOCKS_WARN_ON(!hlock->nest_lock)) 5136 return 0; 5137 5138 if (hlock->class_idx == class - lock_classes) 5139 return 1; 5140 } 5141 5142 return 0; 5143 } 5144 5145 /* @depth must not be zero */ 5146 static struct held_lock *find_held_lock(struct task_struct *curr, 5147 struct lockdep_map *lock, 5148 unsigned int depth, int *idx) 5149 { 5150 struct held_lock *ret, *hlock, *prev_hlock; 5151 int i; 5152 5153 i = depth - 1; 5154 hlock = curr->held_locks + i; 5155 ret = hlock; 5156 if (match_held_lock(hlock, lock)) 5157 goto out; 5158 5159 ret = NULL; 5160 for (i--, prev_hlock = hlock--; 5161 i >= 0; 5162 i--, prev_hlock = hlock--) { 5163 /* 5164 * We must not cross into another context: 5165 */ 5166 if (prev_hlock->irq_context != hlock->irq_context) { 5167 ret = NULL; 5168 break; 5169 } 5170 if (match_held_lock(hlock, lock)) { 5171 ret = hlock; 5172 break; 5173 } 5174 } 5175 5176 out: 5177 *idx = i; 5178 return ret; 5179 } 5180 5181 static int reacquire_held_locks(struct task_struct *curr, unsigned int depth, 5182 int idx, unsigned int *merged) 5183 { 5184 struct held_lock *hlock; 5185 int first_idx = idx; 5186 5187 if (DEBUG_LOCKS_WARN_ON(!irqs_disabled())) 5188 return 0; 5189 5190 for (hlock = curr->held_locks + idx; idx < depth; idx++, hlock++) { 5191 switch (__lock_acquire(hlock->instance, 5192 hlock_class(hlock)->subclass, 5193 hlock->trylock, 5194 hlock->read, hlock->check, 5195 hlock->hardirqs_off, 5196 hlock->nest_lock, hlock->acquire_ip, 5197 hlock->references, hlock->pin_count)) { 5198 case 0: 5199 return 1; 5200 case 1: 5201 break; 5202 case 2: 5203 *merged += (idx == first_idx); 5204 break; 5205 default: 5206 WARN_ON(1); 5207 return 0; 5208 } 5209 } 5210 return 0; 5211 } 5212 5213 static int 5214 __lock_set_class(struct lockdep_map *lock, const char *name, 5215 struct lock_class_key *key, unsigned int subclass, 5216 unsigned long ip) 5217 { 5218 struct task_struct *curr = current; 5219 unsigned int depth, merged = 0; 5220 struct held_lock *hlock; 5221 struct lock_class *class; 5222 int i; 5223 5224 if (unlikely(!debug_locks)) 5225 return 0; 5226 5227 depth = curr->lockdep_depth; 5228 /* 5229 * This function is about (re)setting the class of a held lock, 5230 * yet we're not actually holding any locks. Naughty user! 5231 */ 5232 if (DEBUG_LOCKS_WARN_ON(!depth)) 5233 return 0; 5234 5235 hlock = find_held_lock(curr, lock, depth, &i); 5236 if (!hlock) { 5237 print_unlock_imbalance_bug(curr, lock, ip); 5238 return 0; 5239 } 5240 5241 lockdep_init_map_type(lock, name, key, 0, 5242 lock->wait_type_inner, 5243 lock->wait_type_outer, 5244 lock->lock_type); 5245 class = register_lock_class(lock, subclass, 0); 5246 hlock->class_idx = class - lock_classes; 5247 5248 curr->lockdep_depth = i; 5249 curr->curr_chain_key = hlock->prev_chain_key; 5250 5251 if (reacquire_held_locks(curr, depth, i, &merged)) 5252 return 0; 5253 5254 /* 5255 * I took it apart and put it back together again, except now I have 5256 * these 'spare' parts.. where shall I put them. 5257 */ 5258 if (DEBUG_LOCKS_WARN_ON(curr->lockdep_depth != depth - merged)) 5259 return 0; 5260 return 1; 5261 } 5262 5263 static int __lock_downgrade(struct lockdep_map *lock, unsigned long ip) 5264 { 5265 struct task_struct *curr = current; 5266 unsigned int depth, merged = 0; 5267 struct held_lock *hlock; 5268 int i; 5269 5270 if (unlikely(!debug_locks)) 5271 return 0; 5272 5273 depth = curr->lockdep_depth; 5274 /* 5275 * This function is about (re)setting the class of a held lock, 5276 * yet we're not actually holding any locks. Naughty user! 5277 */ 5278 if (DEBUG_LOCKS_WARN_ON(!depth)) 5279 return 0; 5280 5281 hlock = find_held_lock(curr, lock, depth, &i); 5282 if (!hlock) { 5283 print_unlock_imbalance_bug(curr, lock, ip); 5284 return 0; 5285 } 5286 5287 curr->lockdep_depth = i; 5288 curr->curr_chain_key = hlock->prev_chain_key; 5289 5290 WARN(hlock->read, "downgrading a read lock"); 5291 hlock->read = 1; 5292 hlock->acquire_ip = ip; 5293 5294 if (reacquire_held_locks(curr, depth, i, &merged)) 5295 return 0; 5296 5297 /* Merging can't happen with unchanged classes.. */ 5298 if (DEBUG_LOCKS_WARN_ON(merged)) 5299 return 0; 5300 5301 /* 5302 * I took it apart and put it back together again, except now I have 5303 * these 'spare' parts.. where shall I put them. 5304 */ 5305 if (DEBUG_LOCKS_WARN_ON(curr->lockdep_depth != depth)) 5306 return 0; 5307 5308 return 1; 5309 } 5310 5311 /* 5312 * Remove the lock from the list of currently held locks - this gets 5313 * called on mutex_unlock()/spin_unlock*() (or on a failed 5314 * mutex_lock_interruptible()). 5315 */ 5316 static int 5317 __lock_release(struct lockdep_map *lock, unsigned long ip) 5318 { 5319 struct task_struct *curr = current; 5320 unsigned int depth, merged = 1; 5321 struct held_lock *hlock; 5322 int i; 5323 5324 if (unlikely(!debug_locks)) 5325 return 0; 5326 5327 depth = curr->lockdep_depth; 5328 /* 5329 * So we're all set to release this lock.. wait what lock? We don't 5330 * own any locks, you've been drinking again? 5331 */ 5332 if (depth <= 0) { 5333 print_unlock_imbalance_bug(curr, lock, ip); 5334 return 0; 5335 } 5336 5337 /* 5338 * Check whether the lock exists in the current stack 5339 * of held locks: 5340 */ 5341 hlock = find_held_lock(curr, lock, depth, &i); 5342 if (!hlock) { 5343 print_unlock_imbalance_bug(curr, lock, ip); 5344 return 0; 5345 } 5346 5347 if (hlock->instance == lock) 5348 lock_release_holdtime(hlock); 5349 5350 WARN(hlock->pin_count, "releasing a pinned lock\n"); 5351 5352 if (hlock->references) { 5353 hlock->references--; 5354 if (hlock->references) { 5355 /* 5356 * We had, and after removing one, still have 5357 * references, the current lock stack is still 5358 * valid. We're done! 5359 */ 5360 return 1; 5361 } 5362 } 5363 5364 /* 5365 * We have the right lock to unlock, 'hlock' points to it. 5366 * Now we remove it from the stack, and add back the other 5367 * entries (if any), recalculating the hash along the way: 5368 */ 5369 5370 curr->lockdep_depth = i; 5371 curr->curr_chain_key = hlock->prev_chain_key; 5372 5373 /* 5374 * The most likely case is when the unlock is on the innermost 5375 * lock. In this case, we are done! 5376 */ 5377 if (i == depth-1) 5378 return 1; 5379 5380 if (reacquire_held_locks(curr, depth, i + 1, &merged)) 5381 return 0; 5382 5383 /* 5384 * We had N bottles of beer on the wall, we drank one, but now 5385 * there's not N-1 bottles of beer left on the wall... 5386 * Pouring two of the bottles together is acceptable. 5387 */ 5388 DEBUG_LOCKS_WARN_ON(curr->lockdep_depth != depth - merged); 5389 5390 /* 5391 * Since reacquire_held_locks() would have called check_chain_key() 5392 * indirectly via __lock_acquire(), we don't need to do it again 5393 * on return. 5394 */ 5395 return 0; 5396 } 5397 5398 static __always_inline 5399 int __lock_is_held(const struct lockdep_map *lock, int read) 5400 { 5401 struct task_struct *curr = current; 5402 int i; 5403 5404 for (i = 0; i < curr->lockdep_depth; i++) { 5405 struct held_lock *hlock = curr->held_locks + i; 5406 5407 if (match_held_lock(hlock, lock)) { 5408 if (read == -1 || !!hlock->read == read) 5409 return LOCK_STATE_HELD; 5410 5411 return LOCK_STATE_NOT_HELD; 5412 } 5413 } 5414 5415 return LOCK_STATE_NOT_HELD; 5416 } 5417 5418 static struct pin_cookie __lock_pin_lock(struct lockdep_map *lock) 5419 { 5420 struct pin_cookie cookie = NIL_COOKIE; 5421 struct task_struct *curr = current; 5422 int i; 5423 5424 if (unlikely(!debug_locks)) 5425 return cookie; 5426 5427 for (i = 0; i < curr->lockdep_depth; i++) { 5428 struct held_lock *hlock = curr->held_locks + i; 5429 5430 if (match_held_lock(hlock, lock)) { 5431 /* 5432 * Grab 16bits of randomness; this is sufficient to not 5433 * be guessable and still allows some pin nesting in 5434 * our u32 pin_count. 5435 */ 5436 cookie.val = 1 + (sched_clock() & 0xffff); 5437 hlock->pin_count += cookie.val; 5438 return cookie; 5439 } 5440 } 5441 5442 WARN(1, "pinning an unheld lock\n"); 5443 return cookie; 5444 } 5445 5446 static void __lock_repin_lock(struct lockdep_map *lock, struct pin_cookie cookie) 5447 { 5448 struct task_struct *curr = current; 5449 int i; 5450 5451 if (unlikely(!debug_locks)) 5452 return; 5453 5454 for (i = 0; i < curr->lockdep_depth; i++) { 5455 struct held_lock *hlock = curr->held_locks + i; 5456 5457 if (match_held_lock(hlock, lock)) { 5458 hlock->pin_count += cookie.val; 5459 return; 5460 } 5461 } 5462 5463 WARN(1, "pinning an unheld lock\n"); 5464 } 5465 5466 static void __lock_unpin_lock(struct lockdep_map *lock, struct pin_cookie cookie) 5467 { 5468 struct task_struct *curr = current; 5469 int i; 5470 5471 if (unlikely(!debug_locks)) 5472 return; 5473 5474 for (i = 0; i < curr->lockdep_depth; i++) { 5475 struct held_lock *hlock = curr->held_locks + i; 5476 5477 if (match_held_lock(hlock, lock)) { 5478 if (WARN(!hlock->pin_count, "unpinning an unpinned lock\n")) 5479 return; 5480 5481 hlock->pin_count -= cookie.val; 5482 5483 if (WARN((int)hlock->pin_count < 0, "pin count corrupted\n")) 5484 hlock->pin_count = 0; 5485 5486 return; 5487 } 5488 } 5489 5490 WARN(1, "unpinning an unheld lock\n"); 5491 } 5492 5493 /* 5494 * Check whether we follow the irq-flags state precisely: 5495 */ 5496 static noinstr void check_flags(unsigned long flags) 5497 { 5498 #if defined(CONFIG_PROVE_LOCKING) && defined(CONFIG_DEBUG_LOCKDEP) 5499 if (!debug_locks) 5500 return; 5501 5502 /* Get the warning out.. */ 5503 instrumentation_begin(); 5504 5505 if (irqs_disabled_flags(flags)) { 5506 if (DEBUG_LOCKS_WARN_ON(lockdep_hardirqs_enabled())) { 5507 printk("possible reason: unannotated irqs-off.\n"); 5508 } 5509 } else { 5510 if (DEBUG_LOCKS_WARN_ON(!lockdep_hardirqs_enabled())) { 5511 printk("possible reason: unannotated irqs-on.\n"); 5512 } 5513 } 5514 5515 #ifndef CONFIG_PREEMPT_RT 5516 /* 5517 * We dont accurately track softirq state in e.g. 5518 * hardirq contexts (such as on 4KSTACKS), so only 5519 * check if not in hardirq contexts: 5520 */ 5521 if (!hardirq_count()) { 5522 if (softirq_count()) { 5523 /* like the above, but with softirqs */ 5524 DEBUG_LOCKS_WARN_ON(current->softirqs_enabled); 5525 } else { 5526 /* lick the above, does it taste good? */ 5527 DEBUG_LOCKS_WARN_ON(!current->softirqs_enabled); 5528 } 5529 } 5530 #endif 5531 5532 if (!debug_locks) 5533 print_irqtrace_events(current); 5534 5535 instrumentation_end(); 5536 #endif 5537 } 5538 5539 void lock_set_class(struct lockdep_map *lock, const char *name, 5540 struct lock_class_key *key, unsigned int subclass, 5541 unsigned long ip) 5542 { 5543 unsigned long flags; 5544 5545 if (unlikely(!lockdep_enabled())) 5546 return; 5547 5548 raw_local_irq_save(flags); 5549 lockdep_recursion_inc(); 5550 check_flags(flags); 5551 if (__lock_set_class(lock, name, key, subclass, ip)) 5552 check_chain_key(current); 5553 lockdep_recursion_finish(); 5554 raw_local_irq_restore(flags); 5555 } 5556 EXPORT_SYMBOL_GPL(lock_set_class); 5557 5558 void lock_downgrade(struct lockdep_map *lock, unsigned long ip) 5559 { 5560 unsigned long flags; 5561 5562 if (unlikely(!lockdep_enabled())) 5563 return; 5564 5565 raw_local_irq_save(flags); 5566 lockdep_recursion_inc(); 5567 check_flags(flags); 5568 if (__lock_downgrade(lock, ip)) 5569 check_chain_key(current); 5570 lockdep_recursion_finish(); 5571 raw_local_irq_restore(flags); 5572 } 5573 EXPORT_SYMBOL_GPL(lock_downgrade); 5574 5575 /* NMI context !!! */ 5576 static void verify_lock_unused(struct lockdep_map *lock, struct held_lock *hlock, int subclass) 5577 { 5578 #ifdef CONFIG_PROVE_LOCKING 5579 struct lock_class *class = look_up_lock_class(lock, subclass); 5580 unsigned long mask = LOCKF_USED; 5581 5582 /* if it doesn't have a class (yet), it certainly hasn't been used yet */ 5583 if (!class) 5584 return; 5585 5586 /* 5587 * READ locks only conflict with USED, such that if we only ever use 5588 * READ locks, there is no deadlock possible -- RCU. 5589 */ 5590 if (!hlock->read) 5591 mask |= LOCKF_USED_READ; 5592 5593 if (!(class->usage_mask & mask)) 5594 return; 5595 5596 hlock->class_idx = class - lock_classes; 5597 5598 print_usage_bug(current, hlock, LOCK_USED, LOCK_USAGE_STATES); 5599 #endif 5600 } 5601 5602 static bool lockdep_nmi(void) 5603 { 5604 if (raw_cpu_read(lockdep_recursion)) 5605 return false; 5606 5607 if (!in_nmi()) 5608 return false; 5609 5610 return true; 5611 } 5612 5613 /* 5614 * read_lock() is recursive if: 5615 * 1. We force lockdep think this way in selftests or 5616 * 2. The implementation is not queued read/write lock or 5617 * 3. The locker is at an in_interrupt() context. 5618 */ 5619 bool read_lock_is_recursive(void) 5620 { 5621 return force_read_lock_recursive || 5622 !IS_ENABLED(CONFIG_QUEUED_RWLOCKS) || 5623 in_interrupt(); 5624 } 5625 EXPORT_SYMBOL_GPL(read_lock_is_recursive); 5626 5627 /* 5628 * We are not always called with irqs disabled - do that here, 5629 * and also avoid lockdep recursion: 5630 */ 5631 void lock_acquire(struct lockdep_map *lock, unsigned int subclass, 5632 int trylock, int read, int check, 5633 struct lockdep_map *nest_lock, unsigned long ip) 5634 { 5635 unsigned long flags; 5636 5637 trace_lock_acquire(lock, subclass, trylock, read, check, nest_lock, ip); 5638 5639 if (!debug_locks) 5640 return; 5641 5642 if (unlikely(!lockdep_enabled())) { 5643 /* XXX allow trylock from NMI ?!? */ 5644 if (lockdep_nmi() && !trylock) { 5645 struct held_lock hlock; 5646 5647 hlock.acquire_ip = ip; 5648 hlock.instance = lock; 5649 hlock.nest_lock = nest_lock; 5650 hlock.irq_context = 2; // XXX 5651 hlock.trylock = trylock; 5652 hlock.read = read; 5653 hlock.check = check; 5654 hlock.hardirqs_off = true; 5655 hlock.references = 0; 5656 5657 verify_lock_unused(lock, &hlock, subclass); 5658 } 5659 return; 5660 } 5661 5662 raw_local_irq_save(flags); 5663 check_flags(flags); 5664 5665 lockdep_recursion_inc(); 5666 __lock_acquire(lock, subclass, trylock, read, check, 5667 irqs_disabled_flags(flags), nest_lock, ip, 0, 0); 5668 lockdep_recursion_finish(); 5669 raw_local_irq_restore(flags); 5670 } 5671 EXPORT_SYMBOL_GPL(lock_acquire); 5672 5673 void lock_release(struct lockdep_map *lock, unsigned long ip) 5674 { 5675 unsigned long flags; 5676 5677 trace_lock_release(lock, ip); 5678 5679 if (unlikely(!lockdep_enabled())) 5680 return; 5681 5682 raw_local_irq_save(flags); 5683 check_flags(flags); 5684 5685 lockdep_recursion_inc(); 5686 if (__lock_release(lock, ip)) 5687 check_chain_key(current); 5688 lockdep_recursion_finish(); 5689 raw_local_irq_restore(flags); 5690 } 5691 EXPORT_SYMBOL_GPL(lock_release); 5692 5693 noinstr int lock_is_held_type(const struct lockdep_map *lock, int read) 5694 { 5695 unsigned long flags; 5696 int ret = LOCK_STATE_NOT_HELD; 5697 5698 /* 5699 * Avoid false negative lockdep_assert_held() and 5700 * lockdep_assert_not_held(). 5701 */ 5702 if (unlikely(!lockdep_enabled())) 5703 return LOCK_STATE_UNKNOWN; 5704 5705 raw_local_irq_save(flags); 5706 check_flags(flags); 5707 5708 lockdep_recursion_inc(); 5709 ret = __lock_is_held(lock, read); 5710 lockdep_recursion_finish(); 5711 raw_local_irq_restore(flags); 5712 5713 return ret; 5714 } 5715 EXPORT_SYMBOL_GPL(lock_is_held_type); 5716 NOKPROBE_SYMBOL(lock_is_held_type); 5717 5718 struct pin_cookie lock_pin_lock(struct lockdep_map *lock) 5719 { 5720 struct pin_cookie cookie = NIL_COOKIE; 5721 unsigned long flags; 5722 5723 if (unlikely(!lockdep_enabled())) 5724 return cookie; 5725 5726 raw_local_irq_save(flags); 5727 check_flags(flags); 5728 5729 lockdep_recursion_inc(); 5730 cookie = __lock_pin_lock(lock); 5731 lockdep_recursion_finish(); 5732 raw_local_irq_restore(flags); 5733 5734 return cookie; 5735 } 5736 EXPORT_SYMBOL_GPL(lock_pin_lock); 5737 5738 void lock_repin_lock(struct lockdep_map *lock, struct pin_cookie cookie) 5739 { 5740 unsigned long flags; 5741 5742 if (unlikely(!lockdep_enabled())) 5743 return; 5744 5745 raw_local_irq_save(flags); 5746 check_flags(flags); 5747 5748 lockdep_recursion_inc(); 5749 __lock_repin_lock(lock, cookie); 5750 lockdep_recursion_finish(); 5751 raw_local_irq_restore(flags); 5752 } 5753 EXPORT_SYMBOL_GPL(lock_repin_lock); 5754 5755 void lock_unpin_lock(struct lockdep_map *lock, struct pin_cookie cookie) 5756 { 5757 unsigned long flags; 5758 5759 if (unlikely(!lockdep_enabled())) 5760 return; 5761 5762 raw_local_irq_save(flags); 5763 check_flags(flags); 5764 5765 lockdep_recursion_inc(); 5766 __lock_unpin_lock(lock, cookie); 5767 lockdep_recursion_finish(); 5768 raw_local_irq_restore(flags); 5769 } 5770 EXPORT_SYMBOL_GPL(lock_unpin_lock); 5771 5772 #ifdef CONFIG_LOCK_STAT 5773 static void print_lock_contention_bug(struct task_struct *curr, 5774 struct lockdep_map *lock, 5775 unsigned long ip) 5776 { 5777 if (!debug_locks_off()) 5778 return; 5779 if (debug_locks_silent) 5780 return; 5781 5782 pr_warn("\n"); 5783 pr_warn("=================================\n"); 5784 pr_warn("WARNING: bad contention detected!\n"); 5785 print_kernel_ident(); 5786 pr_warn("---------------------------------\n"); 5787 pr_warn("%s/%d is trying to contend lock (", 5788 curr->comm, task_pid_nr(curr)); 5789 print_lockdep_cache(lock); 5790 pr_cont(") at:\n"); 5791 print_ip_sym(KERN_WARNING, ip); 5792 pr_warn("but there are no locks held!\n"); 5793 pr_warn("\nother info that might help us debug this:\n"); 5794 lockdep_print_held_locks(curr); 5795 5796 pr_warn("\nstack backtrace:\n"); 5797 dump_stack(); 5798 } 5799 5800 static void 5801 __lock_contended(struct lockdep_map *lock, unsigned long ip) 5802 { 5803 struct task_struct *curr = current; 5804 struct held_lock *hlock; 5805 struct lock_class_stats *stats; 5806 unsigned int depth; 5807 int i, contention_point, contending_point; 5808 5809 depth = curr->lockdep_depth; 5810 /* 5811 * Whee, we contended on this lock, except it seems we're not 5812 * actually trying to acquire anything much at all.. 5813 */ 5814 if (DEBUG_LOCKS_WARN_ON(!depth)) 5815 return; 5816 5817 hlock = find_held_lock(curr, lock, depth, &i); 5818 if (!hlock) { 5819 print_lock_contention_bug(curr, lock, ip); 5820 return; 5821 } 5822 5823 if (hlock->instance != lock) 5824 return; 5825 5826 hlock->waittime_stamp = lockstat_clock(); 5827 5828 contention_point = lock_point(hlock_class(hlock)->contention_point, ip); 5829 contending_point = lock_point(hlock_class(hlock)->contending_point, 5830 lock->ip); 5831 5832 stats = get_lock_stats(hlock_class(hlock)); 5833 if (contention_point < LOCKSTAT_POINTS) 5834 stats->contention_point[contention_point]++; 5835 if (contending_point < LOCKSTAT_POINTS) 5836 stats->contending_point[contending_point]++; 5837 if (lock->cpu != smp_processor_id()) 5838 stats->bounces[bounce_contended + !!hlock->read]++; 5839 } 5840 5841 static void 5842 __lock_acquired(struct lockdep_map *lock, unsigned long ip) 5843 { 5844 struct task_struct *curr = current; 5845 struct held_lock *hlock; 5846 struct lock_class_stats *stats; 5847 unsigned int depth; 5848 u64 now, waittime = 0; 5849 int i, cpu; 5850 5851 depth = curr->lockdep_depth; 5852 /* 5853 * Yay, we acquired ownership of this lock we didn't try to 5854 * acquire, how the heck did that happen? 5855 */ 5856 if (DEBUG_LOCKS_WARN_ON(!depth)) 5857 return; 5858 5859 hlock = find_held_lock(curr, lock, depth, &i); 5860 if (!hlock) { 5861 print_lock_contention_bug(curr, lock, _RET_IP_); 5862 return; 5863 } 5864 5865 if (hlock->instance != lock) 5866 return; 5867 5868 cpu = smp_processor_id(); 5869 if (hlock->waittime_stamp) { 5870 now = lockstat_clock(); 5871 waittime = now - hlock->waittime_stamp; 5872 hlock->holdtime_stamp = now; 5873 } 5874 5875 stats = get_lock_stats(hlock_class(hlock)); 5876 if (waittime) { 5877 if (hlock->read) 5878 lock_time_inc(&stats->read_waittime, waittime); 5879 else 5880 lock_time_inc(&stats->write_waittime, waittime); 5881 } 5882 if (lock->cpu != cpu) 5883 stats->bounces[bounce_acquired + !!hlock->read]++; 5884 5885 lock->cpu = cpu; 5886 lock->ip = ip; 5887 } 5888 5889 void lock_contended(struct lockdep_map *lock, unsigned long ip) 5890 { 5891 unsigned long flags; 5892 5893 trace_lock_contended(lock, ip); 5894 5895 if (unlikely(!lock_stat || !lockdep_enabled())) 5896 return; 5897 5898 raw_local_irq_save(flags); 5899 check_flags(flags); 5900 lockdep_recursion_inc(); 5901 __lock_contended(lock, ip); 5902 lockdep_recursion_finish(); 5903 raw_local_irq_restore(flags); 5904 } 5905 EXPORT_SYMBOL_GPL(lock_contended); 5906 5907 void lock_acquired(struct lockdep_map *lock, unsigned long ip) 5908 { 5909 unsigned long flags; 5910 5911 trace_lock_acquired(lock, ip); 5912 5913 if (unlikely(!lock_stat || !lockdep_enabled())) 5914 return; 5915 5916 raw_local_irq_save(flags); 5917 check_flags(flags); 5918 lockdep_recursion_inc(); 5919 __lock_acquired(lock, ip); 5920 lockdep_recursion_finish(); 5921 raw_local_irq_restore(flags); 5922 } 5923 EXPORT_SYMBOL_GPL(lock_acquired); 5924 #endif 5925 5926 /* 5927 * Used by the testsuite, sanitize the validator state 5928 * after a simulated failure: 5929 */ 5930 5931 void lockdep_reset(void) 5932 { 5933 unsigned long flags; 5934 int i; 5935 5936 raw_local_irq_save(flags); 5937 lockdep_init_task(current); 5938 memset(current->held_locks, 0, MAX_LOCK_DEPTH*sizeof(struct held_lock)); 5939 nr_hardirq_chains = 0; 5940 nr_softirq_chains = 0; 5941 nr_process_chains = 0; 5942 debug_locks = 1; 5943 for (i = 0; i < CHAINHASH_SIZE; i++) 5944 INIT_HLIST_HEAD(chainhash_table + i); 5945 raw_local_irq_restore(flags); 5946 } 5947 5948 /* Remove a class from a lock chain. Must be called with the graph lock held. */ 5949 static void remove_class_from_lock_chain(struct pending_free *pf, 5950 struct lock_chain *chain, 5951 struct lock_class *class) 5952 { 5953 #ifdef CONFIG_PROVE_LOCKING 5954 int i; 5955 5956 for (i = chain->base; i < chain->base + chain->depth; i++) { 5957 if (chain_hlock_class_idx(chain_hlocks[i]) != class - lock_classes) 5958 continue; 5959 /* 5960 * Each lock class occurs at most once in a lock chain so once 5961 * we found a match we can break out of this loop. 5962 */ 5963 goto free_lock_chain; 5964 } 5965 /* Since the chain has not been modified, return. */ 5966 return; 5967 5968 free_lock_chain: 5969 free_chain_hlocks(chain->base, chain->depth); 5970 /* Overwrite the chain key for concurrent RCU readers. */ 5971 WRITE_ONCE(chain->chain_key, INITIAL_CHAIN_KEY); 5972 dec_chains(chain->irq_context); 5973 5974 /* 5975 * Note: calling hlist_del_rcu() from inside a 5976 * hlist_for_each_entry_rcu() loop is safe. 5977 */ 5978 hlist_del_rcu(&chain->entry); 5979 __set_bit(chain - lock_chains, pf->lock_chains_being_freed); 5980 nr_zapped_lock_chains++; 5981 #endif 5982 } 5983 5984 /* Must be called with the graph lock held. */ 5985 static void remove_class_from_lock_chains(struct pending_free *pf, 5986 struct lock_class *class) 5987 { 5988 struct lock_chain *chain; 5989 struct hlist_head *head; 5990 int i; 5991 5992 for (i = 0; i < ARRAY_SIZE(chainhash_table); i++) { 5993 head = chainhash_table + i; 5994 hlist_for_each_entry_rcu(chain, head, entry) { 5995 remove_class_from_lock_chain(pf, chain, class); 5996 } 5997 } 5998 } 5999 6000 /* 6001 * Remove all references to a lock class. The caller must hold the graph lock. 6002 */ 6003 static void zap_class(struct pending_free *pf, struct lock_class *class) 6004 { 6005 struct lock_list *entry; 6006 int i; 6007 6008 WARN_ON_ONCE(!class->key); 6009 6010 /* 6011 * Remove all dependencies this lock is 6012 * involved in: 6013 */ 6014 for_each_set_bit(i, list_entries_in_use, ARRAY_SIZE(list_entries)) { 6015 entry = list_entries + i; 6016 if (entry->class != class && entry->links_to != class) 6017 continue; 6018 __clear_bit(i, list_entries_in_use); 6019 nr_list_entries--; 6020 list_del_rcu(&entry->entry); 6021 } 6022 if (list_empty(&class->locks_after) && 6023 list_empty(&class->locks_before)) { 6024 list_move_tail(&class->lock_entry, &pf->zapped); 6025 hlist_del_rcu(&class->hash_entry); 6026 WRITE_ONCE(class->key, NULL); 6027 WRITE_ONCE(class->name, NULL); 6028 nr_lock_classes--; 6029 __clear_bit(class - lock_classes, lock_classes_in_use); 6030 if (class - lock_classes == max_lock_class_idx) 6031 max_lock_class_idx--; 6032 } else { 6033 WARN_ONCE(true, "%s() failed for class %s\n", __func__, 6034 class->name); 6035 } 6036 6037 remove_class_from_lock_chains(pf, class); 6038 nr_zapped_classes++; 6039 } 6040 6041 static void reinit_class(struct lock_class *class) 6042 { 6043 WARN_ON_ONCE(!class->lock_entry.next); 6044 WARN_ON_ONCE(!list_empty(&class->locks_after)); 6045 WARN_ON_ONCE(!list_empty(&class->locks_before)); 6046 memset_startat(class, 0, key); 6047 WARN_ON_ONCE(!class->lock_entry.next); 6048 WARN_ON_ONCE(!list_empty(&class->locks_after)); 6049 WARN_ON_ONCE(!list_empty(&class->locks_before)); 6050 } 6051 6052 static inline int within(const void *addr, void *start, unsigned long size) 6053 { 6054 return addr >= start && addr < start + size; 6055 } 6056 6057 static bool inside_selftest(void) 6058 { 6059 return current == lockdep_selftest_task_struct; 6060 } 6061 6062 /* The caller must hold the graph lock. */ 6063 static struct pending_free *get_pending_free(void) 6064 { 6065 return delayed_free.pf + delayed_free.index; 6066 } 6067 6068 static void free_zapped_rcu(struct rcu_head *cb); 6069 6070 /* 6071 * Schedule an RCU callback if no RCU callback is pending. Must be called with 6072 * the graph lock held. 6073 */ 6074 static void call_rcu_zapped(struct pending_free *pf) 6075 { 6076 WARN_ON_ONCE(inside_selftest()); 6077 6078 if (list_empty(&pf->zapped)) 6079 return; 6080 6081 if (delayed_free.scheduled) 6082 return; 6083 6084 delayed_free.scheduled = true; 6085 6086 WARN_ON_ONCE(delayed_free.pf + delayed_free.index != pf); 6087 delayed_free.index ^= 1; 6088 6089 call_rcu(&delayed_free.rcu_head, free_zapped_rcu); 6090 } 6091 6092 /* The caller must hold the graph lock. May be called from RCU context. */ 6093 static void __free_zapped_classes(struct pending_free *pf) 6094 { 6095 struct lock_class *class; 6096 6097 check_data_structures(); 6098 6099 list_for_each_entry(class, &pf->zapped, lock_entry) 6100 reinit_class(class); 6101 6102 list_splice_init(&pf->zapped, &free_lock_classes); 6103 6104 #ifdef CONFIG_PROVE_LOCKING 6105 bitmap_andnot(lock_chains_in_use, lock_chains_in_use, 6106 pf->lock_chains_being_freed, ARRAY_SIZE(lock_chains)); 6107 bitmap_clear(pf->lock_chains_being_freed, 0, ARRAY_SIZE(lock_chains)); 6108 #endif 6109 } 6110 6111 static void free_zapped_rcu(struct rcu_head *ch) 6112 { 6113 struct pending_free *pf; 6114 unsigned long flags; 6115 6116 if (WARN_ON_ONCE(ch != &delayed_free.rcu_head)) 6117 return; 6118 6119 raw_local_irq_save(flags); 6120 lockdep_lock(); 6121 6122 /* closed head */ 6123 pf = delayed_free.pf + (delayed_free.index ^ 1); 6124 __free_zapped_classes(pf); 6125 delayed_free.scheduled = false; 6126 6127 /* 6128 * If there's anything on the open list, close and start a new callback. 6129 */ 6130 call_rcu_zapped(delayed_free.pf + delayed_free.index); 6131 6132 lockdep_unlock(); 6133 raw_local_irq_restore(flags); 6134 } 6135 6136 /* 6137 * Remove all lock classes from the class hash table and from the 6138 * all_lock_classes list whose key or name is in the address range [start, 6139 * start + size). Move these lock classes to the zapped_classes list. Must 6140 * be called with the graph lock held. 6141 */ 6142 static void __lockdep_free_key_range(struct pending_free *pf, void *start, 6143 unsigned long size) 6144 { 6145 struct lock_class *class; 6146 struct hlist_head *head; 6147 int i; 6148 6149 /* Unhash all classes that were created by a module. */ 6150 for (i = 0; i < CLASSHASH_SIZE; i++) { 6151 head = classhash_table + i; 6152 hlist_for_each_entry_rcu(class, head, hash_entry) { 6153 if (!within(class->key, start, size) && 6154 !within(class->name, start, size)) 6155 continue; 6156 zap_class(pf, class); 6157 } 6158 } 6159 } 6160 6161 /* 6162 * Used in module.c to remove lock classes from memory that is going to be 6163 * freed; and possibly re-used by other modules. 6164 * 6165 * We will have had one synchronize_rcu() before getting here, so we're 6166 * guaranteed nobody will look up these exact classes -- they're properly dead 6167 * but still allocated. 6168 */ 6169 static void lockdep_free_key_range_reg(void *start, unsigned long size) 6170 { 6171 struct pending_free *pf; 6172 unsigned long flags; 6173 6174 init_data_structures_once(); 6175 6176 raw_local_irq_save(flags); 6177 lockdep_lock(); 6178 pf = get_pending_free(); 6179 __lockdep_free_key_range(pf, start, size); 6180 call_rcu_zapped(pf); 6181 lockdep_unlock(); 6182 raw_local_irq_restore(flags); 6183 6184 /* 6185 * Wait for any possible iterators from look_up_lock_class() to pass 6186 * before continuing to free the memory they refer to. 6187 */ 6188 synchronize_rcu(); 6189 } 6190 6191 /* 6192 * Free all lockdep keys in the range [start, start+size). Does not sleep. 6193 * Ignores debug_locks. Must only be used by the lockdep selftests. 6194 */ 6195 static void lockdep_free_key_range_imm(void *start, unsigned long size) 6196 { 6197 struct pending_free *pf = delayed_free.pf; 6198 unsigned long flags; 6199 6200 init_data_structures_once(); 6201 6202 raw_local_irq_save(flags); 6203 lockdep_lock(); 6204 __lockdep_free_key_range(pf, start, size); 6205 __free_zapped_classes(pf); 6206 lockdep_unlock(); 6207 raw_local_irq_restore(flags); 6208 } 6209 6210 void lockdep_free_key_range(void *start, unsigned long size) 6211 { 6212 init_data_structures_once(); 6213 6214 if (inside_selftest()) 6215 lockdep_free_key_range_imm(start, size); 6216 else 6217 lockdep_free_key_range_reg(start, size); 6218 } 6219 6220 /* 6221 * Check whether any element of the @lock->class_cache[] array refers to a 6222 * registered lock class. The caller must hold either the graph lock or the 6223 * RCU read lock. 6224 */ 6225 static bool lock_class_cache_is_registered(struct lockdep_map *lock) 6226 { 6227 struct lock_class *class; 6228 struct hlist_head *head; 6229 int i, j; 6230 6231 for (i = 0; i < CLASSHASH_SIZE; i++) { 6232 head = classhash_table + i; 6233 hlist_for_each_entry_rcu(class, head, hash_entry) { 6234 for (j = 0; j < NR_LOCKDEP_CACHING_CLASSES; j++) 6235 if (lock->class_cache[j] == class) 6236 return true; 6237 } 6238 } 6239 return false; 6240 } 6241 6242 /* The caller must hold the graph lock. Does not sleep. */ 6243 static void __lockdep_reset_lock(struct pending_free *pf, 6244 struct lockdep_map *lock) 6245 { 6246 struct lock_class *class; 6247 int j; 6248 6249 /* 6250 * Remove all classes this lock might have: 6251 */ 6252 for (j = 0; j < MAX_LOCKDEP_SUBCLASSES; j++) { 6253 /* 6254 * If the class exists we look it up and zap it: 6255 */ 6256 class = look_up_lock_class(lock, j); 6257 if (class) 6258 zap_class(pf, class); 6259 } 6260 /* 6261 * Debug check: in the end all mapped classes should 6262 * be gone. 6263 */ 6264 if (WARN_ON_ONCE(lock_class_cache_is_registered(lock))) 6265 debug_locks_off(); 6266 } 6267 6268 /* 6269 * Remove all information lockdep has about a lock if debug_locks == 1. Free 6270 * released data structures from RCU context. 6271 */ 6272 static void lockdep_reset_lock_reg(struct lockdep_map *lock) 6273 { 6274 struct pending_free *pf; 6275 unsigned long flags; 6276 int locked; 6277 6278 raw_local_irq_save(flags); 6279 locked = graph_lock(); 6280 if (!locked) 6281 goto out_irq; 6282 6283 pf = get_pending_free(); 6284 __lockdep_reset_lock(pf, lock); 6285 call_rcu_zapped(pf); 6286 6287 graph_unlock(); 6288 out_irq: 6289 raw_local_irq_restore(flags); 6290 } 6291 6292 /* 6293 * Reset a lock. Does not sleep. Ignores debug_locks. Must only be used by the 6294 * lockdep selftests. 6295 */ 6296 static void lockdep_reset_lock_imm(struct lockdep_map *lock) 6297 { 6298 struct pending_free *pf = delayed_free.pf; 6299 unsigned long flags; 6300 6301 raw_local_irq_save(flags); 6302 lockdep_lock(); 6303 __lockdep_reset_lock(pf, lock); 6304 __free_zapped_classes(pf); 6305 lockdep_unlock(); 6306 raw_local_irq_restore(flags); 6307 } 6308 6309 void lockdep_reset_lock(struct lockdep_map *lock) 6310 { 6311 init_data_structures_once(); 6312 6313 if (inside_selftest()) 6314 lockdep_reset_lock_imm(lock); 6315 else 6316 lockdep_reset_lock_reg(lock); 6317 } 6318 6319 /* 6320 * Unregister a dynamically allocated key. 6321 * 6322 * Unlike lockdep_register_key(), a search is always done to find a matching 6323 * key irrespective of debug_locks to avoid potential invalid access to freed 6324 * memory in lock_class entry. 6325 */ 6326 void lockdep_unregister_key(struct lock_class_key *key) 6327 { 6328 struct hlist_head *hash_head = keyhashentry(key); 6329 struct lock_class_key *k; 6330 struct pending_free *pf; 6331 unsigned long flags; 6332 bool found = false; 6333 6334 might_sleep(); 6335 6336 if (WARN_ON_ONCE(static_obj(key))) 6337 return; 6338 6339 raw_local_irq_save(flags); 6340 lockdep_lock(); 6341 6342 hlist_for_each_entry_rcu(k, hash_head, hash_entry) { 6343 if (k == key) { 6344 hlist_del_rcu(&k->hash_entry); 6345 found = true; 6346 break; 6347 } 6348 } 6349 WARN_ON_ONCE(!found && debug_locks); 6350 if (found) { 6351 pf = get_pending_free(); 6352 __lockdep_free_key_range(pf, key, 1); 6353 call_rcu_zapped(pf); 6354 } 6355 lockdep_unlock(); 6356 raw_local_irq_restore(flags); 6357 6358 /* Wait until is_dynamic_key() has finished accessing k->hash_entry. */ 6359 synchronize_rcu(); 6360 } 6361 EXPORT_SYMBOL_GPL(lockdep_unregister_key); 6362 6363 void __init lockdep_init(void) 6364 { 6365 printk("Lock dependency validator: Copyright (c) 2006 Red Hat, Inc., Ingo Molnar\n"); 6366 6367 printk("... MAX_LOCKDEP_SUBCLASSES: %lu\n", MAX_LOCKDEP_SUBCLASSES); 6368 printk("... MAX_LOCK_DEPTH: %lu\n", MAX_LOCK_DEPTH); 6369 printk("... MAX_LOCKDEP_KEYS: %lu\n", MAX_LOCKDEP_KEYS); 6370 printk("... CLASSHASH_SIZE: %lu\n", CLASSHASH_SIZE); 6371 printk("... MAX_LOCKDEP_ENTRIES: %lu\n", MAX_LOCKDEP_ENTRIES); 6372 printk("... MAX_LOCKDEP_CHAINS: %lu\n", MAX_LOCKDEP_CHAINS); 6373 printk("... CHAINHASH_SIZE: %lu\n", CHAINHASH_SIZE); 6374 6375 printk(" memory used by lock dependency info: %zu kB\n", 6376 (sizeof(lock_classes) + 6377 sizeof(lock_classes_in_use) + 6378 sizeof(classhash_table) + 6379 sizeof(list_entries) + 6380 sizeof(list_entries_in_use) + 6381 sizeof(chainhash_table) + 6382 sizeof(delayed_free) 6383 #ifdef CONFIG_PROVE_LOCKING 6384 + sizeof(lock_cq) 6385 + sizeof(lock_chains) 6386 + sizeof(lock_chains_in_use) 6387 + sizeof(chain_hlocks) 6388 #endif 6389 ) / 1024 6390 ); 6391 6392 #if defined(CONFIG_TRACE_IRQFLAGS) && defined(CONFIG_PROVE_LOCKING) 6393 printk(" memory used for stack traces: %zu kB\n", 6394 (sizeof(stack_trace) + sizeof(stack_trace_hash)) / 1024 6395 ); 6396 #endif 6397 6398 printk(" per task-struct memory footprint: %zu bytes\n", 6399 sizeof(((struct task_struct *)NULL)->held_locks)); 6400 } 6401 6402 static void 6403 print_freed_lock_bug(struct task_struct *curr, const void *mem_from, 6404 const void *mem_to, struct held_lock *hlock) 6405 { 6406 if (!debug_locks_off()) 6407 return; 6408 if (debug_locks_silent) 6409 return; 6410 6411 pr_warn("\n"); 6412 pr_warn("=========================\n"); 6413 pr_warn("WARNING: held lock freed!\n"); 6414 print_kernel_ident(); 6415 pr_warn("-------------------------\n"); 6416 pr_warn("%s/%d is freeing memory %px-%px, with a lock still held there!\n", 6417 curr->comm, task_pid_nr(curr), mem_from, mem_to-1); 6418 print_lock(hlock); 6419 lockdep_print_held_locks(curr); 6420 6421 pr_warn("\nstack backtrace:\n"); 6422 dump_stack(); 6423 } 6424 6425 static inline int not_in_range(const void* mem_from, unsigned long mem_len, 6426 const void* lock_from, unsigned long lock_len) 6427 { 6428 return lock_from + lock_len <= mem_from || 6429 mem_from + mem_len <= lock_from; 6430 } 6431 6432 /* 6433 * Called when kernel memory is freed (or unmapped), or if a lock 6434 * is destroyed or reinitialized - this code checks whether there is 6435 * any held lock in the memory range of <from> to <to>: 6436 */ 6437 void debug_check_no_locks_freed(const void *mem_from, unsigned long mem_len) 6438 { 6439 struct task_struct *curr = current; 6440 struct held_lock *hlock; 6441 unsigned long flags; 6442 int i; 6443 6444 if (unlikely(!debug_locks)) 6445 return; 6446 6447 raw_local_irq_save(flags); 6448 for (i = 0; i < curr->lockdep_depth; i++) { 6449 hlock = curr->held_locks + i; 6450 6451 if (not_in_range(mem_from, mem_len, hlock->instance, 6452 sizeof(*hlock->instance))) 6453 continue; 6454 6455 print_freed_lock_bug(curr, mem_from, mem_from + mem_len, hlock); 6456 break; 6457 } 6458 raw_local_irq_restore(flags); 6459 } 6460 EXPORT_SYMBOL_GPL(debug_check_no_locks_freed); 6461 6462 static void print_held_locks_bug(void) 6463 { 6464 if (!debug_locks_off()) 6465 return; 6466 if (debug_locks_silent) 6467 return; 6468 6469 pr_warn("\n"); 6470 pr_warn("====================================\n"); 6471 pr_warn("WARNING: %s/%d still has locks held!\n", 6472 current->comm, task_pid_nr(current)); 6473 print_kernel_ident(); 6474 pr_warn("------------------------------------\n"); 6475 lockdep_print_held_locks(current); 6476 pr_warn("\nstack backtrace:\n"); 6477 dump_stack(); 6478 } 6479 6480 void debug_check_no_locks_held(void) 6481 { 6482 if (unlikely(current->lockdep_depth > 0)) 6483 print_held_locks_bug(); 6484 } 6485 EXPORT_SYMBOL_GPL(debug_check_no_locks_held); 6486 6487 #ifdef __KERNEL__ 6488 void debug_show_all_locks(void) 6489 { 6490 struct task_struct *g, *p; 6491 6492 if (unlikely(!debug_locks)) { 6493 pr_warn("INFO: lockdep is turned off.\n"); 6494 return; 6495 } 6496 pr_warn("\nShowing all locks held in the system:\n"); 6497 6498 rcu_read_lock(); 6499 for_each_process_thread(g, p) { 6500 if (!p->lockdep_depth) 6501 continue; 6502 lockdep_print_held_locks(p); 6503 touch_nmi_watchdog(); 6504 touch_all_softlockup_watchdogs(); 6505 } 6506 rcu_read_unlock(); 6507 6508 pr_warn("\n"); 6509 pr_warn("=============================================\n\n"); 6510 } 6511 EXPORT_SYMBOL_GPL(debug_show_all_locks); 6512 #endif 6513 6514 /* 6515 * Careful: only use this function if you are sure that 6516 * the task cannot run in parallel! 6517 */ 6518 void debug_show_held_locks(struct task_struct *task) 6519 { 6520 if (unlikely(!debug_locks)) { 6521 printk("INFO: lockdep is turned off.\n"); 6522 return; 6523 } 6524 lockdep_print_held_locks(task); 6525 } 6526 EXPORT_SYMBOL_GPL(debug_show_held_locks); 6527 6528 asmlinkage __visible void lockdep_sys_exit(void) 6529 { 6530 struct task_struct *curr = current; 6531 6532 if (unlikely(curr->lockdep_depth)) { 6533 if (!debug_locks_off()) 6534 return; 6535 pr_warn("\n"); 6536 pr_warn("================================================\n"); 6537 pr_warn("WARNING: lock held when returning to user space!\n"); 6538 print_kernel_ident(); 6539 pr_warn("------------------------------------------------\n"); 6540 pr_warn("%s/%d is leaving the kernel with locks still held!\n", 6541 curr->comm, curr->pid); 6542 lockdep_print_held_locks(curr); 6543 } 6544 6545 /* 6546 * The lock history for each syscall should be independent. So wipe the 6547 * slate clean on return to userspace. 6548 */ 6549 lockdep_invariant_state(false); 6550 } 6551 6552 void lockdep_rcu_suspicious(const char *file, const int line, const char *s) 6553 { 6554 struct task_struct *curr = current; 6555 int dl = READ_ONCE(debug_locks); 6556 6557 /* Note: the following can be executed concurrently, so be careful. */ 6558 pr_warn("\n"); 6559 pr_warn("=============================\n"); 6560 pr_warn("WARNING: suspicious RCU usage\n"); 6561 print_kernel_ident(); 6562 pr_warn("-----------------------------\n"); 6563 pr_warn("%s:%d %s!\n", file, line, s); 6564 pr_warn("\nother info that might help us debug this:\n\n"); 6565 pr_warn("\n%srcu_scheduler_active = %d, debug_locks = %d\n%s", 6566 !rcu_lockdep_current_cpu_online() 6567 ? "RCU used illegally from offline CPU!\n" 6568 : "", 6569 rcu_scheduler_active, dl, 6570 dl ? "" : "Possible false positive due to lockdep disabling via debug_locks = 0\n"); 6571 6572 /* 6573 * If a CPU is in the RCU-free window in idle (ie: in the section 6574 * between ct_idle_enter() and ct_idle_exit(), then RCU 6575 * considers that CPU to be in an "extended quiescent state", 6576 * which means that RCU will be completely ignoring that CPU. 6577 * Therefore, rcu_read_lock() and friends have absolutely no 6578 * effect on a CPU running in that state. In other words, even if 6579 * such an RCU-idle CPU has called rcu_read_lock(), RCU might well 6580 * delete data structures out from under it. RCU really has no 6581 * choice here: we need to keep an RCU-free window in idle where 6582 * the CPU may possibly enter into low power mode. This way we can 6583 * notice an extended quiescent state to other CPUs that started a grace 6584 * period. Otherwise we would delay any grace period as long as we run 6585 * in the idle task. 6586 * 6587 * So complain bitterly if someone does call rcu_read_lock(), 6588 * rcu_read_lock_bh() and so on from extended quiescent states. 6589 */ 6590 if (!rcu_is_watching()) 6591 pr_warn("RCU used illegally from extended quiescent state!\n"); 6592 6593 lockdep_print_held_locks(curr); 6594 pr_warn("\nstack backtrace:\n"); 6595 dump_stack(); 6596 } 6597 EXPORT_SYMBOL_GPL(lockdep_rcu_suspicious); 6598