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 58 #include <asm/sections.h> 59 60 #include "lockdep_internals.h" 61 62 #define CREATE_TRACE_POINTS 63 #include <trace/events/lock.h> 64 65 #ifdef CONFIG_PROVE_LOCKING 66 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 int lock_stat = 1; 74 module_param(lock_stat, int, 0644); 75 #else 76 #define lock_stat 0 77 #endif 78 79 /* 80 * lockdep_lock: protects the lockdep graph, the hashes and the 81 * class/list/hash allocators. 82 * 83 * This is one of the rare exceptions where it's justified 84 * to use a raw spinlock - we really dont want the spinlock 85 * code to recurse back into the lockdep code... 86 */ 87 static arch_spinlock_t lockdep_lock = (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED; 88 static struct task_struct *lockdep_selftest_task_struct; 89 90 static int graph_lock(void) 91 { 92 arch_spin_lock(&lockdep_lock); 93 /* 94 * Make sure that if another CPU detected a bug while 95 * walking the graph we dont change it (while the other 96 * CPU is busy printing out stuff with the graph lock 97 * dropped already) 98 */ 99 if (!debug_locks) { 100 arch_spin_unlock(&lockdep_lock); 101 return 0; 102 } 103 /* prevent any recursions within lockdep from causing deadlocks */ 104 current->lockdep_recursion++; 105 return 1; 106 } 107 108 static inline int graph_unlock(void) 109 { 110 if (debug_locks && !arch_spin_is_locked(&lockdep_lock)) { 111 /* 112 * The lockdep graph lock isn't locked while we expect it to 113 * be, we're confused now, bye! 114 */ 115 return DEBUG_LOCKS_WARN_ON(1); 116 } 117 118 current->lockdep_recursion--; 119 arch_spin_unlock(&lockdep_lock); 120 return 0; 121 } 122 123 /* 124 * Turn lock debugging off and return with 0 if it was off already, 125 * and also release the graph lock: 126 */ 127 static inline int debug_locks_off_graph_unlock(void) 128 { 129 int ret = debug_locks_off(); 130 131 arch_spin_unlock(&lockdep_lock); 132 133 return ret; 134 } 135 136 unsigned long nr_list_entries; 137 static struct lock_list list_entries[MAX_LOCKDEP_ENTRIES]; 138 static DECLARE_BITMAP(list_entries_in_use, MAX_LOCKDEP_ENTRIES); 139 140 /* 141 * All data structures here are protected by the global debug_lock. 142 * 143 * nr_lock_classes is the number of elements of lock_classes[] that is 144 * in use. 145 */ 146 #define KEYHASH_BITS (MAX_LOCKDEP_KEYS_BITS - 1) 147 #define KEYHASH_SIZE (1UL << KEYHASH_BITS) 148 static struct hlist_head lock_keys_hash[KEYHASH_SIZE]; 149 unsigned long nr_lock_classes; 150 #ifndef CONFIG_DEBUG_LOCKDEP 151 static 152 #endif 153 struct lock_class lock_classes[MAX_LOCKDEP_KEYS]; 154 static DECLARE_BITMAP(lock_classes_in_use, MAX_LOCKDEP_KEYS); 155 156 static inline struct lock_class *hlock_class(struct held_lock *hlock) 157 { 158 unsigned int class_idx = hlock->class_idx; 159 160 /* Don't re-read hlock->class_idx, can't use READ_ONCE() on bitfield */ 161 barrier(); 162 163 if (!test_bit(class_idx, lock_classes_in_use)) { 164 /* 165 * Someone passed in garbage, we give up. 166 */ 167 DEBUG_LOCKS_WARN_ON(1); 168 return NULL; 169 } 170 171 /* 172 * At this point, if the passed hlock->class_idx is still garbage, 173 * we just have to live with it 174 */ 175 return lock_classes + class_idx; 176 } 177 178 #ifdef CONFIG_LOCK_STAT 179 static DEFINE_PER_CPU(struct lock_class_stats[MAX_LOCKDEP_KEYS], cpu_lock_stats); 180 181 static inline u64 lockstat_clock(void) 182 { 183 return local_clock(); 184 } 185 186 static int lock_point(unsigned long points[], unsigned long ip) 187 { 188 int i; 189 190 for (i = 0; i < LOCKSTAT_POINTS; i++) { 191 if (points[i] == 0) { 192 points[i] = ip; 193 break; 194 } 195 if (points[i] == ip) 196 break; 197 } 198 199 return i; 200 } 201 202 static void lock_time_inc(struct lock_time *lt, u64 time) 203 { 204 if (time > lt->max) 205 lt->max = time; 206 207 if (time < lt->min || !lt->nr) 208 lt->min = time; 209 210 lt->total += time; 211 lt->nr++; 212 } 213 214 static inline void lock_time_add(struct lock_time *src, struct lock_time *dst) 215 { 216 if (!src->nr) 217 return; 218 219 if (src->max > dst->max) 220 dst->max = src->max; 221 222 if (src->min < dst->min || !dst->nr) 223 dst->min = src->min; 224 225 dst->total += src->total; 226 dst->nr += src->nr; 227 } 228 229 struct lock_class_stats lock_stats(struct lock_class *class) 230 { 231 struct lock_class_stats stats; 232 int cpu, i; 233 234 memset(&stats, 0, sizeof(struct lock_class_stats)); 235 for_each_possible_cpu(cpu) { 236 struct lock_class_stats *pcs = 237 &per_cpu(cpu_lock_stats, cpu)[class - lock_classes]; 238 239 for (i = 0; i < ARRAY_SIZE(stats.contention_point); i++) 240 stats.contention_point[i] += pcs->contention_point[i]; 241 242 for (i = 0; i < ARRAY_SIZE(stats.contending_point); i++) 243 stats.contending_point[i] += pcs->contending_point[i]; 244 245 lock_time_add(&pcs->read_waittime, &stats.read_waittime); 246 lock_time_add(&pcs->write_waittime, &stats.write_waittime); 247 248 lock_time_add(&pcs->read_holdtime, &stats.read_holdtime); 249 lock_time_add(&pcs->write_holdtime, &stats.write_holdtime); 250 251 for (i = 0; i < ARRAY_SIZE(stats.bounces); i++) 252 stats.bounces[i] += pcs->bounces[i]; 253 } 254 255 return stats; 256 } 257 258 void clear_lock_stats(struct lock_class *class) 259 { 260 int cpu; 261 262 for_each_possible_cpu(cpu) { 263 struct lock_class_stats *cpu_stats = 264 &per_cpu(cpu_lock_stats, cpu)[class - lock_classes]; 265 266 memset(cpu_stats, 0, sizeof(struct lock_class_stats)); 267 } 268 memset(class->contention_point, 0, sizeof(class->contention_point)); 269 memset(class->contending_point, 0, sizeof(class->contending_point)); 270 } 271 272 static struct lock_class_stats *get_lock_stats(struct lock_class *class) 273 { 274 return &this_cpu_ptr(cpu_lock_stats)[class - lock_classes]; 275 } 276 277 static void lock_release_holdtime(struct held_lock *hlock) 278 { 279 struct lock_class_stats *stats; 280 u64 holdtime; 281 282 if (!lock_stat) 283 return; 284 285 holdtime = lockstat_clock() - hlock->holdtime_stamp; 286 287 stats = get_lock_stats(hlock_class(hlock)); 288 if (hlock->read) 289 lock_time_inc(&stats->read_holdtime, holdtime); 290 else 291 lock_time_inc(&stats->write_holdtime, holdtime); 292 } 293 #else 294 static inline void lock_release_holdtime(struct held_lock *hlock) 295 { 296 } 297 #endif 298 299 /* 300 * We keep a global list of all lock classes. The list is only accessed with 301 * the lockdep spinlock lock held. free_lock_classes is a list with free 302 * elements. These elements are linked together by the lock_entry member in 303 * struct lock_class. 304 */ 305 LIST_HEAD(all_lock_classes); 306 static LIST_HEAD(free_lock_classes); 307 308 /** 309 * struct pending_free - information about data structures about to be freed 310 * @zapped: Head of a list with struct lock_class elements. 311 * @lock_chains_being_freed: Bitmap that indicates which lock_chains[] elements 312 * are about to be freed. 313 */ 314 struct pending_free { 315 struct list_head zapped; 316 DECLARE_BITMAP(lock_chains_being_freed, MAX_LOCKDEP_CHAINS); 317 }; 318 319 /** 320 * struct delayed_free - data structures used for delayed freeing 321 * 322 * A data structure for delayed freeing of data structures that may be 323 * accessed by RCU readers at the time these were freed. 324 * 325 * @rcu_head: Used to schedule an RCU callback for freeing data structures. 326 * @index: Index of @pf to which freed data structures are added. 327 * @scheduled: Whether or not an RCU callback has been scheduled. 328 * @pf: Array with information about data structures about to be freed. 329 */ 330 static struct delayed_free { 331 struct rcu_head rcu_head; 332 int index; 333 int scheduled; 334 struct pending_free pf[2]; 335 } delayed_free; 336 337 /* 338 * The lockdep classes are in a hash-table as well, for fast lookup: 339 */ 340 #define CLASSHASH_BITS (MAX_LOCKDEP_KEYS_BITS - 1) 341 #define CLASSHASH_SIZE (1UL << CLASSHASH_BITS) 342 #define __classhashfn(key) hash_long((unsigned long)key, CLASSHASH_BITS) 343 #define classhashentry(key) (classhash_table + __classhashfn((key))) 344 345 static struct hlist_head classhash_table[CLASSHASH_SIZE]; 346 347 /* 348 * We put the lock dependency chains into a hash-table as well, to cache 349 * their existence: 350 */ 351 #define CHAINHASH_BITS (MAX_LOCKDEP_CHAINS_BITS-1) 352 #define CHAINHASH_SIZE (1UL << CHAINHASH_BITS) 353 #define __chainhashfn(chain) hash_long(chain, CHAINHASH_BITS) 354 #define chainhashentry(chain) (chainhash_table + __chainhashfn((chain))) 355 356 static struct hlist_head chainhash_table[CHAINHASH_SIZE]; 357 358 /* 359 * The hash key of the lock dependency chains is a hash itself too: 360 * it's a hash of all locks taken up to that lock, including that lock. 361 * It's a 64-bit hash, because it's important for the keys to be 362 * unique. 363 */ 364 static inline u64 iterate_chain_key(u64 key, u32 idx) 365 { 366 u32 k0 = key, k1 = key >> 32; 367 368 __jhash_mix(idx, k0, k1); /* Macro that modifies arguments! */ 369 370 return k0 | (u64)k1 << 32; 371 } 372 373 void lockdep_init_task(struct task_struct *task) 374 { 375 task->lockdep_depth = 0; /* no locks held yet */ 376 task->curr_chain_key = INITIAL_CHAIN_KEY; 377 task->lockdep_recursion = 0; 378 } 379 380 void lockdep_off(void) 381 { 382 current->lockdep_recursion++; 383 } 384 EXPORT_SYMBOL(lockdep_off); 385 386 void lockdep_on(void) 387 { 388 current->lockdep_recursion--; 389 } 390 EXPORT_SYMBOL(lockdep_on); 391 392 void lockdep_set_selftest_task(struct task_struct *task) 393 { 394 lockdep_selftest_task_struct = task; 395 } 396 397 /* 398 * Debugging switches: 399 */ 400 401 #define VERBOSE 0 402 #define VERY_VERBOSE 0 403 404 #if VERBOSE 405 # define HARDIRQ_VERBOSE 1 406 # define SOFTIRQ_VERBOSE 1 407 #else 408 # define HARDIRQ_VERBOSE 0 409 # define SOFTIRQ_VERBOSE 0 410 #endif 411 412 #if VERBOSE || HARDIRQ_VERBOSE || SOFTIRQ_VERBOSE 413 /* 414 * Quick filtering for interesting events: 415 */ 416 static int class_filter(struct lock_class *class) 417 { 418 #if 0 419 /* Example */ 420 if (class->name_version == 1 && 421 !strcmp(class->name, "lockname")) 422 return 1; 423 if (class->name_version == 1 && 424 !strcmp(class->name, "&struct->lockfield")) 425 return 1; 426 #endif 427 /* Filter everything else. 1 would be to allow everything else */ 428 return 0; 429 } 430 #endif 431 432 static int verbose(struct lock_class *class) 433 { 434 #if VERBOSE 435 return class_filter(class); 436 #endif 437 return 0; 438 } 439 440 static void print_lockdep_off(const char *bug_msg) 441 { 442 printk(KERN_DEBUG "%s\n", bug_msg); 443 printk(KERN_DEBUG "turning off the locking correctness validator.\n"); 444 #ifdef CONFIG_LOCK_STAT 445 printk(KERN_DEBUG "Please attach the output of /proc/lock_stat to the bug report\n"); 446 #endif 447 } 448 449 unsigned long nr_stack_trace_entries; 450 451 #ifdef CONFIG_PROVE_LOCKING 452 /** 453 * struct lock_trace - single stack backtrace 454 * @hash_entry: Entry in a stack_trace_hash[] list. 455 * @hash: jhash() of @entries. 456 * @nr_entries: Number of entries in @entries. 457 * @entries: Actual stack backtrace. 458 */ 459 struct lock_trace { 460 struct hlist_node hash_entry; 461 u32 hash; 462 u32 nr_entries; 463 unsigned long entries[0] __aligned(sizeof(unsigned long)); 464 }; 465 #define LOCK_TRACE_SIZE_IN_LONGS \ 466 (sizeof(struct lock_trace) / sizeof(unsigned long)) 467 /* 468 * Stack-trace: sequence of lock_trace structures. Protected by the graph_lock. 469 */ 470 static unsigned long stack_trace[MAX_STACK_TRACE_ENTRIES]; 471 static struct hlist_head stack_trace_hash[STACK_TRACE_HASH_SIZE]; 472 473 static bool traces_identical(struct lock_trace *t1, struct lock_trace *t2) 474 { 475 return t1->hash == t2->hash && t1->nr_entries == t2->nr_entries && 476 memcmp(t1->entries, t2->entries, 477 t1->nr_entries * sizeof(t1->entries[0])) == 0; 478 } 479 480 static struct lock_trace *save_trace(void) 481 { 482 struct lock_trace *trace, *t2; 483 struct hlist_head *hash_head; 484 u32 hash; 485 unsigned int max_entries; 486 487 BUILD_BUG_ON_NOT_POWER_OF_2(STACK_TRACE_HASH_SIZE); 488 BUILD_BUG_ON(LOCK_TRACE_SIZE_IN_LONGS >= MAX_STACK_TRACE_ENTRIES); 489 490 trace = (struct lock_trace *)(stack_trace + nr_stack_trace_entries); 491 max_entries = MAX_STACK_TRACE_ENTRIES - nr_stack_trace_entries - 492 LOCK_TRACE_SIZE_IN_LONGS; 493 trace->nr_entries = stack_trace_save(trace->entries, max_entries, 3); 494 495 if (nr_stack_trace_entries >= MAX_STACK_TRACE_ENTRIES - 496 LOCK_TRACE_SIZE_IN_LONGS - 1) { 497 if (!debug_locks_off_graph_unlock()) 498 return NULL; 499 500 print_lockdep_off("BUG: MAX_STACK_TRACE_ENTRIES too low!"); 501 dump_stack(); 502 503 return NULL; 504 } 505 506 hash = jhash(trace->entries, trace->nr_entries * 507 sizeof(trace->entries[0]), 0); 508 trace->hash = hash; 509 hash_head = stack_trace_hash + (hash & (STACK_TRACE_HASH_SIZE - 1)); 510 hlist_for_each_entry(t2, hash_head, hash_entry) { 511 if (traces_identical(trace, t2)) 512 return t2; 513 } 514 nr_stack_trace_entries += LOCK_TRACE_SIZE_IN_LONGS + trace->nr_entries; 515 hlist_add_head(&trace->hash_entry, hash_head); 516 517 return trace; 518 } 519 520 /* Return the number of stack traces in the stack_trace[] array. */ 521 u64 lockdep_stack_trace_count(void) 522 { 523 struct lock_trace *trace; 524 u64 c = 0; 525 int i; 526 527 for (i = 0; i < ARRAY_SIZE(stack_trace_hash); i++) { 528 hlist_for_each_entry(trace, &stack_trace_hash[i], hash_entry) { 529 c++; 530 } 531 } 532 533 return c; 534 } 535 536 /* Return the number of stack hash chains that have at least one stack trace. */ 537 u64 lockdep_stack_hash_count(void) 538 { 539 u64 c = 0; 540 int i; 541 542 for (i = 0; i < ARRAY_SIZE(stack_trace_hash); i++) 543 if (!hlist_empty(&stack_trace_hash[i])) 544 c++; 545 546 return c; 547 } 548 #endif 549 550 unsigned int nr_hardirq_chains; 551 unsigned int nr_softirq_chains; 552 unsigned int nr_process_chains; 553 unsigned int max_lockdep_depth; 554 555 #ifdef CONFIG_DEBUG_LOCKDEP 556 /* 557 * Various lockdep statistics: 558 */ 559 DEFINE_PER_CPU(struct lockdep_stats, lockdep_stats); 560 #endif 561 562 #ifdef CONFIG_PROVE_LOCKING 563 /* 564 * Locking printouts: 565 */ 566 567 #define __USAGE(__STATE) \ 568 [LOCK_USED_IN_##__STATE] = "IN-"__stringify(__STATE)"-W", \ 569 [LOCK_ENABLED_##__STATE] = __stringify(__STATE)"-ON-W", \ 570 [LOCK_USED_IN_##__STATE##_READ] = "IN-"__stringify(__STATE)"-R",\ 571 [LOCK_ENABLED_##__STATE##_READ] = __stringify(__STATE)"-ON-R", 572 573 static const char *usage_str[] = 574 { 575 #define LOCKDEP_STATE(__STATE) __USAGE(__STATE) 576 #include "lockdep_states.h" 577 #undef LOCKDEP_STATE 578 [LOCK_USED] = "INITIAL USE", 579 }; 580 #endif 581 582 const char *__get_key_name(const struct lockdep_subclass_key *key, char *str) 583 { 584 return kallsyms_lookup((unsigned long)key, NULL, NULL, NULL, str); 585 } 586 587 static inline unsigned long lock_flag(enum lock_usage_bit bit) 588 { 589 return 1UL << bit; 590 } 591 592 static char get_usage_char(struct lock_class *class, enum lock_usage_bit bit) 593 { 594 /* 595 * The usage character defaults to '.' (i.e., irqs disabled and not in 596 * irq context), which is the safest usage category. 597 */ 598 char c = '.'; 599 600 /* 601 * The order of the following usage checks matters, which will 602 * result in the outcome character as follows: 603 * 604 * - '+': irq is enabled and not in irq context 605 * - '-': in irq context and irq is disabled 606 * - '?': in irq context and irq is enabled 607 */ 608 if (class->usage_mask & lock_flag(bit + LOCK_USAGE_DIR_MASK)) { 609 c = '+'; 610 if (class->usage_mask & lock_flag(bit)) 611 c = '?'; 612 } else if (class->usage_mask & lock_flag(bit)) 613 c = '-'; 614 615 return c; 616 } 617 618 void get_usage_chars(struct lock_class *class, char usage[LOCK_USAGE_CHARS]) 619 { 620 int i = 0; 621 622 #define LOCKDEP_STATE(__STATE) \ 623 usage[i++] = get_usage_char(class, LOCK_USED_IN_##__STATE); \ 624 usage[i++] = get_usage_char(class, LOCK_USED_IN_##__STATE##_READ); 625 #include "lockdep_states.h" 626 #undef LOCKDEP_STATE 627 628 usage[i] = '\0'; 629 } 630 631 static void __print_lock_name(struct lock_class *class) 632 { 633 char str[KSYM_NAME_LEN]; 634 const char *name; 635 636 name = class->name; 637 if (!name) { 638 name = __get_key_name(class->key, str); 639 printk(KERN_CONT "%s", name); 640 } else { 641 printk(KERN_CONT "%s", name); 642 if (class->name_version > 1) 643 printk(KERN_CONT "#%d", class->name_version); 644 if (class->subclass) 645 printk(KERN_CONT "/%d", class->subclass); 646 } 647 } 648 649 static void print_lock_name(struct lock_class *class) 650 { 651 char usage[LOCK_USAGE_CHARS]; 652 653 get_usage_chars(class, usage); 654 655 printk(KERN_CONT " ("); 656 __print_lock_name(class); 657 printk(KERN_CONT "){%s}", usage); 658 } 659 660 static void print_lockdep_cache(struct lockdep_map *lock) 661 { 662 const char *name; 663 char str[KSYM_NAME_LEN]; 664 665 name = lock->name; 666 if (!name) 667 name = __get_key_name(lock->key->subkeys, str); 668 669 printk(KERN_CONT "%s", name); 670 } 671 672 static void print_lock(struct held_lock *hlock) 673 { 674 /* 675 * We can be called locklessly through debug_show_all_locks() so be 676 * extra careful, the hlock might have been released and cleared. 677 * 678 * If this indeed happens, lets pretend it does not hurt to continue 679 * to print the lock unless the hlock class_idx does not point to a 680 * registered class. The rationale here is: since we don't attempt 681 * to distinguish whether we are in this situation, if it just 682 * happened we can't count on class_idx to tell either. 683 */ 684 struct lock_class *lock = hlock_class(hlock); 685 686 if (!lock) { 687 printk(KERN_CONT "<RELEASED>\n"); 688 return; 689 } 690 691 printk(KERN_CONT "%px", hlock->instance); 692 print_lock_name(lock); 693 printk(KERN_CONT ", at: %pS\n", (void *)hlock->acquire_ip); 694 } 695 696 static void lockdep_print_held_locks(struct task_struct *p) 697 { 698 int i, depth = READ_ONCE(p->lockdep_depth); 699 700 if (!depth) 701 printk("no locks held by %s/%d.\n", p->comm, task_pid_nr(p)); 702 else 703 printk("%d lock%s held by %s/%d:\n", depth, 704 depth > 1 ? "s" : "", p->comm, task_pid_nr(p)); 705 /* 706 * It's not reliable to print a task's held locks if it's not sleeping 707 * and it's not the current task. 708 */ 709 if (p->state == TASK_RUNNING && p != current) 710 return; 711 for (i = 0; i < depth; i++) { 712 printk(" #%d: ", i); 713 print_lock(p->held_locks + i); 714 } 715 } 716 717 static void print_kernel_ident(void) 718 { 719 printk("%s %.*s %s\n", init_utsname()->release, 720 (int)strcspn(init_utsname()->version, " "), 721 init_utsname()->version, 722 print_tainted()); 723 } 724 725 static int very_verbose(struct lock_class *class) 726 { 727 #if VERY_VERBOSE 728 return class_filter(class); 729 #endif 730 return 0; 731 } 732 733 /* 734 * Is this the address of a static object: 735 */ 736 #ifdef __KERNEL__ 737 static int static_obj(const void *obj) 738 { 739 unsigned long start = (unsigned long) &_stext, 740 end = (unsigned long) &_end, 741 addr = (unsigned long) obj; 742 743 if (arch_is_kernel_initmem_freed(addr)) 744 return 0; 745 746 /* 747 * static variable? 748 */ 749 if ((addr >= start) && (addr < end)) 750 return 1; 751 752 if (arch_is_kernel_data(addr)) 753 return 1; 754 755 /* 756 * in-kernel percpu var? 757 */ 758 if (is_kernel_percpu_address(addr)) 759 return 1; 760 761 /* 762 * module static or percpu var? 763 */ 764 return is_module_address(addr) || is_module_percpu_address(addr); 765 } 766 #endif 767 768 /* 769 * To make lock name printouts unique, we calculate a unique 770 * class->name_version generation counter. The caller must hold the graph 771 * lock. 772 */ 773 static int count_matching_names(struct lock_class *new_class) 774 { 775 struct lock_class *class; 776 int count = 0; 777 778 if (!new_class->name) 779 return 0; 780 781 list_for_each_entry(class, &all_lock_classes, lock_entry) { 782 if (new_class->key - new_class->subclass == class->key) 783 return class->name_version; 784 if (class->name && !strcmp(class->name, new_class->name)) 785 count = max(count, class->name_version); 786 } 787 788 return count + 1; 789 } 790 791 static inline struct lock_class * 792 look_up_lock_class(const struct lockdep_map *lock, unsigned int subclass) 793 { 794 struct lockdep_subclass_key *key; 795 struct hlist_head *hash_head; 796 struct lock_class *class; 797 798 if (unlikely(subclass >= MAX_LOCKDEP_SUBCLASSES)) { 799 debug_locks_off(); 800 printk(KERN_ERR 801 "BUG: looking up invalid subclass: %u\n", subclass); 802 printk(KERN_ERR 803 "turning off the locking correctness validator.\n"); 804 dump_stack(); 805 return NULL; 806 } 807 808 /* 809 * If it is not initialised then it has never been locked, 810 * so it won't be present in the hash table. 811 */ 812 if (unlikely(!lock->key)) 813 return NULL; 814 815 /* 816 * NOTE: the class-key must be unique. For dynamic locks, a static 817 * lock_class_key variable is passed in through the mutex_init() 818 * (or spin_lock_init()) call - which acts as the key. For static 819 * locks we use the lock object itself as the key. 820 */ 821 BUILD_BUG_ON(sizeof(struct lock_class_key) > 822 sizeof(struct lockdep_map)); 823 824 key = lock->key->subkeys + subclass; 825 826 hash_head = classhashentry(key); 827 828 /* 829 * We do an RCU walk of the hash, see lockdep_free_key_range(). 830 */ 831 if (DEBUG_LOCKS_WARN_ON(!irqs_disabled())) 832 return NULL; 833 834 hlist_for_each_entry_rcu(class, hash_head, hash_entry) { 835 if (class->key == key) { 836 /* 837 * Huh! same key, different name? Did someone trample 838 * on some memory? We're most confused. 839 */ 840 WARN_ON_ONCE(class->name != lock->name && 841 lock->key != &__lockdep_no_validate__); 842 return class; 843 } 844 } 845 846 return NULL; 847 } 848 849 /* 850 * Static locks do not have their class-keys yet - for them the key is 851 * the lock object itself. If the lock is in the per cpu area, the 852 * canonical address of the lock (per cpu offset removed) is used. 853 */ 854 static bool assign_lock_key(struct lockdep_map *lock) 855 { 856 unsigned long can_addr, addr = (unsigned long)lock; 857 858 #ifdef __KERNEL__ 859 /* 860 * lockdep_free_key_range() assumes that struct lock_class_key 861 * objects do not overlap. Since we use the address of lock 862 * objects as class key for static objects, check whether the 863 * size of lock_class_key objects does not exceed the size of 864 * the smallest lock object. 865 */ 866 BUILD_BUG_ON(sizeof(struct lock_class_key) > sizeof(raw_spinlock_t)); 867 #endif 868 869 if (__is_kernel_percpu_address(addr, &can_addr)) 870 lock->key = (void *)can_addr; 871 else if (__is_module_percpu_address(addr, &can_addr)) 872 lock->key = (void *)can_addr; 873 else if (static_obj(lock)) 874 lock->key = (void *)lock; 875 else { 876 /* Debug-check: all keys must be persistent! */ 877 debug_locks_off(); 878 pr_err("INFO: trying to register non-static key.\n"); 879 pr_err("the code is fine but needs lockdep annotation.\n"); 880 pr_err("turning off the locking correctness validator.\n"); 881 dump_stack(); 882 return false; 883 } 884 885 return true; 886 } 887 888 #ifdef CONFIG_DEBUG_LOCKDEP 889 890 /* Check whether element @e occurs in list @h */ 891 static bool in_list(struct list_head *e, struct list_head *h) 892 { 893 struct list_head *f; 894 895 list_for_each(f, h) { 896 if (e == f) 897 return true; 898 } 899 900 return false; 901 } 902 903 /* 904 * Check whether entry @e occurs in any of the locks_after or locks_before 905 * lists. 906 */ 907 static bool in_any_class_list(struct list_head *e) 908 { 909 struct lock_class *class; 910 int i; 911 912 for (i = 0; i < ARRAY_SIZE(lock_classes); i++) { 913 class = &lock_classes[i]; 914 if (in_list(e, &class->locks_after) || 915 in_list(e, &class->locks_before)) 916 return true; 917 } 918 return false; 919 } 920 921 static bool class_lock_list_valid(struct lock_class *c, struct list_head *h) 922 { 923 struct lock_list *e; 924 925 list_for_each_entry(e, h, entry) { 926 if (e->links_to != c) { 927 printk(KERN_INFO "class %s: mismatch for lock entry %ld; class %s <> %s", 928 c->name ? : "(?)", 929 (unsigned long)(e - list_entries), 930 e->links_to && e->links_to->name ? 931 e->links_to->name : "(?)", 932 e->class && e->class->name ? e->class->name : 933 "(?)"); 934 return false; 935 } 936 } 937 return true; 938 } 939 940 #ifdef CONFIG_PROVE_LOCKING 941 static u16 chain_hlocks[MAX_LOCKDEP_CHAIN_HLOCKS]; 942 #endif 943 944 static bool check_lock_chain_key(struct lock_chain *chain) 945 { 946 #ifdef CONFIG_PROVE_LOCKING 947 u64 chain_key = INITIAL_CHAIN_KEY; 948 int i; 949 950 for (i = chain->base; i < chain->base + chain->depth; i++) 951 chain_key = iterate_chain_key(chain_key, chain_hlocks[i]); 952 /* 953 * The 'unsigned long long' casts avoid that a compiler warning 954 * is reported when building tools/lib/lockdep. 955 */ 956 if (chain->chain_key != chain_key) { 957 printk(KERN_INFO "chain %lld: key %#llx <> %#llx\n", 958 (unsigned long long)(chain - lock_chains), 959 (unsigned long long)chain->chain_key, 960 (unsigned long long)chain_key); 961 return false; 962 } 963 #endif 964 return true; 965 } 966 967 static bool in_any_zapped_class_list(struct lock_class *class) 968 { 969 struct pending_free *pf; 970 int i; 971 972 for (i = 0, pf = delayed_free.pf; i < ARRAY_SIZE(delayed_free.pf); i++, pf++) { 973 if (in_list(&class->lock_entry, &pf->zapped)) 974 return true; 975 } 976 977 return false; 978 } 979 980 static bool __check_data_structures(void) 981 { 982 struct lock_class *class; 983 struct lock_chain *chain; 984 struct hlist_head *head; 985 struct lock_list *e; 986 int i; 987 988 /* Check whether all classes occur in a lock list. */ 989 for (i = 0; i < ARRAY_SIZE(lock_classes); i++) { 990 class = &lock_classes[i]; 991 if (!in_list(&class->lock_entry, &all_lock_classes) && 992 !in_list(&class->lock_entry, &free_lock_classes) && 993 !in_any_zapped_class_list(class)) { 994 printk(KERN_INFO "class %px/%s is not in any class list\n", 995 class, class->name ? : "(?)"); 996 return false; 997 } 998 } 999 1000 /* Check whether all classes have valid lock lists. */ 1001 for (i = 0; i < ARRAY_SIZE(lock_classes); i++) { 1002 class = &lock_classes[i]; 1003 if (!class_lock_list_valid(class, &class->locks_before)) 1004 return false; 1005 if (!class_lock_list_valid(class, &class->locks_after)) 1006 return false; 1007 } 1008 1009 /* Check the chain_key of all lock chains. */ 1010 for (i = 0; i < ARRAY_SIZE(chainhash_table); i++) { 1011 head = chainhash_table + i; 1012 hlist_for_each_entry_rcu(chain, head, entry) { 1013 if (!check_lock_chain_key(chain)) 1014 return false; 1015 } 1016 } 1017 1018 /* 1019 * Check whether all list entries that are in use occur in a class 1020 * lock list. 1021 */ 1022 for_each_set_bit(i, list_entries_in_use, ARRAY_SIZE(list_entries)) { 1023 e = list_entries + i; 1024 if (!in_any_class_list(&e->entry)) { 1025 printk(KERN_INFO "list entry %d is not in any class list; class %s <> %s\n", 1026 (unsigned int)(e - list_entries), 1027 e->class->name ? : "(?)", 1028 e->links_to->name ? : "(?)"); 1029 return false; 1030 } 1031 } 1032 1033 /* 1034 * Check whether all list entries that are not in use do not occur in 1035 * a class lock list. 1036 */ 1037 for_each_clear_bit(i, list_entries_in_use, ARRAY_SIZE(list_entries)) { 1038 e = list_entries + i; 1039 if (in_any_class_list(&e->entry)) { 1040 printk(KERN_INFO "list entry %d occurs in a class list; class %s <> %s\n", 1041 (unsigned int)(e - list_entries), 1042 e->class && e->class->name ? e->class->name : 1043 "(?)", 1044 e->links_to && e->links_to->name ? 1045 e->links_to->name : "(?)"); 1046 return false; 1047 } 1048 } 1049 1050 return true; 1051 } 1052 1053 int check_consistency = 0; 1054 module_param(check_consistency, int, 0644); 1055 1056 static void check_data_structures(void) 1057 { 1058 static bool once = false; 1059 1060 if (check_consistency && !once) { 1061 if (!__check_data_structures()) { 1062 once = true; 1063 WARN_ON(once); 1064 } 1065 } 1066 } 1067 1068 #else /* CONFIG_DEBUG_LOCKDEP */ 1069 1070 static inline void check_data_structures(void) { } 1071 1072 #endif /* CONFIG_DEBUG_LOCKDEP */ 1073 1074 /* 1075 * Initialize the lock_classes[] array elements, the free_lock_classes list 1076 * and also the delayed_free structure. 1077 */ 1078 static void init_data_structures_once(void) 1079 { 1080 static bool ds_initialized, rcu_head_initialized; 1081 int i; 1082 1083 if (likely(rcu_head_initialized)) 1084 return; 1085 1086 if (system_state >= SYSTEM_SCHEDULING) { 1087 init_rcu_head(&delayed_free.rcu_head); 1088 rcu_head_initialized = true; 1089 } 1090 1091 if (ds_initialized) 1092 return; 1093 1094 ds_initialized = true; 1095 1096 INIT_LIST_HEAD(&delayed_free.pf[0].zapped); 1097 INIT_LIST_HEAD(&delayed_free.pf[1].zapped); 1098 1099 for (i = 0; i < ARRAY_SIZE(lock_classes); i++) { 1100 list_add_tail(&lock_classes[i].lock_entry, &free_lock_classes); 1101 INIT_LIST_HEAD(&lock_classes[i].locks_after); 1102 INIT_LIST_HEAD(&lock_classes[i].locks_before); 1103 } 1104 } 1105 1106 static inline struct hlist_head *keyhashentry(const struct lock_class_key *key) 1107 { 1108 unsigned long hash = hash_long((uintptr_t)key, KEYHASH_BITS); 1109 1110 return lock_keys_hash + hash; 1111 } 1112 1113 /* Register a dynamically allocated key. */ 1114 void lockdep_register_key(struct lock_class_key *key) 1115 { 1116 struct hlist_head *hash_head; 1117 struct lock_class_key *k; 1118 unsigned long flags; 1119 1120 if (WARN_ON_ONCE(static_obj(key))) 1121 return; 1122 hash_head = keyhashentry(key); 1123 1124 raw_local_irq_save(flags); 1125 if (!graph_lock()) 1126 goto restore_irqs; 1127 hlist_for_each_entry_rcu(k, hash_head, hash_entry) { 1128 if (WARN_ON_ONCE(k == key)) 1129 goto out_unlock; 1130 } 1131 hlist_add_head_rcu(&key->hash_entry, hash_head); 1132 out_unlock: 1133 graph_unlock(); 1134 restore_irqs: 1135 raw_local_irq_restore(flags); 1136 } 1137 EXPORT_SYMBOL_GPL(lockdep_register_key); 1138 1139 /* Check whether a key has been registered as a dynamic key. */ 1140 static bool is_dynamic_key(const struct lock_class_key *key) 1141 { 1142 struct hlist_head *hash_head; 1143 struct lock_class_key *k; 1144 bool found = false; 1145 1146 if (WARN_ON_ONCE(static_obj(key))) 1147 return false; 1148 1149 /* 1150 * If lock debugging is disabled lock_keys_hash[] may contain 1151 * pointers to memory that has already been freed. Avoid triggering 1152 * a use-after-free in that case by returning early. 1153 */ 1154 if (!debug_locks) 1155 return true; 1156 1157 hash_head = keyhashentry(key); 1158 1159 rcu_read_lock(); 1160 hlist_for_each_entry_rcu(k, hash_head, hash_entry) { 1161 if (k == key) { 1162 found = true; 1163 break; 1164 } 1165 } 1166 rcu_read_unlock(); 1167 1168 return found; 1169 } 1170 1171 /* 1172 * Register a lock's class in the hash-table, if the class is not present 1173 * yet. Otherwise we look it up. We cache the result in the lock object 1174 * itself, so actual lookup of the hash should be once per lock object. 1175 */ 1176 static struct lock_class * 1177 register_lock_class(struct lockdep_map *lock, unsigned int subclass, int force) 1178 { 1179 struct lockdep_subclass_key *key; 1180 struct hlist_head *hash_head; 1181 struct lock_class *class; 1182 1183 DEBUG_LOCKS_WARN_ON(!irqs_disabled()); 1184 1185 class = look_up_lock_class(lock, subclass); 1186 if (likely(class)) 1187 goto out_set_class_cache; 1188 1189 if (!lock->key) { 1190 if (!assign_lock_key(lock)) 1191 return NULL; 1192 } else if (!static_obj(lock->key) && !is_dynamic_key(lock->key)) { 1193 return NULL; 1194 } 1195 1196 key = lock->key->subkeys + subclass; 1197 hash_head = classhashentry(key); 1198 1199 if (!graph_lock()) { 1200 return NULL; 1201 } 1202 /* 1203 * We have to do the hash-walk again, to avoid races 1204 * with another CPU: 1205 */ 1206 hlist_for_each_entry_rcu(class, hash_head, hash_entry) { 1207 if (class->key == key) 1208 goto out_unlock_set; 1209 } 1210 1211 init_data_structures_once(); 1212 1213 /* Allocate a new lock class and add it to the hash. */ 1214 class = list_first_entry_or_null(&free_lock_classes, typeof(*class), 1215 lock_entry); 1216 if (!class) { 1217 if (!debug_locks_off_graph_unlock()) { 1218 return NULL; 1219 } 1220 1221 print_lockdep_off("BUG: MAX_LOCKDEP_KEYS too low!"); 1222 dump_stack(); 1223 return NULL; 1224 } 1225 nr_lock_classes++; 1226 __set_bit(class - lock_classes, lock_classes_in_use); 1227 debug_atomic_inc(nr_unused_locks); 1228 class->key = key; 1229 class->name = lock->name; 1230 class->subclass = subclass; 1231 WARN_ON_ONCE(!list_empty(&class->locks_before)); 1232 WARN_ON_ONCE(!list_empty(&class->locks_after)); 1233 class->name_version = count_matching_names(class); 1234 /* 1235 * We use RCU's safe list-add method to make 1236 * parallel walking of the hash-list safe: 1237 */ 1238 hlist_add_head_rcu(&class->hash_entry, hash_head); 1239 /* 1240 * Remove the class from the free list and add it to the global list 1241 * of classes. 1242 */ 1243 list_move_tail(&class->lock_entry, &all_lock_classes); 1244 1245 if (verbose(class)) { 1246 graph_unlock(); 1247 1248 printk("\nnew class %px: %s", class->key, class->name); 1249 if (class->name_version > 1) 1250 printk(KERN_CONT "#%d", class->name_version); 1251 printk(KERN_CONT "\n"); 1252 dump_stack(); 1253 1254 if (!graph_lock()) { 1255 return NULL; 1256 } 1257 } 1258 out_unlock_set: 1259 graph_unlock(); 1260 1261 out_set_class_cache: 1262 if (!subclass || force) 1263 lock->class_cache[0] = class; 1264 else if (subclass < NR_LOCKDEP_CACHING_CLASSES) 1265 lock->class_cache[subclass] = class; 1266 1267 /* 1268 * Hash collision, did we smoke some? We found a class with a matching 1269 * hash but the subclass -- which is hashed in -- didn't match. 1270 */ 1271 if (DEBUG_LOCKS_WARN_ON(class->subclass != subclass)) 1272 return NULL; 1273 1274 return class; 1275 } 1276 1277 #ifdef CONFIG_PROVE_LOCKING 1278 /* 1279 * Allocate a lockdep entry. (assumes the graph_lock held, returns 1280 * with NULL on failure) 1281 */ 1282 static struct lock_list *alloc_list_entry(void) 1283 { 1284 int idx = find_first_zero_bit(list_entries_in_use, 1285 ARRAY_SIZE(list_entries)); 1286 1287 if (idx >= ARRAY_SIZE(list_entries)) { 1288 if (!debug_locks_off_graph_unlock()) 1289 return NULL; 1290 1291 print_lockdep_off("BUG: MAX_LOCKDEP_ENTRIES too low!"); 1292 dump_stack(); 1293 return NULL; 1294 } 1295 nr_list_entries++; 1296 __set_bit(idx, list_entries_in_use); 1297 return list_entries + idx; 1298 } 1299 1300 /* 1301 * Add a new dependency to the head of the list: 1302 */ 1303 static int add_lock_to_list(struct lock_class *this, 1304 struct lock_class *links_to, struct list_head *head, 1305 unsigned long ip, int distance, 1306 const struct lock_trace *trace) 1307 { 1308 struct lock_list *entry; 1309 /* 1310 * Lock not present yet - get a new dependency struct and 1311 * add it to the list: 1312 */ 1313 entry = alloc_list_entry(); 1314 if (!entry) 1315 return 0; 1316 1317 entry->class = this; 1318 entry->links_to = links_to; 1319 entry->distance = distance; 1320 entry->trace = trace; 1321 /* 1322 * Both allocation and removal are done under the graph lock; but 1323 * iteration is under RCU-sched; see look_up_lock_class() and 1324 * lockdep_free_key_range(). 1325 */ 1326 list_add_tail_rcu(&entry->entry, head); 1327 1328 return 1; 1329 } 1330 1331 /* 1332 * For good efficiency of modular, we use power of 2 1333 */ 1334 #define MAX_CIRCULAR_QUEUE_SIZE 4096UL 1335 #define CQ_MASK (MAX_CIRCULAR_QUEUE_SIZE-1) 1336 1337 /* 1338 * The circular_queue and helpers are used to implement graph 1339 * breadth-first search (BFS) algorithm, by which we can determine 1340 * whether there is a path from a lock to another. In deadlock checks, 1341 * a path from the next lock to be acquired to a previous held lock 1342 * indicates that adding the <prev> -> <next> lock dependency will 1343 * produce a circle in the graph. Breadth-first search instead of 1344 * depth-first search is used in order to find the shortest (circular) 1345 * path. 1346 */ 1347 struct circular_queue { 1348 struct lock_list *element[MAX_CIRCULAR_QUEUE_SIZE]; 1349 unsigned int front, rear; 1350 }; 1351 1352 static struct circular_queue lock_cq; 1353 1354 unsigned int max_bfs_queue_depth; 1355 1356 static unsigned int lockdep_dependency_gen_id; 1357 1358 static inline void __cq_init(struct circular_queue *cq) 1359 { 1360 cq->front = cq->rear = 0; 1361 lockdep_dependency_gen_id++; 1362 } 1363 1364 static inline int __cq_empty(struct circular_queue *cq) 1365 { 1366 return (cq->front == cq->rear); 1367 } 1368 1369 static inline int __cq_full(struct circular_queue *cq) 1370 { 1371 return ((cq->rear + 1) & CQ_MASK) == cq->front; 1372 } 1373 1374 static inline int __cq_enqueue(struct circular_queue *cq, struct lock_list *elem) 1375 { 1376 if (__cq_full(cq)) 1377 return -1; 1378 1379 cq->element[cq->rear] = elem; 1380 cq->rear = (cq->rear + 1) & CQ_MASK; 1381 return 0; 1382 } 1383 1384 /* 1385 * Dequeue an element from the circular_queue, return a lock_list if 1386 * the queue is not empty, or NULL if otherwise. 1387 */ 1388 static inline struct lock_list * __cq_dequeue(struct circular_queue *cq) 1389 { 1390 struct lock_list * lock; 1391 1392 if (__cq_empty(cq)) 1393 return NULL; 1394 1395 lock = cq->element[cq->front]; 1396 cq->front = (cq->front + 1) & CQ_MASK; 1397 1398 return lock; 1399 } 1400 1401 static inline unsigned int __cq_get_elem_count(struct circular_queue *cq) 1402 { 1403 return (cq->rear - cq->front) & CQ_MASK; 1404 } 1405 1406 static inline void mark_lock_accessed(struct lock_list *lock, 1407 struct lock_list *parent) 1408 { 1409 unsigned long nr; 1410 1411 nr = lock - list_entries; 1412 WARN_ON(nr >= ARRAY_SIZE(list_entries)); /* Out-of-bounds, input fail */ 1413 lock->parent = parent; 1414 lock->class->dep_gen_id = lockdep_dependency_gen_id; 1415 } 1416 1417 static inline unsigned long lock_accessed(struct lock_list *lock) 1418 { 1419 unsigned long nr; 1420 1421 nr = lock - list_entries; 1422 WARN_ON(nr >= ARRAY_SIZE(list_entries)); /* Out-of-bounds, input fail */ 1423 return lock->class->dep_gen_id == lockdep_dependency_gen_id; 1424 } 1425 1426 static inline struct lock_list *get_lock_parent(struct lock_list *child) 1427 { 1428 return child->parent; 1429 } 1430 1431 static inline int get_lock_depth(struct lock_list *child) 1432 { 1433 int depth = 0; 1434 struct lock_list *parent; 1435 1436 while ((parent = get_lock_parent(child))) { 1437 child = parent; 1438 depth++; 1439 } 1440 return depth; 1441 } 1442 1443 /* 1444 * Return the forward or backward dependency list. 1445 * 1446 * @lock: the lock_list to get its class's dependency list 1447 * @offset: the offset to struct lock_class to determine whether it is 1448 * locks_after or locks_before 1449 */ 1450 static inline struct list_head *get_dep_list(struct lock_list *lock, int offset) 1451 { 1452 void *lock_class = lock->class; 1453 1454 return lock_class + offset; 1455 } 1456 1457 /* 1458 * Forward- or backward-dependency search, used for both circular dependency 1459 * checking and hardirq-unsafe/softirq-unsafe checking. 1460 */ 1461 static int __bfs(struct lock_list *source_entry, 1462 void *data, 1463 int (*match)(struct lock_list *entry, void *data), 1464 struct lock_list **target_entry, 1465 int offset) 1466 { 1467 struct lock_list *entry; 1468 struct lock_list *lock; 1469 struct list_head *head; 1470 struct circular_queue *cq = &lock_cq; 1471 int ret = 1; 1472 1473 if (match(source_entry, data)) { 1474 *target_entry = source_entry; 1475 ret = 0; 1476 goto exit; 1477 } 1478 1479 head = get_dep_list(source_entry, offset); 1480 if (list_empty(head)) 1481 goto exit; 1482 1483 __cq_init(cq); 1484 __cq_enqueue(cq, source_entry); 1485 1486 while ((lock = __cq_dequeue(cq))) { 1487 1488 if (!lock->class) { 1489 ret = -2; 1490 goto exit; 1491 } 1492 1493 head = get_dep_list(lock, offset); 1494 1495 DEBUG_LOCKS_WARN_ON(!irqs_disabled()); 1496 1497 list_for_each_entry_rcu(entry, head, entry) { 1498 if (!lock_accessed(entry)) { 1499 unsigned int cq_depth; 1500 mark_lock_accessed(entry, lock); 1501 if (match(entry, data)) { 1502 *target_entry = entry; 1503 ret = 0; 1504 goto exit; 1505 } 1506 1507 if (__cq_enqueue(cq, entry)) { 1508 ret = -1; 1509 goto exit; 1510 } 1511 cq_depth = __cq_get_elem_count(cq); 1512 if (max_bfs_queue_depth < cq_depth) 1513 max_bfs_queue_depth = cq_depth; 1514 } 1515 } 1516 } 1517 exit: 1518 return ret; 1519 } 1520 1521 static inline int __bfs_forwards(struct lock_list *src_entry, 1522 void *data, 1523 int (*match)(struct lock_list *entry, void *data), 1524 struct lock_list **target_entry) 1525 { 1526 return __bfs(src_entry, data, match, target_entry, 1527 offsetof(struct lock_class, locks_after)); 1528 1529 } 1530 1531 static inline int __bfs_backwards(struct lock_list *src_entry, 1532 void *data, 1533 int (*match)(struct lock_list *entry, void *data), 1534 struct lock_list **target_entry) 1535 { 1536 return __bfs(src_entry, data, match, target_entry, 1537 offsetof(struct lock_class, locks_before)); 1538 1539 } 1540 1541 static void print_lock_trace(const struct lock_trace *trace, 1542 unsigned int spaces) 1543 { 1544 stack_trace_print(trace->entries, trace->nr_entries, spaces); 1545 } 1546 1547 /* 1548 * Print a dependency chain entry (this is only done when a deadlock 1549 * has been detected): 1550 */ 1551 static noinline void 1552 print_circular_bug_entry(struct lock_list *target, int depth) 1553 { 1554 if (debug_locks_silent) 1555 return; 1556 printk("\n-> #%u", depth); 1557 print_lock_name(target->class); 1558 printk(KERN_CONT ":\n"); 1559 print_lock_trace(target->trace, 6); 1560 } 1561 1562 static void 1563 print_circular_lock_scenario(struct held_lock *src, 1564 struct held_lock *tgt, 1565 struct lock_list *prt) 1566 { 1567 struct lock_class *source = hlock_class(src); 1568 struct lock_class *target = hlock_class(tgt); 1569 struct lock_class *parent = prt->class; 1570 1571 /* 1572 * A direct locking problem where unsafe_class lock is taken 1573 * directly by safe_class lock, then all we need to show 1574 * is the deadlock scenario, as it is obvious that the 1575 * unsafe lock is taken under the safe lock. 1576 * 1577 * But if there is a chain instead, where the safe lock takes 1578 * an intermediate lock (middle_class) where this lock is 1579 * not the same as the safe lock, then the lock chain is 1580 * used to describe the problem. Otherwise we would need 1581 * to show a different CPU case for each link in the chain 1582 * from the safe_class lock to the unsafe_class lock. 1583 */ 1584 if (parent != source) { 1585 printk("Chain exists of:\n "); 1586 __print_lock_name(source); 1587 printk(KERN_CONT " --> "); 1588 __print_lock_name(parent); 1589 printk(KERN_CONT " --> "); 1590 __print_lock_name(target); 1591 printk(KERN_CONT "\n\n"); 1592 } 1593 1594 printk(" Possible unsafe locking scenario:\n\n"); 1595 printk(" CPU0 CPU1\n"); 1596 printk(" ---- ----\n"); 1597 printk(" lock("); 1598 __print_lock_name(target); 1599 printk(KERN_CONT ");\n"); 1600 printk(" lock("); 1601 __print_lock_name(parent); 1602 printk(KERN_CONT ");\n"); 1603 printk(" lock("); 1604 __print_lock_name(target); 1605 printk(KERN_CONT ");\n"); 1606 printk(" lock("); 1607 __print_lock_name(source); 1608 printk(KERN_CONT ");\n"); 1609 printk("\n *** DEADLOCK ***\n\n"); 1610 } 1611 1612 /* 1613 * When a circular dependency is detected, print the 1614 * header first: 1615 */ 1616 static noinline void 1617 print_circular_bug_header(struct lock_list *entry, unsigned int depth, 1618 struct held_lock *check_src, 1619 struct held_lock *check_tgt) 1620 { 1621 struct task_struct *curr = current; 1622 1623 if (debug_locks_silent) 1624 return; 1625 1626 pr_warn("\n"); 1627 pr_warn("======================================================\n"); 1628 pr_warn("WARNING: possible circular locking dependency detected\n"); 1629 print_kernel_ident(); 1630 pr_warn("------------------------------------------------------\n"); 1631 pr_warn("%s/%d is trying to acquire lock:\n", 1632 curr->comm, task_pid_nr(curr)); 1633 print_lock(check_src); 1634 1635 pr_warn("\nbut task is already holding lock:\n"); 1636 1637 print_lock(check_tgt); 1638 pr_warn("\nwhich lock already depends on the new lock.\n\n"); 1639 pr_warn("\nthe existing dependency chain (in reverse order) is:\n"); 1640 1641 print_circular_bug_entry(entry, depth); 1642 } 1643 1644 static inline int class_equal(struct lock_list *entry, void *data) 1645 { 1646 return entry->class == data; 1647 } 1648 1649 static noinline void print_circular_bug(struct lock_list *this, 1650 struct lock_list *target, 1651 struct held_lock *check_src, 1652 struct held_lock *check_tgt) 1653 { 1654 struct task_struct *curr = current; 1655 struct lock_list *parent; 1656 struct lock_list *first_parent; 1657 int depth; 1658 1659 if (!debug_locks_off_graph_unlock() || debug_locks_silent) 1660 return; 1661 1662 this->trace = save_trace(); 1663 if (!this->trace) 1664 return; 1665 1666 depth = get_lock_depth(target); 1667 1668 print_circular_bug_header(target, depth, check_src, check_tgt); 1669 1670 parent = get_lock_parent(target); 1671 first_parent = parent; 1672 1673 while (parent) { 1674 print_circular_bug_entry(parent, --depth); 1675 parent = get_lock_parent(parent); 1676 } 1677 1678 printk("\nother info that might help us debug this:\n\n"); 1679 print_circular_lock_scenario(check_src, check_tgt, 1680 first_parent); 1681 1682 lockdep_print_held_locks(curr); 1683 1684 printk("\nstack backtrace:\n"); 1685 dump_stack(); 1686 } 1687 1688 static noinline void print_bfs_bug(int ret) 1689 { 1690 if (!debug_locks_off_graph_unlock()) 1691 return; 1692 1693 /* 1694 * Breadth-first-search failed, graph got corrupted? 1695 */ 1696 WARN(1, "lockdep bfs error:%d\n", ret); 1697 } 1698 1699 static int noop_count(struct lock_list *entry, void *data) 1700 { 1701 (*(unsigned long *)data)++; 1702 return 0; 1703 } 1704 1705 static unsigned long __lockdep_count_forward_deps(struct lock_list *this) 1706 { 1707 unsigned long count = 0; 1708 struct lock_list *uninitialized_var(target_entry); 1709 1710 __bfs_forwards(this, (void *)&count, noop_count, &target_entry); 1711 1712 return count; 1713 } 1714 unsigned long lockdep_count_forward_deps(struct lock_class *class) 1715 { 1716 unsigned long ret, flags; 1717 struct lock_list this; 1718 1719 this.parent = NULL; 1720 this.class = class; 1721 1722 raw_local_irq_save(flags); 1723 arch_spin_lock(&lockdep_lock); 1724 ret = __lockdep_count_forward_deps(&this); 1725 arch_spin_unlock(&lockdep_lock); 1726 raw_local_irq_restore(flags); 1727 1728 return ret; 1729 } 1730 1731 static unsigned long __lockdep_count_backward_deps(struct lock_list *this) 1732 { 1733 unsigned long count = 0; 1734 struct lock_list *uninitialized_var(target_entry); 1735 1736 __bfs_backwards(this, (void *)&count, noop_count, &target_entry); 1737 1738 return count; 1739 } 1740 1741 unsigned long lockdep_count_backward_deps(struct lock_class *class) 1742 { 1743 unsigned long ret, flags; 1744 struct lock_list this; 1745 1746 this.parent = NULL; 1747 this.class = class; 1748 1749 raw_local_irq_save(flags); 1750 arch_spin_lock(&lockdep_lock); 1751 ret = __lockdep_count_backward_deps(&this); 1752 arch_spin_unlock(&lockdep_lock); 1753 raw_local_irq_restore(flags); 1754 1755 return ret; 1756 } 1757 1758 /* 1759 * Check that the dependency graph starting at <src> can lead to 1760 * <target> or not. Print an error and return 0 if it does. 1761 */ 1762 static noinline int 1763 check_path(struct lock_class *target, struct lock_list *src_entry, 1764 struct lock_list **target_entry) 1765 { 1766 int ret; 1767 1768 ret = __bfs_forwards(src_entry, (void *)target, class_equal, 1769 target_entry); 1770 1771 if (unlikely(ret < 0)) 1772 print_bfs_bug(ret); 1773 1774 return ret; 1775 } 1776 1777 /* 1778 * Prove that the dependency graph starting at <src> can not 1779 * lead to <target>. If it can, there is a circle when adding 1780 * <target> -> <src> dependency. 1781 * 1782 * Print an error and return 0 if it does. 1783 */ 1784 static noinline int 1785 check_noncircular(struct held_lock *src, struct held_lock *target, 1786 struct lock_trace **const trace) 1787 { 1788 int ret; 1789 struct lock_list *uninitialized_var(target_entry); 1790 struct lock_list src_entry = { 1791 .class = hlock_class(src), 1792 .parent = NULL, 1793 }; 1794 1795 debug_atomic_inc(nr_cyclic_checks); 1796 1797 ret = check_path(hlock_class(target), &src_entry, &target_entry); 1798 1799 if (unlikely(!ret)) { 1800 if (!*trace) { 1801 /* 1802 * If save_trace fails here, the printing might 1803 * trigger a WARN but because of the !nr_entries it 1804 * should not do bad things. 1805 */ 1806 *trace = save_trace(); 1807 } 1808 1809 print_circular_bug(&src_entry, target_entry, src, target); 1810 } 1811 1812 return ret; 1813 } 1814 1815 #ifdef CONFIG_LOCKDEP_SMALL 1816 /* 1817 * Check that the dependency graph starting at <src> can lead to 1818 * <target> or not. If it can, <src> -> <target> dependency is already 1819 * in the graph. 1820 * 1821 * Print an error and return 2 if it does or 1 if it does not. 1822 */ 1823 static noinline int 1824 check_redundant(struct held_lock *src, struct held_lock *target) 1825 { 1826 int ret; 1827 struct lock_list *uninitialized_var(target_entry); 1828 struct lock_list src_entry = { 1829 .class = hlock_class(src), 1830 .parent = NULL, 1831 }; 1832 1833 debug_atomic_inc(nr_redundant_checks); 1834 1835 ret = check_path(hlock_class(target), &src_entry, &target_entry); 1836 1837 if (!ret) { 1838 debug_atomic_inc(nr_redundant); 1839 ret = 2; 1840 } else if (ret < 0) 1841 ret = 0; 1842 1843 return ret; 1844 } 1845 #endif 1846 1847 #ifdef CONFIG_TRACE_IRQFLAGS 1848 1849 static inline int usage_accumulate(struct lock_list *entry, void *mask) 1850 { 1851 *(unsigned long *)mask |= entry->class->usage_mask; 1852 1853 return 0; 1854 } 1855 1856 /* 1857 * Forwards and backwards subgraph searching, for the purposes of 1858 * proving that two subgraphs can be connected by a new dependency 1859 * without creating any illegal irq-safe -> irq-unsafe lock dependency. 1860 */ 1861 1862 static inline int usage_match(struct lock_list *entry, void *mask) 1863 { 1864 return entry->class->usage_mask & *(unsigned long *)mask; 1865 } 1866 1867 /* 1868 * Find a node in the forwards-direction dependency sub-graph starting 1869 * at @root->class that matches @bit. 1870 * 1871 * Return 0 if such a node exists in the subgraph, and put that node 1872 * into *@target_entry. 1873 * 1874 * Return 1 otherwise and keep *@target_entry unchanged. 1875 * Return <0 on error. 1876 */ 1877 static int 1878 find_usage_forwards(struct lock_list *root, unsigned long usage_mask, 1879 struct lock_list **target_entry) 1880 { 1881 int result; 1882 1883 debug_atomic_inc(nr_find_usage_forwards_checks); 1884 1885 result = __bfs_forwards(root, &usage_mask, usage_match, target_entry); 1886 1887 return result; 1888 } 1889 1890 /* 1891 * Find a node in the backwards-direction dependency sub-graph starting 1892 * at @root->class that matches @bit. 1893 * 1894 * Return 0 if such a node exists in the subgraph, and put that node 1895 * into *@target_entry. 1896 * 1897 * Return 1 otherwise and keep *@target_entry unchanged. 1898 * Return <0 on error. 1899 */ 1900 static int 1901 find_usage_backwards(struct lock_list *root, unsigned long usage_mask, 1902 struct lock_list **target_entry) 1903 { 1904 int result; 1905 1906 debug_atomic_inc(nr_find_usage_backwards_checks); 1907 1908 result = __bfs_backwards(root, &usage_mask, usage_match, target_entry); 1909 1910 return result; 1911 } 1912 1913 static void print_lock_class_header(struct lock_class *class, int depth) 1914 { 1915 int bit; 1916 1917 printk("%*s->", depth, ""); 1918 print_lock_name(class); 1919 #ifdef CONFIG_DEBUG_LOCKDEP 1920 printk(KERN_CONT " ops: %lu", debug_class_ops_read(class)); 1921 #endif 1922 printk(KERN_CONT " {\n"); 1923 1924 for (bit = 0; bit < LOCK_USAGE_STATES; bit++) { 1925 if (class->usage_mask & (1 << bit)) { 1926 int len = depth; 1927 1928 len += printk("%*s %s", depth, "", usage_str[bit]); 1929 len += printk(KERN_CONT " at:\n"); 1930 print_lock_trace(class->usage_traces[bit], len); 1931 } 1932 } 1933 printk("%*s }\n", depth, ""); 1934 1935 printk("%*s ... key at: [<%px>] %pS\n", 1936 depth, "", class->key, class->key); 1937 } 1938 1939 /* 1940 * printk the shortest lock dependencies from @start to @end in reverse order: 1941 */ 1942 static void __used 1943 print_shortest_lock_dependencies(struct lock_list *leaf, 1944 struct lock_list *root) 1945 { 1946 struct lock_list *entry = leaf; 1947 int depth; 1948 1949 /*compute depth from generated tree by BFS*/ 1950 depth = get_lock_depth(leaf); 1951 1952 do { 1953 print_lock_class_header(entry->class, depth); 1954 printk("%*s ... acquired at:\n", depth, ""); 1955 print_lock_trace(entry->trace, 2); 1956 printk("\n"); 1957 1958 if (depth == 0 && (entry != root)) { 1959 printk("lockdep:%s bad path found in chain graph\n", __func__); 1960 break; 1961 } 1962 1963 entry = get_lock_parent(entry); 1964 depth--; 1965 } while (entry && (depth >= 0)); 1966 } 1967 1968 static void 1969 print_irq_lock_scenario(struct lock_list *safe_entry, 1970 struct lock_list *unsafe_entry, 1971 struct lock_class *prev_class, 1972 struct lock_class *next_class) 1973 { 1974 struct lock_class *safe_class = safe_entry->class; 1975 struct lock_class *unsafe_class = unsafe_entry->class; 1976 struct lock_class *middle_class = prev_class; 1977 1978 if (middle_class == safe_class) 1979 middle_class = next_class; 1980 1981 /* 1982 * A direct locking problem where unsafe_class lock is taken 1983 * directly by safe_class lock, then all we need to show 1984 * is the deadlock scenario, as it is obvious that the 1985 * unsafe lock is taken under the safe lock. 1986 * 1987 * But if there is a chain instead, where the safe lock takes 1988 * an intermediate lock (middle_class) where this lock is 1989 * not the same as the safe lock, then the lock chain is 1990 * used to describe the problem. Otherwise we would need 1991 * to show a different CPU case for each link in the chain 1992 * from the safe_class lock to the unsafe_class lock. 1993 */ 1994 if (middle_class != unsafe_class) { 1995 printk("Chain exists of:\n "); 1996 __print_lock_name(safe_class); 1997 printk(KERN_CONT " --> "); 1998 __print_lock_name(middle_class); 1999 printk(KERN_CONT " --> "); 2000 __print_lock_name(unsafe_class); 2001 printk(KERN_CONT "\n\n"); 2002 } 2003 2004 printk(" Possible interrupt unsafe locking scenario:\n\n"); 2005 printk(" CPU0 CPU1\n"); 2006 printk(" ---- ----\n"); 2007 printk(" lock("); 2008 __print_lock_name(unsafe_class); 2009 printk(KERN_CONT ");\n"); 2010 printk(" local_irq_disable();\n"); 2011 printk(" lock("); 2012 __print_lock_name(safe_class); 2013 printk(KERN_CONT ");\n"); 2014 printk(" lock("); 2015 __print_lock_name(middle_class); 2016 printk(KERN_CONT ");\n"); 2017 printk(" <Interrupt>\n"); 2018 printk(" lock("); 2019 __print_lock_name(safe_class); 2020 printk(KERN_CONT ");\n"); 2021 printk("\n *** DEADLOCK ***\n\n"); 2022 } 2023 2024 static void 2025 print_bad_irq_dependency(struct task_struct *curr, 2026 struct lock_list *prev_root, 2027 struct lock_list *next_root, 2028 struct lock_list *backwards_entry, 2029 struct lock_list *forwards_entry, 2030 struct held_lock *prev, 2031 struct held_lock *next, 2032 enum lock_usage_bit bit1, 2033 enum lock_usage_bit bit2, 2034 const char *irqclass) 2035 { 2036 if (!debug_locks_off_graph_unlock() || debug_locks_silent) 2037 return; 2038 2039 pr_warn("\n"); 2040 pr_warn("=====================================================\n"); 2041 pr_warn("WARNING: %s-safe -> %s-unsafe lock order detected\n", 2042 irqclass, irqclass); 2043 print_kernel_ident(); 2044 pr_warn("-----------------------------------------------------\n"); 2045 pr_warn("%s/%d [HC%u[%lu]:SC%u[%lu]:HE%u:SE%u] is trying to acquire:\n", 2046 curr->comm, task_pid_nr(curr), 2047 curr->hardirq_context, hardirq_count() >> HARDIRQ_SHIFT, 2048 curr->softirq_context, softirq_count() >> SOFTIRQ_SHIFT, 2049 curr->hardirqs_enabled, 2050 curr->softirqs_enabled); 2051 print_lock(next); 2052 2053 pr_warn("\nand this task is already holding:\n"); 2054 print_lock(prev); 2055 pr_warn("which would create a new lock dependency:\n"); 2056 print_lock_name(hlock_class(prev)); 2057 pr_cont(" ->"); 2058 print_lock_name(hlock_class(next)); 2059 pr_cont("\n"); 2060 2061 pr_warn("\nbut this new dependency connects a %s-irq-safe lock:\n", 2062 irqclass); 2063 print_lock_name(backwards_entry->class); 2064 pr_warn("\n... which became %s-irq-safe at:\n", irqclass); 2065 2066 print_lock_trace(backwards_entry->class->usage_traces[bit1], 1); 2067 2068 pr_warn("\nto a %s-irq-unsafe lock:\n", irqclass); 2069 print_lock_name(forwards_entry->class); 2070 pr_warn("\n... which became %s-irq-unsafe at:\n", irqclass); 2071 pr_warn("..."); 2072 2073 print_lock_trace(forwards_entry->class->usage_traces[bit2], 1); 2074 2075 pr_warn("\nother info that might help us debug this:\n\n"); 2076 print_irq_lock_scenario(backwards_entry, forwards_entry, 2077 hlock_class(prev), hlock_class(next)); 2078 2079 lockdep_print_held_locks(curr); 2080 2081 pr_warn("\nthe dependencies between %s-irq-safe lock and the holding lock:\n", irqclass); 2082 prev_root->trace = save_trace(); 2083 if (!prev_root->trace) 2084 return; 2085 print_shortest_lock_dependencies(backwards_entry, prev_root); 2086 2087 pr_warn("\nthe dependencies between the lock to be acquired"); 2088 pr_warn(" and %s-irq-unsafe lock:\n", irqclass); 2089 next_root->trace = save_trace(); 2090 if (!next_root->trace) 2091 return; 2092 print_shortest_lock_dependencies(forwards_entry, next_root); 2093 2094 pr_warn("\nstack backtrace:\n"); 2095 dump_stack(); 2096 } 2097 2098 static const char *state_names[] = { 2099 #define LOCKDEP_STATE(__STATE) \ 2100 __stringify(__STATE), 2101 #include "lockdep_states.h" 2102 #undef LOCKDEP_STATE 2103 }; 2104 2105 static const char *state_rnames[] = { 2106 #define LOCKDEP_STATE(__STATE) \ 2107 __stringify(__STATE)"-READ", 2108 #include "lockdep_states.h" 2109 #undef LOCKDEP_STATE 2110 }; 2111 2112 static inline const char *state_name(enum lock_usage_bit bit) 2113 { 2114 if (bit & LOCK_USAGE_READ_MASK) 2115 return state_rnames[bit >> LOCK_USAGE_DIR_MASK]; 2116 else 2117 return state_names[bit >> LOCK_USAGE_DIR_MASK]; 2118 } 2119 2120 /* 2121 * The bit number is encoded like: 2122 * 2123 * bit0: 0 exclusive, 1 read lock 2124 * bit1: 0 used in irq, 1 irq enabled 2125 * bit2-n: state 2126 */ 2127 static int exclusive_bit(int new_bit) 2128 { 2129 int state = new_bit & LOCK_USAGE_STATE_MASK; 2130 int dir = new_bit & LOCK_USAGE_DIR_MASK; 2131 2132 /* 2133 * keep state, bit flip the direction and strip read. 2134 */ 2135 return state | (dir ^ LOCK_USAGE_DIR_MASK); 2136 } 2137 2138 /* 2139 * Observe that when given a bitmask where each bitnr is encoded as above, a 2140 * right shift of the mask transforms the individual bitnrs as -1 and 2141 * conversely, a left shift transforms into +1 for the individual bitnrs. 2142 * 2143 * So for all bits whose number have LOCK_ENABLED_* set (bitnr1 == 1), we can 2144 * create the mask with those bit numbers using LOCK_USED_IN_* (bitnr1 == 0) 2145 * instead by subtracting the bit number by 2, or shifting the mask right by 2. 2146 * 2147 * Similarly, bitnr1 == 0 becomes bitnr1 == 1 by adding 2, or shifting left 2. 2148 * 2149 * So split the mask (note that LOCKF_ENABLED_IRQ_ALL|LOCKF_USED_IN_IRQ_ALL is 2150 * all bits set) and recompose with bitnr1 flipped. 2151 */ 2152 static unsigned long invert_dir_mask(unsigned long mask) 2153 { 2154 unsigned long excl = 0; 2155 2156 /* Invert dir */ 2157 excl |= (mask & LOCKF_ENABLED_IRQ_ALL) >> LOCK_USAGE_DIR_MASK; 2158 excl |= (mask & LOCKF_USED_IN_IRQ_ALL) << LOCK_USAGE_DIR_MASK; 2159 2160 return excl; 2161 } 2162 2163 /* 2164 * As above, we clear bitnr0 (LOCK_*_READ off) with bitmask ops. First, for all 2165 * bits with bitnr0 set (LOCK_*_READ), add those with bitnr0 cleared (LOCK_*). 2166 * And then mask out all bitnr0. 2167 */ 2168 static unsigned long exclusive_mask(unsigned long mask) 2169 { 2170 unsigned long excl = invert_dir_mask(mask); 2171 2172 /* Strip read */ 2173 excl |= (excl & LOCKF_IRQ_READ) >> LOCK_USAGE_READ_MASK; 2174 excl &= ~LOCKF_IRQ_READ; 2175 2176 return excl; 2177 } 2178 2179 /* 2180 * Retrieve the _possible_ original mask to which @mask is 2181 * exclusive. Ie: this is the opposite of exclusive_mask(). 2182 * Note that 2 possible original bits can match an exclusive 2183 * bit: one has LOCK_USAGE_READ_MASK set, the other has it 2184 * cleared. So both are returned for each exclusive bit. 2185 */ 2186 static unsigned long original_mask(unsigned long mask) 2187 { 2188 unsigned long excl = invert_dir_mask(mask); 2189 2190 /* Include read in existing usages */ 2191 excl |= (excl & LOCKF_IRQ) << LOCK_USAGE_READ_MASK; 2192 2193 return excl; 2194 } 2195 2196 /* 2197 * Find the first pair of bit match between an original 2198 * usage mask and an exclusive usage mask. 2199 */ 2200 static int find_exclusive_match(unsigned long mask, 2201 unsigned long excl_mask, 2202 enum lock_usage_bit *bitp, 2203 enum lock_usage_bit *excl_bitp) 2204 { 2205 int bit, excl; 2206 2207 for_each_set_bit(bit, &mask, LOCK_USED) { 2208 excl = exclusive_bit(bit); 2209 if (excl_mask & lock_flag(excl)) { 2210 *bitp = bit; 2211 *excl_bitp = excl; 2212 return 0; 2213 } 2214 } 2215 return -1; 2216 } 2217 2218 /* 2219 * Prove that the new dependency does not connect a hardirq-safe(-read) 2220 * lock with a hardirq-unsafe lock - to achieve this we search 2221 * the backwards-subgraph starting at <prev>, and the 2222 * forwards-subgraph starting at <next>: 2223 */ 2224 static int check_irq_usage(struct task_struct *curr, struct held_lock *prev, 2225 struct held_lock *next) 2226 { 2227 unsigned long usage_mask = 0, forward_mask, backward_mask; 2228 enum lock_usage_bit forward_bit = 0, backward_bit = 0; 2229 struct lock_list *uninitialized_var(target_entry1); 2230 struct lock_list *uninitialized_var(target_entry); 2231 struct lock_list this, that; 2232 int ret; 2233 2234 /* 2235 * Step 1: gather all hard/soft IRQs usages backward in an 2236 * accumulated usage mask. 2237 */ 2238 this.parent = NULL; 2239 this.class = hlock_class(prev); 2240 2241 ret = __bfs_backwards(&this, &usage_mask, usage_accumulate, NULL); 2242 if (ret < 0) { 2243 print_bfs_bug(ret); 2244 return 0; 2245 } 2246 2247 usage_mask &= LOCKF_USED_IN_IRQ_ALL; 2248 if (!usage_mask) 2249 return 1; 2250 2251 /* 2252 * Step 2: find exclusive uses forward that match the previous 2253 * backward accumulated mask. 2254 */ 2255 forward_mask = exclusive_mask(usage_mask); 2256 2257 that.parent = NULL; 2258 that.class = hlock_class(next); 2259 2260 ret = find_usage_forwards(&that, forward_mask, &target_entry1); 2261 if (ret < 0) { 2262 print_bfs_bug(ret); 2263 return 0; 2264 } 2265 if (ret == 1) 2266 return ret; 2267 2268 /* 2269 * Step 3: we found a bad match! Now retrieve a lock from the backward 2270 * list whose usage mask matches the exclusive usage mask from the 2271 * lock found on the forward list. 2272 */ 2273 backward_mask = original_mask(target_entry1->class->usage_mask); 2274 2275 ret = find_usage_backwards(&this, backward_mask, &target_entry); 2276 if (ret < 0) { 2277 print_bfs_bug(ret); 2278 return 0; 2279 } 2280 if (DEBUG_LOCKS_WARN_ON(ret == 1)) 2281 return 1; 2282 2283 /* 2284 * Step 4: narrow down to a pair of incompatible usage bits 2285 * and report it. 2286 */ 2287 ret = find_exclusive_match(target_entry->class->usage_mask, 2288 target_entry1->class->usage_mask, 2289 &backward_bit, &forward_bit); 2290 if (DEBUG_LOCKS_WARN_ON(ret == -1)) 2291 return 1; 2292 2293 print_bad_irq_dependency(curr, &this, &that, 2294 target_entry, target_entry1, 2295 prev, next, 2296 backward_bit, forward_bit, 2297 state_name(backward_bit)); 2298 2299 return 0; 2300 } 2301 2302 static void inc_chains(void) 2303 { 2304 if (current->hardirq_context) 2305 nr_hardirq_chains++; 2306 else { 2307 if (current->softirq_context) 2308 nr_softirq_chains++; 2309 else 2310 nr_process_chains++; 2311 } 2312 } 2313 2314 #else 2315 2316 static inline int check_irq_usage(struct task_struct *curr, 2317 struct held_lock *prev, struct held_lock *next) 2318 { 2319 return 1; 2320 } 2321 2322 static inline void inc_chains(void) 2323 { 2324 nr_process_chains++; 2325 } 2326 2327 #endif /* CONFIG_TRACE_IRQFLAGS */ 2328 2329 static void 2330 print_deadlock_scenario(struct held_lock *nxt, struct held_lock *prv) 2331 { 2332 struct lock_class *next = hlock_class(nxt); 2333 struct lock_class *prev = hlock_class(prv); 2334 2335 printk(" Possible unsafe locking scenario:\n\n"); 2336 printk(" CPU0\n"); 2337 printk(" ----\n"); 2338 printk(" lock("); 2339 __print_lock_name(prev); 2340 printk(KERN_CONT ");\n"); 2341 printk(" lock("); 2342 __print_lock_name(next); 2343 printk(KERN_CONT ");\n"); 2344 printk("\n *** DEADLOCK ***\n\n"); 2345 printk(" May be due to missing lock nesting notation\n\n"); 2346 } 2347 2348 static void 2349 print_deadlock_bug(struct task_struct *curr, struct held_lock *prev, 2350 struct held_lock *next) 2351 { 2352 if (!debug_locks_off_graph_unlock() || debug_locks_silent) 2353 return; 2354 2355 pr_warn("\n"); 2356 pr_warn("============================================\n"); 2357 pr_warn("WARNING: possible recursive locking detected\n"); 2358 print_kernel_ident(); 2359 pr_warn("--------------------------------------------\n"); 2360 pr_warn("%s/%d is trying to acquire lock:\n", 2361 curr->comm, task_pid_nr(curr)); 2362 print_lock(next); 2363 pr_warn("\nbut task is already holding lock:\n"); 2364 print_lock(prev); 2365 2366 pr_warn("\nother info that might help us debug this:\n"); 2367 print_deadlock_scenario(next, prev); 2368 lockdep_print_held_locks(curr); 2369 2370 pr_warn("\nstack backtrace:\n"); 2371 dump_stack(); 2372 } 2373 2374 /* 2375 * Check whether we are holding such a class already. 2376 * 2377 * (Note that this has to be done separately, because the graph cannot 2378 * detect such classes of deadlocks.) 2379 * 2380 * Returns: 0 on deadlock detected, 1 on OK, 2 on recursive read 2381 */ 2382 static int 2383 check_deadlock(struct task_struct *curr, struct held_lock *next) 2384 { 2385 struct held_lock *prev; 2386 struct held_lock *nest = NULL; 2387 int i; 2388 2389 for (i = 0; i < curr->lockdep_depth; i++) { 2390 prev = curr->held_locks + i; 2391 2392 if (prev->instance == next->nest_lock) 2393 nest = prev; 2394 2395 if (hlock_class(prev) != hlock_class(next)) 2396 continue; 2397 2398 /* 2399 * Allow read-after-read recursion of the same 2400 * lock class (i.e. read_lock(lock)+read_lock(lock)): 2401 */ 2402 if ((next->read == 2) && prev->read) 2403 return 2; 2404 2405 /* 2406 * We're holding the nest_lock, which serializes this lock's 2407 * nesting behaviour. 2408 */ 2409 if (nest) 2410 return 2; 2411 2412 print_deadlock_bug(curr, prev, next); 2413 return 0; 2414 } 2415 return 1; 2416 } 2417 2418 /* 2419 * There was a chain-cache miss, and we are about to add a new dependency 2420 * to a previous lock. We validate the following rules: 2421 * 2422 * - would the adding of the <prev> -> <next> dependency create a 2423 * circular dependency in the graph? [== circular deadlock] 2424 * 2425 * - does the new prev->next dependency connect any hardirq-safe lock 2426 * (in the full backwards-subgraph starting at <prev>) with any 2427 * hardirq-unsafe lock (in the full forwards-subgraph starting at 2428 * <next>)? [== illegal lock inversion with hardirq contexts] 2429 * 2430 * - does the new prev->next dependency connect any softirq-safe lock 2431 * (in the full backwards-subgraph starting at <prev>) with any 2432 * softirq-unsafe lock (in the full forwards-subgraph starting at 2433 * <next>)? [== illegal lock inversion with softirq contexts] 2434 * 2435 * any of these scenarios could lead to a deadlock. 2436 * 2437 * Then if all the validations pass, we add the forwards and backwards 2438 * dependency. 2439 */ 2440 static int 2441 check_prev_add(struct task_struct *curr, struct held_lock *prev, 2442 struct held_lock *next, int distance, 2443 struct lock_trace **const trace) 2444 { 2445 struct lock_list *entry; 2446 int ret; 2447 2448 if (!hlock_class(prev)->key || !hlock_class(next)->key) { 2449 /* 2450 * The warning statements below may trigger a use-after-free 2451 * of the class name. It is better to trigger a use-after free 2452 * and to have the class name most of the time instead of not 2453 * having the class name available. 2454 */ 2455 WARN_ONCE(!debug_locks_silent && !hlock_class(prev)->key, 2456 "Detected use-after-free of lock class %px/%s\n", 2457 hlock_class(prev), 2458 hlock_class(prev)->name); 2459 WARN_ONCE(!debug_locks_silent && !hlock_class(next)->key, 2460 "Detected use-after-free of lock class %px/%s\n", 2461 hlock_class(next), 2462 hlock_class(next)->name); 2463 return 2; 2464 } 2465 2466 /* 2467 * Prove that the new <prev> -> <next> dependency would not 2468 * create a circular dependency in the graph. (We do this by 2469 * a breadth-first search into the graph starting at <next>, 2470 * and check whether we can reach <prev>.) 2471 * 2472 * The search is limited by the size of the circular queue (i.e., 2473 * MAX_CIRCULAR_QUEUE_SIZE) which keeps track of a breadth of nodes 2474 * in the graph whose neighbours are to be checked. 2475 */ 2476 ret = check_noncircular(next, prev, trace); 2477 if (unlikely(ret <= 0)) 2478 return 0; 2479 2480 if (!check_irq_usage(curr, prev, next)) 2481 return 0; 2482 2483 /* 2484 * For recursive read-locks we do all the dependency checks, 2485 * but we dont store read-triggered dependencies (only 2486 * write-triggered dependencies). This ensures that only the 2487 * write-side dependencies matter, and that if for example a 2488 * write-lock never takes any other locks, then the reads are 2489 * equivalent to a NOP. 2490 */ 2491 if (next->read == 2 || prev->read == 2) 2492 return 1; 2493 /* 2494 * Is the <prev> -> <next> dependency already present? 2495 * 2496 * (this may occur even though this is a new chain: consider 2497 * e.g. the L1 -> L2 -> L3 -> L4 and the L5 -> L1 -> L2 -> L3 2498 * chains - the second one will be new, but L1 already has 2499 * L2 added to its dependency list, due to the first chain.) 2500 */ 2501 list_for_each_entry(entry, &hlock_class(prev)->locks_after, entry) { 2502 if (entry->class == hlock_class(next)) { 2503 if (distance == 1) 2504 entry->distance = 1; 2505 return 1; 2506 } 2507 } 2508 2509 #ifdef CONFIG_LOCKDEP_SMALL 2510 /* 2511 * Is the <prev> -> <next> link redundant? 2512 */ 2513 ret = check_redundant(prev, next); 2514 if (ret != 1) 2515 return ret; 2516 #endif 2517 2518 if (!*trace) { 2519 *trace = save_trace(); 2520 if (!*trace) 2521 return 0; 2522 } 2523 2524 /* 2525 * Ok, all validations passed, add the new lock 2526 * to the previous lock's dependency list: 2527 */ 2528 ret = add_lock_to_list(hlock_class(next), hlock_class(prev), 2529 &hlock_class(prev)->locks_after, 2530 next->acquire_ip, distance, *trace); 2531 2532 if (!ret) 2533 return 0; 2534 2535 ret = add_lock_to_list(hlock_class(prev), hlock_class(next), 2536 &hlock_class(next)->locks_before, 2537 next->acquire_ip, distance, *trace); 2538 if (!ret) 2539 return 0; 2540 2541 return 2; 2542 } 2543 2544 /* 2545 * Add the dependency to all directly-previous locks that are 'relevant'. 2546 * The ones that are relevant are (in increasing distance from curr): 2547 * all consecutive trylock entries and the final non-trylock entry - or 2548 * the end of this context's lock-chain - whichever comes first. 2549 */ 2550 static int 2551 check_prevs_add(struct task_struct *curr, struct held_lock *next) 2552 { 2553 struct lock_trace *trace = NULL; 2554 int depth = curr->lockdep_depth; 2555 struct held_lock *hlock; 2556 2557 /* 2558 * Debugging checks. 2559 * 2560 * Depth must not be zero for a non-head lock: 2561 */ 2562 if (!depth) 2563 goto out_bug; 2564 /* 2565 * At least two relevant locks must exist for this 2566 * to be a head: 2567 */ 2568 if (curr->held_locks[depth].irq_context != 2569 curr->held_locks[depth-1].irq_context) 2570 goto out_bug; 2571 2572 for (;;) { 2573 int distance = curr->lockdep_depth - depth + 1; 2574 hlock = curr->held_locks + depth - 1; 2575 2576 /* 2577 * Only non-recursive-read entries get new dependencies 2578 * added: 2579 */ 2580 if (hlock->read != 2 && hlock->check) { 2581 int ret = check_prev_add(curr, hlock, next, distance, 2582 &trace); 2583 if (!ret) 2584 return 0; 2585 2586 /* 2587 * Stop after the first non-trylock entry, 2588 * as non-trylock entries have added their 2589 * own direct dependencies already, so this 2590 * lock is connected to them indirectly: 2591 */ 2592 if (!hlock->trylock) 2593 break; 2594 } 2595 2596 depth--; 2597 /* 2598 * End of lock-stack? 2599 */ 2600 if (!depth) 2601 break; 2602 /* 2603 * Stop the search if we cross into another context: 2604 */ 2605 if (curr->held_locks[depth].irq_context != 2606 curr->held_locks[depth-1].irq_context) 2607 break; 2608 } 2609 return 1; 2610 out_bug: 2611 if (!debug_locks_off_graph_unlock()) 2612 return 0; 2613 2614 /* 2615 * Clearly we all shouldn't be here, but since we made it we 2616 * can reliable say we messed up our state. See the above two 2617 * gotos for reasons why we could possibly end up here. 2618 */ 2619 WARN_ON(1); 2620 2621 return 0; 2622 } 2623 2624 struct lock_chain lock_chains[MAX_LOCKDEP_CHAINS]; 2625 static DECLARE_BITMAP(lock_chains_in_use, MAX_LOCKDEP_CHAINS); 2626 int nr_chain_hlocks; 2627 static u16 chain_hlocks[MAX_LOCKDEP_CHAIN_HLOCKS]; 2628 2629 struct lock_class *lock_chain_get_class(struct lock_chain *chain, int i) 2630 { 2631 return lock_classes + chain_hlocks[chain->base + i]; 2632 } 2633 2634 /* 2635 * Returns the index of the first held_lock of the current chain 2636 */ 2637 static inline int get_first_held_lock(struct task_struct *curr, 2638 struct held_lock *hlock) 2639 { 2640 int i; 2641 struct held_lock *hlock_curr; 2642 2643 for (i = curr->lockdep_depth - 1; i >= 0; i--) { 2644 hlock_curr = curr->held_locks + i; 2645 if (hlock_curr->irq_context != hlock->irq_context) 2646 break; 2647 2648 } 2649 2650 return ++i; 2651 } 2652 2653 #ifdef CONFIG_DEBUG_LOCKDEP 2654 /* 2655 * Returns the next chain_key iteration 2656 */ 2657 static u64 print_chain_key_iteration(int class_idx, u64 chain_key) 2658 { 2659 u64 new_chain_key = iterate_chain_key(chain_key, class_idx); 2660 2661 printk(" class_idx:%d -> chain_key:%016Lx", 2662 class_idx, 2663 (unsigned long long)new_chain_key); 2664 return new_chain_key; 2665 } 2666 2667 static void 2668 print_chain_keys_held_locks(struct task_struct *curr, struct held_lock *hlock_next) 2669 { 2670 struct held_lock *hlock; 2671 u64 chain_key = INITIAL_CHAIN_KEY; 2672 int depth = curr->lockdep_depth; 2673 int i = get_first_held_lock(curr, hlock_next); 2674 2675 printk("depth: %u (irq_context %u)\n", depth - i + 1, 2676 hlock_next->irq_context); 2677 for (; i < depth; i++) { 2678 hlock = curr->held_locks + i; 2679 chain_key = print_chain_key_iteration(hlock->class_idx, chain_key); 2680 2681 print_lock(hlock); 2682 } 2683 2684 print_chain_key_iteration(hlock_next->class_idx, chain_key); 2685 print_lock(hlock_next); 2686 } 2687 2688 static void print_chain_keys_chain(struct lock_chain *chain) 2689 { 2690 int i; 2691 u64 chain_key = INITIAL_CHAIN_KEY; 2692 int class_id; 2693 2694 printk("depth: %u\n", chain->depth); 2695 for (i = 0; i < chain->depth; i++) { 2696 class_id = chain_hlocks[chain->base + i]; 2697 chain_key = print_chain_key_iteration(class_id, chain_key); 2698 2699 print_lock_name(lock_classes + class_id); 2700 printk("\n"); 2701 } 2702 } 2703 2704 static void print_collision(struct task_struct *curr, 2705 struct held_lock *hlock_next, 2706 struct lock_chain *chain) 2707 { 2708 pr_warn("\n"); 2709 pr_warn("============================\n"); 2710 pr_warn("WARNING: chain_key collision\n"); 2711 print_kernel_ident(); 2712 pr_warn("----------------------------\n"); 2713 pr_warn("%s/%d: ", current->comm, task_pid_nr(current)); 2714 pr_warn("Hash chain already cached but the contents don't match!\n"); 2715 2716 pr_warn("Held locks:"); 2717 print_chain_keys_held_locks(curr, hlock_next); 2718 2719 pr_warn("Locks in cached chain:"); 2720 print_chain_keys_chain(chain); 2721 2722 pr_warn("\nstack backtrace:\n"); 2723 dump_stack(); 2724 } 2725 #endif 2726 2727 /* 2728 * Checks whether the chain and the current held locks are consistent 2729 * in depth and also in content. If they are not it most likely means 2730 * that there was a collision during the calculation of the chain_key. 2731 * Returns: 0 not passed, 1 passed 2732 */ 2733 static int check_no_collision(struct task_struct *curr, 2734 struct held_lock *hlock, 2735 struct lock_chain *chain) 2736 { 2737 #ifdef CONFIG_DEBUG_LOCKDEP 2738 int i, j, id; 2739 2740 i = get_first_held_lock(curr, hlock); 2741 2742 if (DEBUG_LOCKS_WARN_ON(chain->depth != curr->lockdep_depth - (i - 1))) { 2743 print_collision(curr, hlock, chain); 2744 return 0; 2745 } 2746 2747 for (j = 0; j < chain->depth - 1; j++, i++) { 2748 id = curr->held_locks[i].class_idx; 2749 2750 if (DEBUG_LOCKS_WARN_ON(chain_hlocks[chain->base + j] != id)) { 2751 print_collision(curr, hlock, chain); 2752 return 0; 2753 } 2754 } 2755 #endif 2756 return 1; 2757 } 2758 2759 /* 2760 * Given an index that is >= -1, return the index of the next lock chain. 2761 * Return -2 if there is no next lock chain. 2762 */ 2763 long lockdep_next_lockchain(long i) 2764 { 2765 i = find_next_bit(lock_chains_in_use, ARRAY_SIZE(lock_chains), i + 1); 2766 return i < ARRAY_SIZE(lock_chains) ? i : -2; 2767 } 2768 2769 unsigned long lock_chain_count(void) 2770 { 2771 return bitmap_weight(lock_chains_in_use, ARRAY_SIZE(lock_chains)); 2772 } 2773 2774 /* Must be called with the graph lock held. */ 2775 static struct lock_chain *alloc_lock_chain(void) 2776 { 2777 int idx = find_first_zero_bit(lock_chains_in_use, 2778 ARRAY_SIZE(lock_chains)); 2779 2780 if (unlikely(idx >= ARRAY_SIZE(lock_chains))) 2781 return NULL; 2782 __set_bit(idx, lock_chains_in_use); 2783 return lock_chains + idx; 2784 } 2785 2786 /* 2787 * Adds a dependency chain into chain hashtable. And must be called with 2788 * graph_lock held. 2789 * 2790 * Return 0 if fail, and graph_lock is released. 2791 * Return 1 if succeed, with graph_lock held. 2792 */ 2793 static inline int add_chain_cache(struct task_struct *curr, 2794 struct held_lock *hlock, 2795 u64 chain_key) 2796 { 2797 struct lock_class *class = hlock_class(hlock); 2798 struct hlist_head *hash_head = chainhashentry(chain_key); 2799 struct lock_chain *chain; 2800 int i, j; 2801 2802 /* 2803 * The caller must hold the graph lock, ensure we've got IRQs 2804 * disabled to make this an IRQ-safe lock.. for recursion reasons 2805 * lockdep won't complain about its own locking errors. 2806 */ 2807 if (DEBUG_LOCKS_WARN_ON(!irqs_disabled())) 2808 return 0; 2809 2810 chain = alloc_lock_chain(); 2811 if (!chain) { 2812 if (!debug_locks_off_graph_unlock()) 2813 return 0; 2814 2815 print_lockdep_off("BUG: MAX_LOCKDEP_CHAINS too low!"); 2816 dump_stack(); 2817 return 0; 2818 } 2819 chain->chain_key = chain_key; 2820 chain->irq_context = hlock->irq_context; 2821 i = get_first_held_lock(curr, hlock); 2822 chain->depth = curr->lockdep_depth + 1 - i; 2823 2824 BUILD_BUG_ON((1UL << 24) <= ARRAY_SIZE(chain_hlocks)); 2825 BUILD_BUG_ON((1UL << 6) <= ARRAY_SIZE(curr->held_locks)); 2826 BUILD_BUG_ON((1UL << 8*sizeof(chain_hlocks[0])) <= ARRAY_SIZE(lock_classes)); 2827 2828 if (likely(nr_chain_hlocks + chain->depth <= MAX_LOCKDEP_CHAIN_HLOCKS)) { 2829 chain->base = nr_chain_hlocks; 2830 for (j = 0; j < chain->depth - 1; j++, i++) { 2831 int lock_id = curr->held_locks[i].class_idx; 2832 chain_hlocks[chain->base + j] = lock_id; 2833 } 2834 chain_hlocks[chain->base + j] = class - lock_classes; 2835 nr_chain_hlocks += chain->depth; 2836 } else { 2837 if (!debug_locks_off_graph_unlock()) 2838 return 0; 2839 2840 print_lockdep_off("BUG: MAX_LOCKDEP_CHAIN_HLOCKS too low!"); 2841 dump_stack(); 2842 return 0; 2843 } 2844 2845 hlist_add_head_rcu(&chain->entry, hash_head); 2846 debug_atomic_inc(chain_lookup_misses); 2847 inc_chains(); 2848 2849 return 1; 2850 } 2851 2852 /* 2853 * Look up a dependency chain. Must be called with either the graph lock or 2854 * the RCU read lock held. 2855 */ 2856 static inline struct lock_chain *lookup_chain_cache(u64 chain_key) 2857 { 2858 struct hlist_head *hash_head = chainhashentry(chain_key); 2859 struct lock_chain *chain; 2860 2861 hlist_for_each_entry_rcu(chain, hash_head, entry) { 2862 if (READ_ONCE(chain->chain_key) == chain_key) { 2863 debug_atomic_inc(chain_lookup_hits); 2864 return chain; 2865 } 2866 } 2867 return NULL; 2868 } 2869 2870 /* 2871 * If the key is not present yet in dependency chain cache then 2872 * add it and return 1 - in this case the new dependency chain is 2873 * validated. If the key is already hashed, return 0. 2874 * (On return with 1 graph_lock is held.) 2875 */ 2876 static inline int lookup_chain_cache_add(struct task_struct *curr, 2877 struct held_lock *hlock, 2878 u64 chain_key) 2879 { 2880 struct lock_class *class = hlock_class(hlock); 2881 struct lock_chain *chain = lookup_chain_cache(chain_key); 2882 2883 if (chain) { 2884 cache_hit: 2885 if (!check_no_collision(curr, hlock, chain)) 2886 return 0; 2887 2888 if (very_verbose(class)) { 2889 printk("\nhash chain already cached, key: " 2890 "%016Lx tail class: [%px] %s\n", 2891 (unsigned long long)chain_key, 2892 class->key, class->name); 2893 } 2894 2895 return 0; 2896 } 2897 2898 if (very_verbose(class)) { 2899 printk("\nnew hash chain, key: %016Lx tail class: [%px] %s\n", 2900 (unsigned long long)chain_key, class->key, class->name); 2901 } 2902 2903 if (!graph_lock()) 2904 return 0; 2905 2906 /* 2907 * We have to walk the chain again locked - to avoid duplicates: 2908 */ 2909 chain = lookup_chain_cache(chain_key); 2910 if (chain) { 2911 graph_unlock(); 2912 goto cache_hit; 2913 } 2914 2915 if (!add_chain_cache(curr, hlock, chain_key)) 2916 return 0; 2917 2918 return 1; 2919 } 2920 2921 static int validate_chain(struct task_struct *curr, 2922 struct held_lock *hlock, 2923 int chain_head, u64 chain_key) 2924 { 2925 /* 2926 * Trylock needs to maintain the stack of held locks, but it 2927 * does not add new dependencies, because trylock can be done 2928 * in any order. 2929 * 2930 * We look up the chain_key and do the O(N^2) check and update of 2931 * the dependencies only if this is a new dependency chain. 2932 * (If lookup_chain_cache_add() return with 1 it acquires 2933 * graph_lock for us) 2934 */ 2935 if (!hlock->trylock && hlock->check && 2936 lookup_chain_cache_add(curr, hlock, chain_key)) { 2937 /* 2938 * Check whether last held lock: 2939 * 2940 * - is irq-safe, if this lock is irq-unsafe 2941 * - is softirq-safe, if this lock is hardirq-unsafe 2942 * 2943 * And check whether the new lock's dependency graph 2944 * could lead back to the previous lock: 2945 * 2946 * - within the current held-lock stack 2947 * - across our accumulated lock dependency records 2948 * 2949 * any of these scenarios could lead to a deadlock. 2950 */ 2951 /* 2952 * The simple case: does the current hold the same lock 2953 * already? 2954 */ 2955 int ret = check_deadlock(curr, hlock); 2956 2957 if (!ret) 2958 return 0; 2959 /* 2960 * Mark recursive read, as we jump over it when 2961 * building dependencies (just like we jump over 2962 * trylock entries): 2963 */ 2964 if (ret == 2) 2965 hlock->read = 2; 2966 /* 2967 * Add dependency only if this lock is not the head 2968 * of the chain, and if it's not a secondary read-lock: 2969 */ 2970 if (!chain_head && ret != 2) { 2971 if (!check_prevs_add(curr, hlock)) 2972 return 0; 2973 } 2974 2975 graph_unlock(); 2976 } else { 2977 /* after lookup_chain_cache_add(): */ 2978 if (unlikely(!debug_locks)) 2979 return 0; 2980 } 2981 2982 return 1; 2983 } 2984 #else 2985 static inline int validate_chain(struct task_struct *curr, 2986 struct held_lock *hlock, 2987 int chain_head, u64 chain_key) 2988 { 2989 return 1; 2990 } 2991 #endif /* CONFIG_PROVE_LOCKING */ 2992 2993 /* 2994 * We are building curr_chain_key incrementally, so double-check 2995 * it from scratch, to make sure that it's done correctly: 2996 */ 2997 static void check_chain_key(struct task_struct *curr) 2998 { 2999 #ifdef CONFIG_DEBUG_LOCKDEP 3000 struct held_lock *hlock, *prev_hlock = NULL; 3001 unsigned int i; 3002 u64 chain_key = INITIAL_CHAIN_KEY; 3003 3004 for (i = 0; i < curr->lockdep_depth; i++) { 3005 hlock = curr->held_locks + i; 3006 if (chain_key != hlock->prev_chain_key) { 3007 debug_locks_off(); 3008 /* 3009 * We got mighty confused, our chain keys don't match 3010 * with what we expect, someone trample on our task state? 3011 */ 3012 WARN(1, "hm#1, depth: %u [%u], %016Lx != %016Lx\n", 3013 curr->lockdep_depth, i, 3014 (unsigned long long)chain_key, 3015 (unsigned long long)hlock->prev_chain_key); 3016 return; 3017 } 3018 3019 /* 3020 * hlock->class_idx can't go beyond MAX_LOCKDEP_KEYS, but is 3021 * it registered lock class index? 3022 */ 3023 if (DEBUG_LOCKS_WARN_ON(!test_bit(hlock->class_idx, lock_classes_in_use))) 3024 return; 3025 3026 if (prev_hlock && (prev_hlock->irq_context != 3027 hlock->irq_context)) 3028 chain_key = INITIAL_CHAIN_KEY; 3029 chain_key = iterate_chain_key(chain_key, hlock->class_idx); 3030 prev_hlock = hlock; 3031 } 3032 if (chain_key != curr->curr_chain_key) { 3033 debug_locks_off(); 3034 /* 3035 * More smoking hash instead of calculating it, damn see these 3036 * numbers float.. I bet that a pink elephant stepped on my memory. 3037 */ 3038 WARN(1, "hm#2, depth: %u [%u], %016Lx != %016Lx\n", 3039 curr->lockdep_depth, i, 3040 (unsigned long long)chain_key, 3041 (unsigned long long)curr->curr_chain_key); 3042 } 3043 #endif 3044 } 3045 3046 #ifdef CONFIG_PROVE_LOCKING 3047 static int mark_lock(struct task_struct *curr, struct held_lock *this, 3048 enum lock_usage_bit new_bit); 3049 3050 static void print_usage_bug_scenario(struct held_lock *lock) 3051 { 3052 struct lock_class *class = hlock_class(lock); 3053 3054 printk(" Possible unsafe locking scenario:\n\n"); 3055 printk(" CPU0\n"); 3056 printk(" ----\n"); 3057 printk(" lock("); 3058 __print_lock_name(class); 3059 printk(KERN_CONT ");\n"); 3060 printk(" <Interrupt>\n"); 3061 printk(" lock("); 3062 __print_lock_name(class); 3063 printk(KERN_CONT ");\n"); 3064 printk("\n *** DEADLOCK ***\n\n"); 3065 } 3066 3067 static void 3068 print_usage_bug(struct task_struct *curr, struct held_lock *this, 3069 enum lock_usage_bit prev_bit, enum lock_usage_bit new_bit) 3070 { 3071 if (!debug_locks_off_graph_unlock() || debug_locks_silent) 3072 return; 3073 3074 pr_warn("\n"); 3075 pr_warn("================================\n"); 3076 pr_warn("WARNING: inconsistent lock state\n"); 3077 print_kernel_ident(); 3078 pr_warn("--------------------------------\n"); 3079 3080 pr_warn("inconsistent {%s} -> {%s} usage.\n", 3081 usage_str[prev_bit], usage_str[new_bit]); 3082 3083 pr_warn("%s/%d [HC%u[%lu]:SC%u[%lu]:HE%u:SE%u] takes:\n", 3084 curr->comm, task_pid_nr(curr), 3085 trace_hardirq_context(curr), hardirq_count() >> HARDIRQ_SHIFT, 3086 trace_softirq_context(curr), softirq_count() >> SOFTIRQ_SHIFT, 3087 trace_hardirqs_enabled(curr), 3088 trace_softirqs_enabled(curr)); 3089 print_lock(this); 3090 3091 pr_warn("{%s} state was registered at:\n", usage_str[prev_bit]); 3092 print_lock_trace(hlock_class(this)->usage_traces[prev_bit], 1); 3093 3094 print_irqtrace_events(curr); 3095 pr_warn("\nother info that might help us debug this:\n"); 3096 print_usage_bug_scenario(this); 3097 3098 lockdep_print_held_locks(curr); 3099 3100 pr_warn("\nstack backtrace:\n"); 3101 dump_stack(); 3102 } 3103 3104 /* 3105 * Print out an error if an invalid bit is set: 3106 */ 3107 static inline int 3108 valid_state(struct task_struct *curr, struct held_lock *this, 3109 enum lock_usage_bit new_bit, enum lock_usage_bit bad_bit) 3110 { 3111 if (unlikely(hlock_class(this)->usage_mask & (1 << bad_bit))) { 3112 print_usage_bug(curr, this, bad_bit, new_bit); 3113 return 0; 3114 } 3115 return 1; 3116 } 3117 3118 3119 /* 3120 * print irq inversion bug: 3121 */ 3122 static void 3123 print_irq_inversion_bug(struct task_struct *curr, 3124 struct lock_list *root, struct lock_list *other, 3125 struct held_lock *this, int forwards, 3126 const char *irqclass) 3127 { 3128 struct lock_list *entry = other; 3129 struct lock_list *middle = NULL; 3130 int depth; 3131 3132 if (!debug_locks_off_graph_unlock() || debug_locks_silent) 3133 return; 3134 3135 pr_warn("\n"); 3136 pr_warn("========================================================\n"); 3137 pr_warn("WARNING: possible irq lock inversion dependency detected\n"); 3138 print_kernel_ident(); 3139 pr_warn("--------------------------------------------------------\n"); 3140 pr_warn("%s/%d just changed the state of lock:\n", 3141 curr->comm, task_pid_nr(curr)); 3142 print_lock(this); 3143 if (forwards) 3144 pr_warn("but this lock took another, %s-unsafe lock in the past:\n", irqclass); 3145 else 3146 pr_warn("but this lock was taken by another, %s-safe lock in the past:\n", irqclass); 3147 print_lock_name(other->class); 3148 pr_warn("\n\nand interrupts could create inverse lock ordering between them.\n\n"); 3149 3150 pr_warn("\nother info that might help us debug this:\n"); 3151 3152 /* Find a middle lock (if one exists) */ 3153 depth = get_lock_depth(other); 3154 do { 3155 if (depth == 0 && (entry != root)) { 3156 pr_warn("lockdep:%s bad path found in chain graph\n", __func__); 3157 break; 3158 } 3159 middle = entry; 3160 entry = get_lock_parent(entry); 3161 depth--; 3162 } while (entry && entry != root && (depth >= 0)); 3163 if (forwards) 3164 print_irq_lock_scenario(root, other, 3165 middle ? middle->class : root->class, other->class); 3166 else 3167 print_irq_lock_scenario(other, root, 3168 middle ? middle->class : other->class, root->class); 3169 3170 lockdep_print_held_locks(curr); 3171 3172 pr_warn("\nthe shortest dependencies between 2nd lock and 1st lock:\n"); 3173 root->trace = save_trace(); 3174 if (!root->trace) 3175 return; 3176 print_shortest_lock_dependencies(other, root); 3177 3178 pr_warn("\nstack backtrace:\n"); 3179 dump_stack(); 3180 } 3181 3182 /* 3183 * Prove that in the forwards-direction subgraph starting at <this> 3184 * there is no lock matching <mask>: 3185 */ 3186 static int 3187 check_usage_forwards(struct task_struct *curr, struct held_lock *this, 3188 enum lock_usage_bit bit, const char *irqclass) 3189 { 3190 int ret; 3191 struct lock_list root; 3192 struct lock_list *uninitialized_var(target_entry); 3193 3194 root.parent = NULL; 3195 root.class = hlock_class(this); 3196 ret = find_usage_forwards(&root, lock_flag(bit), &target_entry); 3197 if (ret < 0) { 3198 print_bfs_bug(ret); 3199 return 0; 3200 } 3201 if (ret == 1) 3202 return ret; 3203 3204 print_irq_inversion_bug(curr, &root, target_entry, 3205 this, 1, irqclass); 3206 return 0; 3207 } 3208 3209 /* 3210 * Prove that in the backwards-direction subgraph starting at <this> 3211 * there is no lock matching <mask>: 3212 */ 3213 static int 3214 check_usage_backwards(struct task_struct *curr, struct held_lock *this, 3215 enum lock_usage_bit bit, const char *irqclass) 3216 { 3217 int ret; 3218 struct lock_list root; 3219 struct lock_list *uninitialized_var(target_entry); 3220 3221 root.parent = NULL; 3222 root.class = hlock_class(this); 3223 ret = find_usage_backwards(&root, lock_flag(bit), &target_entry); 3224 if (ret < 0) { 3225 print_bfs_bug(ret); 3226 return 0; 3227 } 3228 if (ret == 1) 3229 return ret; 3230 3231 print_irq_inversion_bug(curr, &root, target_entry, 3232 this, 0, irqclass); 3233 return 0; 3234 } 3235 3236 void print_irqtrace_events(struct task_struct *curr) 3237 { 3238 printk("irq event stamp: %u\n", curr->irq_events); 3239 printk("hardirqs last enabled at (%u): [<%px>] %pS\n", 3240 curr->hardirq_enable_event, (void *)curr->hardirq_enable_ip, 3241 (void *)curr->hardirq_enable_ip); 3242 printk("hardirqs last disabled at (%u): [<%px>] %pS\n", 3243 curr->hardirq_disable_event, (void *)curr->hardirq_disable_ip, 3244 (void *)curr->hardirq_disable_ip); 3245 printk("softirqs last enabled at (%u): [<%px>] %pS\n", 3246 curr->softirq_enable_event, (void *)curr->softirq_enable_ip, 3247 (void *)curr->softirq_enable_ip); 3248 printk("softirqs last disabled at (%u): [<%px>] %pS\n", 3249 curr->softirq_disable_event, (void *)curr->softirq_disable_ip, 3250 (void *)curr->softirq_disable_ip); 3251 } 3252 3253 static int HARDIRQ_verbose(struct lock_class *class) 3254 { 3255 #if HARDIRQ_VERBOSE 3256 return class_filter(class); 3257 #endif 3258 return 0; 3259 } 3260 3261 static int SOFTIRQ_verbose(struct lock_class *class) 3262 { 3263 #if SOFTIRQ_VERBOSE 3264 return class_filter(class); 3265 #endif 3266 return 0; 3267 } 3268 3269 #define STRICT_READ_CHECKS 1 3270 3271 static int (*state_verbose_f[])(struct lock_class *class) = { 3272 #define LOCKDEP_STATE(__STATE) \ 3273 __STATE##_verbose, 3274 #include "lockdep_states.h" 3275 #undef LOCKDEP_STATE 3276 }; 3277 3278 static inline int state_verbose(enum lock_usage_bit bit, 3279 struct lock_class *class) 3280 { 3281 return state_verbose_f[bit >> LOCK_USAGE_DIR_MASK](class); 3282 } 3283 3284 typedef int (*check_usage_f)(struct task_struct *, struct held_lock *, 3285 enum lock_usage_bit bit, const char *name); 3286 3287 static int 3288 mark_lock_irq(struct task_struct *curr, struct held_lock *this, 3289 enum lock_usage_bit new_bit) 3290 { 3291 int excl_bit = exclusive_bit(new_bit); 3292 int read = new_bit & LOCK_USAGE_READ_MASK; 3293 int dir = new_bit & LOCK_USAGE_DIR_MASK; 3294 3295 /* 3296 * mark USED_IN has to look forwards -- to ensure no dependency 3297 * has ENABLED state, which would allow recursion deadlocks. 3298 * 3299 * mark ENABLED has to look backwards -- to ensure no dependee 3300 * has USED_IN state, which, again, would allow recursion deadlocks. 3301 */ 3302 check_usage_f usage = dir ? 3303 check_usage_backwards : check_usage_forwards; 3304 3305 /* 3306 * Validate that this particular lock does not have conflicting 3307 * usage states. 3308 */ 3309 if (!valid_state(curr, this, new_bit, excl_bit)) 3310 return 0; 3311 3312 /* 3313 * Validate that the lock dependencies don't have conflicting usage 3314 * states. 3315 */ 3316 if ((!read || STRICT_READ_CHECKS) && 3317 !usage(curr, this, excl_bit, state_name(new_bit & ~LOCK_USAGE_READ_MASK))) 3318 return 0; 3319 3320 /* 3321 * Check for read in write conflicts 3322 */ 3323 if (!read) { 3324 if (!valid_state(curr, this, new_bit, excl_bit + LOCK_USAGE_READ_MASK)) 3325 return 0; 3326 3327 if (STRICT_READ_CHECKS && 3328 !usage(curr, this, excl_bit + LOCK_USAGE_READ_MASK, 3329 state_name(new_bit + LOCK_USAGE_READ_MASK))) 3330 return 0; 3331 } 3332 3333 if (state_verbose(new_bit, hlock_class(this))) 3334 return 2; 3335 3336 return 1; 3337 } 3338 3339 /* 3340 * Mark all held locks with a usage bit: 3341 */ 3342 static int 3343 mark_held_locks(struct task_struct *curr, enum lock_usage_bit base_bit) 3344 { 3345 struct held_lock *hlock; 3346 int i; 3347 3348 for (i = 0; i < curr->lockdep_depth; i++) { 3349 enum lock_usage_bit hlock_bit = base_bit; 3350 hlock = curr->held_locks + i; 3351 3352 if (hlock->read) 3353 hlock_bit += LOCK_USAGE_READ_MASK; 3354 3355 BUG_ON(hlock_bit >= LOCK_USAGE_STATES); 3356 3357 if (!hlock->check) 3358 continue; 3359 3360 if (!mark_lock(curr, hlock, hlock_bit)) 3361 return 0; 3362 } 3363 3364 return 1; 3365 } 3366 3367 /* 3368 * Hardirqs will be enabled: 3369 */ 3370 static void __trace_hardirqs_on_caller(unsigned long ip) 3371 { 3372 struct task_struct *curr = current; 3373 3374 /* we'll do an OFF -> ON transition: */ 3375 curr->hardirqs_enabled = 1; 3376 3377 /* 3378 * We are going to turn hardirqs on, so set the 3379 * usage bit for all held locks: 3380 */ 3381 if (!mark_held_locks(curr, LOCK_ENABLED_HARDIRQ)) 3382 return; 3383 /* 3384 * If we have softirqs enabled, then set the usage 3385 * bit for all held locks. (disabled hardirqs prevented 3386 * this bit from being set before) 3387 */ 3388 if (curr->softirqs_enabled) 3389 if (!mark_held_locks(curr, LOCK_ENABLED_SOFTIRQ)) 3390 return; 3391 3392 curr->hardirq_enable_ip = ip; 3393 curr->hardirq_enable_event = ++curr->irq_events; 3394 debug_atomic_inc(hardirqs_on_events); 3395 } 3396 3397 void lockdep_hardirqs_on(unsigned long ip) 3398 { 3399 if (unlikely(!debug_locks || current->lockdep_recursion)) 3400 return; 3401 3402 if (unlikely(current->hardirqs_enabled)) { 3403 /* 3404 * Neither irq nor preemption are disabled here 3405 * so this is racy by nature but losing one hit 3406 * in a stat is not a big deal. 3407 */ 3408 __debug_atomic_inc(redundant_hardirqs_on); 3409 return; 3410 } 3411 3412 /* 3413 * We're enabling irqs and according to our state above irqs weren't 3414 * already enabled, yet we find the hardware thinks they are in fact 3415 * enabled.. someone messed up their IRQ state tracing. 3416 */ 3417 if (DEBUG_LOCKS_WARN_ON(!irqs_disabled())) 3418 return; 3419 3420 /* 3421 * See the fine text that goes along with this variable definition. 3422 */ 3423 if (DEBUG_LOCKS_WARN_ON(early_boot_irqs_disabled)) 3424 return; 3425 3426 /* 3427 * Can't allow enabling interrupts while in an interrupt handler, 3428 * that's general bad form and such. Recursion, limited stack etc.. 3429 */ 3430 if (DEBUG_LOCKS_WARN_ON(current->hardirq_context)) 3431 return; 3432 3433 current->lockdep_recursion = 1; 3434 __trace_hardirqs_on_caller(ip); 3435 current->lockdep_recursion = 0; 3436 } 3437 NOKPROBE_SYMBOL(lockdep_hardirqs_on); 3438 3439 /* 3440 * Hardirqs were disabled: 3441 */ 3442 void lockdep_hardirqs_off(unsigned long ip) 3443 { 3444 struct task_struct *curr = current; 3445 3446 if (unlikely(!debug_locks || current->lockdep_recursion)) 3447 return; 3448 3449 /* 3450 * So we're supposed to get called after you mask local IRQs, but for 3451 * some reason the hardware doesn't quite think you did a proper job. 3452 */ 3453 if (DEBUG_LOCKS_WARN_ON(!irqs_disabled())) 3454 return; 3455 3456 if (curr->hardirqs_enabled) { 3457 /* 3458 * We have done an ON -> OFF transition: 3459 */ 3460 curr->hardirqs_enabled = 0; 3461 curr->hardirq_disable_ip = ip; 3462 curr->hardirq_disable_event = ++curr->irq_events; 3463 debug_atomic_inc(hardirqs_off_events); 3464 } else 3465 debug_atomic_inc(redundant_hardirqs_off); 3466 } 3467 NOKPROBE_SYMBOL(lockdep_hardirqs_off); 3468 3469 /* 3470 * Softirqs will be enabled: 3471 */ 3472 void trace_softirqs_on(unsigned long ip) 3473 { 3474 struct task_struct *curr = current; 3475 3476 if (unlikely(!debug_locks || current->lockdep_recursion)) 3477 return; 3478 3479 /* 3480 * We fancy IRQs being disabled here, see softirq.c, avoids 3481 * funny state and nesting things. 3482 */ 3483 if (DEBUG_LOCKS_WARN_ON(!irqs_disabled())) 3484 return; 3485 3486 if (curr->softirqs_enabled) { 3487 debug_atomic_inc(redundant_softirqs_on); 3488 return; 3489 } 3490 3491 current->lockdep_recursion = 1; 3492 /* 3493 * We'll do an OFF -> ON transition: 3494 */ 3495 curr->softirqs_enabled = 1; 3496 curr->softirq_enable_ip = ip; 3497 curr->softirq_enable_event = ++curr->irq_events; 3498 debug_atomic_inc(softirqs_on_events); 3499 /* 3500 * We are going to turn softirqs on, so set the 3501 * usage bit for all held locks, if hardirqs are 3502 * enabled too: 3503 */ 3504 if (curr->hardirqs_enabled) 3505 mark_held_locks(curr, LOCK_ENABLED_SOFTIRQ); 3506 current->lockdep_recursion = 0; 3507 } 3508 3509 /* 3510 * Softirqs were disabled: 3511 */ 3512 void trace_softirqs_off(unsigned long ip) 3513 { 3514 struct task_struct *curr = current; 3515 3516 if (unlikely(!debug_locks || current->lockdep_recursion)) 3517 return; 3518 3519 /* 3520 * We fancy IRQs being disabled here, see softirq.c 3521 */ 3522 if (DEBUG_LOCKS_WARN_ON(!irqs_disabled())) 3523 return; 3524 3525 if (curr->softirqs_enabled) { 3526 /* 3527 * We have done an ON -> OFF transition: 3528 */ 3529 curr->softirqs_enabled = 0; 3530 curr->softirq_disable_ip = ip; 3531 curr->softirq_disable_event = ++curr->irq_events; 3532 debug_atomic_inc(softirqs_off_events); 3533 /* 3534 * Whoops, we wanted softirqs off, so why aren't they? 3535 */ 3536 DEBUG_LOCKS_WARN_ON(!softirq_count()); 3537 } else 3538 debug_atomic_inc(redundant_softirqs_off); 3539 } 3540 3541 static int 3542 mark_usage(struct task_struct *curr, struct held_lock *hlock, int check) 3543 { 3544 if (!check) 3545 goto lock_used; 3546 3547 /* 3548 * If non-trylock use in a hardirq or softirq context, then 3549 * mark the lock as used in these contexts: 3550 */ 3551 if (!hlock->trylock) { 3552 if (hlock->read) { 3553 if (curr->hardirq_context) 3554 if (!mark_lock(curr, hlock, 3555 LOCK_USED_IN_HARDIRQ_READ)) 3556 return 0; 3557 if (curr->softirq_context) 3558 if (!mark_lock(curr, hlock, 3559 LOCK_USED_IN_SOFTIRQ_READ)) 3560 return 0; 3561 } else { 3562 if (curr->hardirq_context) 3563 if (!mark_lock(curr, hlock, LOCK_USED_IN_HARDIRQ)) 3564 return 0; 3565 if (curr->softirq_context) 3566 if (!mark_lock(curr, hlock, LOCK_USED_IN_SOFTIRQ)) 3567 return 0; 3568 } 3569 } 3570 if (!hlock->hardirqs_off) { 3571 if (hlock->read) { 3572 if (!mark_lock(curr, hlock, 3573 LOCK_ENABLED_HARDIRQ_READ)) 3574 return 0; 3575 if (curr->softirqs_enabled) 3576 if (!mark_lock(curr, hlock, 3577 LOCK_ENABLED_SOFTIRQ_READ)) 3578 return 0; 3579 } else { 3580 if (!mark_lock(curr, hlock, 3581 LOCK_ENABLED_HARDIRQ)) 3582 return 0; 3583 if (curr->softirqs_enabled) 3584 if (!mark_lock(curr, hlock, 3585 LOCK_ENABLED_SOFTIRQ)) 3586 return 0; 3587 } 3588 } 3589 3590 lock_used: 3591 /* mark it as used: */ 3592 if (!mark_lock(curr, hlock, LOCK_USED)) 3593 return 0; 3594 3595 return 1; 3596 } 3597 3598 static inline unsigned int task_irq_context(struct task_struct *task) 3599 { 3600 return 2 * !!task->hardirq_context + !!task->softirq_context; 3601 } 3602 3603 static int separate_irq_context(struct task_struct *curr, 3604 struct held_lock *hlock) 3605 { 3606 unsigned int depth = curr->lockdep_depth; 3607 3608 /* 3609 * Keep track of points where we cross into an interrupt context: 3610 */ 3611 if (depth) { 3612 struct held_lock *prev_hlock; 3613 3614 prev_hlock = curr->held_locks + depth-1; 3615 /* 3616 * If we cross into another context, reset the 3617 * hash key (this also prevents the checking and the 3618 * adding of the dependency to 'prev'): 3619 */ 3620 if (prev_hlock->irq_context != hlock->irq_context) 3621 return 1; 3622 } 3623 return 0; 3624 } 3625 3626 /* 3627 * Mark a lock with a usage bit, and validate the state transition: 3628 */ 3629 static int mark_lock(struct task_struct *curr, struct held_lock *this, 3630 enum lock_usage_bit new_bit) 3631 { 3632 unsigned int new_mask = 1 << new_bit, ret = 1; 3633 3634 if (new_bit >= LOCK_USAGE_STATES) { 3635 DEBUG_LOCKS_WARN_ON(1); 3636 return 0; 3637 } 3638 3639 /* 3640 * If already set then do not dirty the cacheline, 3641 * nor do any checks: 3642 */ 3643 if (likely(hlock_class(this)->usage_mask & new_mask)) 3644 return 1; 3645 3646 if (!graph_lock()) 3647 return 0; 3648 /* 3649 * Make sure we didn't race: 3650 */ 3651 if (unlikely(hlock_class(this)->usage_mask & new_mask)) { 3652 graph_unlock(); 3653 return 1; 3654 } 3655 3656 hlock_class(this)->usage_mask |= new_mask; 3657 3658 if (!(hlock_class(this)->usage_traces[new_bit] = save_trace())) 3659 return 0; 3660 3661 switch (new_bit) { 3662 case LOCK_USED: 3663 debug_atomic_dec(nr_unused_locks); 3664 break; 3665 default: 3666 ret = mark_lock_irq(curr, this, new_bit); 3667 if (!ret) 3668 return 0; 3669 } 3670 3671 graph_unlock(); 3672 3673 /* 3674 * We must printk outside of the graph_lock: 3675 */ 3676 if (ret == 2) { 3677 printk("\nmarked lock as {%s}:\n", usage_str[new_bit]); 3678 print_lock(this); 3679 print_irqtrace_events(curr); 3680 dump_stack(); 3681 } 3682 3683 return ret; 3684 } 3685 3686 #else /* CONFIG_PROVE_LOCKING */ 3687 3688 static inline int 3689 mark_usage(struct task_struct *curr, struct held_lock *hlock, int check) 3690 { 3691 return 1; 3692 } 3693 3694 static inline unsigned int task_irq_context(struct task_struct *task) 3695 { 3696 return 0; 3697 } 3698 3699 static inline int separate_irq_context(struct task_struct *curr, 3700 struct held_lock *hlock) 3701 { 3702 return 0; 3703 } 3704 3705 #endif /* CONFIG_PROVE_LOCKING */ 3706 3707 /* 3708 * Initialize a lock instance's lock-class mapping info: 3709 */ 3710 void lockdep_init_map(struct lockdep_map *lock, const char *name, 3711 struct lock_class_key *key, int subclass) 3712 { 3713 int i; 3714 3715 for (i = 0; i < NR_LOCKDEP_CACHING_CLASSES; i++) 3716 lock->class_cache[i] = NULL; 3717 3718 #ifdef CONFIG_LOCK_STAT 3719 lock->cpu = raw_smp_processor_id(); 3720 #endif 3721 3722 /* 3723 * Can't be having no nameless bastards around this place! 3724 */ 3725 if (DEBUG_LOCKS_WARN_ON(!name)) { 3726 lock->name = "NULL"; 3727 return; 3728 } 3729 3730 lock->name = name; 3731 3732 /* 3733 * No key, no joy, we need to hash something. 3734 */ 3735 if (DEBUG_LOCKS_WARN_ON(!key)) 3736 return; 3737 /* 3738 * Sanity check, the lock-class key must either have been allocated 3739 * statically or must have been registered as a dynamic key. 3740 */ 3741 if (!static_obj(key) && !is_dynamic_key(key)) { 3742 if (debug_locks) 3743 printk(KERN_ERR "BUG: key %px has not been registered!\n", key); 3744 DEBUG_LOCKS_WARN_ON(1); 3745 return; 3746 } 3747 lock->key = key; 3748 3749 if (unlikely(!debug_locks)) 3750 return; 3751 3752 if (subclass) { 3753 unsigned long flags; 3754 3755 if (DEBUG_LOCKS_WARN_ON(current->lockdep_recursion)) 3756 return; 3757 3758 raw_local_irq_save(flags); 3759 current->lockdep_recursion = 1; 3760 register_lock_class(lock, subclass, 1); 3761 current->lockdep_recursion = 0; 3762 raw_local_irq_restore(flags); 3763 } 3764 } 3765 EXPORT_SYMBOL_GPL(lockdep_init_map); 3766 3767 struct lock_class_key __lockdep_no_validate__; 3768 EXPORT_SYMBOL_GPL(__lockdep_no_validate__); 3769 3770 static void 3771 print_lock_nested_lock_not_held(struct task_struct *curr, 3772 struct held_lock *hlock, 3773 unsigned long ip) 3774 { 3775 if (!debug_locks_off()) 3776 return; 3777 if (debug_locks_silent) 3778 return; 3779 3780 pr_warn("\n"); 3781 pr_warn("==================================\n"); 3782 pr_warn("WARNING: Nested lock was not taken\n"); 3783 print_kernel_ident(); 3784 pr_warn("----------------------------------\n"); 3785 3786 pr_warn("%s/%d is trying to lock:\n", curr->comm, task_pid_nr(curr)); 3787 print_lock(hlock); 3788 3789 pr_warn("\nbut this task is not holding:\n"); 3790 pr_warn("%s\n", hlock->nest_lock->name); 3791 3792 pr_warn("\nstack backtrace:\n"); 3793 dump_stack(); 3794 3795 pr_warn("\nother info that might help us debug this:\n"); 3796 lockdep_print_held_locks(curr); 3797 3798 pr_warn("\nstack backtrace:\n"); 3799 dump_stack(); 3800 } 3801 3802 static int __lock_is_held(const struct lockdep_map *lock, int read); 3803 3804 /* 3805 * This gets called for every mutex_lock*()/spin_lock*() operation. 3806 * We maintain the dependency maps and validate the locking attempt: 3807 * 3808 * The callers must make sure that IRQs are disabled before calling it, 3809 * otherwise we could get an interrupt which would want to take locks, 3810 * which would end up in lockdep again. 3811 */ 3812 static int __lock_acquire(struct lockdep_map *lock, unsigned int subclass, 3813 int trylock, int read, int check, int hardirqs_off, 3814 struct lockdep_map *nest_lock, unsigned long ip, 3815 int references, int pin_count) 3816 { 3817 struct task_struct *curr = current; 3818 struct lock_class *class = NULL; 3819 struct held_lock *hlock; 3820 unsigned int depth; 3821 int chain_head = 0; 3822 int class_idx; 3823 u64 chain_key; 3824 3825 if (unlikely(!debug_locks)) 3826 return 0; 3827 3828 if (!prove_locking || lock->key == &__lockdep_no_validate__) 3829 check = 0; 3830 3831 if (subclass < NR_LOCKDEP_CACHING_CLASSES) 3832 class = lock->class_cache[subclass]; 3833 /* 3834 * Not cached? 3835 */ 3836 if (unlikely(!class)) { 3837 class = register_lock_class(lock, subclass, 0); 3838 if (!class) 3839 return 0; 3840 } 3841 3842 debug_class_ops_inc(class); 3843 3844 if (very_verbose(class)) { 3845 printk("\nacquire class [%px] %s", class->key, class->name); 3846 if (class->name_version > 1) 3847 printk(KERN_CONT "#%d", class->name_version); 3848 printk(KERN_CONT "\n"); 3849 dump_stack(); 3850 } 3851 3852 /* 3853 * Add the lock to the list of currently held locks. 3854 * (we dont increase the depth just yet, up until the 3855 * dependency checks are done) 3856 */ 3857 depth = curr->lockdep_depth; 3858 /* 3859 * Ran out of static storage for our per-task lock stack again have we? 3860 */ 3861 if (DEBUG_LOCKS_WARN_ON(depth >= MAX_LOCK_DEPTH)) 3862 return 0; 3863 3864 class_idx = class - lock_classes; 3865 3866 if (depth) { 3867 hlock = curr->held_locks + depth - 1; 3868 if (hlock->class_idx == class_idx && nest_lock) { 3869 if (!references) 3870 references++; 3871 3872 if (!hlock->references) 3873 hlock->references++; 3874 3875 hlock->references += references; 3876 3877 /* Overflow */ 3878 if (DEBUG_LOCKS_WARN_ON(hlock->references < references)) 3879 return 0; 3880 3881 return 2; 3882 } 3883 } 3884 3885 hlock = curr->held_locks + depth; 3886 /* 3887 * Plain impossible, we just registered it and checked it weren't no 3888 * NULL like.. I bet this mushroom I ate was good! 3889 */ 3890 if (DEBUG_LOCKS_WARN_ON(!class)) 3891 return 0; 3892 hlock->class_idx = class_idx; 3893 hlock->acquire_ip = ip; 3894 hlock->instance = lock; 3895 hlock->nest_lock = nest_lock; 3896 hlock->irq_context = task_irq_context(curr); 3897 hlock->trylock = trylock; 3898 hlock->read = read; 3899 hlock->check = check; 3900 hlock->hardirqs_off = !!hardirqs_off; 3901 hlock->references = references; 3902 #ifdef CONFIG_LOCK_STAT 3903 hlock->waittime_stamp = 0; 3904 hlock->holdtime_stamp = lockstat_clock(); 3905 #endif 3906 hlock->pin_count = pin_count; 3907 3908 /* Initialize the lock usage bit */ 3909 if (!mark_usage(curr, hlock, check)) 3910 return 0; 3911 3912 /* 3913 * Calculate the chain hash: it's the combined hash of all the 3914 * lock keys along the dependency chain. We save the hash value 3915 * at every step so that we can get the current hash easily 3916 * after unlock. The chain hash is then used to cache dependency 3917 * results. 3918 * 3919 * The 'key ID' is what is the most compact key value to drive 3920 * the hash, not class->key. 3921 */ 3922 /* 3923 * Whoops, we did it again.. class_idx is invalid. 3924 */ 3925 if (DEBUG_LOCKS_WARN_ON(!test_bit(class_idx, lock_classes_in_use))) 3926 return 0; 3927 3928 chain_key = curr->curr_chain_key; 3929 if (!depth) { 3930 /* 3931 * How can we have a chain hash when we ain't got no keys?! 3932 */ 3933 if (DEBUG_LOCKS_WARN_ON(chain_key != INITIAL_CHAIN_KEY)) 3934 return 0; 3935 chain_head = 1; 3936 } 3937 3938 hlock->prev_chain_key = chain_key; 3939 if (separate_irq_context(curr, hlock)) { 3940 chain_key = INITIAL_CHAIN_KEY; 3941 chain_head = 1; 3942 } 3943 chain_key = iterate_chain_key(chain_key, class_idx); 3944 3945 if (nest_lock && !__lock_is_held(nest_lock, -1)) { 3946 print_lock_nested_lock_not_held(curr, hlock, ip); 3947 return 0; 3948 } 3949 3950 if (!debug_locks_silent) { 3951 WARN_ON_ONCE(depth && !hlock_class(hlock - 1)->key); 3952 WARN_ON_ONCE(!hlock_class(hlock)->key); 3953 } 3954 3955 if (!validate_chain(curr, hlock, chain_head, chain_key)) 3956 return 0; 3957 3958 curr->curr_chain_key = chain_key; 3959 curr->lockdep_depth++; 3960 check_chain_key(curr); 3961 #ifdef CONFIG_DEBUG_LOCKDEP 3962 if (unlikely(!debug_locks)) 3963 return 0; 3964 #endif 3965 if (unlikely(curr->lockdep_depth >= MAX_LOCK_DEPTH)) { 3966 debug_locks_off(); 3967 print_lockdep_off("BUG: MAX_LOCK_DEPTH too low!"); 3968 printk(KERN_DEBUG "depth: %i max: %lu!\n", 3969 curr->lockdep_depth, MAX_LOCK_DEPTH); 3970 3971 lockdep_print_held_locks(current); 3972 debug_show_all_locks(); 3973 dump_stack(); 3974 3975 return 0; 3976 } 3977 3978 if (unlikely(curr->lockdep_depth > max_lockdep_depth)) 3979 max_lockdep_depth = curr->lockdep_depth; 3980 3981 return 1; 3982 } 3983 3984 static void print_unlock_imbalance_bug(struct task_struct *curr, 3985 struct lockdep_map *lock, 3986 unsigned long ip) 3987 { 3988 if (!debug_locks_off()) 3989 return; 3990 if (debug_locks_silent) 3991 return; 3992 3993 pr_warn("\n"); 3994 pr_warn("=====================================\n"); 3995 pr_warn("WARNING: bad unlock balance detected!\n"); 3996 print_kernel_ident(); 3997 pr_warn("-------------------------------------\n"); 3998 pr_warn("%s/%d is trying to release lock (", 3999 curr->comm, task_pid_nr(curr)); 4000 print_lockdep_cache(lock); 4001 pr_cont(") at:\n"); 4002 print_ip_sym(ip); 4003 pr_warn("but there are no more locks to release!\n"); 4004 pr_warn("\nother info that might help us debug this:\n"); 4005 lockdep_print_held_locks(curr); 4006 4007 pr_warn("\nstack backtrace:\n"); 4008 dump_stack(); 4009 } 4010 4011 static int match_held_lock(const struct held_lock *hlock, 4012 const struct lockdep_map *lock) 4013 { 4014 if (hlock->instance == lock) 4015 return 1; 4016 4017 if (hlock->references) { 4018 const struct lock_class *class = lock->class_cache[0]; 4019 4020 if (!class) 4021 class = look_up_lock_class(lock, 0); 4022 4023 /* 4024 * If look_up_lock_class() failed to find a class, we're trying 4025 * to test if we hold a lock that has never yet been acquired. 4026 * Clearly if the lock hasn't been acquired _ever_, we're not 4027 * holding it either, so report failure. 4028 */ 4029 if (!class) 4030 return 0; 4031 4032 /* 4033 * References, but not a lock we're actually ref-counting? 4034 * State got messed up, follow the sites that change ->references 4035 * and try to make sense of it. 4036 */ 4037 if (DEBUG_LOCKS_WARN_ON(!hlock->nest_lock)) 4038 return 0; 4039 4040 if (hlock->class_idx == class - lock_classes) 4041 return 1; 4042 } 4043 4044 return 0; 4045 } 4046 4047 /* @depth must not be zero */ 4048 static struct held_lock *find_held_lock(struct task_struct *curr, 4049 struct lockdep_map *lock, 4050 unsigned int depth, int *idx) 4051 { 4052 struct held_lock *ret, *hlock, *prev_hlock; 4053 int i; 4054 4055 i = depth - 1; 4056 hlock = curr->held_locks + i; 4057 ret = hlock; 4058 if (match_held_lock(hlock, lock)) 4059 goto out; 4060 4061 ret = NULL; 4062 for (i--, prev_hlock = hlock--; 4063 i >= 0; 4064 i--, prev_hlock = hlock--) { 4065 /* 4066 * We must not cross into another context: 4067 */ 4068 if (prev_hlock->irq_context != hlock->irq_context) { 4069 ret = NULL; 4070 break; 4071 } 4072 if (match_held_lock(hlock, lock)) { 4073 ret = hlock; 4074 break; 4075 } 4076 } 4077 4078 out: 4079 *idx = i; 4080 return ret; 4081 } 4082 4083 static int reacquire_held_locks(struct task_struct *curr, unsigned int depth, 4084 int idx, unsigned int *merged) 4085 { 4086 struct held_lock *hlock; 4087 int first_idx = idx; 4088 4089 if (DEBUG_LOCKS_WARN_ON(!irqs_disabled())) 4090 return 0; 4091 4092 for (hlock = curr->held_locks + idx; idx < depth; idx++, hlock++) { 4093 switch (__lock_acquire(hlock->instance, 4094 hlock_class(hlock)->subclass, 4095 hlock->trylock, 4096 hlock->read, hlock->check, 4097 hlock->hardirqs_off, 4098 hlock->nest_lock, hlock->acquire_ip, 4099 hlock->references, hlock->pin_count)) { 4100 case 0: 4101 return 1; 4102 case 1: 4103 break; 4104 case 2: 4105 *merged += (idx == first_idx); 4106 break; 4107 default: 4108 WARN_ON(1); 4109 return 0; 4110 } 4111 } 4112 return 0; 4113 } 4114 4115 static int 4116 __lock_set_class(struct lockdep_map *lock, const char *name, 4117 struct lock_class_key *key, unsigned int subclass, 4118 unsigned long ip) 4119 { 4120 struct task_struct *curr = current; 4121 unsigned int depth, merged = 0; 4122 struct held_lock *hlock; 4123 struct lock_class *class; 4124 int i; 4125 4126 if (unlikely(!debug_locks)) 4127 return 0; 4128 4129 depth = curr->lockdep_depth; 4130 /* 4131 * This function is about (re)setting the class of a held lock, 4132 * yet we're not actually holding any locks. Naughty user! 4133 */ 4134 if (DEBUG_LOCKS_WARN_ON(!depth)) 4135 return 0; 4136 4137 hlock = find_held_lock(curr, lock, depth, &i); 4138 if (!hlock) { 4139 print_unlock_imbalance_bug(curr, lock, ip); 4140 return 0; 4141 } 4142 4143 lockdep_init_map(lock, name, key, 0); 4144 class = register_lock_class(lock, subclass, 0); 4145 hlock->class_idx = class - lock_classes; 4146 4147 curr->lockdep_depth = i; 4148 curr->curr_chain_key = hlock->prev_chain_key; 4149 4150 if (reacquire_held_locks(curr, depth, i, &merged)) 4151 return 0; 4152 4153 /* 4154 * I took it apart and put it back together again, except now I have 4155 * these 'spare' parts.. where shall I put them. 4156 */ 4157 if (DEBUG_LOCKS_WARN_ON(curr->lockdep_depth != depth - merged)) 4158 return 0; 4159 return 1; 4160 } 4161 4162 static int __lock_downgrade(struct lockdep_map *lock, unsigned long ip) 4163 { 4164 struct task_struct *curr = current; 4165 unsigned int depth, merged = 0; 4166 struct held_lock *hlock; 4167 int i; 4168 4169 if (unlikely(!debug_locks)) 4170 return 0; 4171 4172 depth = curr->lockdep_depth; 4173 /* 4174 * This function is about (re)setting the class of a held lock, 4175 * yet we're not actually holding any locks. Naughty user! 4176 */ 4177 if (DEBUG_LOCKS_WARN_ON(!depth)) 4178 return 0; 4179 4180 hlock = find_held_lock(curr, lock, depth, &i); 4181 if (!hlock) { 4182 print_unlock_imbalance_bug(curr, lock, ip); 4183 return 0; 4184 } 4185 4186 curr->lockdep_depth = i; 4187 curr->curr_chain_key = hlock->prev_chain_key; 4188 4189 WARN(hlock->read, "downgrading a read lock"); 4190 hlock->read = 1; 4191 hlock->acquire_ip = ip; 4192 4193 if (reacquire_held_locks(curr, depth, i, &merged)) 4194 return 0; 4195 4196 /* Merging can't happen with unchanged classes.. */ 4197 if (DEBUG_LOCKS_WARN_ON(merged)) 4198 return 0; 4199 4200 /* 4201 * I took it apart and put it back together again, except now I have 4202 * these 'spare' parts.. where shall I put them. 4203 */ 4204 if (DEBUG_LOCKS_WARN_ON(curr->lockdep_depth != depth)) 4205 return 0; 4206 4207 return 1; 4208 } 4209 4210 /* 4211 * Remove the lock to the list of currently held locks - this gets 4212 * called on mutex_unlock()/spin_unlock*() (or on a failed 4213 * mutex_lock_interruptible()). 4214 * 4215 * @nested is an hysterical artifact, needs a tree wide cleanup. 4216 */ 4217 static int 4218 __lock_release(struct lockdep_map *lock, unsigned long ip) 4219 { 4220 struct task_struct *curr = current; 4221 unsigned int depth, merged = 1; 4222 struct held_lock *hlock; 4223 int i; 4224 4225 if (unlikely(!debug_locks)) 4226 return 0; 4227 4228 depth = curr->lockdep_depth; 4229 /* 4230 * So we're all set to release this lock.. wait what lock? We don't 4231 * own any locks, you've been drinking again? 4232 */ 4233 if (depth <= 0) { 4234 print_unlock_imbalance_bug(curr, lock, ip); 4235 return 0; 4236 } 4237 4238 /* 4239 * Check whether the lock exists in the current stack 4240 * of held locks: 4241 */ 4242 hlock = find_held_lock(curr, lock, depth, &i); 4243 if (!hlock) { 4244 print_unlock_imbalance_bug(curr, lock, ip); 4245 return 0; 4246 } 4247 4248 if (hlock->instance == lock) 4249 lock_release_holdtime(hlock); 4250 4251 WARN(hlock->pin_count, "releasing a pinned lock\n"); 4252 4253 if (hlock->references) { 4254 hlock->references--; 4255 if (hlock->references) { 4256 /* 4257 * We had, and after removing one, still have 4258 * references, the current lock stack is still 4259 * valid. We're done! 4260 */ 4261 return 1; 4262 } 4263 } 4264 4265 /* 4266 * We have the right lock to unlock, 'hlock' points to it. 4267 * Now we remove it from the stack, and add back the other 4268 * entries (if any), recalculating the hash along the way: 4269 */ 4270 4271 curr->lockdep_depth = i; 4272 curr->curr_chain_key = hlock->prev_chain_key; 4273 4274 /* 4275 * The most likely case is when the unlock is on the innermost 4276 * lock. In this case, we are done! 4277 */ 4278 if (i == depth-1) 4279 return 1; 4280 4281 if (reacquire_held_locks(curr, depth, i + 1, &merged)) 4282 return 0; 4283 4284 /* 4285 * We had N bottles of beer on the wall, we drank one, but now 4286 * there's not N-1 bottles of beer left on the wall... 4287 * Pouring two of the bottles together is acceptable. 4288 */ 4289 DEBUG_LOCKS_WARN_ON(curr->lockdep_depth != depth - merged); 4290 4291 /* 4292 * Since reacquire_held_locks() would have called check_chain_key() 4293 * indirectly via __lock_acquire(), we don't need to do it again 4294 * on return. 4295 */ 4296 return 0; 4297 } 4298 4299 static nokprobe_inline 4300 int __lock_is_held(const struct lockdep_map *lock, int read) 4301 { 4302 struct task_struct *curr = current; 4303 int i; 4304 4305 for (i = 0; i < curr->lockdep_depth; i++) { 4306 struct held_lock *hlock = curr->held_locks + i; 4307 4308 if (match_held_lock(hlock, lock)) { 4309 if (read == -1 || hlock->read == read) 4310 return 1; 4311 4312 return 0; 4313 } 4314 } 4315 4316 return 0; 4317 } 4318 4319 static struct pin_cookie __lock_pin_lock(struct lockdep_map *lock) 4320 { 4321 struct pin_cookie cookie = NIL_COOKIE; 4322 struct task_struct *curr = current; 4323 int i; 4324 4325 if (unlikely(!debug_locks)) 4326 return cookie; 4327 4328 for (i = 0; i < curr->lockdep_depth; i++) { 4329 struct held_lock *hlock = curr->held_locks + i; 4330 4331 if (match_held_lock(hlock, lock)) { 4332 /* 4333 * Grab 16bits of randomness; this is sufficient to not 4334 * be guessable and still allows some pin nesting in 4335 * our u32 pin_count. 4336 */ 4337 cookie.val = 1 + (prandom_u32() >> 16); 4338 hlock->pin_count += cookie.val; 4339 return cookie; 4340 } 4341 } 4342 4343 WARN(1, "pinning an unheld lock\n"); 4344 return cookie; 4345 } 4346 4347 static void __lock_repin_lock(struct lockdep_map *lock, struct pin_cookie cookie) 4348 { 4349 struct task_struct *curr = current; 4350 int i; 4351 4352 if (unlikely(!debug_locks)) 4353 return; 4354 4355 for (i = 0; i < curr->lockdep_depth; i++) { 4356 struct held_lock *hlock = curr->held_locks + i; 4357 4358 if (match_held_lock(hlock, lock)) { 4359 hlock->pin_count += cookie.val; 4360 return; 4361 } 4362 } 4363 4364 WARN(1, "pinning an unheld lock\n"); 4365 } 4366 4367 static void __lock_unpin_lock(struct lockdep_map *lock, struct pin_cookie cookie) 4368 { 4369 struct task_struct *curr = current; 4370 int i; 4371 4372 if (unlikely(!debug_locks)) 4373 return; 4374 4375 for (i = 0; i < curr->lockdep_depth; i++) { 4376 struct held_lock *hlock = curr->held_locks + i; 4377 4378 if (match_held_lock(hlock, lock)) { 4379 if (WARN(!hlock->pin_count, "unpinning an unpinned lock\n")) 4380 return; 4381 4382 hlock->pin_count -= cookie.val; 4383 4384 if (WARN((int)hlock->pin_count < 0, "pin count corrupted\n")) 4385 hlock->pin_count = 0; 4386 4387 return; 4388 } 4389 } 4390 4391 WARN(1, "unpinning an unheld lock\n"); 4392 } 4393 4394 /* 4395 * Check whether we follow the irq-flags state precisely: 4396 */ 4397 static void check_flags(unsigned long flags) 4398 { 4399 #if defined(CONFIG_PROVE_LOCKING) && defined(CONFIG_DEBUG_LOCKDEP) 4400 if (!debug_locks) 4401 return; 4402 4403 if (irqs_disabled_flags(flags)) { 4404 if (DEBUG_LOCKS_WARN_ON(current->hardirqs_enabled)) { 4405 printk("possible reason: unannotated irqs-off.\n"); 4406 } 4407 } else { 4408 if (DEBUG_LOCKS_WARN_ON(!current->hardirqs_enabled)) { 4409 printk("possible reason: unannotated irqs-on.\n"); 4410 } 4411 } 4412 4413 /* 4414 * We dont accurately track softirq state in e.g. 4415 * hardirq contexts (such as on 4KSTACKS), so only 4416 * check if not in hardirq contexts: 4417 */ 4418 if (!hardirq_count()) { 4419 if (softirq_count()) { 4420 /* like the above, but with softirqs */ 4421 DEBUG_LOCKS_WARN_ON(current->softirqs_enabled); 4422 } else { 4423 /* lick the above, does it taste good? */ 4424 DEBUG_LOCKS_WARN_ON(!current->softirqs_enabled); 4425 } 4426 } 4427 4428 if (!debug_locks) 4429 print_irqtrace_events(current); 4430 #endif 4431 } 4432 4433 void lock_set_class(struct lockdep_map *lock, const char *name, 4434 struct lock_class_key *key, unsigned int subclass, 4435 unsigned long ip) 4436 { 4437 unsigned long flags; 4438 4439 if (unlikely(current->lockdep_recursion)) 4440 return; 4441 4442 raw_local_irq_save(flags); 4443 current->lockdep_recursion = 1; 4444 check_flags(flags); 4445 if (__lock_set_class(lock, name, key, subclass, ip)) 4446 check_chain_key(current); 4447 current->lockdep_recursion = 0; 4448 raw_local_irq_restore(flags); 4449 } 4450 EXPORT_SYMBOL_GPL(lock_set_class); 4451 4452 void lock_downgrade(struct lockdep_map *lock, unsigned long ip) 4453 { 4454 unsigned long flags; 4455 4456 if (unlikely(current->lockdep_recursion)) 4457 return; 4458 4459 raw_local_irq_save(flags); 4460 current->lockdep_recursion = 1; 4461 check_flags(flags); 4462 if (__lock_downgrade(lock, ip)) 4463 check_chain_key(current); 4464 current->lockdep_recursion = 0; 4465 raw_local_irq_restore(flags); 4466 } 4467 EXPORT_SYMBOL_GPL(lock_downgrade); 4468 4469 /* 4470 * We are not always called with irqs disabled - do that here, 4471 * and also avoid lockdep recursion: 4472 */ 4473 void lock_acquire(struct lockdep_map *lock, unsigned int subclass, 4474 int trylock, int read, int check, 4475 struct lockdep_map *nest_lock, unsigned long ip) 4476 { 4477 unsigned long flags; 4478 4479 if (unlikely(current->lockdep_recursion)) 4480 return; 4481 4482 raw_local_irq_save(flags); 4483 check_flags(flags); 4484 4485 current->lockdep_recursion = 1; 4486 trace_lock_acquire(lock, subclass, trylock, read, check, nest_lock, ip); 4487 __lock_acquire(lock, subclass, trylock, read, check, 4488 irqs_disabled_flags(flags), nest_lock, ip, 0, 0); 4489 current->lockdep_recursion = 0; 4490 raw_local_irq_restore(flags); 4491 } 4492 EXPORT_SYMBOL_GPL(lock_acquire); 4493 4494 void lock_release(struct lockdep_map *lock, int nested, 4495 unsigned long ip) 4496 { 4497 unsigned long flags; 4498 4499 if (unlikely(current->lockdep_recursion)) 4500 return; 4501 4502 raw_local_irq_save(flags); 4503 check_flags(flags); 4504 current->lockdep_recursion = 1; 4505 trace_lock_release(lock, ip); 4506 if (__lock_release(lock, ip)) 4507 check_chain_key(current); 4508 current->lockdep_recursion = 0; 4509 raw_local_irq_restore(flags); 4510 } 4511 EXPORT_SYMBOL_GPL(lock_release); 4512 4513 int lock_is_held_type(const struct lockdep_map *lock, int read) 4514 { 4515 unsigned long flags; 4516 int ret = 0; 4517 4518 if (unlikely(current->lockdep_recursion)) 4519 return 1; /* avoid false negative lockdep_assert_held() */ 4520 4521 raw_local_irq_save(flags); 4522 check_flags(flags); 4523 4524 current->lockdep_recursion = 1; 4525 ret = __lock_is_held(lock, read); 4526 current->lockdep_recursion = 0; 4527 raw_local_irq_restore(flags); 4528 4529 return ret; 4530 } 4531 EXPORT_SYMBOL_GPL(lock_is_held_type); 4532 NOKPROBE_SYMBOL(lock_is_held_type); 4533 4534 struct pin_cookie lock_pin_lock(struct lockdep_map *lock) 4535 { 4536 struct pin_cookie cookie = NIL_COOKIE; 4537 unsigned long flags; 4538 4539 if (unlikely(current->lockdep_recursion)) 4540 return cookie; 4541 4542 raw_local_irq_save(flags); 4543 check_flags(flags); 4544 4545 current->lockdep_recursion = 1; 4546 cookie = __lock_pin_lock(lock); 4547 current->lockdep_recursion = 0; 4548 raw_local_irq_restore(flags); 4549 4550 return cookie; 4551 } 4552 EXPORT_SYMBOL_GPL(lock_pin_lock); 4553 4554 void lock_repin_lock(struct lockdep_map *lock, struct pin_cookie cookie) 4555 { 4556 unsigned long flags; 4557 4558 if (unlikely(current->lockdep_recursion)) 4559 return; 4560 4561 raw_local_irq_save(flags); 4562 check_flags(flags); 4563 4564 current->lockdep_recursion = 1; 4565 __lock_repin_lock(lock, cookie); 4566 current->lockdep_recursion = 0; 4567 raw_local_irq_restore(flags); 4568 } 4569 EXPORT_SYMBOL_GPL(lock_repin_lock); 4570 4571 void lock_unpin_lock(struct lockdep_map *lock, struct pin_cookie cookie) 4572 { 4573 unsigned long flags; 4574 4575 if (unlikely(current->lockdep_recursion)) 4576 return; 4577 4578 raw_local_irq_save(flags); 4579 check_flags(flags); 4580 4581 current->lockdep_recursion = 1; 4582 __lock_unpin_lock(lock, cookie); 4583 current->lockdep_recursion = 0; 4584 raw_local_irq_restore(flags); 4585 } 4586 EXPORT_SYMBOL_GPL(lock_unpin_lock); 4587 4588 #ifdef CONFIG_LOCK_STAT 4589 static void print_lock_contention_bug(struct task_struct *curr, 4590 struct lockdep_map *lock, 4591 unsigned long ip) 4592 { 4593 if (!debug_locks_off()) 4594 return; 4595 if (debug_locks_silent) 4596 return; 4597 4598 pr_warn("\n"); 4599 pr_warn("=================================\n"); 4600 pr_warn("WARNING: bad contention detected!\n"); 4601 print_kernel_ident(); 4602 pr_warn("---------------------------------\n"); 4603 pr_warn("%s/%d is trying to contend lock (", 4604 curr->comm, task_pid_nr(curr)); 4605 print_lockdep_cache(lock); 4606 pr_cont(") at:\n"); 4607 print_ip_sym(ip); 4608 pr_warn("but there are no locks held!\n"); 4609 pr_warn("\nother info that might help us debug this:\n"); 4610 lockdep_print_held_locks(curr); 4611 4612 pr_warn("\nstack backtrace:\n"); 4613 dump_stack(); 4614 } 4615 4616 static void 4617 __lock_contended(struct lockdep_map *lock, unsigned long ip) 4618 { 4619 struct task_struct *curr = current; 4620 struct held_lock *hlock; 4621 struct lock_class_stats *stats; 4622 unsigned int depth; 4623 int i, contention_point, contending_point; 4624 4625 depth = curr->lockdep_depth; 4626 /* 4627 * Whee, we contended on this lock, except it seems we're not 4628 * actually trying to acquire anything much at all.. 4629 */ 4630 if (DEBUG_LOCKS_WARN_ON(!depth)) 4631 return; 4632 4633 hlock = find_held_lock(curr, lock, depth, &i); 4634 if (!hlock) { 4635 print_lock_contention_bug(curr, lock, ip); 4636 return; 4637 } 4638 4639 if (hlock->instance != lock) 4640 return; 4641 4642 hlock->waittime_stamp = lockstat_clock(); 4643 4644 contention_point = lock_point(hlock_class(hlock)->contention_point, ip); 4645 contending_point = lock_point(hlock_class(hlock)->contending_point, 4646 lock->ip); 4647 4648 stats = get_lock_stats(hlock_class(hlock)); 4649 if (contention_point < LOCKSTAT_POINTS) 4650 stats->contention_point[contention_point]++; 4651 if (contending_point < LOCKSTAT_POINTS) 4652 stats->contending_point[contending_point]++; 4653 if (lock->cpu != smp_processor_id()) 4654 stats->bounces[bounce_contended + !!hlock->read]++; 4655 } 4656 4657 static void 4658 __lock_acquired(struct lockdep_map *lock, unsigned long ip) 4659 { 4660 struct task_struct *curr = current; 4661 struct held_lock *hlock; 4662 struct lock_class_stats *stats; 4663 unsigned int depth; 4664 u64 now, waittime = 0; 4665 int i, cpu; 4666 4667 depth = curr->lockdep_depth; 4668 /* 4669 * Yay, we acquired ownership of this lock we didn't try to 4670 * acquire, how the heck did that happen? 4671 */ 4672 if (DEBUG_LOCKS_WARN_ON(!depth)) 4673 return; 4674 4675 hlock = find_held_lock(curr, lock, depth, &i); 4676 if (!hlock) { 4677 print_lock_contention_bug(curr, lock, _RET_IP_); 4678 return; 4679 } 4680 4681 if (hlock->instance != lock) 4682 return; 4683 4684 cpu = smp_processor_id(); 4685 if (hlock->waittime_stamp) { 4686 now = lockstat_clock(); 4687 waittime = now - hlock->waittime_stamp; 4688 hlock->holdtime_stamp = now; 4689 } 4690 4691 trace_lock_acquired(lock, ip); 4692 4693 stats = get_lock_stats(hlock_class(hlock)); 4694 if (waittime) { 4695 if (hlock->read) 4696 lock_time_inc(&stats->read_waittime, waittime); 4697 else 4698 lock_time_inc(&stats->write_waittime, waittime); 4699 } 4700 if (lock->cpu != cpu) 4701 stats->bounces[bounce_acquired + !!hlock->read]++; 4702 4703 lock->cpu = cpu; 4704 lock->ip = ip; 4705 } 4706 4707 void lock_contended(struct lockdep_map *lock, unsigned long ip) 4708 { 4709 unsigned long flags; 4710 4711 if (unlikely(!lock_stat || !debug_locks)) 4712 return; 4713 4714 if (unlikely(current->lockdep_recursion)) 4715 return; 4716 4717 raw_local_irq_save(flags); 4718 check_flags(flags); 4719 current->lockdep_recursion = 1; 4720 trace_lock_contended(lock, ip); 4721 __lock_contended(lock, ip); 4722 current->lockdep_recursion = 0; 4723 raw_local_irq_restore(flags); 4724 } 4725 EXPORT_SYMBOL_GPL(lock_contended); 4726 4727 void lock_acquired(struct lockdep_map *lock, unsigned long ip) 4728 { 4729 unsigned long flags; 4730 4731 if (unlikely(!lock_stat || !debug_locks)) 4732 return; 4733 4734 if (unlikely(current->lockdep_recursion)) 4735 return; 4736 4737 raw_local_irq_save(flags); 4738 check_flags(flags); 4739 current->lockdep_recursion = 1; 4740 __lock_acquired(lock, ip); 4741 current->lockdep_recursion = 0; 4742 raw_local_irq_restore(flags); 4743 } 4744 EXPORT_SYMBOL_GPL(lock_acquired); 4745 #endif 4746 4747 /* 4748 * Used by the testsuite, sanitize the validator state 4749 * after a simulated failure: 4750 */ 4751 4752 void lockdep_reset(void) 4753 { 4754 unsigned long flags; 4755 int i; 4756 4757 raw_local_irq_save(flags); 4758 lockdep_init_task(current); 4759 memset(current->held_locks, 0, MAX_LOCK_DEPTH*sizeof(struct held_lock)); 4760 nr_hardirq_chains = 0; 4761 nr_softirq_chains = 0; 4762 nr_process_chains = 0; 4763 debug_locks = 1; 4764 for (i = 0; i < CHAINHASH_SIZE; i++) 4765 INIT_HLIST_HEAD(chainhash_table + i); 4766 raw_local_irq_restore(flags); 4767 } 4768 4769 /* Remove a class from a lock chain. Must be called with the graph lock held. */ 4770 static void remove_class_from_lock_chain(struct pending_free *pf, 4771 struct lock_chain *chain, 4772 struct lock_class *class) 4773 { 4774 #ifdef CONFIG_PROVE_LOCKING 4775 struct lock_chain *new_chain; 4776 u64 chain_key; 4777 int i; 4778 4779 for (i = chain->base; i < chain->base + chain->depth; i++) { 4780 if (chain_hlocks[i] != class - lock_classes) 4781 continue; 4782 /* The code below leaks one chain_hlock[] entry. */ 4783 if (--chain->depth > 0) { 4784 memmove(&chain_hlocks[i], &chain_hlocks[i + 1], 4785 (chain->base + chain->depth - i) * 4786 sizeof(chain_hlocks[0])); 4787 } 4788 /* 4789 * Each lock class occurs at most once in a lock chain so once 4790 * we found a match we can break out of this loop. 4791 */ 4792 goto recalc; 4793 } 4794 /* Since the chain has not been modified, return. */ 4795 return; 4796 4797 recalc: 4798 chain_key = INITIAL_CHAIN_KEY; 4799 for (i = chain->base; i < chain->base + chain->depth; i++) 4800 chain_key = iterate_chain_key(chain_key, chain_hlocks[i]); 4801 if (chain->depth && chain->chain_key == chain_key) 4802 return; 4803 /* Overwrite the chain key for concurrent RCU readers. */ 4804 WRITE_ONCE(chain->chain_key, chain_key); 4805 /* 4806 * Note: calling hlist_del_rcu() from inside a 4807 * hlist_for_each_entry_rcu() loop is safe. 4808 */ 4809 hlist_del_rcu(&chain->entry); 4810 __set_bit(chain - lock_chains, pf->lock_chains_being_freed); 4811 if (chain->depth == 0) 4812 return; 4813 /* 4814 * If the modified lock chain matches an existing lock chain, drop 4815 * the modified lock chain. 4816 */ 4817 if (lookup_chain_cache(chain_key)) 4818 return; 4819 new_chain = alloc_lock_chain(); 4820 if (WARN_ON_ONCE(!new_chain)) { 4821 debug_locks_off(); 4822 return; 4823 } 4824 *new_chain = *chain; 4825 hlist_add_head_rcu(&new_chain->entry, chainhashentry(chain_key)); 4826 #endif 4827 } 4828 4829 /* Must be called with the graph lock held. */ 4830 static void remove_class_from_lock_chains(struct pending_free *pf, 4831 struct lock_class *class) 4832 { 4833 struct lock_chain *chain; 4834 struct hlist_head *head; 4835 int i; 4836 4837 for (i = 0; i < ARRAY_SIZE(chainhash_table); i++) { 4838 head = chainhash_table + i; 4839 hlist_for_each_entry_rcu(chain, head, entry) { 4840 remove_class_from_lock_chain(pf, chain, class); 4841 } 4842 } 4843 } 4844 4845 /* 4846 * Remove all references to a lock class. The caller must hold the graph lock. 4847 */ 4848 static void zap_class(struct pending_free *pf, struct lock_class *class) 4849 { 4850 struct lock_list *entry; 4851 int i; 4852 4853 WARN_ON_ONCE(!class->key); 4854 4855 /* 4856 * Remove all dependencies this lock is 4857 * involved in: 4858 */ 4859 for_each_set_bit(i, list_entries_in_use, ARRAY_SIZE(list_entries)) { 4860 entry = list_entries + i; 4861 if (entry->class != class && entry->links_to != class) 4862 continue; 4863 __clear_bit(i, list_entries_in_use); 4864 nr_list_entries--; 4865 list_del_rcu(&entry->entry); 4866 } 4867 if (list_empty(&class->locks_after) && 4868 list_empty(&class->locks_before)) { 4869 list_move_tail(&class->lock_entry, &pf->zapped); 4870 hlist_del_rcu(&class->hash_entry); 4871 WRITE_ONCE(class->key, NULL); 4872 WRITE_ONCE(class->name, NULL); 4873 nr_lock_classes--; 4874 __clear_bit(class - lock_classes, lock_classes_in_use); 4875 } else { 4876 WARN_ONCE(true, "%s() failed for class %s\n", __func__, 4877 class->name); 4878 } 4879 4880 remove_class_from_lock_chains(pf, class); 4881 } 4882 4883 static void reinit_class(struct lock_class *class) 4884 { 4885 void *const p = class; 4886 const unsigned int offset = offsetof(struct lock_class, key); 4887 4888 WARN_ON_ONCE(!class->lock_entry.next); 4889 WARN_ON_ONCE(!list_empty(&class->locks_after)); 4890 WARN_ON_ONCE(!list_empty(&class->locks_before)); 4891 memset(p + offset, 0, sizeof(*class) - offset); 4892 WARN_ON_ONCE(!class->lock_entry.next); 4893 WARN_ON_ONCE(!list_empty(&class->locks_after)); 4894 WARN_ON_ONCE(!list_empty(&class->locks_before)); 4895 } 4896 4897 static inline int within(const void *addr, void *start, unsigned long size) 4898 { 4899 return addr >= start && addr < start + size; 4900 } 4901 4902 static bool inside_selftest(void) 4903 { 4904 return current == lockdep_selftest_task_struct; 4905 } 4906 4907 /* The caller must hold the graph lock. */ 4908 static struct pending_free *get_pending_free(void) 4909 { 4910 return delayed_free.pf + delayed_free.index; 4911 } 4912 4913 static void free_zapped_rcu(struct rcu_head *cb); 4914 4915 /* 4916 * Schedule an RCU callback if no RCU callback is pending. Must be called with 4917 * the graph lock held. 4918 */ 4919 static void call_rcu_zapped(struct pending_free *pf) 4920 { 4921 WARN_ON_ONCE(inside_selftest()); 4922 4923 if (list_empty(&pf->zapped)) 4924 return; 4925 4926 if (delayed_free.scheduled) 4927 return; 4928 4929 delayed_free.scheduled = true; 4930 4931 WARN_ON_ONCE(delayed_free.pf + delayed_free.index != pf); 4932 delayed_free.index ^= 1; 4933 4934 call_rcu(&delayed_free.rcu_head, free_zapped_rcu); 4935 } 4936 4937 /* The caller must hold the graph lock. May be called from RCU context. */ 4938 static void __free_zapped_classes(struct pending_free *pf) 4939 { 4940 struct lock_class *class; 4941 4942 check_data_structures(); 4943 4944 list_for_each_entry(class, &pf->zapped, lock_entry) 4945 reinit_class(class); 4946 4947 list_splice_init(&pf->zapped, &free_lock_classes); 4948 4949 #ifdef CONFIG_PROVE_LOCKING 4950 bitmap_andnot(lock_chains_in_use, lock_chains_in_use, 4951 pf->lock_chains_being_freed, ARRAY_SIZE(lock_chains)); 4952 bitmap_clear(pf->lock_chains_being_freed, 0, ARRAY_SIZE(lock_chains)); 4953 #endif 4954 } 4955 4956 static void free_zapped_rcu(struct rcu_head *ch) 4957 { 4958 struct pending_free *pf; 4959 unsigned long flags; 4960 4961 if (WARN_ON_ONCE(ch != &delayed_free.rcu_head)) 4962 return; 4963 4964 raw_local_irq_save(flags); 4965 arch_spin_lock(&lockdep_lock); 4966 current->lockdep_recursion = 1; 4967 4968 /* closed head */ 4969 pf = delayed_free.pf + (delayed_free.index ^ 1); 4970 __free_zapped_classes(pf); 4971 delayed_free.scheduled = false; 4972 4973 /* 4974 * If there's anything on the open list, close and start a new callback. 4975 */ 4976 call_rcu_zapped(delayed_free.pf + delayed_free.index); 4977 4978 current->lockdep_recursion = 0; 4979 arch_spin_unlock(&lockdep_lock); 4980 raw_local_irq_restore(flags); 4981 } 4982 4983 /* 4984 * Remove all lock classes from the class hash table and from the 4985 * all_lock_classes list whose key or name is in the address range [start, 4986 * start + size). Move these lock classes to the zapped_classes list. Must 4987 * be called with the graph lock held. 4988 */ 4989 static void __lockdep_free_key_range(struct pending_free *pf, void *start, 4990 unsigned long size) 4991 { 4992 struct lock_class *class; 4993 struct hlist_head *head; 4994 int i; 4995 4996 /* Unhash all classes that were created by a module. */ 4997 for (i = 0; i < CLASSHASH_SIZE; i++) { 4998 head = classhash_table + i; 4999 hlist_for_each_entry_rcu(class, head, hash_entry) { 5000 if (!within(class->key, start, size) && 5001 !within(class->name, start, size)) 5002 continue; 5003 zap_class(pf, class); 5004 } 5005 } 5006 } 5007 5008 /* 5009 * Used in module.c to remove lock classes from memory that is going to be 5010 * freed; and possibly re-used by other modules. 5011 * 5012 * We will have had one synchronize_rcu() before getting here, so we're 5013 * guaranteed nobody will look up these exact classes -- they're properly dead 5014 * but still allocated. 5015 */ 5016 static void lockdep_free_key_range_reg(void *start, unsigned long size) 5017 { 5018 struct pending_free *pf; 5019 unsigned long flags; 5020 5021 init_data_structures_once(); 5022 5023 raw_local_irq_save(flags); 5024 arch_spin_lock(&lockdep_lock); 5025 current->lockdep_recursion = 1; 5026 pf = get_pending_free(); 5027 __lockdep_free_key_range(pf, start, size); 5028 call_rcu_zapped(pf); 5029 current->lockdep_recursion = 0; 5030 arch_spin_unlock(&lockdep_lock); 5031 raw_local_irq_restore(flags); 5032 5033 /* 5034 * Wait for any possible iterators from look_up_lock_class() to pass 5035 * before continuing to free the memory they refer to. 5036 */ 5037 synchronize_rcu(); 5038 } 5039 5040 /* 5041 * Free all lockdep keys in the range [start, start+size). Does not sleep. 5042 * Ignores debug_locks. Must only be used by the lockdep selftests. 5043 */ 5044 static void lockdep_free_key_range_imm(void *start, unsigned long size) 5045 { 5046 struct pending_free *pf = delayed_free.pf; 5047 unsigned long flags; 5048 5049 init_data_structures_once(); 5050 5051 raw_local_irq_save(flags); 5052 arch_spin_lock(&lockdep_lock); 5053 __lockdep_free_key_range(pf, start, size); 5054 __free_zapped_classes(pf); 5055 arch_spin_unlock(&lockdep_lock); 5056 raw_local_irq_restore(flags); 5057 } 5058 5059 void lockdep_free_key_range(void *start, unsigned long size) 5060 { 5061 init_data_structures_once(); 5062 5063 if (inside_selftest()) 5064 lockdep_free_key_range_imm(start, size); 5065 else 5066 lockdep_free_key_range_reg(start, size); 5067 } 5068 5069 /* 5070 * Check whether any element of the @lock->class_cache[] array refers to a 5071 * registered lock class. The caller must hold either the graph lock or the 5072 * RCU read lock. 5073 */ 5074 static bool lock_class_cache_is_registered(struct lockdep_map *lock) 5075 { 5076 struct lock_class *class; 5077 struct hlist_head *head; 5078 int i, j; 5079 5080 for (i = 0; i < CLASSHASH_SIZE; i++) { 5081 head = classhash_table + i; 5082 hlist_for_each_entry_rcu(class, head, hash_entry) { 5083 for (j = 0; j < NR_LOCKDEP_CACHING_CLASSES; j++) 5084 if (lock->class_cache[j] == class) 5085 return true; 5086 } 5087 } 5088 return false; 5089 } 5090 5091 /* The caller must hold the graph lock. Does not sleep. */ 5092 static void __lockdep_reset_lock(struct pending_free *pf, 5093 struct lockdep_map *lock) 5094 { 5095 struct lock_class *class; 5096 int j; 5097 5098 /* 5099 * Remove all classes this lock might have: 5100 */ 5101 for (j = 0; j < MAX_LOCKDEP_SUBCLASSES; j++) { 5102 /* 5103 * If the class exists we look it up and zap it: 5104 */ 5105 class = look_up_lock_class(lock, j); 5106 if (class) 5107 zap_class(pf, class); 5108 } 5109 /* 5110 * Debug check: in the end all mapped classes should 5111 * be gone. 5112 */ 5113 if (WARN_ON_ONCE(lock_class_cache_is_registered(lock))) 5114 debug_locks_off(); 5115 } 5116 5117 /* 5118 * Remove all information lockdep has about a lock if debug_locks == 1. Free 5119 * released data structures from RCU context. 5120 */ 5121 static void lockdep_reset_lock_reg(struct lockdep_map *lock) 5122 { 5123 struct pending_free *pf; 5124 unsigned long flags; 5125 int locked; 5126 5127 raw_local_irq_save(flags); 5128 locked = graph_lock(); 5129 if (!locked) 5130 goto out_irq; 5131 5132 pf = get_pending_free(); 5133 __lockdep_reset_lock(pf, lock); 5134 call_rcu_zapped(pf); 5135 5136 graph_unlock(); 5137 out_irq: 5138 raw_local_irq_restore(flags); 5139 } 5140 5141 /* 5142 * Reset a lock. Does not sleep. Ignores debug_locks. Must only be used by the 5143 * lockdep selftests. 5144 */ 5145 static void lockdep_reset_lock_imm(struct lockdep_map *lock) 5146 { 5147 struct pending_free *pf = delayed_free.pf; 5148 unsigned long flags; 5149 5150 raw_local_irq_save(flags); 5151 arch_spin_lock(&lockdep_lock); 5152 __lockdep_reset_lock(pf, lock); 5153 __free_zapped_classes(pf); 5154 arch_spin_unlock(&lockdep_lock); 5155 raw_local_irq_restore(flags); 5156 } 5157 5158 void lockdep_reset_lock(struct lockdep_map *lock) 5159 { 5160 init_data_structures_once(); 5161 5162 if (inside_selftest()) 5163 lockdep_reset_lock_imm(lock); 5164 else 5165 lockdep_reset_lock_reg(lock); 5166 } 5167 5168 /* Unregister a dynamically allocated key. */ 5169 void lockdep_unregister_key(struct lock_class_key *key) 5170 { 5171 struct hlist_head *hash_head = keyhashentry(key); 5172 struct lock_class_key *k; 5173 struct pending_free *pf; 5174 unsigned long flags; 5175 bool found = false; 5176 5177 might_sleep(); 5178 5179 if (WARN_ON_ONCE(static_obj(key))) 5180 return; 5181 5182 raw_local_irq_save(flags); 5183 if (!graph_lock()) 5184 goto out_irq; 5185 5186 pf = get_pending_free(); 5187 hlist_for_each_entry_rcu(k, hash_head, hash_entry) { 5188 if (k == key) { 5189 hlist_del_rcu(&k->hash_entry); 5190 found = true; 5191 break; 5192 } 5193 } 5194 WARN_ON_ONCE(!found); 5195 __lockdep_free_key_range(pf, key, 1); 5196 call_rcu_zapped(pf); 5197 graph_unlock(); 5198 out_irq: 5199 raw_local_irq_restore(flags); 5200 5201 /* Wait until is_dynamic_key() has finished accessing k->hash_entry. */ 5202 synchronize_rcu(); 5203 } 5204 EXPORT_SYMBOL_GPL(lockdep_unregister_key); 5205 5206 void __init lockdep_init(void) 5207 { 5208 printk("Lock dependency validator: Copyright (c) 2006 Red Hat, Inc., Ingo Molnar\n"); 5209 5210 printk("... MAX_LOCKDEP_SUBCLASSES: %lu\n", MAX_LOCKDEP_SUBCLASSES); 5211 printk("... MAX_LOCK_DEPTH: %lu\n", MAX_LOCK_DEPTH); 5212 printk("... MAX_LOCKDEP_KEYS: %lu\n", MAX_LOCKDEP_KEYS); 5213 printk("... CLASSHASH_SIZE: %lu\n", CLASSHASH_SIZE); 5214 printk("... MAX_LOCKDEP_ENTRIES: %lu\n", MAX_LOCKDEP_ENTRIES); 5215 printk("... MAX_LOCKDEP_CHAINS: %lu\n", MAX_LOCKDEP_CHAINS); 5216 printk("... CHAINHASH_SIZE: %lu\n", CHAINHASH_SIZE); 5217 5218 printk(" memory used by lock dependency info: %zu kB\n", 5219 (sizeof(lock_classes) + 5220 sizeof(lock_classes_in_use) + 5221 sizeof(classhash_table) + 5222 sizeof(list_entries) + 5223 sizeof(list_entries_in_use) + 5224 sizeof(chainhash_table) + 5225 sizeof(delayed_free) 5226 #ifdef CONFIG_PROVE_LOCKING 5227 + sizeof(lock_cq) 5228 + sizeof(lock_chains) 5229 + sizeof(lock_chains_in_use) 5230 + sizeof(chain_hlocks) 5231 #endif 5232 ) / 1024 5233 ); 5234 5235 #if defined(CONFIG_TRACE_IRQFLAGS) && defined(CONFIG_PROVE_LOCKING) 5236 printk(" memory used for stack traces: %zu kB\n", 5237 (sizeof(stack_trace) + sizeof(stack_trace_hash)) / 1024 5238 ); 5239 #endif 5240 5241 printk(" per task-struct memory footprint: %zu bytes\n", 5242 sizeof(((struct task_struct *)NULL)->held_locks)); 5243 } 5244 5245 static void 5246 print_freed_lock_bug(struct task_struct *curr, const void *mem_from, 5247 const void *mem_to, struct held_lock *hlock) 5248 { 5249 if (!debug_locks_off()) 5250 return; 5251 if (debug_locks_silent) 5252 return; 5253 5254 pr_warn("\n"); 5255 pr_warn("=========================\n"); 5256 pr_warn("WARNING: held lock freed!\n"); 5257 print_kernel_ident(); 5258 pr_warn("-------------------------\n"); 5259 pr_warn("%s/%d is freeing memory %px-%px, with a lock still held there!\n", 5260 curr->comm, task_pid_nr(curr), mem_from, mem_to-1); 5261 print_lock(hlock); 5262 lockdep_print_held_locks(curr); 5263 5264 pr_warn("\nstack backtrace:\n"); 5265 dump_stack(); 5266 } 5267 5268 static inline int not_in_range(const void* mem_from, unsigned long mem_len, 5269 const void* lock_from, unsigned long lock_len) 5270 { 5271 return lock_from + lock_len <= mem_from || 5272 mem_from + mem_len <= lock_from; 5273 } 5274 5275 /* 5276 * Called when kernel memory is freed (or unmapped), or if a lock 5277 * is destroyed or reinitialized - this code checks whether there is 5278 * any held lock in the memory range of <from> to <to>: 5279 */ 5280 void debug_check_no_locks_freed(const void *mem_from, unsigned long mem_len) 5281 { 5282 struct task_struct *curr = current; 5283 struct held_lock *hlock; 5284 unsigned long flags; 5285 int i; 5286 5287 if (unlikely(!debug_locks)) 5288 return; 5289 5290 raw_local_irq_save(flags); 5291 for (i = 0; i < curr->lockdep_depth; i++) { 5292 hlock = curr->held_locks + i; 5293 5294 if (not_in_range(mem_from, mem_len, hlock->instance, 5295 sizeof(*hlock->instance))) 5296 continue; 5297 5298 print_freed_lock_bug(curr, mem_from, mem_from + mem_len, hlock); 5299 break; 5300 } 5301 raw_local_irq_restore(flags); 5302 } 5303 EXPORT_SYMBOL_GPL(debug_check_no_locks_freed); 5304 5305 static void print_held_locks_bug(void) 5306 { 5307 if (!debug_locks_off()) 5308 return; 5309 if (debug_locks_silent) 5310 return; 5311 5312 pr_warn("\n"); 5313 pr_warn("====================================\n"); 5314 pr_warn("WARNING: %s/%d still has locks held!\n", 5315 current->comm, task_pid_nr(current)); 5316 print_kernel_ident(); 5317 pr_warn("------------------------------------\n"); 5318 lockdep_print_held_locks(current); 5319 pr_warn("\nstack backtrace:\n"); 5320 dump_stack(); 5321 } 5322 5323 void debug_check_no_locks_held(void) 5324 { 5325 if (unlikely(current->lockdep_depth > 0)) 5326 print_held_locks_bug(); 5327 } 5328 EXPORT_SYMBOL_GPL(debug_check_no_locks_held); 5329 5330 #ifdef __KERNEL__ 5331 void debug_show_all_locks(void) 5332 { 5333 struct task_struct *g, *p; 5334 5335 if (unlikely(!debug_locks)) { 5336 pr_warn("INFO: lockdep is turned off.\n"); 5337 return; 5338 } 5339 pr_warn("\nShowing all locks held in the system:\n"); 5340 5341 rcu_read_lock(); 5342 for_each_process_thread(g, p) { 5343 if (!p->lockdep_depth) 5344 continue; 5345 lockdep_print_held_locks(p); 5346 touch_nmi_watchdog(); 5347 touch_all_softlockup_watchdogs(); 5348 } 5349 rcu_read_unlock(); 5350 5351 pr_warn("\n"); 5352 pr_warn("=============================================\n\n"); 5353 } 5354 EXPORT_SYMBOL_GPL(debug_show_all_locks); 5355 #endif 5356 5357 /* 5358 * Careful: only use this function if you are sure that 5359 * the task cannot run in parallel! 5360 */ 5361 void debug_show_held_locks(struct task_struct *task) 5362 { 5363 if (unlikely(!debug_locks)) { 5364 printk("INFO: lockdep is turned off.\n"); 5365 return; 5366 } 5367 lockdep_print_held_locks(task); 5368 } 5369 EXPORT_SYMBOL_GPL(debug_show_held_locks); 5370 5371 asmlinkage __visible void lockdep_sys_exit(void) 5372 { 5373 struct task_struct *curr = current; 5374 5375 if (unlikely(curr->lockdep_depth)) { 5376 if (!debug_locks_off()) 5377 return; 5378 pr_warn("\n"); 5379 pr_warn("================================================\n"); 5380 pr_warn("WARNING: lock held when returning to user space!\n"); 5381 print_kernel_ident(); 5382 pr_warn("------------------------------------------------\n"); 5383 pr_warn("%s/%d is leaving the kernel with locks still held!\n", 5384 curr->comm, curr->pid); 5385 lockdep_print_held_locks(curr); 5386 } 5387 5388 /* 5389 * The lock history for each syscall should be independent. So wipe the 5390 * slate clean on return to userspace. 5391 */ 5392 lockdep_invariant_state(false); 5393 } 5394 5395 void lockdep_rcu_suspicious(const char *file, const int line, const char *s) 5396 { 5397 struct task_struct *curr = current; 5398 5399 /* Note: the following can be executed concurrently, so be careful. */ 5400 pr_warn("\n"); 5401 pr_warn("=============================\n"); 5402 pr_warn("WARNING: suspicious RCU usage\n"); 5403 print_kernel_ident(); 5404 pr_warn("-----------------------------\n"); 5405 pr_warn("%s:%d %s!\n", file, line, s); 5406 pr_warn("\nother info that might help us debug this:\n\n"); 5407 pr_warn("\n%srcu_scheduler_active = %d, debug_locks = %d\n", 5408 !rcu_lockdep_current_cpu_online() 5409 ? "RCU used illegally from offline CPU!\n" 5410 : !rcu_is_watching() 5411 ? "RCU used illegally from idle CPU!\n" 5412 : "", 5413 rcu_scheduler_active, debug_locks); 5414 5415 /* 5416 * If a CPU is in the RCU-free window in idle (ie: in the section 5417 * between rcu_idle_enter() and rcu_idle_exit(), then RCU 5418 * considers that CPU to be in an "extended quiescent state", 5419 * which means that RCU will be completely ignoring that CPU. 5420 * Therefore, rcu_read_lock() and friends have absolutely no 5421 * effect on a CPU running in that state. In other words, even if 5422 * such an RCU-idle CPU has called rcu_read_lock(), RCU might well 5423 * delete data structures out from under it. RCU really has no 5424 * choice here: we need to keep an RCU-free window in idle where 5425 * the CPU may possibly enter into low power mode. This way we can 5426 * notice an extended quiescent state to other CPUs that started a grace 5427 * period. Otherwise we would delay any grace period as long as we run 5428 * in the idle task. 5429 * 5430 * So complain bitterly if someone does call rcu_read_lock(), 5431 * rcu_read_lock_bh() and so on from extended quiescent states. 5432 */ 5433 if (!rcu_is_watching()) 5434 pr_warn("RCU used illegally from extended quiescent state!\n"); 5435 5436 lockdep_print_held_locks(curr); 5437 pr_warn("\nstack backtrace:\n"); 5438 dump_stack(); 5439 } 5440 EXPORT_SYMBOL_GPL(lockdep_rcu_suspicious); 5441