projects/linux·blob v5.13 · refs

6438 lines · 162.4 KiB

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 <[email protected]>
10 * Copyright (C) 2007 Red Hat, Inc., Peter Zijlstra
11 *
12 * this code maps all the lock dependencies as they occur in a live kernel
13 * and will warn about the following classes of locking bugs:
14 *
15 * - lock inversion scenarios
16 * - circular lock dependencies
17 * - hardirq/softirq safe/unsafe locking bugs
18 *
19 * Bugs are reported even if the current locking scenario does not cause
20 * any deadlock at this point.
21 *
22 * I.e. if anytime in the past two locks were taken in a different order,
23 * even if it happened for another task, even if those were different
24 * locks (but of the same class as this lock), this code will detect it.
25 *
26 * Thanks to Arjan van de Ven for coming up with the initial idea of
27 * mapping lock dependencies runtime.
28 */
29#define DISABLE_BRANCH_PROFILING
30#include <linux/mutex.h>
31#include <linux/sched.h>
32#include <linux/sched/clock.h>
33#include <linux/sched/task.h>
34#include <linux/sched/mm.h>
35#include <linux/delay.h>
36#include <linux/module.h>
37#include <linux/proc_fs.h>
38#include <linux/seq_file.h>
39#include <linux/spinlock.h>
40#include <linux/kallsyms.h>
41#include <linux/interrupt.h>
42#include <linux/stacktrace.h>
43#include <linux/debug_locks.h>
44#include <linux/irqflags.h>
45#include <linux/utsname.h>
46#include <linux/hash.h>
47#include <linux/ftrace.h>
48#include <linux/stringify.h>
49#include <linux/bitmap.h>
50#include <linux/bitops.h>
51#include <linux/gfp.h>
52#include <linux/random.h>
53#include <linux/jhash.h>
54#include <linux/nmi.h>
55#include <linux/rcupdate.h>
56#include <linux/kprobes.h>
57#include <linux/lockdep.h>
58
59#include <asm/sections.h>
60
61#include "lockdep_internals.h"
62
63#define CREATE_TRACE_POINTS
64#include <trace/events/lock.h>
65
66#ifdef CONFIG_PROVE_LOCKING
67int prove_locking = 1;
68module_param(prove_locking, int, 0644);
69#else
70#define prove_locking 0
71#endif
72
73#ifdef CONFIG_LOCK_STAT
74int lock_stat = 1;
75module_param(lock_stat, int, 0644);
76#else
77#define lock_stat 0
78#endif
79
80DEFINE_PER_CPU(unsigned int, lockdep_recursion);
81EXPORT_PER_CPU_SYMBOL_GPL(lockdep_recursion);
82
83static __always_inline bool lockdep_enabled(void)
84{
85 if (!debug_locks)
86 return false;
87
88 if (this_cpu_read(lockdep_recursion))
89 return false;
90
91 if (current->lockdep_recursion)
92 return false;
93
94 return true;
95}
96
97/*
98 * lockdep_lock: protects the lockdep graph, the hashes and the
99 * class/list/hash allocators.
100 *
101 * This is one of the rare exceptions where it's justified
102 * to use a raw spinlock - we really dont want the spinlock
103 * code to recurse back into the lockdep code...
104 */
105static arch_spinlock_t __lock = (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
106static struct task_struct *__owner;
107
108static inline void lockdep_lock(void)
109{
110 DEBUG_LOCKS_WARN_ON(!irqs_disabled());
111
112 __this_cpu_inc(lockdep_recursion);
113 arch_spin_lock(&__lock);
114 __owner = current;
115}
116
117static inline void lockdep_unlock(void)
118{
119 DEBUG_LOCKS_WARN_ON(!irqs_disabled());
120
121 if (debug_locks && DEBUG_LOCKS_WARN_ON(__owner != current))
122 return;
123
124 __owner = NULL;
125 arch_spin_unlock(&__lock);
126 __this_cpu_dec(lockdep_recursion);
127}
128
129static inline bool lockdep_assert_locked(void)
130{
131 return DEBUG_LOCKS_WARN_ON(__owner != current);
132}
133
134static struct task_struct *lockdep_selftest_task_struct;
135
136
137static int graph_lock(void)
138{
139 lockdep_lock();
140 /*
141 * Make sure that if another CPU detected a bug while
142 * walking the graph we dont change it (while the other
143 * CPU is busy printing out stuff with the graph lock
144 * dropped already)
145 */
146 if (!debug_locks) {
147 lockdep_unlock();
148 return 0;
149 }
150 return 1;
151}
152
153static inline void graph_unlock(void)
154{
155 lockdep_unlock();
156}
157
158/*
159 * Turn lock debugging off and return with 0 if it was off already,
160 * and also release the graph lock:
161 */
162static inline int debug_locks_off_graph_unlock(void)
163{
164 int ret = debug_locks_off();
165
166 lockdep_unlock();
167
168 return ret;
169}
170
171unsigned long nr_list_entries;
172static struct lock_list list_entries[MAX_LOCKDEP_ENTRIES];
173static DECLARE_BITMAP(list_entries_in_use, MAX_LOCKDEP_ENTRIES);
174
175/*
176 * All data structures here are protected by the global debug_lock.
177 *
178 * nr_lock_classes is the number of elements of lock_classes[] that is
179 * in use.
180 */
181#define KEYHASH_BITS (MAX_LOCKDEP_KEYS_BITS - 1)
182#define KEYHASH_SIZE (1UL << KEYHASH_BITS)
183static struct hlist_head lock_keys_hash[KEYHASH_SIZE];
184unsigned long nr_lock_classes;
185unsigned long nr_zapped_classes;
186#ifndef CONFIG_DEBUG_LOCKDEP
187static
188#endif
189struct lock_class lock_classes[MAX_LOCKDEP_KEYS];
190static DECLARE_BITMAP(lock_classes_in_use, MAX_LOCKDEP_KEYS);
191
192static inline struct lock_class *hlock_class(struct held_lock *hlock)
193{
194 unsigned int class_idx = hlock->class_idx;
195
196 /* Don't re-read hlock->class_idx, can't use READ_ONCE() on bitfield */
197 barrier();
198
199 if (!test_bit(class_idx, lock_classes_in_use)) {
200 /*
201 * Someone passed in garbage, we give up.
202 */
203 DEBUG_LOCKS_WARN_ON(1);
204 return NULL;
205 }
206
207 /*
208 * At this point, if the passed hlock->class_idx is still garbage,
209 * we just have to live with it
210 */
211 return lock_classes + class_idx;
212}
213
214#ifdef CONFIG_LOCK_STAT
215static DEFINE_PER_CPU(struct lock_class_stats[MAX_LOCKDEP_KEYS], cpu_lock_stats);
216
217static inline u64 lockstat_clock(void)
218{
219 return local_clock();
220}
221
222static int lock_point(unsigned long points[], unsigned long ip)
223{
224 int i;
225
226 for (i = 0; i < LOCKSTAT_POINTS; i++) {
227 if (points[i] == 0) {
228 points[i] = ip;
229 break;
230 }
231 if (points[i] == ip)
232 break;
233 }
234
235 return i;
236}
237
238static void lock_time_inc(struct lock_time *lt, u64 time)
239{
240 if (time > lt->max)
241 lt->max = time;
242
243 if (time < lt->min || !lt->nr)
244 lt->min = time;
245
246 lt->total += time;
247 lt->nr++;
248}
249
250static inline void lock_time_add(struct lock_time *src, struct lock_time *dst)
251{
252 if (!src->nr)
253 return;
254
255 if (src->max > dst->max)
256 dst->max = src->max;
257
258 if (src->min < dst->min || !dst->nr)
259 dst->min = src->min;
260
261 dst->total += src->total;
262 dst->nr += src->nr;
263}
264
265struct lock_class_stats lock_stats(struct lock_class *class)
266{
267 struct lock_class_stats stats;
268 int cpu, i;
269
270 memset(&stats, 0, sizeof(struct lock_class_stats));
271 for_each_possible_cpu(cpu) {
272 struct lock_class_stats *pcs =
273 &per_cpu(cpu_lock_stats, cpu)[class - lock_classes];
274
275 for (i = 0; i < ARRAY_SIZE(stats.contention_point); i++)
276 stats.contention_point[i] += pcs->contention_point[i];
277
278 for (i = 0; i < ARRAY_SIZE(stats.contending_point); i++)
279 stats.contending_point[i] += pcs->contending_point[i];
280
281 lock_time_add(&pcs->read_waittime, &stats.read_waittime);
282 lock_time_add(&pcs->write_waittime, &stats.write_waittime);
283
284 lock_time_add(&pcs->read_holdtime, &stats.read_holdtime);
285 lock_time_add(&pcs->write_holdtime, &stats.write_holdtime);
286
287 for (i = 0; i < ARRAY_SIZE(stats.bounces); i++)
288 stats.bounces[i] += pcs->bounces[i];
289 }
290
291 return stats;
292}
293
294void clear_lock_stats(struct lock_class *class)
295{
296 int cpu;
297
298 for_each_possible_cpu(cpu) {
299 struct lock_class_stats *cpu_stats =
300 &per_cpu(cpu_lock_stats, cpu)[class - lock_classes];
301
302 memset(cpu_stats, 0, sizeof(struct lock_class_stats));
303 }
304 memset(class->contention_point, 0, sizeof(class->contention_point));
305 memset(class->contending_point, 0, sizeof(class->contending_point));
306}
307
308static struct lock_class_stats *get_lock_stats(struct lock_class *class)
309{
310 return &this_cpu_ptr(cpu_lock_stats)[class - lock_classes];
311}
312
313static void lock_release_holdtime(struct held_lock *hlock)
314{
315 struct lock_class_stats *stats;
316 u64 holdtime;
317
318 if (!lock_stat)
319 return;
320
321 holdtime = lockstat_clock() - hlock->holdtime_stamp;
322
323 stats = get_lock_stats(hlock_class(hlock));
324 if (hlock->read)
325 lock_time_inc(&stats->read_holdtime, holdtime);
326 else
327 lock_time_inc(&stats->write_holdtime, holdtime);
328}
329#else
330static inline void lock_release_holdtime(struct held_lock *hlock)
331{
332}
333#endif
334
335/*
336 * We keep a global list of all lock classes. The list is only accessed with
337 * the lockdep spinlock lock held. free_lock_classes is a list with free
338 * elements. These elements are linked together by the lock_entry member in
339 * struct lock_class.
340 */
341LIST_HEAD(all_lock_classes);
342static LIST_HEAD(free_lock_classes);
343
344/**
345 * struct pending_free - information about data structures about to be freed
346 * @zapped: Head of a list with struct lock_class elements.
347 * @lock_chains_being_freed: Bitmap that indicates which lock_chains[] elements
348 * are about to be freed.
349 */
350struct pending_free {
351 struct list_head zapped;
352 DECLARE_BITMAP(lock_chains_being_freed, MAX_LOCKDEP_CHAINS);
353};
354
355/**
356 * struct delayed_free - data structures used for delayed freeing
357 *
358 * A data structure for delayed freeing of data structures that may be
359 * accessed by RCU readers at the time these were freed.
360 *
361 * @rcu_head: Used to schedule an RCU callback for freeing data structures.
362 * @index: Index of @pf to which freed data structures are added.
363 * @scheduled: Whether or not an RCU callback has been scheduled.
364 * @pf: Array with information about data structures about to be freed.
365 */
366static struct delayed_free {
367 struct rcu_head rcu_head;
368 int index;
369 int scheduled;
370 struct pending_free pf[2];
371} delayed_free;
372
373/*
374 * The lockdep classes are in a hash-table as well, for fast lookup:
375 */
376#define CLASSHASH_BITS (MAX_LOCKDEP_KEYS_BITS - 1)
377#define CLASSHASH_SIZE (1UL << CLASSHASH_BITS)
378#define __classhashfn(key) hash_long((unsigned long)key, CLASSHASH_BITS)
379#define classhashentry(key) (classhash_table + __classhashfn((key)))
380
381static struct hlist_head classhash_table[CLASSHASH_SIZE];
382
383/*
384 * We put the lock dependency chains into a hash-table as well, to cache
385 * their existence:
386 */
387#define CHAINHASH_BITS (MAX_LOCKDEP_CHAINS_BITS-1)
388#define CHAINHASH_SIZE (1UL << CHAINHASH_BITS)
389#define __chainhashfn(chain) hash_long(chain, CHAINHASH_BITS)
390#define chainhashentry(chain) (chainhash_table + __chainhashfn((chain)))
391
392static struct hlist_head chainhash_table[CHAINHASH_SIZE];
393
394/*
395 * the id of held_lock
396 */
397static inline u16 hlock_id(struct held_lock *hlock)
398{
399 BUILD_BUG_ON(MAX_LOCKDEP_KEYS_BITS + 2 > 16);
400
401 return (hlock->class_idx | (hlock->read << MAX_LOCKDEP_KEYS_BITS));
402}
403
404static inline unsigned int chain_hlock_class_idx(u16 hlock_id)
405{
406 return hlock_id & (MAX_LOCKDEP_KEYS - 1);
407}
408
409/*
410 * The hash key of the lock dependency chains is a hash itself too:
411 * it's a hash of all locks taken up to that lock, including that lock.
412 * It's a 64-bit hash, because it's important for the keys to be
413 * unique.
414 */
415static inline u64 iterate_chain_key(u64 key, u32 idx)
416{
417 u32 k0 = key, k1 = key >> 32;
418
419 __jhash_mix(idx, k0, k1); /* Macro that modifies arguments! */
420
421 return k0 | (u64)k1 << 32;
422}
423
424void lockdep_init_task(struct task_struct *task)
425{
426 task->lockdep_depth = 0; /* no locks held yet */
427 task->curr_chain_key = INITIAL_CHAIN_KEY;
428 task->lockdep_recursion = 0;
429}
430
431static __always_inline void lockdep_recursion_inc(void)
432{
433 __this_cpu_inc(lockdep_recursion);
434}
435
436static __always_inline void lockdep_recursion_finish(void)
437{
438 if (WARN_ON_ONCE(__this_cpu_dec_return(lockdep_recursion)))
439 __this_cpu_write(lockdep_recursion, 0);
440}
441
442void lockdep_set_selftest_task(struct task_struct *task)
443{
444 lockdep_selftest_task_struct = task;
445}
446
447/*
448 * Debugging switches:
449 */
450
451#define VERBOSE 0
452#define VERY_VERBOSE 0
453
454#if VERBOSE
455# define HARDIRQ_VERBOSE 1
456# define SOFTIRQ_VERBOSE 1
457#else
458# define HARDIRQ_VERBOSE 0
459# define SOFTIRQ_VERBOSE 0
460#endif
461
462#if VERBOSE || HARDIRQ_VERBOSE || SOFTIRQ_VERBOSE
463/*
464 * Quick filtering for interesting events:
465 */
466static int class_filter(struct lock_class *class)
467{
468#if 0
469 /* Example */
470 if (class->name_version == 1 &&
471 !strcmp(class->name, "lockname"))
472 return 1;
473 if (class->name_version == 1 &&
474 !strcmp(class->name, "&struct->lockfield"))
475 return 1;
476#endif
477 /* Filter everything else. 1 would be to allow everything else */
478 return 0;
479}
480#endif
481
482static int verbose(struct lock_class *class)
483{
484#if VERBOSE
485 return class_filter(class);
486#endif
487 return 0;
488}
489
490static void print_lockdep_off(const char *bug_msg)
491{
492 printk(KERN_DEBUG "%s\n", bug_msg);
493 printk(KERN_DEBUG "turning off the locking correctness validator.\n");
494#ifdef CONFIG_LOCK_STAT
495 printk(KERN_DEBUG "Please attach the output of /proc/lock_stat to the bug report\n");
496#endif
497}
498
499unsigned long nr_stack_trace_entries;
500
501#ifdef CONFIG_PROVE_LOCKING
502/**
503 * struct lock_trace - single stack backtrace
504 * @hash_entry: Entry in a stack_trace_hash[] list.
505 * @hash: jhash() of @entries.
506 * @nr_entries: Number of entries in @entries.
507 * @entries: Actual stack backtrace.
508 */
509struct lock_trace {
510 struct hlist_node hash_entry;
511 u32 hash;
512 u32 nr_entries;
513 unsigned long entries[] __aligned(sizeof(unsigned long));
514};
515#define LOCK_TRACE_SIZE_IN_LONGS \
516 (sizeof(struct lock_trace) / sizeof(unsigned long))
517/*
518 * Stack-trace: sequence of lock_trace structures. Protected by the graph_lock.
519 */
520static unsigned long stack_trace[MAX_STACK_TRACE_ENTRIES];
521static struct hlist_head stack_trace_hash[STACK_TRACE_HASH_SIZE];
522
523static bool traces_identical(struct lock_trace *t1, struct lock_trace *t2)
524{
525 return t1->hash == t2->hash && t1->nr_entries == t2->nr_entries &&
526 memcmp(t1->entries, t2->entries,
527 t1->nr_entries * sizeof(t1->entries[0])) == 0;
528}
529
530static struct lock_trace *save_trace(void)
531{
532 struct lock_trace *trace, *t2;
533 struct hlist_head *hash_head;
534 u32 hash;
535 int max_entries;
536
537 BUILD_BUG_ON_NOT_POWER_OF_2(STACK_TRACE_HASH_SIZE);
538 BUILD_BUG_ON(LOCK_TRACE_SIZE_IN_LONGS >= MAX_STACK_TRACE_ENTRIES);
539
540 trace = (struct lock_trace *)(stack_trace + nr_stack_trace_entries);
541 max_entries = MAX_STACK_TRACE_ENTRIES - nr_stack_trace_entries -
542 LOCK_TRACE_SIZE_IN_LONGS;
543
544 if (max_entries <= 0) {
545 if (!debug_locks_off_graph_unlock())
546 return NULL;
547
548 print_lockdep_off("BUG: MAX_STACK_TRACE_ENTRIES too low!");
549 dump_stack();
550
551 return NULL;
552 }
553 trace->nr_entries = stack_trace_save(trace->entries, max_entries, 3);
554
555 hash = jhash(trace->entries, trace->nr_entries *
556 sizeof(trace->entries[0]), 0);
557 trace->hash = hash;
558 hash_head = stack_trace_hash + (hash & (STACK_TRACE_HASH_SIZE - 1));
559 hlist_for_each_entry(t2, hash_head, hash_entry) {
560 if (traces_identical(trace, t2))
561 return t2;
562 }
563 nr_stack_trace_entries += LOCK_TRACE_SIZE_IN_LONGS + trace->nr_entries;
564 hlist_add_head(&trace->hash_entry, hash_head);
565
566 return trace;
567}
568
569/* Return the number of stack traces in the stack_trace[] array. */
570u64 lockdep_stack_trace_count(void)
571{
572 struct lock_trace *trace;
573 u64 c = 0;
574 int i;
575
576 for (i = 0; i < ARRAY_SIZE(stack_trace_hash); i++) {
577 hlist_for_each_entry(trace, &stack_trace_hash[i], hash_entry) {
578 c++;
579 }
580 }
581
582 return c;
583}
584
585/* Return the number of stack hash chains that have at least one stack trace. */
586u64 lockdep_stack_hash_count(void)
587{
588 u64 c = 0;
589 int i;
590
591 for (i = 0; i < ARRAY_SIZE(stack_trace_hash); i++)
592 if (!hlist_empty(&stack_trace_hash[i]))
593 c++;
594
595 return c;
596}
597#endif
598
599unsigned int nr_hardirq_chains;
600unsigned int nr_softirq_chains;
601unsigned int nr_process_chains;
602unsigned int max_lockdep_depth;
603
604#ifdef CONFIG_DEBUG_LOCKDEP
605/*
606 * Various lockdep statistics:
607 */
608DEFINE_PER_CPU(struct lockdep_stats, lockdep_stats);
609#endif
610
611#ifdef CONFIG_PROVE_LOCKING
612/*
613 * Locking printouts:
614 */
615
616#define __USAGE(__STATE) \
617 [LOCK_USED_IN_##__STATE] = "IN-"__stringify(__STATE)"-W", \
618 [LOCK_ENABLED_##__STATE] = __stringify(__STATE)"-ON-W", \
619 [LOCK_USED_IN_##__STATE##_READ] = "IN-"__stringify(__STATE)"-R",\
620 [LOCK_ENABLED_##__STATE##_READ] = __stringify(__STATE)"-ON-R",
621
622static const char *usage_str[] =
623{
624#define LOCKDEP_STATE(__STATE) __USAGE(__STATE)
625#include "lockdep_states.h"
626#undef LOCKDEP_STATE
627 [LOCK_USED] = "INITIAL USE",
628 [LOCK_USED_READ] = "INITIAL READ USE",
629 /* abused as string storage for verify_lock_unused() */
630 [LOCK_USAGE_STATES] = "IN-NMI",
631};
632#endif
633
634const char *__get_key_name(const struct lockdep_subclass_key *key, char *str)
635{
636 return kallsyms_lookup((unsigned long)key, NULL, NULL, NULL, str);
637}
638
639static inline unsigned long lock_flag(enum lock_usage_bit bit)
640{
641 return 1UL << bit;
642}
643
644static char get_usage_char(struct lock_class *class, enum lock_usage_bit bit)
645{
646 /*
647 * The usage character defaults to '.' (i.e., irqs disabled and not in
648 * irq context), which is the safest usage category.
649 */
650 char c = '.';
651
652 /*
653 * The order of the following usage checks matters, which will
654 * result in the outcome character as follows:
655 *
656 * - '+': irq is enabled and not in irq context
657 * - '-': in irq context and irq is disabled
658 * - '?': in irq context and irq is enabled
659 */
660 if (class->usage_mask & lock_flag(bit + LOCK_USAGE_DIR_MASK)) {
661 c = '+';
662 if (class->usage_mask & lock_flag(bit))
663 c = '?';
664 } else if (class->usage_mask & lock_flag(bit))
665 c = '-';
666
667 return c;
668}
669
670void get_usage_chars(struct lock_class *class, char usage[LOCK_USAGE_CHARS])
671{
672 int i = 0;
673
674#define LOCKDEP_STATE(__STATE) \
675 usage[i++] = get_usage_char(class, LOCK_USED_IN_##__STATE); \
676 usage[i++] = get_usage_char(class, LOCK_USED_IN_##__STATE##_READ);
677#include "lockdep_states.h"
678#undef LOCKDEP_STATE
679
680 usage[i] = '\0';
681}
682
683static void __print_lock_name(struct lock_class *class)
684{
685 char str[KSYM_NAME_LEN];
686 const char *name;
687
688 name = class->name;
689 if (!name) {
690 name = __get_key_name(class->key, str);
691 printk(KERN_CONT "%s", name);
692 } else {
693 printk(KERN_CONT "%s", name);
694 if (class->name_version > 1)
695 printk(KERN_CONT "#%d", class->name_version);
696 if (class->subclass)
697 printk(KERN_CONT "/%d", class->subclass);
698 }
699}
700
701static void print_lock_name(struct lock_class *class)
702{
703 char usage[LOCK_USAGE_CHARS];
704
705 get_usage_chars(class, usage);
706
707 printk(KERN_CONT " (");
708 __print_lock_name(class);
709 printk(KERN_CONT "){%s}-{%d:%d}", usage,
710 class->wait_type_outer ?: class->wait_type_inner,
711 class->wait_type_inner);
712}
713
714static void print_lockdep_cache(struct lockdep_map *lock)
715{
716 const char *name;
717 char str[KSYM_NAME_LEN];
718
719 name = lock->name;
720 if (!name)
721 name = __get_key_name(lock->key->subkeys, str);
722
723 printk(KERN_CONT "%s", name);
724}
725
726static void print_lock(struct held_lock *hlock)
727{
728 /*
729 * We can be called locklessly through debug_show_all_locks() so be
730 * extra careful, the hlock might have been released and cleared.
731 *
732 * If this indeed happens, lets pretend it does not hurt to continue
733 * to print the lock unless the hlock class_idx does not point to a
734 * registered class. The rationale here is: since we don't attempt
735 * to distinguish whether we are in this situation, if it just
736 * happened we can't count on class_idx to tell either.
737 */
738 struct lock_class *lock = hlock_class(hlock);
739
740 if (!lock) {
741 printk(KERN_CONT "<RELEASED>\n");
742 return;
743 }
744
745 printk(KERN_CONT "%px", hlock->instance);
746 print_lock_name(lock);
747 printk(KERN_CONT ", at: %pS\n", (void *)hlock->acquire_ip);
748}
749
750static void lockdep_print_held_locks(struct task_struct *p)
751{
752 int i, depth = READ_ONCE(p->lockdep_depth);
753
754 if (!depth)
755 printk("no locks held by %s/%d.\n", p->comm, task_pid_nr(p));
756 else
757 printk("%d lock%s held by %s/%d:\n", depth,
758 depth > 1 ? "s" : "", p->comm, task_pid_nr(p));
759 /*
760 * It's not reliable to print a task's held locks if it's not sleeping
761 * and it's not the current task.
762 */
763 if (p->state == TASK_RUNNING && p != current)
764 return;
765 for (i = 0; i < depth; i++) {
766 printk(" #%d: ", i);
767 print_lock(p->held_locks + i);
768 }
769}
770
771static void print_kernel_ident(void)
772{
773 printk("%s %.*s %s\n", init_utsname()->release,
774 (int)strcspn(init_utsname()->version, " "),
775 init_utsname()->version,
776 print_tainted());
777}
778
779static int very_verbose(struct lock_class *class)
780{
781#if VERY_VERBOSE
782 return class_filter(class);
783#endif
784 return 0;
785}
786
787/*
788 * Is this the address of a static object:
789 */
790#ifdef __KERNEL__
791static int static_obj(const void *obj)
792{
793 unsigned long start = (unsigned long) &_stext,
794 end = (unsigned long) &_end,
795 addr = (unsigned long) obj;
796
797 if (arch_is_kernel_initmem_freed(addr))
798 return 0;
799
800 /*
801 * static variable?
802 */
803 if ((addr >= start) && (addr < end))
804 return 1;
805
806 if (arch_is_kernel_data(addr))
807 return 1;
808
809 /*
810 * in-kernel percpu var?
811 */
812 if (is_kernel_percpu_address(addr))
813 return 1;
814
815 /*
816 * module static or percpu var?
817 */
818 return is_module_address(addr) || is_module_percpu_address(addr);
819}
820#endif
821
822/*
823 * To make lock name printouts unique, we calculate a unique
824 * class->name_version generation counter. The caller must hold the graph
825 * lock.
826 */
827static int count_matching_names(struct lock_class *new_class)
828{
829 struct lock_class *class;
830 int count = 0;
831
832 if (!new_class->name)
833 return 0;
834
835 list_for_each_entry(class, &all_lock_classes, lock_entry) {
836 if (new_class->key - new_class->subclass == class->key)
837 return class->name_version;
838 if (class->name && !strcmp(class->name, new_class->name))
839 count = max(count, class->name_version);
840 }
841
842 return count + 1;
843}
844
845/* used from NMI context -- must be lockless */
846static noinstr struct lock_class *
847look_up_lock_class(const struct lockdep_map *lock, unsigned int subclass)
848{
849 struct lockdep_subclass_key *key;
850 struct hlist_head *hash_head;
851 struct lock_class *class;
852
853 if (unlikely(subclass >= MAX_LOCKDEP_SUBCLASSES)) {
854 instrumentation_begin();
855 debug_locks_off();
856 printk(KERN_ERR
857 "BUG: looking up invalid subclass: %u\n", subclass);
858 printk(KERN_ERR
859 "turning off the locking correctness validator.\n");
860 dump_stack();
861 instrumentation_end();
862 return NULL;
863 }
864
865 /*
866 * If it is not initialised then it has never been locked,
867 * so it won't be present in the hash table.
868 */
869 if (unlikely(!lock->key))
870 return NULL;
871
872 /*
873 * NOTE: the class-key must be unique. For dynamic locks, a static
874 * lock_class_key variable is passed in through the mutex_init()
875 * (or spin_lock_init()) call - which acts as the key. For static
876 * locks we use the lock object itself as the key.
877 */
878 BUILD_BUG_ON(sizeof(struct lock_class_key) >
879 sizeof(struct lockdep_map));
880
881 key = lock->key->subkeys + subclass;
882
883 hash_head = classhashentry(key);
884
885 /*
886 * We do an RCU walk of the hash, see lockdep_free_key_range().
887 */
888 if (DEBUG_LOCKS_WARN_ON(!irqs_disabled()))
889 return NULL;
890
891 hlist_for_each_entry_rcu(class, hash_head, hash_entry) {
892 if (class->key == key) {
893 /*
894 * Huh! same key, different name? Did someone trample
895 * on some memory? We're most confused.
896 */
897 WARN_ON_ONCE(class->name != lock->name &&
898 lock->key != &__lockdep_no_validate__);
899 return class;
900 }
901 }
902
903 return NULL;
904}
905
906/*
907 * Static locks do not have their class-keys yet - for them the key is
908 * the lock object itself. If the lock is in the per cpu area, the
909 * canonical address of the lock (per cpu offset removed) is used.
910 */
911static bool assign_lock_key(struct lockdep_map *lock)
912{
913 unsigned long can_addr, addr = (unsigned long)lock;
914
915#ifdef __KERNEL__
916 /*
917 * lockdep_free_key_range() assumes that struct lock_class_key
918 * objects do not overlap. Since we use the address of lock
919 * objects as class key for static objects, check whether the
920 * size of lock_class_key objects does not exceed the size of
921 * the smallest lock object.
922 */
923 BUILD_BUG_ON(sizeof(struct lock_class_key) > sizeof(raw_spinlock_t));
924#endif
925
926 if (__is_kernel_percpu_address(addr, &can_addr))
927 lock->key = (void *)can_addr;
928 else if (__is_module_percpu_address(addr, &can_addr))
929 lock->key = (void *)can_addr;
930 else if (static_obj(lock))
931 lock->key = (void *)lock;
932 else {
933 /* Debug-check: all keys must be persistent! */
934 debug_locks_off();
935 pr_err("INFO: trying to register non-static key.\n");
936 pr_err("The code is fine but needs lockdep annotation, or maybe\n");
937 pr_err("you didn't initialize this object before use?\n");
938 pr_err("turning off the locking correctness validator.\n");
939 dump_stack();
940 return false;
941 }
942
943 return true;
944}
945
946#ifdef CONFIG_DEBUG_LOCKDEP
947
948/* Check whether element @e occurs in list @h */
949static bool in_list(struct list_head *e, struct list_head *h)
950{
951 struct list_head *f;
952
953 list_for_each(f, h) {
954 if (e == f)
955 return true;
956 }
957
958 return false;
959}
960
961/*
962 * Check whether entry @e occurs in any of the locks_after or locks_before
963 * lists.
964 */
965static bool in_any_class_list(struct list_head *e)
966{
967 struct lock_class *class;
968 int i;
969
970 for (i = 0; i < ARRAY_SIZE(lock_classes); i++) {
971 class = &lock_classes[i];
972 if (in_list(e, &class->locks_after) ||
973 in_list(e, &class->locks_before))
974 return true;
975 }
976 return false;
977}
978
979static bool class_lock_list_valid(struct lock_class *c, struct list_head *h)
980{
981 struct lock_list *e;
982
983 list_for_each_entry(e, h, entry) {
984 if (e->links_to != c) {
985 printk(KERN_INFO "class %s: mismatch for lock entry %ld; class %s <> %s",
986 c->name ? : "(?)",
987 (unsigned long)(e - list_entries),
988 e->links_to && e->links_to->name ?
989 e->links_to->name : "(?)",
990 e->class && e->class->name ? e->class->name :
991 "(?)");
992 return false;
993 }
994 }
995 return true;
996}
997
998#ifdef CONFIG_PROVE_LOCKING
999static u16 chain_hlocks[MAX_LOCKDEP_CHAIN_HLOCKS];
1000#endif
1001
1002static bool check_lock_chain_key(struct lock_chain *chain)
1003{
1004#ifdef CONFIG_PROVE_LOCKING
1005 u64 chain_key = INITIAL_CHAIN_KEY;
1006 int i;
1007
1008 for (i = chain->base; i < chain->base + chain->depth; i++)
1009 chain_key = iterate_chain_key(chain_key, chain_hlocks[i]);
1010 /*
1011 * The 'unsigned long long' casts avoid that a compiler warning
1012 * is reported when building tools/lib/lockdep.
1013 */
1014 if (chain->chain_key != chain_key) {
1015 printk(KERN_INFO "chain %lld: key %#llx <> %#llx\n",
1016 (unsigned long long)(chain - lock_chains),
1017 (unsigned long long)chain->chain_key,
1018 (unsigned long long)chain_key);
1019 return false;
1020 }
1021#endif
1022 return true;
1023}
1024
1025static bool in_any_zapped_class_list(struct lock_class *class)
1026{
1027 struct pending_free *pf;
1028 int i;
1029
1030 for (i = 0, pf = delayed_free.pf; i < ARRAY_SIZE(delayed_free.pf); i++, pf++) {
1031 if (in_list(&class->lock_entry, &pf->zapped))
1032 return true;
1033 }
1034
1035 return false;
1036}
1037
1038static bool __check_data_structures(void)
1039{
1040 struct lock_class *class;
1041 struct lock_chain *chain;
1042 struct hlist_head *head;
1043 struct lock_list *e;
1044 int i;
1045
1046 /* Check whether all classes occur in a lock list. */
1047 for (i = 0; i < ARRAY_SIZE(lock_classes); i++) {
1048 class = &lock_classes[i];
1049 if (!in_list(&class->lock_entry, &all_lock_classes) &&
1050 !in_list(&class->lock_entry, &free_lock_classes) &&
1051 !in_any_zapped_class_list(class)) {
1052 printk(KERN_INFO "class %px/%s is not in any class list\n",
1053 class, class->name ? : "(?)");
1054 return false;
1055 }
1056 }
1057
1058 /* Check whether all classes have valid lock lists. */
1059 for (i = 0; i < ARRAY_SIZE(lock_classes); i++) {
1060 class = &lock_classes[i];
1061 if (!class_lock_list_valid(class, &class->locks_before))
1062 return false;
1063 if (!class_lock_list_valid(class, &class->locks_after))
1064 return false;
1065 }
1066
1067 /* Check the chain_key of all lock chains. */
1068 for (i = 0; i < ARRAY_SIZE(chainhash_table); i++) {
1069 head = chainhash_table + i;
1070 hlist_for_each_entry_rcu(chain, head, entry) {
1071 if (!check_lock_chain_key(chain))
1072 return false;
1073 }
1074 }
1075
1076 /*
1077 * Check whether all list entries that are in use occur in a class
1078 * lock list.
1079 */
1080 for_each_set_bit(i, list_entries_in_use, ARRAY_SIZE(list_entries)) {
1081 e = list_entries + i;
1082 if (!in_any_class_list(&e->entry)) {
1083 printk(KERN_INFO "list entry %d is not in any class list; class %s <> %s\n",
1084 (unsigned int)(e - list_entries),
1085 e->class->name ? : "(?)",
1086 e->links_to->name ? : "(?)");
1087 return false;
1088 }
1089 }
1090
1091 /*
1092 * Check whether all list entries that are not in use do not occur in
1093 * a class lock list.
1094 */
1095 for_each_clear_bit(i, list_entries_in_use, ARRAY_SIZE(list_entries)) {
1096 e = list_entries + i;
1097 if (in_any_class_list(&e->entry)) {
1098 printk(KERN_INFO "list entry %d occurs in a class list; class %s <> %s\n",
1099 (unsigned int)(e - list_entries),
1100 e->class && e->class->name ? e->class->name :
1101 "(?)",
1102 e->links_to && e->links_to->name ?
1103 e->links_to->name : "(?)");
1104 return false;
1105 }
1106 }
1107
1108 return true;
1109}
1110
1111int check_consistency = 0;
1112module_param(check_consistency, int, 0644);
1113
1114static void check_data_structures(void)
1115{
1116 static bool once = false;
1117
1118 if (check_consistency && !once) {
1119 if (!__check_data_structures()) {
1120 once = true;
1121 WARN_ON(once);
1122 }
1123 }
1124}
1125
1126#else /* CONFIG_DEBUG_LOCKDEP */
1127
1128static inline void check_data_structures(void) { }
1129
1130#endif /* CONFIG_DEBUG_LOCKDEP */
1131
1132static void init_chain_block_buckets(void);
1133
1134/*
1135 * Initialize the lock_classes[] array elements, the free_lock_classes list
1136 * and also the delayed_free structure.
1137 */
1138static void init_data_structures_once(void)
1139{
1140 static bool __read_mostly ds_initialized, rcu_head_initialized;
1141 int i;
1142
1143 if (likely(rcu_head_initialized))
1144 return;
1145
1146 if (system_state >= SYSTEM_SCHEDULING) {
1147 init_rcu_head(&delayed_free.rcu_head);
1148 rcu_head_initialized = true;
1149 }
1150
1151 if (ds_initialized)
1152 return;
1153
1154 ds_initialized = true;
1155
1156 INIT_LIST_HEAD(&delayed_free.pf[0].zapped);
1157 INIT_LIST_HEAD(&delayed_free.pf[1].zapped);
1158
1159 for (i = 0; i < ARRAY_SIZE(lock_classes); i++) {
1160 list_add_tail(&lock_classes[i].lock_entry, &free_lock_classes);
1161 INIT_LIST_HEAD(&lock_classes[i].locks_after);
1162 INIT_LIST_HEAD(&lock_classes[i].locks_before);
1163 }
1164 init_chain_block_buckets();
1165}
1166
1167static inline struct hlist_head *keyhashentry(const struct lock_class_key *key)
1168{
1169 unsigned long hash = hash_long((uintptr_t)key, KEYHASH_BITS);
1170
1171 return lock_keys_hash + hash;
1172}
1173
1174/* Register a dynamically allocated key. */
1175void lockdep_register_key(struct lock_class_key *key)
1176{
1177 struct hlist_head *hash_head;
1178 struct lock_class_key *k;
1179 unsigned long flags;
1180
1181 if (WARN_ON_ONCE(static_obj(key)))
1182 return;
1183 hash_head = keyhashentry(key);
1184
1185 raw_local_irq_save(flags);
1186 if (!graph_lock())
1187 goto restore_irqs;
1188 hlist_for_each_entry_rcu(k, hash_head, hash_entry) {
1189 if (WARN_ON_ONCE(k == key))
1190 goto out_unlock;
1191 }
1192 hlist_add_head_rcu(&key->hash_entry, hash_head);
1193out_unlock:
1194 graph_unlock();
1195restore_irqs:
1196 raw_local_irq_restore(flags);
1197}
1198EXPORT_SYMBOL_GPL(lockdep_register_key);
1199
1200/* Check whether a key has been registered as a dynamic key. */
1201static bool is_dynamic_key(const struct lock_class_key *key)
1202{
1203 struct hlist_head *hash_head;
1204 struct lock_class_key *k;
1205 bool found = false;
1206
1207 if (WARN_ON_ONCE(static_obj(key)))
1208 return false;
1209
1210 /*
1211 * If lock debugging is disabled lock_keys_hash[] may contain
1212 * pointers to memory that has already been freed. Avoid triggering
1213 * a use-after-free in that case by returning early.
1214 */
1215 if (!debug_locks)
1216 return true;
1217
1218 hash_head = keyhashentry(key);
1219
1220 rcu_read_lock();
1221 hlist_for_each_entry_rcu(k, hash_head, hash_entry) {
1222 if (k == key) {
1223 found = true;
1224 break;
1225 }
1226 }
1227 rcu_read_unlock();
1228
1229 return found;
1230}
1231
1232/*
1233 * Register a lock's class in the hash-table, if the class is not present
1234 * yet. Otherwise we look it up. We cache the result in the lock object
1235 * itself, so actual lookup of the hash should be once per lock object.
1236 */
1237static struct lock_class *
1238register_lock_class(struct lockdep_map *lock, unsigned int subclass, int force)
1239{
1240 struct lockdep_subclass_key *key;
1241 struct hlist_head *hash_head;
1242 struct lock_class *class;
1243
1244 DEBUG_LOCKS_WARN_ON(!irqs_disabled());
1245
1246 class = look_up_lock_class(lock, subclass);
1247 if (likely(class))
1248 goto out_set_class_cache;
1249
1250 if (!lock->key) {
1251 if (!assign_lock_key(lock))
1252 return NULL;
1253 } else if (!static_obj(lock->key) && !is_dynamic_key(lock->key)) {
1254 return NULL;
1255 }
1256
1257 key = lock->key->subkeys + subclass;
1258 hash_head = classhashentry(key);
1259
1260 if (!graph_lock()) {
1261 return NULL;
1262 }
1263 /*
1264 * We have to do the hash-walk again, to avoid races
1265 * with another CPU:
1266 */
1267 hlist_for_each_entry_rcu(class, hash_head, hash_entry) {
1268 if (class->key == key)
1269 goto out_unlock_set;
1270 }
1271
1272 init_data_structures_once();
1273
1274 /* Allocate a new lock class and add it to the hash. */
1275 class = list_first_entry_or_null(&free_lock_classes, typeof(*class),
1276 lock_entry);
1277 if (!class) {
1278 if (!debug_locks_off_graph_unlock()) {
1279 return NULL;
1280 }
1281
1282 print_lockdep_off("BUG: MAX_LOCKDEP_KEYS too low!");
1283 dump_stack();
1284 return NULL;
1285 }
1286 nr_lock_classes++;
1287 __set_bit(class - lock_classes, lock_classes_in_use);
1288 debug_atomic_inc(nr_unused_locks);
1289 class->key = key;
1290 class->name = lock->name;
1291 class->subclass = subclass;
1292 WARN_ON_ONCE(!list_empty(&class->locks_before));
1293 WARN_ON_ONCE(!list_empty(&class->locks_after));
1294 class->name_version = count_matching_names(class);
1295 class->wait_type_inner = lock->wait_type_inner;
1296 class->wait_type_outer = lock->wait_type_outer;
1297 class->lock_type = lock->lock_type;
1298 /*
1299 * We use RCU's safe list-add method to make
1300 * parallel walking of the hash-list safe:
1301 */
1302 hlist_add_head_rcu(&class->hash_entry, hash_head);
1303 /*
1304 * Remove the class from the free list and add it to the global list
1305 * of classes.
1306 */
1307 list_move_tail(&class->lock_entry, &all_lock_classes);
1308
1309 if (verbose(class)) {
1310 graph_unlock();
1311
1312 printk("\nnew class %px: %s", class->key, class->name);
1313 if (class->name_version > 1)
1314 printk(KERN_CONT "#%d", class->name_version);
1315 printk(KERN_CONT "\n");
1316 dump_stack();
1317
1318 if (!graph_lock()) {
1319 return NULL;
1320 }
1321 }
1322out_unlock_set:
1323 graph_unlock();
1324
1325out_set_class_cache:
1326 if (!subclass || force)
1327 lock->class_cache[0] = class;
1328 else if (subclass < NR_LOCKDEP_CACHING_CLASSES)
1329 lock->class_cache[subclass] = class;
1330
1331 /*
1332 * Hash collision, did we smoke some? We found a class with a matching
1333 * hash but the subclass -- which is hashed in -- didn't match.
1334 */
1335 if (DEBUG_LOCKS_WARN_ON(class->subclass != subclass))
1336 return NULL;
1337
1338 return class;
1339}
1340
1341#ifdef CONFIG_PROVE_LOCKING
1342/*
1343 * Allocate a lockdep entry. (assumes the graph_lock held, returns
1344 * with NULL on failure)
1345 */
1346static struct lock_list *alloc_list_entry(void)
1347{
1348 int idx = find_first_zero_bit(list_entries_in_use,
1349 ARRAY_SIZE(list_entries));
1350
1351 if (idx >= ARRAY_SIZE(list_entries)) {
1352 if (!debug_locks_off_graph_unlock())
1353 return NULL;
1354
1355 print_lockdep_off("BUG: MAX_LOCKDEP_ENTRIES too low!");
1356 dump_stack();
1357 return NULL;
1358 }
1359 nr_list_entries++;
1360 __set_bit(idx, list_entries_in_use);
1361 return list_entries + idx;
1362}
1363
1364/*
1365 * Add a new dependency to the head of the list:
1366 */
1367static int add_lock_to_list(struct lock_class *this,
1368 struct lock_class *links_to, struct list_head *head,
1369 unsigned long ip, u16 distance, u8 dep,
1370 const struct lock_trace *trace)
1371{
1372 struct lock_list *entry;
1373 /*
1374 * Lock not present yet - get a new dependency struct and
1375 * add it to the list:
1376 */
1377 entry = alloc_list_entry();
1378 if (!entry)
1379 return 0;
1380
1381 entry->class = this;
1382 entry->links_to = links_to;
1383 entry->dep = dep;
1384 entry->distance = distance;
1385 entry->trace = trace;
1386 /*
1387 * Both allocation and removal are done under the graph lock; but
1388 * iteration is under RCU-sched; see look_up_lock_class() and
1389 * lockdep_free_key_range().
1390 */
1391 list_add_tail_rcu(&entry->entry, head);
1392
1393 return 1;
1394}
1395
1396/*
1397 * For good efficiency of modular, we use power of 2
1398 */
1399#define MAX_CIRCULAR_QUEUE_SIZE (1UL << CONFIG_LOCKDEP_CIRCULAR_QUEUE_BITS)
1400#define CQ_MASK (MAX_CIRCULAR_QUEUE_SIZE-1)
1401
1402/*
1403 * The circular_queue and helpers are used to implement graph
1404 * breadth-first search (BFS) algorithm, by which we can determine
1405 * whether there is a path from a lock to another. In deadlock checks,
1406 * a path from the next lock to be acquired to a previous held lock
1407 * indicates that adding the <prev> -> <next> lock dependency will
1408 * produce a circle in the graph. Breadth-first search instead of
1409 * depth-first search is used in order to find the shortest (circular)
1410 * path.
1411 */
1412struct circular_queue {
1413 struct lock_list *element[MAX_CIRCULAR_QUEUE_SIZE];
1414 unsigned int front, rear;
1415};
1416
1417static struct circular_queue lock_cq;
1418
1419unsigned int max_bfs_queue_depth;
1420
1421static unsigned int lockdep_dependency_gen_id;
1422
1423static inline void __cq_init(struct circular_queue *cq)
1424{
1425 cq->front = cq->rear = 0;
1426 lockdep_dependency_gen_id++;
1427}
1428
1429static inline int __cq_empty(struct circular_queue *cq)
1430{
1431 return (cq->front == cq->rear);
1432}
1433
1434static inline int __cq_full(struct circular_queue *cq)
1435{
1436 return ((cq->rear + 1) & CQ_MASK) == cq->front;
1437}
1438
1439static inline int __cq_enqueue(struct circular_queue *cq, struct lock_list *elem)
1440{
1441 if (__cq_full(cq))
1442 return -1;
1443
1444 cq->element[cq->rear] = elem;
1445 cq->rear = (cq->rear + 1) & CQ_MASK;
1446 return 0;
1447}
1448
1449/*
1450 * Dequeue an element from the circular_queue, return a lock_list if
1451 * the queue is not empty, or NULL if otherwise.
1452 */
1453static inline struct lock_list * __cq_dequeue(struct circular_queue *cq)
1454{
1455 struct lock_list * lock;
1456
1457 if (__cq_empty(cq))
1458 return NULL;
1459
1460 lock = cq->element[cq->front];
1461 cq->front = (cq->front + 1) & CQ_MASK;
1462
1463 return lock;
1464}
1465
1466static inline unsigned int __cq_get_elem_count(struct circular_queue *cq)
1467{
1468 return (cq->rear - cq->front) & CQ_MASK;
1469}
1470
1471static inline void mark_lock_accessed(struct lock_list *lock)
1472{
1473 lock->class->dep_gen_id = lockdep_dependency_gen_id;
1474}
1475
1476static inline void visit_lock_entry(struct lock_list *lock,
1477 struct lock_list *parent)
1478{
1479 lock->parent = parent;
1480}
1481
1482static inline unsigned long lock_accessed(struct lock_list *lock)
1483{
1484 return lock->class->dep_gen_id == lockdep_dependency_gen_id;
1485}
1486
1487static inline struct lock_list *get_lock_parent(struct lock_list *child)
1488{
1489 return child->parent;
1490}
1491
1492static inline int get_lock_depth(struct lock_list *child)
1493{
1494 int depth = 0;
1495 struct lock_list *parent;
1496
1497 while ((parent = get_lock_parent(child))) {
1498 child = parent;
1499 depth++;
1500 }
1501 return depth;
1502}
1503
1504/*
1505 * Return the forward or backward dependency list.
1506 *
1507 * @lock: the lock_list to get its class's dependency list
1508 * @offset: the offset to struct lock_class to determine whether it is
1509 * locks_after or locks_before
1510 */
1511static inline struct list_head *get_dep_list(struct lock_list *lock, int offset)
1512{
1513 void *lock_class = lock->class;
1514
1515 return lock_class + offset;
1516}
1517/*
1518 * Return values of a bfs search:
1519 *
1520 * BFS_E* indicates an error
1521 * BFS_R* indicates a result (match or not)
1522 *
1523 * BFS_EINVALIDNODE: Find a invalid node in the graph.
1524 *
1525 * BFS_EQUEUEFULL: The queue is full while doing the bfs.
1526 *
1527 * BFS_RMATCH: Find the matched node in the graph, and put that node into
1528 * *@target_entry.
1529 *
1530 * BFS_RNOMATCH: Haven't found the matched node and keep *@target_entry
1531 * _unchanged_.
1532 */
1533enum bfs_result {
1534 BFS_EINVALIDNODE = -2,
1535 BFS_EQUEUEFULL = -1,
1536 BFS_RMATCH = 0,
1537 BFS_RNOMATCH = 1,
1538};
1539
1540/*
1541 * bfs_result < 0 means error
1542 */
1543static inline bool bfs_error(enum bfs_result res)
1544{
1545 return res < 0;
1546}
1547
1548/*
1549 * DEP_*_BIT in lock_list::dep
1550 *
1551 * For dependency @prev -> @next:
1552 *
1553 * SR: @prev is shared reader (->read != 0) and @next is recursive reader
1554 * (->read == 2)
1555 * ER: @prev is exclusive locker (->read == 0) and @next is recursive reader
1556 * SN: @prev is shared reader and @next is non-recursive locker (->read != 2)
1557 * EN: @prev is exclusive locker and @next is non-recursive locker
1558 *
1559 * Note that we define the value of DEP_*_BITs so that:
1560 * bit0 is prev->read == 0
1561 * bit1 is next->read != 2
1562 */
1563#define DEP_SR_BIT (0 + (0 << 1)) /* 0 */
1564#define DEP_ER_BIT (1 + (0 << 1)) /* 1 */
1565#define DEP_SN_BIT (0 + (1 << 1)) /* 2 */
1566#define DEP_EN_BIT (1 + (1 << 1)) /* 3 */
1567
1568#define DEP_SR_MASK (1U << (DEP_SR_BIT))
1569#define DEP_ER_MASK (1U << (DEP_ER_BIT))
1570#define DEP_SN_MASK (1U << (DEP_SN_BIT))
1571#define DEP_EN_MASK (1U << (DEP_EN_BIT))
1572
1573static inline unsigned int
1574__calc_dep_bit(struct held_lock *prev, struct held_lock *next)
1575{
1576 return (prev->read == 0) + ((next->read != 2) << 1);
1577}
1578
1579static inline u8 calc_dep(struct held_lock *prev, struct held_lock *next)
1580{
1581 return 1U << __calc_dep_bit(prev, next);
1582}
1583
1584/*
1585 * calculate the dep_bit for backwards edges. We care about whether @prev is
1586 * shared and whether @next is recursive.
1587 */
1588static inline unsigned int
1589__calc_dep_bitb(struct held_lock *prev, struct held_lock *next)
1590{
1591 return (next->read != 2) + ((prev->read == 0) << 1);
1592}
1593
1594static inline u8 calc_depb(struct held_lock *prev, struct held_lock *next)
1595{
1596 return 1U << __calc_dep_bitb(prev, next);
1597}
1598
1599/*
1600 * Initialize a lock_list entry @lock belonging to @class as the root for a BFS
1601 * search.
1602 */
1603static inline void __bfs_init_root(struct lock_list *lock,
1604 struct lock_class *class)
1605{
1606 lock->class = class;
1607 lock->parent = NULL;
1608 lock->only_xr = 0;
1609}
1610
1611/*
1612 * Initialize a lock_list entry @lock based on a lock acquisition @hlock as the
1613 * root for a BFS search.
1614 *
1615 * ->only_xr of the initial lock node is set to @hlock->read == 2, to make sure
1616 * that <prev> -> @hlock and @hlock -> <whatever __bfs() found> is not -(*R)->
1617 * and -(S*)->.
1618 */
1619static inline void bfs_init_root(struct lock_list *lock,
1620 struct held_lock *hlock)
1621{
1622 __bfs_init_root(lock, hlock_class(hlock));
1623 lock->only_xr = (hlock->read == 2);
1624}
1625
1626/*
1627 * Similar to bfs_init_root() but initialize the root for backwards BFS.
1628 *
1629 * ->only_xr of the initial lock node is set to @hlock->read != 0, to make sure
1630 * that <next> -> @hlock and @hlock -> <whatever backwards BFS found> is not
1631 * -(*S)-> and -(R*)-> (reverse order of -(*R)-> and -(S*)->).
1632 */
1633static inline void bfs_init_rootb(struct lock_list *lock,
1634 struct held_lock *hlock)
1635{
1636 __bfs_init_root(lock, hlock_class(hlock));
1637 lock->only_xr = (hlock->read != 0);
1638}
1639
1640static inline struct lock_list *__bfs_next(struct lock_list *lock, int offset)
1641{
1642 if (!lock || !lock->parent)
1643 return NULL;
1644
1645 return list_next_or_null_rcu(get_dep_list(lock->parent, offset),
1646 &lock->entry, struct lock_list, entry);
1647}
1648
1649/*
1650 * Breadth-First Search to find a strong path in the dependency graph.
1651 *
1652 * @source_entry: the source of the path we are searching for.
1653 * @data: data used for the second parameter of @match function
1654 * @match: match function for the search
1655 * @target_entry: pointer to the target of a matched path
1656 * @offset: the offset to struct lock_class to determine whether it is
1657 * locks_after or locks_before
1658 *
1659 * We may have multiple edges (considering different kinds of dependencies,
1660 * e.g. ER and SN) between two nodes in the dependency graph. But
1661 * only the strong dependency path in the graph is relevant to deadlocks. A
1662 * strong dependency path is a dependency path that doesn't have two adjacent
1663 * dependencies as -(*R)-> -(S*)->, please see:
1664 *
1665 * Documentation/locking/lockdep-design.rst
1666 *
1667 * for more explanation of the definition of strong dependency paths
1668 *
1669 * In __bfs(), we only traverse in the strong dependency path:
1670 *
1671 * In lock_list::only_xr, we record whether the previous dependency only
1672 * has -(*R)-> in the search, and if it does (prev only has -(*R)->), we
1673 * filter out any -(S*)-> in the current dependency and after that, the
1674 * ->only_xr is set according to whether we only have -(*R)-> left.
1675 */
1676static enum bfs_result __bfs(struct lock_list *source_entry,
1677 void *data,
1678 bool (*match)(struct lock_list *entry, void *data),
1679 bool (*skip)(struct lock_list *entry, void *data),
1680 struct lock_list **target_entry,
1681 int offset)
1682{
1683 struct circular_queue *cq = &lock_cq;
1684 struct lock_list *lock = NULL;
1685 struct lock_list *entry;
1686 struct list_head *head;
1687 unsigned int cq_depth;
1688 bool first;
1689
1690 lockdep_assert_locked();
1691
1692 __cq_init(cq);
1693 __cq_enqueue(cq, source_entry);
1694
1695 while ((lock = __bfs_next(lock, offset)) || (lock = __cq_dequeue(cq))) {
1696 if (!lock->class)
1697 return BFS_EINVALIDNODE;
1698
1699 /*
1700 * Step 1: check whether we already finish on this one.
1701 *
1702 * If we have visited all the dependencies from this @lock to
1703 * others (iow, if we have visited all lock_list entries in
1704 * @lock->class->locks_{after,before}) we skip, otherwise go
1705 * and visit all the dependencies in the list and mark this
1706 * list accessed.
1707 */
1708 if (lock_accessed(lock))
1709 continue;
1710 else
1711 mark_lock_accessed(lock);
1712
1713 /*
1714 * Step 2: check whether prev dependency and this form a strong
1715 * dependency path.
1716 */
1717 if (lock->parent) { /* Parent exists, check prev dependency */
1718 u8 dep = lock->dep;
1719 bool prev_only_xr = lock->parent->only_xr;
1720
1721 /*
1722 * Mask out all -(S*)-> if we only have *R in previous
1723 * step, because -(*R)-> -(S*)-> don't make up a strong
1724 * dependency.
1725 */
1726 if (prev_only_xr)
1727 dep &= ~(DEP_SR_MASK | DEP_SN_MASK);
1728
1729 /* If nothing left, we skip */
1730 if (!dep)
1731 continue;
1732
1733 /* If there are only -(*R)-> left, set that for the next step */
1734 lock->only_xr = !(dep & (DEP_SN_MASK | DEP_EN_MASK));
1735 }
1736
1737 /*
1738 * Step 3: we haven't visited this and there is a strong
1739 * dependency path to this, so check with @match.
1740 * If @skip is provide and returns true, we skip this
1741 * lock (and any path this lock is in).
1742 */
1743 if (skip && skip(lock, data))
1744 continue;
1745
1746 if (match(lock, data)) {
1747 *target_entry = lock;
1748 return BFS_RMATCH;
1749 }
1750
1751 /*
1752 * Step 4: if not match, expand the path by adding the
1753 * forward or backwards dependencies in the search
1754 *
1755 */
1756 first = true;
1757 head = get_dep_list(lock, offset);
1758 list_for_each_entry_rcu(entry, head, entry) {
1759 visit_lock_entry(entry, lock);
1760
1761 /*
1762 * Note we only enqueue the first of the list into the
1763 * queue, because we can always find a sibling
1764 * dependency from one (see __bfs_next()), as a result
1765 * the space of queue is saved.
1766 */
1767 if (!first)
1768 continue;
1769
1770 first = false;
1771
1772 if (__cq_enqueue(cq, entry))
1773 return BFS_EQUEUEFULL;
1774
1775 cq_depth = __cq_get_elem_count(cq);
1776 if (max_bfs_queue_depth < cq_depth)
1777 max_bfs_queue_depth = cq_depth;
1778 }
1779 }
1780
1781 return BFS_RNOMATCH;
1782}
1783
1784static inline enum bfs_result
1785__bfs_forwards(struct lock_list *src_entry,
1786 void *data,
1787 bool (*match)(struct lock_list *entry, void *data),
1788 bool (*skip)(struct lock_list *entry, void *data),
1789 struct lock_list **target_entry)
1790{
1791 return __bfs(src_entry, data, match, skip, target_entry,
1792 offsetof(struct lock_class, locks_after));
1793
1794}
1795
1796static inline enum bfs_result
1797__bfs_backwards(struct lock_list *src_entry,
1798 void *data,
1799 bool (*match)(struct lock_list *entry, void *data),
1800 bool (*skip)(struct lock_list *entry, void *data),
1801 struct lock_list **target_entry)
1802{
1803 return __bfs(src_entry, data, match, skip, target_entry,
1804 offsetof(struct lock_class, locks_before));
1805
1806}
1807
1808static void print_lock_trace(const struct lock_trace *trace,
1809 unsigned int spaces)
1810{
1811 stack_trace_print(trace->entries, trace->nr_entries, spaces);
1812}
1813
1814/*
1815 * Print a dependency chain entry (this is only done when a deadlock
1816 * has been detected):
1817 */
1818static noinline void
1819print_circular_bug_entry(struct lock_list *target, int depth)
1820{
1821 if (debug_locks_silent)
1822 return;
1823 printk("\n-> #%u", depth);
1824 print_lock_name(target->class);
1825 printk(KERN_CONT ":\n");
1826 print_lock_trace(target->trace, 6);
1827}
1828
1829static void
1830print_circular_lock_scenario(struct held_lock *src,
1831 struct held_lock *tgt,
1832 struct lock_list *prt)
1833{
1834 struct lock_class *source = hlock_class(src);
1835 struct lock_class *target = hlock_class(tgt);
1836 struct lock_class *parent = prt->class;
1837
1838 /*
1839 * A direct locking problem where unsafe_class lock is taken
1840 * directly by safe_class lock, then all we need to show
1841 * is the deadlock scenario, as it is obvious that the
1842 * unsafe lock is taken under the safe lock.
1843 *
1844 * But if there is a chain instead, where the safe lock takes
1845 * an intermediate lock (middle_class) where this lock is
1846 * not the same as the safe lock, then the lock chain is
1847 * used to describe the problem. Otherwise we would need
1848 * to show a different CPU case for each link in the chain
1849 * from the safe_class lock to the unsafe_class lock.
1850 */
1851 if (parent != source) {
1852 printk("Chain exists of:\n ");
1853 __print_lock_name(source);
1854 printk(KERN_CONT " --> ");
1855 __print_lock_name(parent);
1856 printk(KERN_CONT " --> ");
1857 __print_lock_name(target);
1858 printk(KERN_CONT "\n\n");
1859 }
1860
1861 printk(" Possible unsafe locking scenario:\n\n");
1862 printk(" CPU0 CPU1\n");
1863 printk(" ---- ----\n");
1864 printk(" lock(");
1865 __print_lock_name(target);
1866 printk(KERN_CONT ");\n");
1867 printk(" lock(");
1868 __print_lock_name(parent);
1869 printk(KERN_CONT ");\n");
1870 printk(" lock(");
1871 __print_lock_name(target);
1872 printk(KERN_CONT ");\n");
1873 printk(" lock(");
1874 __print_lock_name(source);
1875 printk(KERN_CONT ");\n");
1876 printk("\n *** DEADLOCK ***\n\n");
1877}
1878
1879/*
1880 * When a circular dependency is detected, print the
1881 * header first:
1882 */
1883static noinline void
1884print_circular_bug_header(struct lock_list *entry, unsigned int depth,
1885 struct held_lock *check_src,
1886 struct held_lock *check_tgt)
1887{
1888 struct task_struct *curr = current;
1889
1890 if (debug_locks_silent)
1891 return;
1892
1893 pr_warn("\n");
1894 pr_warn("======================================================\n");
1895 pr_warn("WARNING: possible circular locking dependency detected\n");
1896 print_kernel_ident();
1897 pr_warn("------------------------------------------------------\n");
1898 pr_warn("%s/%d is trying to acquire lock:\n",
1899 curr->comm, task_pid_nr(curr));
1900 print_lock(check_src);
1901
1902 pr_warn("\nbut task is already holding lock:\n");
1903
1904 print_lock(check_tgt);
1905 pr_warn("\nwhich lock already depends on the new lock.\n\n");
1906 pr_warn("\nthe existing dependency chain (in reverse order) is:\n");
1907
1908 print_circular_bug_entry(entry, depth);
1909}
1910
1911/*
1912 * We are about to add A -> B into the dependency graph, and in __bfs() a
1913 * strong dependency path A -> .. -> B is found: hlock_class equals
1914 * entry->class.
1915 *
1916 * If A -> .. -> B can replace A -> B in any __bfs() search (means the former
1917 * is _stronger_ than or equal to the latter), we consider A -> B as redundant.
1918 * For example if A -> .. -> B is -(EN)-> (i.e. A -(E*)-> .. -(*N)-> B), and A
1919 * -> B is -(ER)-> or -(EN)->, then we don't need to add A -> B into the
1920 * dependency graph, as any strong path ..-> A -> B ->.. we can get with
1921 * having dependency A -> B, we could already get a equivalent path ..-> A ->
1922 * .. -> B -> .. with A -> .. -> B. Therefore A -> B is redundant.
1923 *
1924 * We need to make sure both the start and the end of A -> .. -> B is not
1925 * weaker than A -> B. For the start part, please see the comment in
1926 * check_redundant(). For the end part, we need:
1927 *
1928 * Either
1929 *
1930 * a) A -> B is -(*R)-> (everything is not weaker than that)
1931 *
1932 * or
1933 *
1934 * b) A -> .. -> B is -(*N)-> (nothing is stronger than this)
1935 *
1936 */
1937static inline bool hlock_equal(struct lock_list *entry, void *data)
1938{
1939 struct held_lock *hlock = (struct held_lock *)data;
1940
1941 return hlock_class(hlock) == entry->class && /* Found A -> .. -> B */
1942 (hlock->read == 2 || /* A -> B is -(*R)-> */
1943 !entry->only_xr); /* A -> .. -> B is -(*N)-> */
1944}
1945
1946/*
1947 * We are about to add B -> A into the dependency graph, and in __bfs() a
1948 * strong dependency path A -> .. -> B is found: hlock_class equals
1949 * entry->class.
1950 *
1951 * We will have a deadlock case (conflict) if A -> .. -> B -> A is a strong
1952 * dependency cycle, that means:
1953 *
1954 * Either
1955 *
1956 * a) B -> A is -(E*)->
1957 *
1958 * or
1959 *
1960 * b) A -> .. -> B is -(*N)-> (i.e. A -> .. -(*N)-> B)
1961 *
1962 * as then we don't have -(*R)-> -(S*)-> in the cycle.
1963 */
1964static inline bool hlock_conflict(struct lock_list *entry, void *data)
1965{
1966 struct held_lock *hlock = (struct held_lock *)data;
1967
1968 return hlock_class(hlock) == entry->class && /* Found A -> .. -> B */
1969 (hlock->read == 0 || /* B -> A is -(E*)-> */
1970 !entry->only_xr); /* A -> .. -> B is -(*N)-> */
1971}
1972
1973static noinline void print_circular_bug(struct lock_list *this,
1974 struct lock_list *target,
1975 struct held_lock *check_src,
1976 struct held_lock *check_tgt)
1977{
1978 struct task_struct *curr = current;
1979 struct lock_list *parent;
1980 struct lock_list *first_parent;
1981 int depth;
1982
1983 if (!debug_locks_off_graph_unlock() || debug_locks_silent)
1984 return;
1985
1986 this->trace = save_trace();
1987 if (!this->trace)
1988 return;
1989
1990 depth = get_lock_depth(target);
1991
1992 print_circular_bug_header(target, depth, check_src, check_tgt);
1993
1994 parent = get_lock_parent(target);
1995 first_parent = parent;
1996
1997 while (parent) {
1998 print_circular_bug_entry(parent, --depth);
1999 parent = get_lock_parent(parent);
2000 }
2001
2002 printk("\nother info that might help us debug this:\n\n");
2003 print_circular_lock_scenario(check_src, check_tgt,
2004 first_parent);
2005
2006 lockdep_print_held_locks(curr);
2007
2008 printk("\nstack backtrace:\n");
2009 dump_stack();
2010}
2011
2012static noinline void print_bfs_bug(int ret)
2013{
2014 if (!debug_locks_off_graph_unlock())
2015 return;
2016
2017 /*
2018 * Breadth-first-search failed, graph got corrupted?
2019 */
2020 WARN(1, "lockdep bfs error:%d\n", ret);
2021}
2022
2023static bool noop_count(struct lock_list *entry, void *data)
2024{
2025 (*(unsigned long *)data)++;
2026 return false;
2027}
2028
2029static unsigned long __lockdep_count_forward_deps(struct lock_list *this)
2030{
2031 unsigned long count = 0;
2032 struct lock_list *target_entry;
2033
2034 __bfs_forwards(this, (void *)&count, noop_count, NULL, &target_entry);
2035
2036 return count;
2037}
2038unsigned long lockdep_count_forward_deps(struct lock_class *class)
2039{
2040 unsigned long ret, flags;
2041 struct lock_list this;
2042
2043 __bfs_init_root(&this, class);
2044
2045 raw_local_irq_save(flags);
2046 lockdep_lock();
2047 ret = __lockdep_count_forward_deps(&this);
2048 lockdep_unlock();
2049 raw_local_irq_restore(flags);
2050
2051 return ret;
2052}
2053
2054static unsigned long __lockdep_count_backward_deps(struct lock_list *this)
2055{
2056 unsigned long count = 0;
2057 struct lock_list *target_entry;
2058
2059 __bfs_backwards(this, (void *)&count, noop_count, NULL, &target_entry);
2060
2061 return count;
2062}
2063
2064unsigned long lockdep_count_backward_deps(struct lock_class *class)
2065{
2066 unsigned long ret, flags;
2067 struct lock_list this;
2068
2069 __bfs_init_root(&this, class);
2070
2071 raw_local_irq_save(flags);
2072 lockdep_lock();
2073 ret = __lockdep_count_backward_deps(&this);
2074 lockdep_unlock();
2075 raw_local_irq_restore(flags);
2076
2077 return ret;
2078}
2079
2080/*
2081 * Check that the dependency graph starting at <src> can lead to
2082 * <target> or not.
2083 */
2084static noinline enum bfs_result
2085check_path(struct held_lock *target, struct lock_list *src_entry,
2086 bool (*match)(struct lock_list *entry, void *data),
2087 bool (*skip)(struct lock_list *entry, void *data),
2088 struct lock_list **target_entry)
2089{
2090 enum bfs_result ret;
2091
2092 ret = __bfs_forwards(src_entry, target, match, skip, target_entry);
2093
2094 if (unlikely(bfs_error(ret)))
2095 print_bfs_bug(ret);
2096
2097 return ret;
2098}
2099
2100/*
2101 * Prove that the dependency graph starting at <src> can not
2102 * lead to <target>. If it can, there is a circle when adding
2103 * <target> -> <src> dependency.
2104 *
2105 * Print an error and return BFS_RMATCH if it does.
2106 */
2107static noinline enum bfs_result
2108check_noncircular(struct held_lock *src, struct held_lock *target,
2109 struct lock_trace **const trace)
2110{
2111 enum bfs_result ret;
2112 struct lock_list *target_entry;
2113 struct lock_list src_entry;
2114
2115 bfs_init_root(&src_entry, src);
2116
2117 debug_atomic_inc(nr_cyclic_checks);
2118
2119 ret = check_path(target, &src_entry, hlock_conflict, NULL, &target_entry);
2120
2121 if (unlikely(ret == BFS_RMATCH)) {
2122 if (!*trace) {
2123 /*
2124 * If save_trace fails here, the printing might
2125 * trigger a WARN but because of the !nr_entries it
2126 * should not do bad things.
2127 */
2128 *trace = save_trace();
2129 }
2130
2131 print_circular_bug(&src_entry, target_entry, src, target);
2132 }
2133
2134 return ret;
2135}
2136
2137#ifdef CONFIG_TRACE_IRQFLAGS
2138
2139/*
2140 * Forwards and backwards subgraph searching, for the purposes of
2141 * proving that two subgraphs can be connected by a new dependency
2142 * without creating any illegal irq-safe -> irq-unsafe lock dependency.
2143 *
2144 * A irq safe->unsafe deadlock happens with the following conditions:
2145 *
2146 * 1) We have a strong dependency path A -> ... -> B
2147 *
2148 * 2) and we have ENABLED_IRQ usage of B and USED_IN_IRQ usage of A, therefore
2149 * irq can create a new dependency B -> A (consider the case that a holder
2150 * of B gets interrupted by an irq whose handler will try to acquire A).
2151 *
2152 * 3) the dependency circle A -> ... -> B -> A we get from 1) and 2) is a
2153 * strong circle:
2154 *
2155 * For the usage bits of B:
2156 * a) if A -> B is -(*N)->, then B -> A could be any type, so any
2157 * ENABLED_IRQ usage suffices.
2158 * b) if A -> B is -(*R)->, then B -> A must be -(E*)->, so only
2159 * ENABLED_IRQ_*_READ usage suffices.
2160 *
2161 * For the usage bits of A:
2162 * c) if A -> B is -(E*)->, then B -> A could be any type, so any
2163 * USED_IN_IRQ usage suffices.
2164 * d) if A -> B is -(S*)->, then B -> A must be -(*N)->, so only
2165 * USED_IN_IRQ_*_READ usage suffices.
2166 */
2167
2168/*
2169 * There is a strong dependency path in the dependency graph: A -> B, and now
2170 * we need to decide which usage bit of A should be accumulated to detect
2171 * safe->unsafe bugs.
2172 *
2173 * Note that usage_accumulate() is used in backwards search, so ->only_xr
2174 * stands for whether A -> B only has -(S*)-> (in this case ->only_xr is true).
2175 *
2176 * As above, if only_xr is false, which means A -> B has -(E*)-> dependency
2177 * path, any usage of A should be considered. Otherwise, we should only
2178 * consider _READ usage.
2179 */
2180static inline bool usage_accumulate(struct lock_list *entry, void *mask)
2181{
2182 if (!entry->only_xr)
2183 *(unsigned long *)mask |= entry->class->usage_mask;
2184 else /* Mask out _READ usage bits */
2185 *(unsigned long *)mask |= (entry->class->usage_mask & LOCKF_IRQ);
2186
2187 return false;
2188}
2189
2190/*
2191 * There is a strong dependency path in the dependency graph: A -> B, and now
2192 * we need to decide which usage bit of B conflicts with the usage bits of A,
2193 * i.e. which usage bit of B may introduce safe->unsafe deadlocks.
2194 *
2195 * As above, if only_xr is false, which means A -> B has -(*N)-> dependency
2196 * path, any usage of B should be considered. Otherwise, we should only
2197 * consider _READ usage.
2198 */
2199static inline bool usage_match(struct lock_list *entry, void *mask)
2200{
2201 if (!entry->only_xr)
2202 return !!(entry->class->usage_mask & *(unsigned long *)mask);
2203 else /* Mask out _READ usage bits */
2204 return !!((entry->class->usage_mask & LOCKF_IRQ) & *(unsigned long *)mask);
2205}
2206
2207static inline bool usage_skip(struct lock_list *entry, void *mask)
2208{
2209 /*
2210 * Skip local_lock() for irq inversion detection.
2211 *
2212 * For !RT, local_lock() is not a real lock, so it won't carry any
2213 * dependency.
2214 *
2215 * For RT, an irq inversion happens when we have lock A and B, and on
2216 * some CPU we can have:
2217 *
2218 * lock(A);
2219 * <interrupted>
2220 * lock(B);
2221 *
2222 * where lock(B) cannot sleep, and we have a dependency B -> ... -> A.
2223 *
2224 * Now we prove local_lock() cannot exist in that dependency. First we
2225 * have the observation for any lock chain L1 -> ... -> Ln, for any
2226 * 1 <= i <= n, Li.inner_wait_type <= L1.inner_wait_type, otherwise
2227 * wait context check will complain. And since B is not a sleep lock,
2228 * therefore B.inner_wait_type >= 2, and since the inner_wait_type of
2229 * local_lock() is 3, which is greater than 2, therefore there is no
2230 * way the local_lock() exists in the dependency B -> ... -> A.
2231 *
2232 * As a result, we will skip local_lock(), when we search for irq
2233 * inversion bugs.
2234 */
2235 if (entry->class->lock_type == LD_LOCK_PERCPU) {
2236 if (DEBUG_LOCKS_WARN_ON(entry->class->wait_type_inner < LD_WAIT_CONFIG))
2237 return false;
2238
2239 return true;
2240 }
2241
2242 return false;
2243}
2244
2245/*
2246 * Find a node in the forwards-direction dependency sub-graph starting
2247 * at @root->class that matches @bit.
2248 *
2249 * Return BFS_MATCH if such a node exists in the subgraph, and put that node
2250 * into *@target_entry.
2251 */
2252static enum bfs_result
2253find_usage_forwards(struct lock_list *root, unsigned long usage_mask,
2254 struct lock_list **target_entry)
2255{
2256 enum bfs_result result;
2257
2258 debug_atomic_inc(nr_find_usage_forwards_checks);
2259
2260 result = __bfs_forwards(root, &usage_mask, usage_match, usage_skip, target_entry);
2261
2262 return result;
2263}
2264
2265/*
2266 * Find a node in the backwards-direction dependency sub-graph starting
2267 * at @root->class that matches @bit.
2268 */
2269static enum bfs_result
2270find_usage_backwards(struct lock_list *root, unsigned long usage_mask,
2271 struct lock_list **target_entry)
2272{
2273 enum bfs_result result;
2274
2275 debug_atomic_inc(nr_find_usage_backwards_checks);
2276
2277 result = __bfs_backwards(root, &usage_mask, usage_match, usage_skip, target_entry);
2278
2279 return result;
2280}
2281
2282static void print_lock_class_header(struct lock_class *class, int depth)
2283{
2284 int bit;
2285
2286 printk("%*s->", depth, "");
2287 print_lock_name(class);
2288#ifdef CONFIG_DEBUG_LOCKDEP
2289 printk(KERN_CONT " ops: %lu", debug_class_ops_read(class));
2290#endif
2291 printk(KERN_CONT " {\n");
2292
2293 for (bit = 0; bit < LOCK_TRACE_STATES; bit++) {
2294 if (class->usage_mask & (1 << bit)) {
2295 int len = depth;
2296
2297 len += printk("%*s %s", depth, "", usage_str[bit]);
2298 len += printk(KERN_CONT " at:\n");
2299 print_lock_trace(class->usage_traces[bit], len);
2300 }
2301 }
2302 printk("%*s }\n", depth, "");
2303
2304 printk("%*s ... key at: [<%px>] %pS\n",
2305 depth, "", class->key, class->key);
2306}
2307
2308/*
2309 * printk the shortest lock dependencies from @start to @end in reverse order:
2310 */
2311static void __used
2312print_shortest_lock_dependencies(struct lock_list *leaf,
2313 struct lock_list *root)
2314{
2315 struct lock_list *entry = leaf;
2316 int depth;
2317
2318 /*compute depth from generated tree by BFS*/
2319 depth = get_lock_depth(leaf);
2320
2321 do {
2322 print_lock_class_header(entry->class, depth);
2323 printk("%*s ... acquired at:\n", depth, "");
2324 print_lock_trace(entry->trace, 2);
2325 printk("\n");
2326
2327 if (depth == 0 && (entry != root)) {
2328 printk("lockdep:%s bad path found in chain graph\n", __func__);
2329 break;
2330 }
2331
2332 entry = get_lock_parent(entry);
2333 depth--;
2334 } while (entry && (depth >= 0));
2335}
2336
2337static void
2338print_irq_lock_scenario(struct lock_list *safe_entry,
2339 struct lock_list *unsafe_entry,
2340 struct lock_class *prev_class,
2341 struct lock_class *next_class)
2342{
2343 struct lock_class *safe_class = safe_entry->class;
2344 struct lock_class *unsafe_class = unsafe_entry->class;
2345 struct lock_class *middle_class = prev_class;
2346
2347 if (middle_class == safe_class)
2348 middle_class = next_class;
2349
2350 /*
2351 * A direct locking problem where unsafe_class lock is taken
2352 * directly by safe_class lock, then all we need to show
2353 * is the deadlock scenario, as it is obvious that the
2354 * unsafe lock is taken under the safe lock.
2355 *
2356 * But if there is a chain instead, where the safe lock takes
2357 * an intermediate lock (middle_class) where this lock is
2358 * not the same as the safe lock, then the lock chain is
2359 * used to describe the problem. Otherwise we would need
2360 * to show a different CPU case for each link in the chain
2361 * from the safe_class lock to the unsafe_class lock.
2362 */
2363 if (middle_class != unsafe_class) {
2364 printk("Chain exists of:\n ");
2365 __print_lock_name(safe_class);
2366 printk(KERN_CONT " --> ");
2367 __print_lock_name(middle_class);
2368 printk(KERN_CONT " --> ");
2369 __print_lock_name(unsafe_class);
2370 printk(KERN_CONT "\n\n");
2371 }
2372
2373 printk(" Possible interrupt unsafe locking scenario:\n\n");
2374 printk(" CPU0 CPU1\n");
2375 printk(" ---- ----\n");
2376 printk(" lock(");
2377 __print_lock_name(unsafe_class);
2378 printk(KERN_CONT ");\n");
2379 printk(" local_irq_disable();\n");
2380 printk(" lock(");
2381 __print_lock_name(safe_class);
2382 printk(KERN_CONT ");\n");
2383 printk(" lock(");
2384 __print_lock_name(middle_class);
2385 printk(KERN_CONT ");\n");
2386 printk(" <Interrupt>\n");
2387 printk(" lock(");
2388 __print_lock_name(safe_class);
2389 printk(KERN_CONT ");\n");
2390 printk("\n *** DEADLOCK ***\n\n");
2391}
2392
2393static void
2394print_bad_irq_dependency(struct task_struct *curr,
2395 struct lock_list *prev_root,
2396 struct lock_list *next_root,
2397 struct lock_list *backwards_entry,
2398 struct lock_list *forwards_entry,
2399 struct held_lock *prev,
2400 struct held_lock *next,
2401 enum lock_usage_bit bit1,
2402 enum lock_usage_bit bit2,
2403 const char *irqclass)
2404{
2405 if (!debug_locks_off_graph_unlock() || debug_locks_silent)
2406 return;
2407
2408 pr_warn("\n");
2409 pr_warn("=====================================================\n");
2410 pr_warn("WARNING: %s-safe -> %s-unsafe lock order detected\n",
2411 irqclass, irqclass);
2412 print_kernel_ident();
2413 pr_warn("-----------------------------------------------------\n");
2414 pr_warn("%s/%d [HC%u[%lu]:SC%u[%lu]:HE%u:SE%u] is trying to acquire:\n",
2415 curr->comm, task_pid_nr(curr),
2416 lockdep_hardirq_context(), hardirq_count() >> HARDIRQ_SHIFT,
2417 curr->softirq_context, softirq_count() >> SOFTIRQ_SHIFT,
2418 lockdep_hardirqs_enabled(),
2419 curr->softirqs_enabled);
2420 print_lock(next);
2421
2422 pr_warn("\nand this task is already holding:\n");
2423 print_lock(prev);
2424 pr_warn("which would create a new lock dependency:\n");
2425 print_lock_name(hlock_class(prev));
2426 pr_cont(" ->");
2427 print_lock_name(hlock_class(next));
2428 pr_cont("\n");
2429
2430 pr_warn("\nbut this new dependency connects a %s-irq-safe lock:\n",
2431 irqclass);
2432 print_lock_name(backwards_entry->class);
2433 pr_warn("\n... which became %s-irq-safe at:\n", irqclass);
2434
2435 print_lock_trace(backwards_entry->class->usage_traces[bit1], 1);
2436
2437 pr_warn("\nto a %s-irq-unsafe lock:\n", irqclass);
2438 print_lock_name(forwards_entry->class);
2439 pr_warn("\n... which became %s-irq-unsafe at:\n", irqclass);
2440 pr_warn("...");
2441
2442 print_lock_trace(forwards_entry->class->usage_traces[bit2], 1);
2443
2444 pr_warn("\nother info that might help us debug this:\n\n");
2445 print_irq_lock_scenario(backwards_entry, forwards_entry,
2446 hlock_class(prev), hlock_class(next));
2447
2448 lockdep_print_held_locks(curr);
2449
2450 pr_warn("\nthe dependencies between %s-irq-safe lock and the holding lock:\n", irqclass);
2451 prev_root->trace = save_trace();
2452 if (!prev_root->trace)
2453 return;
2454 print_shortest_lock_dependencies(backwards_entry, prev_root);
2455
2456 pr_warn("\nthe dependencies between the lock to be acquired");
2457 pr_warn(" and %s-irq-unsafe lock:\n", irqclass);
2458 next_root->trace = save_trace();
2459 if (!next_root->trace)
2460 return;
2461 print_shortest_lock_dependencies(forwards_entry, next_root);
2462
2463 pr_warn("\nstack backtrace:\n");
2464 dump_stack();
2465}
2466
2467static const char *state_names[] = {
2468#define LOCKDEP_STATE(__STATE) \
2469 __stringify(__STATE),
2470#include "lockdep_states.h"
2471#undef LOCKDEP_STATE
2472};
2473
2474static const char *state_rnames[] = {
2475#define LOCKDEP_STATE(__STATE) \
2476 __stringify(__STATE)"-READ",
2477#include "lockdep_states.h"
2478#undef LOCKDEP_STATE
2479};
2480
2481static inline const char *state_name(enum lock_usage_bit bit)
2482{
2483 if (bit & LOCK_USAGE_READ_MASK)
2484 return state_rnames[bit >> LOCK_USAGE_DIR_MASK];
2485 else
2486 return state_names[bit >> LOCK_USAGE_DIR_MASK];
2487}
2488
2489/*
2490 * The bit number is encoded like:
2491 *
2492 * bit0: 0 exclusive, 1 read lock
2493 * bit1: 0 used in irq, 1 irq enabled
2494 * bit2-n: state
2495 */
2496static int exclusive_bit(int new_bit)
2497{
2498 int state = new_bit & LOCK_USAGE_STATE_MASK;
2499 int dir = new_bit & LOCK_USAGE_DIR_MASK;
2500
2501 /*
2502 * keep state, bit flip the direction and strip read.
2503 */
2504 return state | (dir ^ LOCK_USAGE_DIR_MASK);
2505}
2506
2507/*
2508 * Observe that when given a bitmask where each bitnr is encoded as above, a
2509 * right shift of the mask transforms the individual bitnrs as -1 and
2510 * conversely, a left shift transforms into +1 for the individual bitnrs.
2511 *
2512 * So for all bits whose number have LOCK_ENABLED_* set (bitnr1 == 1), we can
2513 * create the mask with those bit numbers using LOCK_USED_IN_* (bitnr1 == 0)
2514 * instead by subtracting the bit number by 2, or shifting the mask right by 2.
2515 *
2516 * Similarly, bitnr1 == 0 becomes bitnr1 == 1 by adding 2, or shifting left 2.
2517 *
2518 * So split the mask (note that LOCKF_ENABLED_IRQ_ALL|LOCKF_USED_IN_IRQ_ALL is
2519 * all bits set) and recompose with bitnr1 flipped.
2520 */
2521static unsigned long invert_dir_mask(unsigned long mask)
2522{
2523 unsigned long excl = 0;
2524
2525 /* Invert dir */
2526 excl |= (mask & LOCKF_ENABLED_IRQ_ALL) >> LOCK_USAGE_DIR_MASK;
2527 excl |= (mask & LOCKF_USED_IN_IRQ_ALL) << LOCK_USAGE_DIR_MASK;
2528
2529 return excl;
2530}
2531
2532/*
2533 * Note that a LOCK_ENABLED_IRQ_*_READ usage and a LOCK_USED_IN_IRQ_*_READ
2534 * usage may cause deadlock too, for example:
2535 *
2536 * P1 P2
2537 * <irq disabled>
2538 * write_lock(l1); <irq enabled>
2539 * read_lock(l2);
2540 * write_lock(l2);
2541 * <in irq>
2542 * read_lock(l1);
2543 *
2544 * , in above case, l1 will be marked as LOCK_USED_IN_IRQ_HARDIRQ_READ and l2
2545 * will marked as LOCK_ENABLE_IRQ_HARDIRQ_READ, and this is a possible
2546 * deadlock.
2547 *
2548 * In fact, all of the following cases may cause deadlocks:
2549 *
2550 * LOCK_USED_IN_IRQ_* -> LOCK_ENABLED_IRQ_*
2551 * LOCK_USED_IN_IRQ_*_READ -> LOCK_ENABLED_IRQ_*
2552 * LOCK_USED_IN_IRQ_* -> LOCK_ENABLED_IRQ_*_READ
2553 * LOCK_USED_IN_IRQ_*_READ -> LOCK_ENABLED_IRQ_*_READ
2554 *
2555 * As a result, to calculate the "exclusive mask", first we invert the
2556 * direction (USED_IN/ENABLED) of the original mask, and 1) for all bits with
2557 * bitnr0 set (LOCK_*_READ), add those with bitnr0 cleared (LOCK_*). 2) for all
2558 * bits with bitnr0 cleared (LOCK_*_READ), add those with bitnr0 set (LOCK_*).
2559 */
2560static unsigned long exclusive_mask(unsigned long mask)
2561{
2562 unsigned long excl = invert_dir_mask(mask);
2563
2564 excl |= (excl & LOCKF_IRQ_READ) >> LOCK_USAGE_READ_MASK;
2565 excl |= (excl & LOCKF_IRQ) << LOCK_USAGE_READ_MASK;
2566
2567 return excl;
2568}
2569
2570/*
2571 * Retrieve the _possible_ original mask to which @mask is
2572 * exclusive. Ie: this is the opposite of exclusive_mask().
2573 * Note that 2 possible original bits can match an exclusive
2574 * bit: one has LOCK_USAGE_READ_MASK set, the other has it
2575 * cleared. So both are returned for each exclusive bit.
2576 */
2577static unsigned long original_mask(unsigned long mask)
2578{
2579 unsigned long excl = invert_dir_mask(mask);
2580
2581 /* Include read in existing usages */
2582 excl |= (excl & LOCKF_IRQ_READ) >> LOCK_USAGE_READ_MASK;
2583 excl |= (excl & LOCKF_IRQ) << LOCK_USAGE_READ_MASK;
2584
2585 return excl;
2586}
2587
2588/*
2589 * Find the first pair of bit match between an original
2590 * usage mask and an exclusive usage mask.
2591 */
2592static int find_exclusive_match(unsigned long mask,
2593 unsigned long excl_mask,
2594 enum lock_usage_bit *bitp,
2595 enum lock_usage_bit *excl_bitp)
2596{
2597 int bit, excl, excl_read;
2598
2599 for_each_set_bit(bit, &mask, LOCK_USED) {
2600 /*
2601 * exclusive_bit() strips the read bit, however,
2602 * LOCK_ENABLED_IRQ_*_READ may cause deadlocks too, so we need
2603 * to search excl | LOCK_USAGE_READ_MASK as well.
2604 */
2605 excl = exclusive_bit(bit);
2606 excl_read = excl | LOCK_USAGE_READ_MASK;
2607 if (excl_mask & lock_flag(excl)) {
2608 *bitp = bit;
2609 *excl_bitp = excl;
2610 return 0;
2611 } else if (excl_mask & lock_flag(excl_read)) {
2612 *bitp = bit;
2613 *excl_bitp = excl_read;
2614 return 0;
2615 }
2616 }
2617 return -1;
2618}
2619
2620/*
2621 * Prove that the new dependency does not connect a hardirq-safe(-read)
2622 * lock with a hardirq-unsafe lock - to achieve this we search
2623 * the backwards-subgraph starting at <prev>, and the
2624 * forwards-subgraph starting at <next>:
2625 */
2626static int check_irq_usage(struct task_struct *curr, struct held_lock *prev,
2627 struct held_lock *next)
2628{
2629 unsigned long usage_mask = 0, forward_mask, backward_mask;
2630 enum lock_usage_bit forward_bit = 0, backward_bit = 0;
2631 struct lock_list *target_entry1;
2632 struct lock_list *target_entry;
2633 struct lock_list this, that;
2634 enum bfs_result ret;
2635
2636 /*
2637 * Step 1: gather all hard/soft IRQs usages backward in an
2638 * accumulated usage mask.
2639 */
2640 bfs_init_rootb(&this, prev);
2641
2642 ret = __bfs_backwards(&this, &usage_mask, usage_accumulate, usage_skip, NULL);
2643 if (bfs_error(ret)) {
2644 print_bfs_bug(ret);
2645 return 0;
2646 }
2647
2648 usage_mask &= LOCKF_USED_IN_IRQ_ALL;
2649 if (!usage_mask)
2650 return 1;
2651
2652 /*
2653 * Step 2: find exclusive uses forward that match the previous
2654 * backward accumulated mask.
2655 */
2656 forward_mask = exclusive_mask(usage_mask);
2657
2658 bfs_init_root(&that, next);
2659
2660 ret = find_usage_forwards(&that, forward_mask, &target_entry1);
2661 if (bfs_error(ret)) {
2662 print_bfs_bug(ret);
2663 return 0;
2664 }
2665 if (ret == BFS_RNOMATCH)
2666 return 1;
2667
2668 /*
2669 * Step 3: we found a bad match! Now retrieve a lock from the backward
2670 * list whose usage mask matches the exclusive usage mask from the
2671 * lock found on the forward list.
2672 */
2673 backward_mask = original_mask(target_entry1->class->usage_mask);
2674
2675 ret = find_usage_backwards(&this, backward_mask, &target_entry);
2676 if (bfs_error(ret)) {
2677 print_bfs_bug(ret);
2678 return 0;
2679 }
2680 if (DEBUG_LOCKS_WARN_ON(ret == BFS_RNOMATCH))
2681 return 1;
2682
2683 /*
2684 * Step 4: narrow down to a pair of incompatible usage bits
2685 * and report it.
2686 */
2687 ret = find_exclusive_match(target_entry->class->usage_mask,
2688 target_entry1->class->usage_mask,
2689 &backward_bit, &forward_bit);
2690 if (DEBUG_LOCKS_WARN_ON(ret == -1))
2691 return 1;
2692
2693 print_bad_irq_dependency(curr, &this, &that,
2694 target_entry, target_entry1,
2695 prev, next,
2696 backward_bit, forward_bit,
2697 state_name(backward_bit));
2698
2699 return 0;
2700}
2701
2702#else
2703
2704static inline int check_irq_usage(struct task_struct *curr,
2705 struct held_lock *prev, struct held_lock *next)
2706{
2707 return 1;
2708}
2709
2710static inline bool usage_skip(struct lock_list *entry, void *mask)
2711{
2712 return false;
2713}
2714
2715#endif /* CONFIG_TRACE_IRQFLAGS */
2716
2717#ifdef CONFIG_LOCKDEP_SMALL
2718/*
2719 * Check that the dependency graph starting at <src> can lead to
2720 * <target> or not. If it can, <src> -> <target> dependency is already
2721 * in the graph.
2722 *
2723 * Return BFS_RMATCH if it does, or BFS_RMATCH if it does not, return BFS_E* if
2724 * any error appears in the bfs search.
2725 */
2726static noinline enum bfs_result
2727check_redundant(struct held_lock *src, struct held_lock *target)
2728{
2729 enum bfs_result ret;
2730 struct lock_list *target_entry;
2731 struct lock_list src_entry;
2732
2733 bfs_init_root(&src_entry, src);
2734 /*
2735 * Special setup for check_redundant().
2736 *
2737 * To report redundant, we need to find a strong dependency path that
2738 * is equal to or stronger than <src> -> <target>. So if <src> is E,
2739 * we need to let __bfs() only search for a path starting at a -(E*)->,
2740 * we achieve this by setting the initial node's ->only_xr to true in
2741 * that case. And if <prev> is S, we set initial ->only_xr to false
2742 * because both -(S*)-> (equal) and -(E*)-> (stronger) are redundant.
2743 */
2744 src_entry.only_xr = src->read == 0;
2745
2746 debug_atomic_inc(nr_redundant_checks);
2747
2748 /*
2749 * Note: we skip local_lock() for redundant check, because as the
2750 * comment in usage_skip(), A -> local_lock() -> B and A -> B are not
2751 * the same.
2752 */
2753 ret = check_path(target, &src_entry, hlock_equal, usage_skip, &target_entry);
2754
2755 if (ret == BFS_RMATCH)
2756 debug_atomic_inc(nr_redundant);
2757
2758 return ret;
2759}
2760
2761#else
2762
2763static inline enum bfs_result
2764check_redundant(struct held_lock *src, struct held_lock *target)
2765{
2766 return BFS_RNOMATCH;
2767}
2768
2769#endif
2770
2771static void inc_chains(int irq_context)
2772{
2773 if (irq_context & LOCK_CHAIN_HARDIRQ_CONTEXT)
2774 nr_hardirq_chains++;
2775 else if (irq_context & LOCK_CHAIN_SOFTIRQ_CONTEXT)
2776 nr_softirq_chains++;
2777 else
2778 nr_process_chains++;
2779}
2780
2781static void dec_chains(int irq_context)
2782{
2783 if (irq_context & LOCK_CHAIN_HARDIRQ_CONTEXT)
2784 nr_hardirq_chains--;
2785 else if (irq_context & LOCK_CHAIN_SOFTIRQ_CONTEXT)
2786 nr_softirq_chains--;
2787 else
2788 nr_process_chains--;
2789}
2790
2791static void
2792print_deadlock_scenario(struct held_lock *nxt, struct held_lock *prv)
2793{
2794 struct lock_class *next = hlock_class(nxt);
2795 struct lock_class *prev = hlock_class(prv);
2796
2797 printk(" Possible unsafe locking scenario:\n\n");
2798 printk(" CPU0\n");
2799 printk(" ----\n");
2800 printk(" lock(");
2801 __print_lock_name(prev);
2802 printk(KERN_CONT ");\n");
2803 printk(" lock(");
2804 __print_lock_name(next);
2805 printk(KERN_CONT ");\n");
2806 printk("\n *** DEADLOCK ***\n\n");
2807 printk(" May be due to missing lock nesting notation\n\n");
2808}
2809
2810static void
2811print_deadlock_bug(struct task_struct *curr, struct held_lock *prev,
2812 struct held_lock *next)
2813{
2814 if (!debug_locks_off_graph_unlock() || debug_locks_silent)
2815 return;
2816
2817 pr_warn("\n");
2818 pr_warn("============================================\n");
2819 pr_warn("WARNING: possible recursive locking detected\n");
2820 print_kernel_ident();
2821 pr_warn("--------------------------------------------\n");
2822 pr_warn("%s/%d is trying to acquire lock:\n",
2823 curr->comm, task_pid_nr(curr));
2824 print_lock(next);
2825 pr_warn("\nbut task is already holding lock:\n");
2826 print_lock(prev);
2827
2828 pr_warn("\nother info that might help us debug this:\n");
2829 print_deadlock_scenario(next, prev);
2830 lockdep_print_held_locks(curr);
2831
2832 pr_warn("\nstack backtrace:\n");
2833 dump_stack();
2834}
2835
2836/*
2837 * Check whether we are holding such a class already.
2838 *
2839 * (Note that this has to be done separately, because the graph cannot
2840 * detect such classes of deadlocks.)
2841 *
2842 * Returns: 0 on deadlock detected, 1 on OK, 2 if another lock with the same
2843 * lock class is held but nest_lock is also held, i.e. we rely on the
2844 * nest_lock to avoid the deadlock.
2845 */
2846static int
2847check_deadlock(struct task_struct *curr, struct held_lock *next)
2848{
2849 struct held_lock *prev;
2850 struct held_lock *nest = NULL;
2851 int i;
2852
2853 for (i = 0; i < curr->lockdep_depth; i++) {
2854 prev = curr->held_locks + i;
2855
2856 if (prev->instance == next->nest_lock)
2857 nest = prev;
2858
2859 if (hlock_class(prev) != hlock_class(next))
2860 continue;
2861
2862 /*
2863 * Allow read-after-read recursion of the same
2864 * lock class (i.e. read_lock(lock)+read_lock(lock)):
2865 */
2866 if ((next->read == 2) && prev->read)
2867 continue;
2868
2869 /*
2870 * We're holding the nest_lock, which serializes this lock's
2871 * nesting behaviour.
2872 */
2873 if (nest)
2874 return 2;
2875
2876 print_deadlock_bug(curr, prev, next);
2877 return 0;
2878 }
2879 return 1;
2880}
2881
2882/*
2883 * There was a chain-cache miss, and we are about to add a new dependency
2884 * to a previous lock. We validate the following rules:
2885 *
2886 * - would the adding of the <prev> -> <next> dependency create a
2887 * circular dependency in the graph? [== circular deadlock]
2888 *
2889 * - does the new prev->next dependency connect any hardirq-safe lock
2890 * (in the full backwards-subgraph starting at <prev>) with any
2891 * hardirq-unsafe lock (in the full forwards-subgraph starting at
2892 * <next>)? [== illegal lock inversion with hardirq contexts]
2893 *
2894 * - does the new prev->next dependency connect any softirq-safe lock
2895 * (in the full backwards-subgraph starting at <prev>) with any
2896 * softirq-unsafe lock (in the full forwards-subgraph starting at
2897 * <next>)? [== illegal lock inversion with softirq contexts]
2898 *
2899 * any of these scenarios could lead to a deadlock.
2900 *
2901 * Then if all the validations pass, we add the forwards and backwards
2902 * dependency.
2903 */
2904static int
2905check_prev_add(struct task_struct *curr, struct held_lock *prev,
2906 struct held_lock *next, u16 distance,
2907 struct lock_trace **const trace)
2908{
2909 struct lock_list *entry;
2910 enum bfs_result ret;
2911
2912 if (!hlock_class(prev)->key || !hlock_class(next)->key) {
2913 /*
2914 * The warning statements below may trigger a use-after-free
2915 * of the class name. It is better to trigger a use-after free
2916 * and to have the class name most of the time instead of not
2917 * having the class name available.
2918 */
2919 WARN_ONCE(!debug_locks_silent && !hlock_class(prev)->key,
2920 "Detected use-after-free of lock class %px/%s\n",
2921 hlock_class(prev),
2922 hlock_class(prev)->name);
2923 WARN_ONCE(!debug_locks_silent && !hlock_class(next)->key,
2924 "Detected use-after-free of lock class %px/%s\n",
2925 hlock_class(next),
2926 hlock_class(next)->name);
2927 return 2;
2928 }
2929
2930 /*
2931 * Prove that the new <prev> -> <next> dependency would not
2932 * create a circular dependency in the graph. (We do this by
2933 * a breadth-first search into the graph starting at <next>,
2934 * and check whether we can reach <prev>.)
2935 *
2936 * The search is limited by the size of the circular queue (i.e.,
2937 * MAX_CIRCULAR_QUEUE_SIZE) which keeps track of a breadth of nodes
2938 * in the graph whose neighbours are to be checked.
2939 */
2940 ret = check_noncircular(next, prev, trace);
2941 if (unlikely(bfs_error(ret) || ret == BFS_RMATCH))
2942 return 0;
2943
2944 if (!check_irq_usage(curr, prev, next))
2945 return 0;
2946
2947 /*
2948 * Is the <prev> -> <next> dependency already present?
2949 *
2950 * (this may occur even though this is a new chain: consider
2951 * e.g. the L1 -> L2 -> L3 -> L4 and the L5 -> L1 -> L2 -> L3
2952 * chains - the second one will be new, but L1 already has
2953 * L2 added to its dependency list, due to the first chain.)
2954 */
2955 list_for_each_entry(entry, &hlock_class(prev)->locks_after, entry) {
2956 if (entry->class == hlock_class(next)) {
2957 if (distance == 1)
2958 entry->distance = 1;
2959 entry->dep |= calc_dep(prev, next);
2960
2961 /*
2962 * Also, update the reverse dependency in @next's
2963 * ->locks_before list.
2964 *
2965 * Here we reuse @entry as the cursor, which is fine
2966 * because we won't go to the next iteration of the
2967 * outer loop:
2968 *
2969 * For normal cases, we return in the inner loop.
2970 *
2971 * If we fail to return, we have inconsistency, i.e.
2972 * <prev>::locks_after contains <next> while
2973 * <next>::locks_before doesn't contain <prev>. In
2974 * that case, we return after the inner and indicate
2975 * something is wrong.
2976 */
2977 list_for_each_entry(entry, &hlock_class(next)->locks_before, entry) {
2978 if (entry->class == hlock_class(prev)) {
2979 if (distance == 1)
2980 entry->distance = 1;
2981 entry->dep |= calc_depb(prev, next);
2982 return 1;
2983 }
2984 }
2985
2986 /* <prev> is not found in <next>::locks_before */
2987 return 0;
2988 }
2989 }
2990
2991 /*
2992 * Is the <prev> -> <next> link redundant?
2993 */
2994 ret = check_redundant(prev, next);
2995 if (bfs_error(ret))
2996 return 0;
2997 else if (ret == BFS_RMATCH)
2998 return 2;
2999
3000 if (!*trace) {
3001 *trace = save_trace();
3002 if (!*trace)
3003 return 0;
3004 }
3005
3006 /*
3007 * Ok, all validations passed, add the new lock
3008 * to the previous lock's dependency list:
3009 */
3010 ret = add_lock_to_list(hlock_class(next), hlock_class(prev),
3011 &hlock_class(prev)->locks_after,
3012 next->acquire_ip, distance,
3013 calc_dep(prev, next),
3014 *trace);
3015
3016 if (!ret)
3017 return 0;
3018
3019 ret = add_lock_to_list(hlock_class(prev), hlock_class(next),
3020 &hlock_class(next)->locks_before,
3021 next->acquire_ip, distance,
3022 calc_depb(prev, next),
3023 *trace);
3024 if (!ret)
3025 return 0;
3026
3027 return 2;
3028}
3029
3030/*
3031 * Add the dependency to all directly-previous locks that are 'relevant'.
3032 * The ones that are relevant are (in increasing distance from curr):
3033 * all consecutive trylock entries and the final non-trylock entry - or
3034 * the end of this context's lock-chain - whichever comes first.
3035 */
3036static int
3037check_prevs_add(struct task_struct *curr, struct held_lock *next)
3038{
3039 struct lock_trace *trace = NULL;
3040 int depth = curr->lockdep_depth;
3041 struct held_lock *hlock;
3042
3043 /*
3044 * Debugging checks.
3045 *
3046 * Depth must not be zero for a non-head lock:
3047 */
3048 if (!depth)
3049 goto out_bug;
3050 /*
3051 * At least two relevant locks must exist for this
3052 * to be a head:
3053 */
3054 if (curr->held_locks[depth].irq_context !=
3055 curr->held_locks[depth-1].irq_context)
3056 goto out_bug;
3057
3058 for (;;) {
3059 u16 distance = curr->lockdep_depth - depth + 1;
3060 hlock = curr->held_locks + depth - 1;
3061
3062 if (hlock->check) {
3063 int ret = check_prev_add(curr, hlock, next, distance, &trace);
3064 if (!ret)
3065 return 0;
3066
3067 /*
3068 * Stop after the first non-trylock entry,
3069 * as non-trylock entries have added their
3070 * own direct dependencies already, so this
3071 * lock is connected to them indirectly:
3072 */
3073 if (!hlock->trylock)
3074 break;
3075 }
3076
3077 depth--;
3078 /*
3079 * End of lock-stack?
3080 */
3081 if (!depth)
3082 break;
3083 /*
3084 * Stop the search if we cross into another context:
3085 */
3086 if (curr->held_locks[depth].irq_context !=
3087 curr->held_locks[depth-1].irq_context)
3088 break;
3089 }
3090 return 1;
3091out_bug:
3092 if (!debug_locks_off_graph_unlock())
3093 return 0;
3094
3095 /*
3096 * Clearly we all shouldn't be here, but since we made it we
3097 * can reliable say we messed up our state. See the above two
3098 * gotos for reasons why we could possibly end up here.
3099 */
3100 WARN_ON(1);
3101
3102 return 0;
3103}
3104
3105struct lock_chain lock_chains[MAX_LOCKDEP_CHAINS];
3106static DECLARE_BITMAP(lock_chains_in_use, MAX_LOCKDEP_CHAINS);
3107static u16 chain_hlocks[MAX_LOCKDEP_CHAIN_HLOCKS];
3108unsigned long nr_zapped_lock_chains;
3109unsigned int nr_free_chain_hlocks; /* Free chain_hlocks in buckets */
3110unsigned int nr_lost_chain_hlocks; /* Lost chain_hlocks */
3111unsigned int nr_large_chain_blocks; /* size > MAX_CHAIN_BUCKETS */
3112
3113/*
3114 * The first 2 chain_hlocks entries in the chain block in the bucket
3115 * list contains the following meta data:
3116 *
3117 * entry[0]:
3118 * Bit 15 - always set to 1 (it is not a class index)
3119 * Bits 0-14 - upper 15 bits of the next block index
3120 * entry[1] - lower 16 bits of next block index
3121 *
3122 * A next block index of all 1 bits means it is the end of the list.
3123 *
3124 * On the unsized bucket (bucket-0), the 3rd and 4th entries contain
3125 * the chain block size:
3126 *
3127 * entry[2] - upper 16 bits of the chain block size
3128 * entry[3] - lower 16 bits of the chain block size
3129 */
3130#define MAX_CHAIN_BUCKETS 16
3131#define CHAIN_BLK_FLAG (1U << 15)
3132#define CHAIN_BLK_LIST_END 0xFFFFU
3133
3134static int chain_block_buckets[MAX_CHAIN_BUCKETS];
3135
3136static inline int size_to_bucket(int size)
3137{
3138 if (size > MAX_CHAIN_BUCKETS)
3139 return 0;
3140
3141 return size - 1;
3142}
3143
3144/*
3145 * Iterate all the chain blocks in a bucket.
3146 */
3147#define for_each_chain_block(bucket, prev, curr) \
3148 for ((prev) = -1, (curr) = chain_block_buckets[bucket]; \
3149 (curr) >= 0; \
3150 (prev) = (curr), (curr) = chain_block_next(curr))
3151
3152/*
3153 * next block or -1
3154 */
3155static inline int chain_block_next(int offset)
3156{
3157 int next = chain_hlocks[offset];
3158
3159 WARN_ON_ONCE(!(next & CHAIN_BLK_FLAG));
3160
3161 if (next == CHAIN_BLK_LIST_END)
3162 return -1;
3163
3164 next &= ~CHAIN_BLK_FLAG;
3165 next <<= 16;
3166 next |= chain_hlocks[offset + 1];
3167
3168 return next;
3169}
3170
3171/*
3172 * bucket-0 only
3173 */
3174static inline int chain_block_size(int offset)
3175{
3176 return (chain_hlocks[offset + 2] << 16) | chain_hlocks[offset + 3];
3177}
3178
3179static inline void init_chain_block(int offset, int next, int bucket, int size)
3180{
3181 chain_hlocks[offset] = (next >> 16) | CHAIN_BLK_FLAG;
3182 chain_hlocks[offset + 1] = (u16)next;
3183
3184 if (size && !bucket) {
3185 chain_hlocks[offset + 2] = size >> 16;
3186 chain_hlocks[offset + 3] = (u16)size;
3187 }
3188}
3189
3190static inline void add_chain_block(int offset, int size)
3191{
3192 int bucket = size_to_bucket(size);
3193 int next = chain_block_buckets[bucket];
3194 int prev, curr;
3195
3196 if (unlikely(size < 2)) {
3197 /*
3198 * We can't store single entries on the freelist. Leak them.
3199 *
3200 * One possible way out would be to uniquely mark them, other
3201 * than with CHAIN_BLK_FLAG, such that we can recover them when
3202 * the block before it is re-added.
3203 */
3204 if (size)
3205 nr_lost_chain_hlocks++;
3206 return;
3207 }
3208
3209 nr_free_chain_hlocks += size;
3210 if (!bucket) {
3211 nr_large_chain_blocks++;
3212
3213 /*
3214 * Variable sized, sort large to small.
3215 */
3216 for_each_chain_block(0, prev, curr) {
3217 if (size >= chain_block_size(curr))
3218 break;
3219 }
3220 init_chain_block(offset, curr, 0, size);
3221 if (prev < 0)
3222 chain_block_buckets[0] = offset;
3223 else
3224 init_chain_block(prev, offset, 0, 0);
3225 return;
3226 }
3227 /*
3228 * Fixed size, add to head.
3229 */
3230 init_chain_block(offset, next, bucket, size);
3231 chain_block_buckets[bucket] = offset;
3232}
3233
3234/*
3235 * Only the first block in the list can be deleted.
3236 *
3237 * For the variable size bucket[0], the first block (the largest one) is
3238 * returned, broken up and put back into the pool. So if a chain block of
3239 * length > MAX_CHAIN_BUCKETS is ever used and zapped, it will just be
3240 * queued up after the primordial chain block and never be used until the
3241 * hlock entries in the primordial chain block is almost used up. That
3242 * causes fragmentation and reduce allocation efficiency. That can be
3243 * monitored by looking at the "large chain blocks" number in lockdep_stats.
3244 */
3245static inline void del_chain_block(int bucket, int size, int next)
3246{
3247 nr_free_chain_hlocks -= size;
3248 chain_block_buckets[bucket] = next;
3249
3250 if (!bucket)
3251 nr_large_chain_blocks--;
3252}
3253
3254static void init_chain_block_buckets(void)
3255{
3256 int i;
3257
3258 for (i = 0; i < MAX_CHAIN_BUCKETS; i++)
3259 chain_block_buckets[i] = -1;
3260
3261 add_chain_block(0, ARRAY_SIZE(chain_hlocks));
3262}
3263
3264/*
3265 * Return offset of a chain block of the right size or -1 if not found.
3266 *
3267 * Fairly simple worst-fit allocator with the addition of a number of size
3268 * specific free lists.
3269 */
3270static int alloc_chain_hlocks(int req)
3271{
3272 int bucket, curr, size;
3273
3274 /*
3275 * We rely on the MSB to act as an escape bit to denote freelist
3276 * pointers. Make sure this bit isn't set in 'normal' class_idx usage.
3277 */
3278 BUILD_BUG_ON((MAX_LOCKDEP_KEYS-1) & CHAIN_BLK_FLAG);
3279
3280 init_data_structures_once();
3281
3282 if (nr_free_chain_hlocks < req)
3283 return -1;
3284
3285 /*
3286 * We require a minimum of 2 (u16) entries to encode a freelist
3287 * 'pointer'.
3288 */
3289 req = max(req, 2);
3290 bucket = size_to_bucket(req);
3291 curr = chain_block_buckets[bucket];
3292
3293 if (bucket) {
3294 if (curr >= 0) {
3295 del_chain_block(bucket, req, chain_block_next(curr));
3296 return curr;
3297 }
3298 /* Try bucket 0 */
3299 curr = chain_block_buckets[0];
3300 }
3301
3302 /*
3303 * The variable sized freelist is sorted by size; the first entry is
3304 * the largest. Use it if it fits.
3305 */
3306 if (curr >= 0) {
3307 size = chain_block_size(curr);
3308 if (likely(size >= req)) {
3309 del_chain_block(0, size, chain_block_next(curr));
3310 add_chain_block(curr + req, size - req);
3311 return curr;
3312 }
3313 }
3314
3315 /*
3316 * Last resort, split a block in a larger sized bucket.
3317 */
3318 for (size = MAX_CHAIN_BUCKETS; size > req; size--) {
3319 bucket = size_to_bucket(size);
3320 curr = chain_block_buckets[bucket];
3321 if (curr < 0)
3322 continue;
3323
3324 del_chain_block(bucket, size, chain_block_next(curr));
3325 add_chain_block(curr + req, size - req);
3326 return curr;
3327 }
3328
3329 return -1;
3330}
3331
3332static inline void free_chain_hlocks(int base, int size)
3333{
3334 add_chain_block(base, max(size, 2));
3335}
3336
3337struct lock_class *lock_chain_get_class(struct lock_chain *chain, int i)
3338{
3339 u16 chain_hlock = chain_hlocks[chain->base + i];
3340 unsigned int class_idx = chain_hlock_class_idx(chain_hlock);
3341
3342 return lock_classes + class_idx - 1;
3343}
3344
3345/*
3346 * Returns the index of the first held_lock of the current chain
3347 */
3348static inline int get_first_held_lock(struct task_struct *curr,
3349 struct held_lock *hlock)
3350{
3351 int i;
3352 struct held_lock *hlock_curr;
3353
3354 for (i = curr->lockdep_depth - 1; i >= 0; i--) {
3355 hlock_curr = curr->held_locks + i;
3356 if (hlock_curr->irq_context != hlock->irq_context)
3357 break;
3358
3359 }
3360
3361 return ++i;
3362}
3363
3364#ifdef CONFIG_DEBUG_LOCKDEP
3365/*
3366 * Returns the next chain_key iteration
3367 */
3368static u64 print_chain_key_iteration(u16 hlock_id, u64 chain_key)
3369{
3370 u64 new_chain_key = iterate_chain_key(chain_key, hlock_id);
3371
3372 printk(" hlock_id:%d -> chain_key:%016Lx",
3373 (unsigned int)hlock_id,
3374 (unsigned long long)new_chain_key);
3375 return new_chain_key;
3376}
3377
3378static void
3379print_chain_keys_held_locks(struct task_struct *curr, struct held_lock *hlock_next)
3380{
3381 struct held_lock *hlock;
3382 u64 chain_key = INITIAL_CHAIN_KEY;
3383 int depth = curr->lockdep_depth;
3384 int i = get_first_held_lock(curr, hlock_next);
3385
3386 printk("depth: %u (irq_context %u)\n", depth - i + 1,
3387 hlock_next->irq_context);
3388 for (; i < depth; i++) {
3389 hlock = curr->held_locks + i;
3390 chain_key = print_chain_key_iteration(hlock_id(hlock), chain_key);
3391
3392 print_lock(hlock);
3393 }
3394
3395 print_chain_key_iteration(hlock_id(hlock_next), chain_key);
3396 print_lock(hlock_next);
3397}
3398
3399static void print_chain_keys_chain(struct lock_chain *chain)
3400{
3401 int i;
3402 u64 chain_key = INITIAL_CHAIN_KEY;
3403 u16 hlock_id;
3404
3405 printk("depth: %u\n", chain->depth);
3406 for (i = 0; i < chain->depth; i++) {
3407 hlock_id = chain_hlocks[chain->base + i];
3408 chain_key = print_chain_key_iteration(hlock_id, chain_key);
3409
3410 print_lock_name(lock_classes + chain_hlock_class_idx(hlock_id) - 1);
3411 printk("\n");
3412 }
3413}
3414
3415static void print_collision(struct task_struct *curr,
3416 struct held_lock *hlock_next,
3417 struct lock_chain *chain)
3418{
3419 pr_warn("\n");
3420 pr_warn("============================\n");
3421 pr_warn("WARNING: chain_key collision\n");
3422 print_kernel_ident();
3423 pr_warn("----------------------------\n");
3424 pr_warn("%s/%d: ", current->comm, task_pid_nr(current));
3425 pr_warn("Hash chain already cached but the contents don't match!\n");
3426
3427 pr_warn("Held locks:");
3428 print_chain_keys_held_locks(curr, hlock_next);
3429
3430 pr_warn("Locks in cached chain:");
3431 print_chain_keys_chain(chain);
3432
3433 pr_warn("\nstack backtrace:\n");
3434 dump_stack();
3435}
3436#endif
3437
3438/*
3439 * Checks whether the chain and the current held locks are consistent
3440 * in depth and also in content. If they are not it most likely means
3441 * that there was a collision during the calculation of the chain_key.
3442 * Returns: 0 not passed, 1 passed
3443 */
3444static int check_no_collision(struct task_struct *curr,
3445 struct held_lock *hlock,
3446 struct lock_chain *chain)
3447{
3448#ifdef CONFIG_DEBUG_LOCKDEP
3449 int i, j, id;
3450
3451 i = get_first_held_lock(curr, hlock);
3452
3453 if (DEBUG_LOCKS_WARN_ON(chain->depth != curr->lockdep_depth - (i - 1))) {
3454 print_collision(curr, hlock, chain);
3455 return 0;
3456 }
3457
3458 for (j = 0; j < chain->depth - 1; j++, i++) {
3459 id = hlock_id(&curr->held_locks[i]);
3460
3461 if (DEBUG_LOCKS_WARN_ON(chain_hlocks[chain->base + j] != id)) {
3462 print_collision(curr, hlock, chain);
3463 return 0;
3464 }
3465 }
3466#endif
3467 return 1;
3468}
3469
3470/*
3471 * Given an index that is >= -1, return the index of the next lock chain.
3472 * Return -2 if there is no next lock chain.
3473 */
3474long lockdep_next_lockchain(long i)
3475{
3476 i = find_next_bit(lock_chains_in_use, ARRAY_SIZE(lock_chains), i + 1);
3477 return i < ARRAY_SIZE(lock_chains) ? i : -2;
3478}
3479
3480unsigned long lock_chain_count(void)
3481{
3482 return bitmap_weight(lock_chains_in_use, ARRAY_SIZE(lock_chains));
3483}
3484
3485/* Must be called with the graph lock held. */
3486static struct lock_chain *alloc_lock_chain(void)
3487{
3488 int idx = find_first_zero_bit(lock_chains_in_use,
3489 ARRAY_SIZE(lock_chains));
3490
3491 if (unlikely(idx >= ARRAY_SIZE(lock_chains)))
3492 return NULL;
3493 __set_bit(idx, lock_chains_in_use);
3494 return lock_chains + idx;
3495}
3496
3497/*
3498 * Adds a dependency chain into chain hashtable. And must be called with
3499 * graph_lock held.
3500 *
3501 * Return 0 if fail, and graph_lock is released.
3502 * Return 1 if succeed, with graph_lock held.
3503 */
3504static inline int add_chain_cache(struct task_struct *curr,
3505 struct held_lock *hlock,
3506 u64 chain_key)
3507{
3508 struct hlist_head *hash_head = chainhashentry(chain_key);
3509 struct lock_chain *chain;
3510 int i, j;
3511
3512 /*
3513 * The caller must hold the graph lock, ensure we've got IRQs
3514 * disabled to make this an IRQ-safe lock.. for recursion reasons
3515 * lockdep won't complain about its own locking errors.
3516 */
3517 if (lockdep_assert_locked())
3518 return 0;
3519
3520 chain = alloc_lock_chain();
3521 if (!chain) {
3522 if (!debug_locks_off_graph_unlock())
3523 return 0;
3524
3525 print_lockdep_off("BUG: MAX_LOCKDEP_CHAINS too low!");
3526 dump_stack();
3527 return 0;
3528 }
3529 chain->chain_key = chain_key;
3530 chain->irq_context = hlock->irq_context;
3531 i = get_first_held_lock(curr, hlock);
3532 chain->depth = curr->lockdep_depth + 1 - i;
3533
3534 BUILD_BUG_ON((1UL << 24) <= ARRAY_SIZE(chain_hlocks));
3535 BUILD_BUG_ON((1UL << 6) <= ARRAY_SIZE(curr->held_locks));
3536 BUILD_BUG_ON((1UL << 8*sizeof(chain_hlocks[0])) <= ARRAY_SIZE(lock_classes));
3537
3538 j = alloc_chain_hlocks(chain->depth);
3539 if (j < 0) {
3540 if (!debug_locks_off_graph_unlock())
3541 return 0;
3542
3543 print_lockdep_off("BUG: MAX_LOCKDEP_CHAIN_HLOCKS too low!");
3544 dump_stack();
3545 return 0;
3546 }
3547
3548 chain->base = j;
3549 for (j = 0; j < chain->depth - 1; j++, i++) {
3550 int lock_id = hlock_id(curr->held_locks + i);
3551
3552 chain_hlocks[chain->base + j] = lock_id;
3553 }
3554 chain_hlocks[chain->base + j] = hlock_id(hlock);
3555 hlist_add_head_rcu(&chain->entry, hash_head);
3556 debug_atomic_inc(chain_lookup_misses);
3557 inc_chains(chain->irq_context);
3558
3559 return 1;
3560}
3561
3562/*
3563 * Look up a dependency chain. Must be called with either the graph lock or
3564 * the RCU read lock held.
3565 */
3566static inline struct lock_chain *lookup_chain_cache(u64 chain_key)
3567{
3568 struct hlist_head *hash_head = chainhashentry(chain_key);
3569 struct lock_chain *chain;
3570
3571 hlist_for_each_entry_rcu(chain, hash_head, entry) {
3572 if (READ_ONCE(chain->chain_key) == chain_key) {
3573 debug_atomic_inc(chain_lookup_hits);
3574 return chain;
3575 }
3576 }
3577 return NULL;
3578}
3579
3580/*
3581 * If the key is not present yet in dependency chain cache then
3582 * add it and return 1 - in this case the new dependency chain is
3583 * validated. If the key is already hashed, return 0.
3584 * (On return with 1 graph_lock is held.)
3585 */
3586static inline int lookup_chain_cache_add(struct task_struct *curr,
3587 struct held_lock *hlock,
3588 u64 chain_key)
3589{
3590 struct lock_class *class = hlock_class(hlock);
3591 struct lock_chain *chain = lookup_chain_cache(chain_key);
3592
3593 if (chain) {
3594cache_hit:
3595 if (!check_no_collision(curr, hlock, chain))
3596 return 0;
3597
3598 if (very_verbose(class)) {
3599 printk("\nhash chain already cached, key: "
3600 "%016Lx tail class: [%px] %s\n",
3601 (unsigned long long)chain_key,
3602 class->key, class->name);
3603 }
3604
3605 return 0;
3606 }
3607
3608 if (very_verbose(class)) {
3609 printk("\nnew hash chain, key: %016Lx tail class: [%px] %s\n",
3610 (unsigned long long)chain_key, class->key, class->name);
3611 }
3612
3613 if (!graph_lock())
3614 return 0;
3615
3616 /*
3617 * We have to walk the chain again locked - to avoid duplicates:
3618 */
3619 chain = lookup_chain_cache(chain_key);
3620 if (chain) {
3621 graph_unlock();
3622 goto cache_hit;
3623 }
3624
3625 if (!add_chain_cache(curr, hlock, chain_key))
3626 return 0;
3627
3628 return 1;
3629}
3630
3631static int validate_chain(struct task_struct *curr,
3632 struct held_lock *hlock,
3633 int chain_head, u64 chain_key)
3634{
3635 /*
3636 * Trylock needs to maintain the stack of held locks, but it
3637 * does not add new dependencies, because trylock can be done
3638 * in any order.
3639 *
3640 * We look up the chain_key and do the O(N^2) check and update of
3641 * the dependencies only if this is a new dependency chain.
3642 * (If lookup_chain_cache_add() return with 1 it acquires
3643 * graph_lock for us)
3644 */
3645 if (!hlock->trylock && hlock->check &&
3646 lookup_chain_cache_add(curr, hlock, chain_key)) {
3647 /*
3648 * Check whether last held lock:
3649 *
3650 * - is irq-safe, if this lock is irq-unsafe
3651 * - is softirq-safe, if this lock is hardirq-unsafe
3652 *
3653 * And check whether the new lock's dependency graph
3654 * could lead back to the previous lock:
3655 *
3656 * - within the current held-lock stack
3657 * - across our accumulated lock dependency records
3658 *
3659 * any of these scenarios could lead to a deadlock.
3660 */
3661 /*
3662 * The simple case: does the current hold the same lock
3663 * already?
3664 */
3665 int ret = check_deadlock(curr, hlock);
3666
3667 if (!ret)
3668 return 0;
3669 /*
3670 * Add dependency only if this lock is not the head
3671 * of the chain, and if the new lock introduces no more
3672 * lock dependency (because we already hold a lock with the
3673 * same lock class) nor deadlock (because the nest_lock
3674 * serializes nesting locks), see the comments for
3675 * check_deadlock().
3676 */
3677 if (!chain_head && ret != 2) {
3678 if (!check_prevs_add(curr, hlock))
3679 return 0;
3680 }
3681
3682 graph_unlock();
3683 } else {
3684 /* after lookup_chain_cache_add(): */
3685 if (unlikely(!debug_locks))
3686 return 0;
3687 }
3688
3689 return 1;
3690}
3691#else
3692static inline int validate_chain(struct task_struct *curr,
3693 struct held_lock *hlock,
3694 int chain_head, u64 chain_key)
3695{
3696 return 1;
3697}
3698
3699static void init_chain_block_buckets(void) { }
3700#endif /* CONFIG_PROVE_LOCKING */
3701
3702/*
3703 * We are building curr_chain_key incrementally, so double-check
3704 * it from scratch, to make sure that it's done correctly:
3705 */
3706static void check_chain_key(struct task_struct *curr)
3707{
3708#ifdef CONFIG_DEBUG_LOCKDEP
3709 struct held_lock *hlock, *prev_hlock = NULL;
3710 unsigned int i;
3711 u64 chain_key = INITIAL_CHAIN_KEY;
3712
3713 for (i = 0; i < curr->lockdep_depth; i++) {
3714 hlock = curr->held_locks + i;
3715 if (chain_key != hlock->prev_chain_key) {
3716 debug_locks_off();
3717 /*
3718 * We got mighty confused, our chain keys don't match
3719 * with what we expect, someone trample on our task state?
3720 */
3721 WARN(1, "hm#1, depth: %u [%u], %016Lx != %016Lx\n",
3722 curr->lockdep_depth, i,
3723 (unsigned long long)chain_key,
3724 (unsigned long long)hlock->prev_chain_key);
3725 return;
3726 }
3727
3728 /*
3729 * hlock->class_idx can't go beyond MAX_LOCKDEP_KEYS, but is
3730 * it registered lock class index?
3731 */
3732 if (DEBUG_LOCKS_WARN_ON(!test_bit(hlock->class_idx, lock_classes_in_use)))
3733 return;
3734
3735 if (prev_hlock && (prev_hlock->irq_context !=
3736 hlock->irq_context))
3737 chain_key = INITIAL_CHAIN_KEY;
3738 chain_key = iterate_chain_key(chain_key, hlock_id(hlock));
3739 prev_hlock = hlock;
3740 }
3741 if (chain_key != curr->curr_chain_key) {
3742 debug_locks_off();
3743 /*
3744 * More smoking hash instead of calculating it, damn see these
3745 * numbers float.. I bet that a pink elephant stepped on my memory.
3746 */
3747 WARN(1, "hm#2, depth: %u [%u], %016Lx != %016Lx\n",
3748 curr->lockdep_depth, i,
3749 (unsigned long long)chain_key,
3750 (unsigned long long)curr->curr_chain_key);
3751 }
3752#endif
3753}
3754
3755#ifdef CONFIG_PROVE_LOCKING
3756static int mark_lock(struct task_struct *curr, struct held_lock *this,
3757 enum lock_usage_bit new_bit);
3758
3759static void print_usage_bug_scenario(struct held_lock *lock)
3760{
3761 struct lock_class *class = hlock_class(lock);
3762
3763 printk(" Possible unsafe locking scenario:\n\n");
3764 printk(" CPU0\n");
3765 printk(" ----\n");
3766 printk(" lock(");
3767 __print_lock_name(class);
3768 printk(KERN_CONT ");\n");
3769 printk(" <Interrupt>\n");
3770 printk(" lock(");
3771 __print_lock_name(class);
3772 printk(KERN_CONT ");\n");
3773 printk("\n *** DEADLOCK ***\n\n");
3774}
3775
3776static void
3777print_usage_bug(struct task_struct *curr, struct held_lock *this,
3778 enum lock_usage_bit prev_bit, enum lock_usage_bit new_bit)
3779{
3780 if (!debug_locks_off() || debug_locks_silent)
3781 return;
3782
3783 pr_warn("\n");
3784 pr_warn("================================\n");
3785 pr_warn("WARNING: inconsistent lock state\n");
3786 print_kernel_ident();
3787 pr_warn("--------------------------------\n");
3788
3789 pr_warn("inconsistent {%s} -> {%s} usage.\n",
3790 usage_str[prev_bit], usage_str[new_bit]);
3791
3792 pr_warn("%s/%d [HC%u[%lu]:SC%u[%lu]:HE%u:SE%u] takes:\n",
3793 curr->comm, task_pid_nr(curr),
3794 lockdep_hardirq_context(), hardirq_count() >> HARDIRQ_SHIFT,
3795 lockdep_softirq_context(curr), softirq_count() >> SOFTIRQ_SHIFT,
3796 lockdep_hardirqs_enabled(),
3797 lockdep_softirqs_enabled(curr));
3798 print_lock(this);
3799
3800 pr_warn("{%s} state was registered at:\n", usage_str[prev_bit]);
3801 print_lock_trace(hlock_class(this)->usage_traces[prev_bit], 1);
3802
3803 print_irqtrace_events(curr);
3804 pr_warn("\nother info that might help us debug this:\n");
3805 print_usage_bug_scenario(this);
3806
3807 lockdep_print_held_locks(curr);
3808
3809 pr_warn("\nstack backtrace:\n");
3810 dump_stack();
3811}
3812
3813/*
3814 * Print out an error if an invalid bit is set:
3815 */
3816static inline int
3817valid_state(struct task_struct *curr, struct held_lock *this,
3818 enum lock_usage_bit new_bit, enum lock_usage_bit bad_bit)
3819{
3820 if (unlikely(hlock_class(this)->usage_mask & (1 << bad_bit))) {
3821 graph_unlock();
3822 print_usage_bug(curr, this, bad_bit, new_bit);
3823 return 0;
3824 }
3825 return 1;
3826}
3827
3828
3829/*
3830 * print irq inversion bug:
3831 */
3832static void
3833print_irq_inversion_bug(struct task_struct *curr,
3834 struct lock_list *root, struct lock_list *other,
3835 struct held_lock *this, int forwards,
3836 const char *irqclass)
3837{
3838 struct lock_list *entry = other;
3839 struct lock_list *middle = NULL;
3840 int depth;
3841
3842 if (!debug_locks_off_graph_unlock() || debug_locks_silent)
3843 return;
3844
3845 pr_warn("\n");
3846 pr_warn("========================================================\n");
3847 pr_warn("WARNING: possible irq lock inversion dependency detected\n");
3848 print_kernel_ident();
3849 pr_warn("--------------------------------------------------------\n");
3850 pr_warn("%s/%d just changed the state of lock:\n",
3851 curr->comm, task_pid_nr(curr));
3852 print_lock(this);
3853 if (forwards)
3854 pr_warn("but this lock took another, %s-unsafe lock in the past:\n", irqclass);
3855 else
3856 pr_warn("but this lock was taken by another, %s-safe lock in the past:\n", irqclass);
3857 print_lock_name(other->class);
3858 pr_warn("\n\nand interrupts could create inverse lock ordering between them.\n\n");
3859
3860 pr_warn("\nother info that might help us debug this:\n");
3861
3862 /* Find a middle lock (if one exists) */
3863 depth = get_lock_depth(other);
3864 do {
3865 if (depth == 0 && (entry != root)) {
3866 pr_warn("lockdep:%s bad path found in chain graph\n", __func__);
3867 break;
3868 }
3869 middle = entry;
3870 entry = get_lock_parent(entry);
3871 depth--;
3872 } while (entry && entry != root && (depth >= 0));
3873 if (forwards)
3874 print_irq_lock_scenario(root, other,
3875 middle ? middle->class : root->class, other->class);
3876 else
3877 print_irq_lock_scenario(other, root,
3878 middle ? middle->class : other->class, root->class);
3879
3880 lockdep_print_held_locks(curr);
3881
3882 pr_warn("\nthe shortest dependencies between 2nd lock and 1st lock:\n");
3883 root->trace = save_trace();
3884 if (!root->trace)
3885 return;
3886 print_shortest_lock_dependencies(other, root);
3887
3888 pr_warn("\nstack backtrace:\n");
3889 dump_stack();
3890}
3891
3892/*
3893 * Prove that in the forwards-direction subgraph starting at <this>
3894 * there is no lock matching <mask>:
3895 */
3896static int
3897check_usage_forwards(struct task_struct *curr, struct held_lock *this,
3898 enum lock_usage_bit bit)
3899{
3900 enum bfs_result ret;
3901 struct lock_list root;
3902 struct lock_list *target_entry;
3903 enum lock_usage_bit read_bit = bit + LOCK_USAGE_READ_MASK;
3904 unsigned usage_mask = lock_flag(bit) | lock_flag(read_bit);
3905
3906 bfs_init_root(&root, this);
3907 ret = find_usage_forwards(&root, usage_mask, &target_entry);
3908 if (bfs_error(ret)) {
3909 print_bfs_bug(ret);
3910 return 0;
3911 }
3912 if (ret == BFS_RNOMATCH)
3913 return 1;
3914
3915 /* Check whether write or read usage is the match */
3916 if (target_entry->class->usage_mask & lock_flag(bit)) {
3917 print_irq_inversion_bug(curr, &root, target_entry,
3918 this, 1, state_name(bit));
3919 } else {
3920 print_irq_inversion_bug(curr, &root, target_entry,
3921 this, 1, state_name(read_bit));
3922 }
3923
3924 return 0;
3925}
3926
3927/*
3928 * Prove that in the backwards-direction subgraph starting at <this>
3929 * there is no lock matching <mask>:
3930 */
3931static int
3932check_usage_backwards(struct task_struct *curr, struct held_lock *this,
3933 enum lock_usage_bit bit)
3934{
3935 enum bfs_result ret;
3936 struct lock_list root;
3937 struct lock_list *target_entry;
3938 enum lock_usage_bit read_bit = bit + LOCK_USAGE_READ_MASK;
3939 unsigned usage_mask = lock_flag(bit) | lock_flag(read_bit);
3940
3941 bfs_init_rootb(&root, this);
3942 ret = find_usage_backwards(&root, usage_mask, &target_entry);
3943 if (bfs_error(ret)) {
3944 print_bfs_bug(ret);
3945 return 0;
3946 }
3947 if (ret == BFS_RNOMATCH)
3948 return 1;
3949
3950 /* Check whether write or read usage is the match */
3951 if (target_entry->class->usage_mask & lock_flag(bit)) {
3952 print_irq_inversion_bug(curr, &root, target_entry,
3953 this, 0, state_name(bit));
3954 } else {
3955 print_irq_inversion_bug(curr, &root, target_entry,
3956 this, 0, state_name(read_bit));
3957 }
3958
3959 return 0;
3960}
3961
3962void print_irqtrace_events(struct task_struct *curr)
3963{
3964 const struct irqtrace_events *trace = &curr->irqtrace;
3965
3966 printk("irq event stamp: %u\n", trace->irq_events);
3967 printk("hardirqs last enabled at (%u): [<%px>] %pS\n",
3968 trace->hardirq_enable_event, (void *)trace->hardirq_enable_ip,
3969 (void *)trace->hardirq_enable_ip);
3970 printk("hardirqs last disabled at (%u): [<%px>] %pS\n",
3971 trace->hardirq_disable_event, (void *)trace->hardirq_disable_ip,
3972 (void *)trace->hardirq_disable_ip);
3973 printk("softirqs last enabled at (%u): [<%px>] %pS\n",
3974 trace->softirq_enable_event, (void *)trace->softirq_enable_ip,
3975 (void *)trace->softirq_enable_ip);
3976 printk("softirqs last disabled at (%u): [<%px>] %pS\n",
3977 trace->softirq_disable_event, (void *)trace->softirq_disable_ip,
3978 (void *)trace->softirq_disable_ip);
3979}
3980
3981static int HARDIRQ_verbose(struct lock_class *class)
3982{
3983#if HARDIRQ_VERBOSE
3984 return class_filter(class);
3985#endif
3986 return 0;
3987}
3988
3989static int SOFTIRQ_verbose(struct lock_class *class)
3990{
3991#if SOFTIRQ_VERBOSE
3992 return class_filter(class);
3993#endif
3994 return 0;
3995}
3996
3997static int (*state_verbose_f[])(struct lock_class *class) = {
3998#define LOCKDEP_STATE(__STATE) \
3999 __STATE##_verbose,
4000#include "lockdep_states.h"
4001#undef LOCKDEP_STATE
4002};
4003
4004static inline int state_verbose(enum lock_usage_bit bit,
4005 struct lock_class *class)
4006{
4007 return state_verbose_f[bit >> LOCK_USAGE_DIR_MASK](class);
4008}
4009
4010typedef int (*check_usage_f)(struct task_struct *, struct held_lock *,
4011 enum lock_usage_bit bit, const char *name);
4012
4013static int
4014mark_lock_irq(struct task_struct *curr, struct held_lock *this,
4015 enum lock_usage_bit new_bit)
4016{
4017 int excl_bit = exclusive_bit(new_bit);
4018 int read = new_bit & LOCK_USAGE_READ_MASK;
4019 int dir = new_bit & LOCK_USAGE_DIR_MASK;
4020
4021 /*
4022 * Validate that this particular lock does not have conflicting
4023 * usage states.
4024 */
4025 if (!valid_state(curr, this, new_bit, excl_bit))
4026 return 0;
4027
4028 /*
4029 * Check for read in write conflicts
4030 */
4031 if (!read && !valid_state(curr, this, new_bit,
4032 excl_bit + LOCK_USAGE_READ_MASK))
4033 return 0;
4034
4035
4036 /*
4037 * Validate that the lock dependencies don't have conflicting usage
4038 * states.
4039 */
4040 if (dir) {
4041 /*
4042 * mark ENABLED has to look backwards -- to ensure no dependee
4043 * has USED_IN state, which, again, would allow recursion deadlocks.
4044 */
4045 if (!check_usage_backwards(curr, this, excl_bit))
4046 return 0;
4047 } else {
4048 /*
4049 * mark USED_IN has to look forwards -- to ensure no dependency
4050 * has ENABLED state, which would allow recursion deadlocks.
4051 */
4052 if (!check_usage_forwards(curr, this, excl_bit))
4053 return 0;
4054 }
4055
4056 if (state_verbose(new_bit, hlock_class(this)))
4057 return 2;
4058
4059 return 1;
4060}
4061
4062/*
4063 * Mark all held locks with a usage bit:
4064 */
4065static int
4066mark_held_locks(struct task_struct *curr, enum lock_usage_bit base_bit)
4067{
4068 struct held_lock *hlock;
4069 int i;
4070
4071 for (i = 0; i < curr->lockdep_depth; i++) {
4072 enum lock_usage_bit hlock_bit = base_bit;
4073 hlock = curr->held_locks + i;
4074
4075 if (hlock->read)
4076 hlock_bit += LOCK_USAGE_READ_MASK;
4077
4078 BUG_ON(hlock_bit >= LOCK_USAGE_STATES);
4079
4080 if (!hlock->check)
4081 continue;
4082
4083 if (!mark_lock(curr, hlock, hlock_bit))
4084 return 0;
4085 }
4086
4087 return 1;
4088}
4089
4090/*
4091 * Hardirqs will be enabled:
4092 */
4093static void __trace_hardirqs_on_caller(void)
4094{
4095 struct task_struct *curr = current;
4096
4097 /*
4098 * We are going to turn hardirqs on, so set the
4099 * usage bit for all held locks:
4100 */
4101 if (!mark_held_locks(curr, LOCK_ENABLED_HARDIRQ))
4102 return;
4103 /*
4104 * If we have softirqs enabled, then set the usage
4105 * bit for all held locks. (disabled hardirqs prevented
4106 * this bit from being set before)
4107 */
4108 if (curr->softirqs_enabled)
4109 mark_held_locks(curr, LOCK_ENABLED_SOFTIRQ);
4110}
4111
4112/**
4113 * lockdep_hardirqs_on_prepare - Prepare for enabling interrupts
4114 * @ip: Caller address
4115 *
4116 * Invoked before a possible transition to RCU idle from exit to user or
4117 * guest mode. This ensures that all RCU operations are done before RCU
4118 * stops watching. After the RCU transition lockdep_hardirqs_on() has to be
4119 * invoked to set the final state.
4120 */
4121void lockdep_hardirqs_on_prepare(unsigned long ip)
4122{
4123 if (unlikely(!debug_locks))
4124 return;
4125
4126 /*
4127 * NMIs do not (and cannot) track lock dependencies, nothing to do.
4128 */
4129 if (unlikely(in_nmi()))
4130 return;
4131
4132 if (unlikely(this_cpu_read(lockdep_recursion)))
4133 return;
4134
4135 if (unlikely(lockdep_hardirqs_enabled())) {
4136 /*
4137 * Neither irq nor preemption are disabled here
4138 * so this is racy by nature but losing one hit
4139 * in a stat is not a big deal.
4140 */
4141 __debug_atomic_inc(redundant_hardirqs_on);
4142 return;
4143 }
4144
4145 /*
4146 * We're enabling irqs and according to our state above irqs weren't
4147 * already enabled, yet we find the hardware thinks they are in fact
4148 * enabled.. someone messed up their IRQ state tracing.
4149 */
4150 if (DEBUG_LOCKS_WARN_ON(!irqs_disabled()))
4151 return;
4152
4153 /*
4154 * See the fine text that goes along with this variable definition.
4155 */
4156 if (DEBUG_LOCKS_WARN_ON(early_boot_irqs_disabled))
4157 return;
4158
4159 /*
4160 * Can't allow enabling interrupts while in an interrupt handler,
4161 * that's general bad form and such. Recursion, limited stack etc..
4162 */
4163 if (DEBUG_LOCKS_WARN_ON(lockdep_hardirq_context()))
4164 return;
4165
4166 current->hardirq_chain_key = current->curr_chain_key;
4167
4168 lockdep_recursion_inc();
4169 __trace_hardirqs_on_caller();
4170 lockdep_recursion_finish();
4171}
4172EXPORT_SYMBOL_GPL(lockdep_hardirqs_on_prepare);
4173
4174void noinstr lockdep_hardirqs_on(unsigned long ip)
4175{
4176 struct irqtrace_events *trace = &current->irqtrace;
4177
4178 if (unlikely(!debug_locks))
4179 return;
4180
4181 /*
4182 * NMIs can happen in the middle of local_irq_{en,dis}able() where the
4183 * tracking state and hardware state are out of sync.
4184 *
4185 * NMIs must save lockdep_hardirqs_enabled() to restore IRQ state from,
4186 * and not rely on hardware state like normal interrupts.
4187 */
4188 if (unlikely(in_nmi())) {
4189 if (!IS_ENABLED(CONFIG_TRACE_IRQFLAGS_NMI))
4190 return;
4191
4192 /*
4193 * Skip:
4194 * - recursion check, because NMI can hit lockdep;
4195 * - hardware state check, because above;
4196 * - chain_key check, see lockdep_hardirqs_on_prepare().
4197 */
4198 goto skip_checks;
4199 }
4200
4201 if (unlikely(this_cpu_read(lockdep_recursion)))
4202 return;
4203
4204 if (lockdep_hardirqs_enabled()) {
4205 /*
4206 * Neither irq nor preemption are disabled here
4207 * so this is racy by nature but losing one hit
4208 * in a stat is not a big deal.
4209 */
4210 __debug_atomic_inc(redundant_hardirqs_on);
4211 return;
4212 }
4213
4214 /*
4215 * We're enabling irqs and according to our state above irqs weren't
4216 * already enabled, yet we find the hardware thinks they are in fact
4217 * enabled.. someone messed up their IRQ state tracing.
4218 */
4219 if (DEBUG_LOCKS_WARN_ON(!irqs_disabled()))
4220 return;
4221
4222 /*
4223 * Ensure the lock stack remained unchanged between
4224 * lockdep_hardirqs_on_prepare() and lockdep_hardirqs_on().
4225 */
4226 DEBUG_LOCKS_WARN_ON(current->hardirq_chain_key !=
4227 current->curr_chain_key);
4228
4229skip_checks:
4230 /* we'll do an OFF -> ON transition: */
4231 __this_cpu_write(hardirqs_enabled, 1);
4232 trace->hardirq_enable_ip = ip;
4233 trace->hardirq_enable_event = ++trace->irq_events;
4234 debug_atomic_inc(hardirqs_on_events);
4235}
4236EXPORT_SYMBOL_GPL(lockdep_hardirqs_on);
4237
4238/*
4239 * Hardirqs were disabled:
4240 */
4241void noinstr lockdep_hardirqs_off(unsigned long ip)
4242{
4243 if (unlikely(!debug_locks))
4244 return;
4245
4246 /*
4247 * Matching lockdep_hardirqs_on(), allow NMIs in the middle of lockdep;
4248 * they will restore the software state. This ensures the software
4249 * state is consistent inside NMIs as well.
4250 */
4251 if (in_nmi()) {
4252 if (!IS_ENABLED(CONFIG_TRACE_IRQFLAGS_NMI))
4253 return;
4254 } else if (__this_cpu_read(lockdep_recursion))
4255 return;
4256
4257 /*
4258 * So we're supposed to get called after you mask local IRQs, but for
4259 * some reason the hardware doesn't quite think you did a proper job.
4260 */
4261 if (DEBUG_LOCKS_WARN_ON(!irqs_disabled()))
4262 return;
4263
4264 if (lockdep_hardirqs_enabled()) {
4265 struct irqtrace_events *trace = &current->irqtrace;
4266
4267 /*
4268 * We have done an ON -> OFF transition:
4269 */
4270 __this_cpu_write(hardirqs_enabled, 0);
4271 trace->hardirq_disable_ip = ip;
4272 trace->hardirq_disable_event = ++trace->irq_events;
4273 debug_atomic_inc(hardirqs_off_events);
4274 } else {
4275 debug_atomic_inc(redundant_hardirqs_off);
4276 }
4277}
4278EXPORT_SYMBOL_GPL(lockdep_hardirqs_off);
4279
4280/*
4281 * Softirqs will be enabled:
4282 */
4283void lockdep_softirqs_on(unsigned long ip)
4284{
4285 struct irqtrace_events *trace = &current->irqtrace;
4286
4287 if (unlikely(!lockdep_enabled()))
4288 return;
4289
4290 /*
4291 * We fancy IRQs being disabled here, see softirq.c, avoids
4292 * funny state and nesting things.
4293 */
4294 if (DEBUG_LOCKS_WARN_ON(!irqs_disabled()))
4295 return;
4296
4297 if (current->softirqs_enabled) {
4298 debug_atomic_inc(redundant_softirqs_on);
4299 return;
4300 }
4301
4302 lockdep_recursion_inc();
4303 /*
4304 * We'll do an OFF -> ON transition:
4305 */
4306 current->softirqs_enabled = 1;
4307 trace->softirq_enable_ip = ip;
4308 trace->softirq_enable_event = ++trace->irq_events;
4309 debug_atomic_inc(softirqs_on_events);
4310 /*
4311 * We are going to turn softirqs on, so set the
4312 * usage bit for all held locks, if hardirqs are
4313 * enabled too:
4314 */
4315 if (lockdep_hardirqs_enabled())
4316 mark_held_locks(current, LOCK_ENABLED_SOFTIRQ);
4317 lockdep_recursion_finish();
4318}
4319
4320/*
4321 * Softirqs were disabled:
4322 */
4323void lockdep_softirqs_off(unsigned long ip)
4324{
4325 if (unlikely(!lockdep_enabled()))
4326 return;
4327
4328 /*
4329 * We fancy IRQs being disabled here, see softirq.c
4330 */
4331 if (DEBUG_LOCKS_WARN_ON(!irqs_disabled()))
4332 return;
4333
4334 if (current->softirqs_enabled) {
4335 struct irqtrace_events *trace = &current->irqtrace;
4336
4337 /*
4338 * We have done an ON -> OFF transition:
4339 */
4340 current->softirqs_enabled = 0;
4341 trace->softirq_disable_ip = ip;
4342 trace->softirq_disable_event = ++trace->irq_events;
4343 debug_atomic_inc(softirqs_off_events);
4344 /*
4345 * Whoops, we wanted softirqs off, so why aren't they?
4346 */
4347 DEBUG_LOCKS_WARN_ON(!softirq_count());
4348 } else
4349 debug_atomic_inc(redundant_softirqs_off);
4350}
4351
4352static int
4353mark_usage(struct task_struct *curr, struct held_lock *hlock, int check)
4354{
4355 if (!check)
4356 goto lock_used;
4357
4358 /*
4359 * If non-trylock use in a hardirq or softirq context, then
4360 * mark the lock as used in these contexts:
4361 */
4362 if (!hlock->trylock) {
4363 if (hlock->read) {
4364 if (lockdep_hardirq_context())
4365 if (!mark_lock(curr, hlock,
4366 LOCK_USED_IN_HARDIRQ_READ))
4367 return 0;
4368 if (curr->softirq_context)
4369 if (!mark_lock(curr, hlock,
4370 LOCK_USED_IN_SOFTIRQ_READ))
4371 return 0;
4372 } else {
4373 if (lockdep_hardirq_context())
4374 if (!mark_lock(curr, hlock, LOCK_USED_IN_HARDIRQ))
4375 return 0;
4376 if (curr->softirq_context)
4377 if (!mark_lock(curr, hlock, LOCK_USED_IN_SOFTIRQ))
4378 return 0;
4379 }
4380 }
4381 if (!hlock->hardirqs_off) {
4382 if (hlock->read) {
4383 if (!mark_lock(curr, hlock,
4384 LOCK_ENABLED_HARDIRQ_READ))
4385 return 0;
4386 if (curr->softirqs_enabled)
4387 if (!mark_lock(curr, hlock,
4388 LOCK_ENABLED_SOFTIRQ_READ))
4389 return 0;
4390 } else {
4391 if (!mark_lock(curr, hlock,
4392 LOCK_ENABLED_HARDIRQ))
4393 return 0;
4394 if (curr->softirqs_enabled)
4395 if (!mark_lock(curr, hlock,
4396 LOCK_ENABLED_SOFTIRQ))
4397 return 0;
4398 }
4399 }
4400
4401lock_used:
4402 /* mark it as used: */
4403 if (!mark_lock(curr, hlock, LOCK_USED))
4404 return 0;
4405
4406 return 1;
4407}
4408
4409static inline unsigned int task_irq_context(struct task_struct *task)
4410{
4411 return LOCK_CHAIN_HARDIRQ_CONTEXT * !!lockdep_hardirq_context() +
4412 LOCK_CHAIN_SOFTIRQ_CONTEXT * !!task->softirq_context;
4413}
4414
4415static int separate_irq_context(struct task_struct *curr,
4416 struct held_lock *hlock)
4417{
4418 unsigned int depth = curr->lockdep_depth;
4419
4420 /*
4421 * Keep track of points where we cross into an interrupt context:
4422 */
4423 if (depth) {
4424 struct held_lock *prev_hlock;
4425
4426 prev_hlock = curr->held_locks + depth-1;
4427 /*
4428 * If we cross into another context, reset the
4429 * hash key (this also prevents the checking and the
4430 * adding of the dependency to 'prev'):
4431 */
4432 if (prev_hlock->irq_context != hlock->irq_context)
4433 return 1;
4434 }
4435 return 0;
4436}
4437
4438/*
4439 * Mark a lock with a usage bit, and validate the state transition:
4440 */
4441static int mark_lock(struct task_struct *curr, struct held_lock *this,
4442 enum lock_usage_bit new_bit)
4443{
4444 unsigned int new_mask, ret = 1;
4445
4446 if (new_bit >= LOCK_USAGE_STATES) {
4447 DEBUG_LOCKS_WARN_ON(1);
4448 return 0;
4449 }
4450
4451 if (new_bit == LOCK_USED && this->read)
4452 new_bit = LOCK_USED_READ;
4453
4454 new_mask = 1 << new_bit;
4455
4456 /*
4457 * If already set then do not dirty the cacheline,
4458 * nor do any checks:
4459 */
4460 if (likely(hlock_class(this)->usage_mask & new_mask))
4461 return 1;
4462
4463 if (!graph_lock())
4464 return 0;
4465 /*
4466 * Make sure we didn't race:
4467 */
4468 if (unlikely(hlock_class(this)->usage_mask & new_mask))
4469 goto unlock;
4470
4471 if (!hlock_class(this)->usage_mask)
4472 debug_atomic_dec(nr_unused_locks);
4473
4474 hlock_class(this)->usage_mask |= new_mask;
4475
4476 if (new_bit < LOCK_TRACE_STATES) {
4477 if (!(hlock_class(this)->usage_traces[new_bit] = save_trace()))
4478 return 0;
4479 }
4480
4481 if (new_bit < LOCK_USED) {
4482 ret = mark_lock_irq(curr, this, new_bit);
4483 if (!ret)
4484 return 0;
4485 }
4486
4487unlock:
4488 graph_unlock();
4489
4490 /*
4491 * We must printk outside of the graph_lock:
4492 */
4493 if (ret == 2) {
4494 printk("\nmarked lock as {%s}:\n", usage_str[new_bit]);
4495 print_lock(this);
4496 print_irqtrace_events(curr);
4497 dump_stack();
4498 }
4499
4500 return ret;
4501}
4502
4503static inline short task_wait_context(struct task_struct *curr)
4504{
4505 /*
4506 * Set appropriate wait type for the context; for IRQs we have to take
4507 * into account force_irqthread as that is implied by PREEMPT_RT.
4508 */
4509 if (lockdep_hardirq_context()) {
4510 /*
4511 * Check if force_irqthreads will run us threaded.
4512 */
4513 if (curr->hardirq_threaded || curr->irq_config)
4514 return LD_WAIT_CONFIG;
4515
4516 return LD_WAIT_SPIN;
4517 } else if (curr->softirq_context) {
4518 /*
4519 * Softirqs are always threaded.
4520 */
4521 return LD_WAIT_CONFIG;
4522 }
4523
4524 return LD_WAIT_MAX;
4525}
4526
4527static int
4528print_lock_invalid_wait_context(struct task_struct *curr,
4529 struct held_lock *hlock)
4530{
4531 short curr_inner;
4532
4533 if (!debug_locks_off())
4534 return 0;
4535 if (debug_locks_silent)
4536 return 0;
4537
4538 pr_warn("\n");
4539 pr_warn("=============================\n");
4540 pr_warn("[ BUG: Invalid wait context ]\n");
4541 print_kernel_ident();
4542 pr_warn("-----------------------------\n");
4543
4544 pr_warn("%s/%d is trying to lock:\n", curr->comm, task_pid_nr(curr));
4545 print_lock(hlock);
4546
4547 pr_warn("other info that might help us debug this:\n");
4548
4549 curr_inner = task_wait_context(curr);
4550 pr_warn("context-{%d:%d}\n", curr_inner, curr_inner);
4551
4552 lockdep_print_held_locks(curr);
4553
4554 pr_warn("stack backtrace:\n");
4555 dump_stack();
4556
4557 return 0;
4558}
4559
4560/*
4561 * Verify the wait_type context.
4562 *
4563 * This check validates we takes locks in the right wait-type order; that is it
4564 * ensures that we do not take mutexes inside spinlocks and do not attempt to
4565 * acquire spinlocks inside raw_spinlocks and the sort.
4566 *
4567 * The entire thing is slightly more complex because of RCU, RCU is a lock that
4568 * can be taken from (pretty much) any context but also has constraints.
4569 * However when taken in a stricter environment the RCU lock does not loosen
4570 * the constraints.
4571 *
4572 * Therefore we must look for the strictest environment in the lock stack and
4573 * compare that to the lock we're trying to acquire.
4574 */
4575static int check_wait_context(struct task_struct *curr, struct held_lock *next)
4576{
4577 u8 next_inner = hlock_class(next)->wait_type_inner;
4578 u8 next_outer = hlock_class(next)->wait_type_outer;
4579 u8 curr_inner;
4580 int depth;
4581
4582 if (!curr->lockdep_depth || !next_inner || next->trylock)
4583 return 0;
4584
4585 if (!next_outer)
4586 next_outer = next_inner;
4587
4588 /*
4589 * Find start of current irq_context..
4590 */
4591 for (depth = curr->lockdep_depth - 1; depth >= 0; depth--) {
4592 struct held_lock *prev = curr->held_locks + depth;
4593 if (prev->irq_context != next->irq_context)
4594 break;
4595 }
4596 depth++;
4597
4598 curr_inner = task_wait_context(curr);
4599
4600 for (; depth < curr->lockdep_depth; depth++) {
4601 struct held_lock *prev = curr->held_locks + depth;
4602 u8 prev_inner = hlock_class(prev)->wait_type_inner;
4603
4604 if (prev_inner) {
4605 /*
4606 * We can have a bigger inner than a previous one
4607 * when outer is smaller than inner, as with RCU.
4608 *
4609 * Also due to trylocks.
4610 */
4611 curr_inner = min(curr_inner, prev_inner);
4612 }
4613 }
4614
4615 if (next_outer > curr_inner)
4616 return print_lock_invalid_wait_context(curr, next);
4617
4618 return 0;
4619}
4620
4621#else /* CONFIG_PROVE_LOCKING */
4622
4623static inline int
4624mark_usage(struct task_struct *curr, struct held_lock *hlock, int check)
4625{
4626 return 1;
4627}
4628
4629static inline unsigned int task_irq_context(struct task_struct *task)
4630{
4631 return 0;
4632}
4633
4634static inline int separate_irq_context(struct task_struct *curr,
4635 struct held_lock *hlock)
4636{
4637 return 0;
4638}
4639
4640static inline int check_wait_context(struct task_struct *curr,
4641 struct held_lock *next)
4642{
4643 return 0;
4644}
4645
4646#endif /* CONFIG_PROVE_LOCKING */
4647
4648/*
4649 * Initialize a lock instance's lock-class mapping info:
4650 */
4651void lockdep_init_map_type(struct lockdep_map *lock, const char *name,
4652 struct lock_class_key *key, int subclass,
4653 u8 inner, u8 outer, u8 lock_type)
4654{
4655 int i;
4656
4657 for (i = 0; i < NR_LOCKDEP_CACHING_CLASSES; i++)
4658 lock->class_cache[i] = NULL;
4659
4660#ifdef CONFIG_LOCK_STAT
4661 lock->cpu = raw_smp_processor_id();
4662#endif
4663
4664 /*
4665 * Can't be having no nameless bastards around this place!
4666 */
4667 if (DEBUG_LOCKS_WARN_ON(!name)) {
4668 lock->name = "NULL";
4669 return;
4670 }
4671
4672 lock->name = name;
4673
4674 lock->wait_type_outer = outer;
4675 lock->wait_type_inner = inner;
4676 lock->lock_type = lock_type;
4677
4678 /*
4679 * No key, no joy, we need to hash something.
4680 */
4681 if (DEBUG_LOCKS_WARN_ON(!key))
4682 return;
4683 /*
4684 * Sanity check, the lock-class key must either have been allocated
4685 * statically or must have been registered as a dynamic key.
4686 */
4687 if (!static_obj(key) && !is_dynamic_key(key)) {
4688 if (debug_locks)
4689 printk(KERN_ERR "BUG: key %px has not been registered!\n", key);
4690 DEBUG_LOCKS_WARN_ON(1);
4691 return;
4692 }
4693 lock->key = key;
4694
4695 if (unlikely(!debug_locks))
4696 return;
4697
4698 if (subclass) {
4699 unsigned long flags;
4700
4701 if (DEBUG_LOCKS_WARN_ON(!lockdep_enabled()))
4702 return;
4703
4704 raw_local_irq_save(flags);
4705 lockdep_recursion_inc();
4706 register_lock_class(lock, subclass, 1);
4707 lockdep_recursion_finish();
4708 raw_local_irq_restore(flags);
4709 }
4710}
4711EXPORT_SYMBOL_GPL(lockdep_init_map_type);
4712
4713struct lock_class_key __lockdep_no_validate__;
4714EXPORT_SYMBOL_GPL(__lockdep_no_validate__);
4715
4716static void
4717print_lock_nested_lock_not_held(struct task_struct *curr,
4718 struct held_lock *hlock,
4719 unsigned long ip)
4720{
4721 if (!debug_locks_off())
4722 return;
4723 if (debug_locks_silent)
4724 return;
4725
4726 pr_warn("\n");
4727 pr_warn("==================================\n");
4728 pr_warn("WARNING: Nested lock was not taken\n");
4729 print_kernel_ident();
4730 pr_warn("----------------------------------\n");
4731
4732 pr_warn("%s/%d is trying to lock:\n", curr->comm, task_pid_nr(curr));
4733 print_lock(hlock);
4734
4735 pr_warn("\nbut this task is not holding:\n");
4736 pr_warn("%s\n", hlock->nest_lock->name);
4737
4738 pr_warn("\nstack backtrace:\n");
4739 dump_stack();
4740
4741 pr_warn("\nother info that might help us debug this:\n");
4742 lockdep_print_held_locks(curr);
4743
4744 pr_warn("\nstack backtrace:\n");
4745 dump_stack();
4746}
4747
4748static int __lock_is_held(const struct lockdep_map *lock, int read);
4749
4750/*
4751 * This gets called for every mutex_lock*()/spin_lock*() operation.
4752 * We maintain the dependency maps and validate the locking attempt:
4753 *
4754 * The callers must make sure that IRQs are disabled before calling it,
4755 * otherwise we could get an interrupt which would want to take locks,
4756 * which would end up in lockdep again.
4757 */
4758static int __lock_acquire(struct lockdep_map *lock, unsigned int subclass,
4759 int trylock, int read, int check, int hardirqs_off,
4760 struct lockdep_map *nest_lock, unsigned long ip,
4761 int references, int pin_count)
4762{
4763 struct task_struct *curr = current;
4764 struct lock_class *class = NULL;
4765 struct held_lock *hlock;
4766 unsigned int depth;
4767 int chain_head = 0;
4768 int class_idx;
4769 u64 chain_key;
4770
4771 if (unlikely(!debug_locks))
4772 return 0;
4773
4774 if (!prove_locking || lock->key == &__lockdep_no_validate__)
4775 check = 0;
4776
4777 if (subclass < NR_LOCKDEP_CACHING_CLASSES)
4778 class = lock->class_cache[subclass];
4779 /*
4780 * Not cached?
4781 */
4782 if (unlikely(!class)) {
4783 class = register_lock_class(lock, subclass, 0);
4784 if (!class)
4785 return 0;
4786 }
4787
4788 debug_class_ops_inc(class);
4789
4790 if (very_verbose(class)) {
4791 printk("\nacquire class [%px] %s", class->key, class->name);
4792 if (class->name_version > 1)
4793 printk(KERN_CONT "#%d", class->name_version);
4794 printk(KERN_CONT "\n");
4795 dump_stack();
4796 }
4797
4798 /*
4799 * Add the lock to the list of currently held locks.
4800 * (we dont increase the depth just yet, up until the
4801 * dependency checks are done)
4802 */
4803 depth = curr->lockdep_depth;
4804 /*
4805 * Ran out of static storage for our per-task lock stack again have we?
4806 */
4807 if (DEBUG_LOCKS_WARN_ON(depth >= MAX_LOCK_DEPTH))
4808 return 0;
4809
4810 class_idx = class - lock_classes;
4811
4812 if (depth) { /* we're holding locks */
4813 hlock = curr->held_locks + depth - 1;
4814 if (hlock->class_idx == class_idx && nest_lock) {
4815 if (!references)
4816 references++;
4817
4818 if (!hlock->references)
4819 hlock->references++;
4820
4821 hlock->references += references;
4822
4823 /* Overflow */
4824 if (DEBUG_LOCKS_WARN_ON(hlock->references < references))
4825 return 0;
4826
4827 return 2;
4828 }
4829 }
4830
4831 hlock = curr->held_locks + depth;
4832 /*
4833 * Plain impossible, we just registered it and checked it weren't no
4834 * NULL like.. I bet this mushroom I ate was good!
4835 */
4836 if (DEBUG_LOCKS_WARN_ON(!class))
4837 return 0;
4838 hlock->class_idx = class_idx;
4839 hlock->acquire_ip = ip;
4840 hlock->instance = lock;
4841 hlock->nest_lock = nest_lock;
4842 hlock->irq_context = task_irq_context(curr);
4843 hlock->trylock = trylock;
4844 hlock->read = read;
4845 hlock->check = check;
4846 hlock->hardirqs_off = !!hardirqs_off;
4847 hlock->references = references;
4848#ifdef CONFIG_LOCK_STAT
4849 hlock->waittime_stamp = 0;
4850 hlock->holdtime_stamp = lockstat_clock();
4851#endif
4852 hlock->pin_count = pin_count;
4853
4854 if (check_wait_context(curr, hlock))
4855 return 0;
4856
4857 /* Initialize the lock usage bit */
4858 if (!mark_usage(curr, hlock, check))
4859 return 0;
4860
4861 /*
4862 * Calculate the chain hash: it's the combined hash of all the
4863 * lock keys along the dependency chain. We save the hash value
4864 * at every step so that we can get the current hash easily
4865 * after unlock. The chain hash is then used to cache dependency
4866 * results.
4867 *
4868 * The 'key ID' is what is the most compact key value to drive
4869 * the hash, not class->key.
4870 */
4871 /*
4872 * Whoops, we did it again.. class_idx is invalid.
4873 */
4874 if (DEBUG_LOCKS_WARN_ON(!test_bit(class_idx, lock_classes_in_use)))
4875 return 0;
4876
4877 chain_key = curr->curr_chain_key;
4878 if (!depth) {
4879 /*
4880 * How can we have a chain hash when we ain't got no keys?!
4881 */
4882 if (DEBUG_LOCKS_WARN_ON(chain_key != INITIAL_CHAIN_KEY))
4883 return 0;
4884 chain_head = 1;
4885 }
4886
4887 hlock->prev_chain_key = chain_key;
4888 if (separate_irq_context(curr, hlock)) {
4889 chain_key = INITIAL_CHAIN_KEY;
4890 chain_head = 1;
4891 }
4892 chain_key = iterate_chain_key(chain_key, hlock_id(hlock));
4893
4894 if (nest_lock && !__lock_is_held(nest_lock, -1)) {
4895 print_lock_nested_lock_not_held(curr, hlock, ip);
4896 return 0;
4897 }
4898
4899 if (!debug_locks_silent) {
4900 WARN_ON_ONCE(depth && !hlock_class(hlock - 1)->key);
4901 WARN_ON_ONCE(!hlock_class(hlock)->key);
4902 }
4903
4904 if (!validate_chain(curr, hlock, chain_head, chain_key))
4905 return 0;
4906
4907 curr->curr_chain_key = chain_key;
4908 curr->lockdep_depth++;
4909 check_chain_key(curr);
4910#ifdef CONFIG_DEBUG_LOCKDEP
4911 if (unlikely(!debug_locks))
4912 return 0;
4913#endif
4914 if (unlikely(curr->lockdep_depth >= MAX_LOCK_DEPTH)) {
4915 debug_locks_off();
4916 print_lockdep_off("BUG: MAX_LOCK_DEPTH too low!");
4917 printk(KERN_DEBUG "depth: %i max: %lu!\n",
4918 curr->lockdep_depth, MAX_LOCK_DEPTH);
4919
4920 lockdep_print_held_locks(current);
4921 debug_show_all_locks();
4922 dump_stack();
4923
4924 return 0;
4925 }
4926
4927 if (unlikely(curr->lockdep_depth > max_lockdep_depth))
4928 max_lockdep_depth = curr->lockdep_depth;
4929
4930 return 1;
4931}
4932
4933static void print_unlock_imbalance_bug(struct task_struct *curr,
4934 struct lockdep_map *lock,
4935 unsigned long ip)
4936{
4937 if (!debug_locks_off())
4938 return;
4939 if (debug_locks_silent)
4940 return;
4941
4942 pr_warn("\n");
4943 pr_warn("=====================================\n");
4944 pr_warn("WARNING: bad unlock balance detected!\n");
4945 print_kernel_ident();
4946 pr_warn("-------------------------------------\n");
4947 pr_warn("%s/%d is trying to release lock (",
4948 curr->comm, task_pid_nr(curr));
4949 print_lockdep_cache(lock);
4950 pr_cont(") at:\n");
4951 print_ip_sym(KERN_WARNING, ip);
4952 pr_warn("but there are no more locks to release!\n");
4953 pr_warn("\nother info that might help us debug this:\n");
4954 lockdep_print_held_locks(curr);
4955
4956 pr_warn("\nstack backtrace:\n");
4957 dump_stack();
4958}
4959
4960static noinstr int match_held_lock(const struct held_lock *hlock,
4961 const struct lockdep_map *lock)
4962{
4963 if (hlock->instance == lock)
4964 return 1;
4965
4966 if (hlock->references) {
4967 const struct lock_class *class = lock->class_cache[0];
4968
4969 if (!class)
4970 class = look_up_lock_class(lock, 0);
4971
4972 /*
4973 * If look_up_lock_class() failed to find a class, we're trying
4974 * to test if we hold a lock that has never yet been acquired.
4975 * Clearly if the lock hasn't been acquired _ever_, we're not
4976 * holding it either, so report failure.
4977 */
4978 if (!class)
4979 return 0;
4980
4981 /*
4982 * References, but not a lock we're actually ref-counting?
4983 * State got messed up, follow the sites that change ->references
4984 * and try to make sense of it.
4985 */
4986 if (DEBUG_LOCKS_WARN_ON(!hlock->nest_lock))
4987 return 0;
4988
4989 if (hlock->class_idx == class - lock_classes)
4990 return 1;
4991 }
4992
4993 return 0;
4994}
4995
4996/* @depth must not be zero */
4997static struct held_lock *find_held_lock(struct task_struct *curr,
4998 struct lockdep_map *lock,
4999 unsigned int depth, int *idx)
5000{
5001 struct held_lock *ret, *hlock, *prev_hlock;
5002 int i;
5003
5004 i = depth - 1;
5005 hlock = curr->held_locks + i;
5006 ret = hlock;
5007 if (match_held_lock(hlock, lock))
5008 goto out;
5009
5010 ret = NULL;
5011 for (i--, prev_hlock = hlock--;
5012 i >= 0;
5013 i--, prev_hlock = hlock--) {
5014 /*
5015 * We must not cross into another context:
5016 */
5017 if (prev_hlock->irq_context != hlock->irq_context) {
5018 ret = NULL;
5019 break;
5020 }
5021 if (match_held_lock(hlock, lock)) {
5022 ret = hlock;
5023 break;
5024 }
5025 }
5026
5027out:
5028 *idx = i;
5029 return ret;
5030}
5031
5032static int reacquire_held_locks(struct task_struct *curr, unsigned int depth,
5033 int idx, unsigned int *merged)
5034{
5035 struct held_lock *hlock;
5036 int first_idx = idx;
5037
5038 if (DEBUG_LOCKS_WARN_ON(!irqs_disabled()))
5039 return 0;
5040
5041 for (hlock = curr->held_locks + idx; idx < depth; idx++, hlock++) {
5042 switch (__lock_acquire(hlock->instance,
5043 hlock_class(hlock)->subclass,
5044 hlock->trylock,
5045 hlock->read, hlock->check,
5046 hlock->hardirqs_off,
5047 hlock->nest_lock, hlock->acquire_ip,
5048 hlock->references, hlock->pin_count)) {
5049 case 0:
5050 return 1;
5051 case 1:
5052 break;
5053 case 2:
5054 *merged += (idx == first_idx);
5055 break;
5056 default:
5057 WARN_ON(1);
5058 return 0;
5059 }
5060 }
5061 return 0;
5062}
5063
5064static int
5065__lock_set_class(struct lockdep_map *lock, const char *name,
5066 struct lock_class_key *key, unsigned int subclass,
5067 unsigned long ip)
5068{
5069 struct task_struct *curr = current;
5070 unsigned int depth, merged = 0;
5071 struct held_lock *hlock;
5072 struct lock_class *class;
5073 int i;
5074
5075 if (unlikely(!debug_locks))
5076 return 0;
5077
5078 depth = curr->lockdep_depth;
5079 /*
5080 * This function is about (re)setting the class of a held lock,
5081 * yet we're not actually holding any locks. Naughty user!
5082 */
5083 if (DEBUG_LOCKS_WARN_ON(!depth))
5084 return 0;
5085
5086 hlock = find_held_lock(curr, lock, depth, &i);
5087 if (!hlock) {
5088 print_unlock_imbalance_bug(curr, lock, ip);
5089 return 0;
5090 }
5091
5092 lockdep_init_map_waits(lock, name, key, 0,
5093 lock->wait_type_inner,
5094 lock->wait_type_outer);
5095 class = register_lock_class(lock, subclass, 0);
5096 hlock->class_idx = class - lock_classes;
5097
5098 curr->lockdep_depth = i;
5099 curr->curr_chain_key = hlock->prev_chain_key;
5100
5101 if (reacquire_held_locks(curr, depth, i, &merged))
5102 return 0;
5103
5104 /*
5105 * I took it apart and put it back together again, except now I have
5106 * these 'spare' parts.. where shall I put them.
5107 */
5108 if (DEBUG_LOCKS_WARN_ON(curr->lockdep_depth != depth - merged))
5109 return 0;
5110 return 1;
5111}
5112
5113static int __lock_downgrade(struct lockdep_map *lock, unsigned long ip)
5114{
5115 struct task_struct *curr = current;
5116 unsigned int depth, merged = 0;
5117 struct held_lock *hlock;
5118 int i;
5119
5120 if (unlikely(!debug_locks))
5121 return 0;
5122
5123 depth = curr->lockdep_depth;
5124 /*
5125 * This function is about (re)setting the class of a held lock,
5126 * yet we're not actually holding any locks. Naughty user!
5127 */
5128 if (DEBUG_LOCKS_WARN_ON(!depth))
5129 return 0;
5130
5131 hlock = find_held_lock(curr, lock, depth, &i);
5132 if (!hlock) {
5133 print_unlock_imbalance_bug(curr, lock, ip);
5134 return 0;
5135 }
5136
5137 curr->lockdep_depth = i;
5138 curr->curr_chain_key = hlock->prev_chain_key;
5139
5140 WARN(hlock->read, "downgrading a read lock");
5141 hlock->read = 1;
5142 hlock->acquire_ip = ip;
5143
5144 if (reacquire_held_locks(curr, depth, i, &merged))
5145 return 0;
5146
5147 /* Merging can't happen with unchanged classes.. */
5148 if (DEBUG_LOCKS_WARN_ON(merged))
5149 return 0;
5150
5151 /*
5152 * I took it apart and put it back together again, except now I have
5153 * these 'spare' parts.. where shall I put them.
5154 */
5155 if (DEBUG_LOCKS_WARN_ON(curr->lockdep_depth != depth))
5156 return 0;
5157
5158 return 1;
5159}
5160
5161/*
5162 * Remove the lock from the list of currently held locks - this gets
5163 * called on mutex_unlock()/spin_unlock*() (or on a failed
5164 * mutex_lock_interruptible()).
5165 */
5166static int
5167__lock_release(struct lockdep_map *lock, unsigned long ip)
5168{
5169 struct task_struct *curr = current;
5170 unsigned int depth, merged = 1;
5171 struct held_lock *hlock;
5172 int i;
5173
5174 if (unlikely(!debug_locks))
5175 return 0;
5176
5177 depth = curr->lockdep_depth;
5178 /*
5179 * So we're all set to release this lock.. wait what lock? We don't
5180 * own any locks, you've been drinking again?
5181 */
5182 if (depth <= 0) {
5183 print_unlock_imbalance_bug(curr, lock, ip);
5184 return 0;
5185 }
5186
5187 /*
5188 * Check whether the lock exists in the current stack
5189 * of held locks:
5190 */
5191 hlock = find_held_lock(curr, lock, depth, &i);
5192 if (!hlock) {
5193 print_unlock_imbalance_bug(curr, lock, ip);
5194 return 0;
5195 }
5196
5197 if (hlock->instance == lock)
5198 lock_release_holdtime(hlock);
5199
5200 WARN(hlock->pin_count, "releasing a pinned lock\n");
5201
5202 if (hlock->references) {
5203 hlock->references--;
5204 if (hlock->references) {
5205 /*
5206 * We had, and after removing one, still have
5207 * references, the current lock stack is still
5208 * valid. We're done!
5209 */
5210 return 1;
5211 }
5212 }
5213
5214 /*
5215 * We have the right lock to unlock, 'hlock' points to it.
5216 * Now we remove it from the stack, and add back the other
5217 * entries (if any), recalculating the hash along the way:
5218 */
5219
5220 curr->lockdep_depth = i;
5221 curr->curr_chain_key = hlock->prev_chain_key;
5222
5223 /*
5224 * The most likely case is when the unlock is on the innermost
5225 * lock. In this case, we are done!
5226 */
5227 if (i == depth-1)
5228 return 1;
5229
5230 if (reacquire_held_locks(curr, depth, i + 1, &merged))
5231 return 0;
5232
5233 /*
5234 * We had N bottles of beer on the wall, we drank one, but now
5235 * there's not N-1 bottles of beer left on the wall...
5236 * Pouring two of the bottles together is acceptable.
5237 */
5238 DEBUG_LOCKS_WARN_ON(curr->lockdep_depth != depth - merged);
5239
5240 /*
5241 * Since reacquire_held_locks() would have called check_chain_key()
5242 * indirectly via __lock_acquire(), we don't need to do it again
5243 * on return.
5244 */
5245 return 0;
5246}
5247
5248static __always_inline
5249int __lock_is_held(const struct lockdep_map *lock, int read)
5250{
5251 struct task_struct *curr = current;
5252 int i;
5253
5254 for (i = 0; i < curr->lockdep_depth; i++) {
5255 struct held_lock *hlock = curr->held_locks + i;
5256
5257 if (match_held_lock(hlock, lock)) {
5258 if (read == -1 || hlock->read == read)
5259 return LOCK_STATE_HELD;
5260
5261 return LOCK_STATE_NOT_HELD;
5262 }
5263 }
5264
5265 return LOCK_STATE_NOT_HELD;
5266}
5267
5268static struct pin_cookie __lock_pin_lock(struct lockdep_map *lock)
5269{
5270 struct pin_cookie cookie = NIL_COOKIE;
5271 struct task_struct *curr = current;
5272 int i;
5273
5274 if (unlikely(!debug_locks))
5275 return cookie;
5276
5277 for (i = 0; i < curr->lockdep_depth; i++) {
5278 struct held_lock *hlock = curr->held_locks + i;
5279
5280 if (match_held_lock(hlock, lock)) {
5281 /*
5282 * Grab 16bits of randomness; this is sufficient to not
5283 * be guessable and still allows some pin nesting in
5284 * our u32 pin_count.
5285 */
5286 cookie.val = 1 + (prandom_u32() >> 16);
5287 hlock->pin_count += cookie.val;
5288 return cookie;
5289 }
5290 }
5291
5292 WARN(1, "pinning an unheld lock\n");
5293 return cookie;
5294}
5295
5296static void __lock_repin_lock(struct lockdep_map *lock, struct pin_cookie cookie)
5297{
5298 struct task_struct *curr = current;
5299 int i;
5300
5301 if (unlikely(!debug_locks))
5302 return;
5303
5304 for (i = 0; i < curr->lockdep_depth; i++) {
5305 struct held_lock *hlock = curr->held_locks + i;
5306
5307 if (match_held_lock(hlock, lock)) {
5308 hlock->pin_count += cookie.val;
5309 return;
5310 }
5311 }
5312
5313 WARN(1, "pinning an unheld lock\n");
5314}
5315
5316static void __lock_unpin_lock(struct lockdep_map *lock, struct pin_cookie cookie)
5317{
5318 struct task_struct *curr = current;
5319 int i;
5320
5321 if (unlikely(!debug_locks))
5322 return;
5323
5324 for (i = 0; i < curr->lockdep_depth; i++) {
5325 struct held_lock *hlock = curr->held_locks + i;
5326
5327 if (match_held_lock(hlock, lock)) {
5328 if (WARN(!hlock->pin_count, "unpinning an unpinned lock\n"))
5329 return;
5330
5331 hlock->pin_count -= cookie.val;
5332
5333 if (WARN((int)hlock->pin_count < 0, "pin count corrupted\n"))
5334 hlock->pin_count = 0;
5335
5336 return;
5337 }
5338 }
5339
5340 WARN(1, "unpinning an unheld lock\n");
5341}
5342
5343/*
5344 * Check whether we follow the irq-flags state precisely:
5345 */
5346static noinstr void check_flags(unsigned long flags)
5347{
5348#if defined(CONFIG_PROVE_LOCKING) && defined(CONFIG_DEBUG_LOCKDEP)
5349 if (!debug_locks)
5350 return;
5351
5352 /* Get the warning out.. */
5353 instrumentation_begin();
5354
5355 if (irqs_disabled_flags(flags)) {
5356 if (DEBUG_LOCKS_WARN_ON(lockdep_hardirqs_enabled())) {
5357 printk("possible reason: unannotated irqs-off.\n");
5358 }
5359 } else {
5360 if (DEBUG_LOCKS_WARN_ON(!lockdep_hardirqs_enabled())) {
5361 printk("possible reason: unannotated irqs-on.\n");
5362 }
5363 }
5364
5365 /*
5366 * We dont accurately track softirq state in e.g.
5367 * hardirq contexts (such as on 4KSTACKS), so only
5368 * check if not in hardirq contexts:
5369 */
5370 if (!hardirq_count()) {
5371 if (softirq_count()) {
5372 /* like the above, but with softirqs */
5373 DEBUG_LOCKS_WARN_ON(current->softirqs_enabled);
5374 } else {
5375 /* lick the above, does it taste good? */
5376 DEBUG_LOCKS_WARN_ON(!current->softirqs_enabled);
5377 }
5378 }
5379
5380 if (!debug_locks)
5381 print_irqtrace_events(current);
5382
5383 instrumentation_end();
5384#endif
5385}
5386
5387void lock_set_class(struct lockdep_map *lock, const char *name,
5388 struct lock_class_key *key, unsigned int subclass,
5389 unsigned long ip)
5390{
5391 unsigned long flags;
5392
5393 if (unlikely(!lockdep_enabled()))
5394 return;
5395
5396 raw_local_irq_save(flags);
5397 lockdep_recursion_inc();
5398 check_flags(flags);
5399 if (__lock_set_class(lock, name, key, subclass, ip))
5400 check_chain_key(current);
5401 lockdep_recursion_finish();
5402 raw_local_irq_restore(flags);
5403}
5404EXPORT_SYMBOL_GPL(lock_set_class);
5405
5406void lock_downgrade(struct lockdep_map *lock, unsigned long ip)
5407{
5408 unsigned long flags;
5409
5410 if (unlikely(!lockdep_enabled()))
5411 return;
5412
5413 raw_local_irq_save(flags);
5414 lockdep_recursion_inc();
5415 check_flags(flags);
5416 if (__lock_downgrade(lock, ip))
5417 check_chain_key(current);
5418 lockdep_recursion_finish();
5419 raw_local_irq_restore(flags);
5420}
5421EXPORT_SYMBOL_GPL(lock_downgrade);
5422
5423/* NMI context !!! */
5424static void verify_lock_unused(struct lockdep_map *lock, struct held_lock *hlock, int subclass)
5425{
5426#ifdef CONFIG_PROVE_LOCKING
5427 struct lock_class *class = look_up_lock_class(lock, subclass);
5428 unsigned long mask = LOCKF_USED;
5429
5430 /* if it doesn't have a class (yet), it certainly hasn't been used yet */
5431 if (!class)
5432 return;
5433
5434 /*
5435 * READ locks only conflict with USED, such that if we only ever use
5436 * READ locks, there is no deadlock possible -- RCU.
5437 */
5438 if (!hlock->read)
5439 mask |= LOCKF_USED_READ;
5440
5441 if (!(class->usage_mask & mask))
5442 return;
5443
5444 hlock->class_idx = class - lock_classes;
5445
5446 print_usage_bug(current, hlock, LOCK_USED, LOCK_USAGE_STATES);
5447#endif
5448}
5449
5450static bool lockdep_nmi(void)
5451{
5452 if (raw_cpu_read(lockdep_recursion))
5453 return false;
5454
5455 if (!in_nmi())
5456 return false;
5457
5458 return true;
5459}
5460
5461/*
5462 * read_lock() is recursive if:
5463 * 1. We force lockdep think this way in selftests or
5464 * 2. The implementation is not queued read/write lock or
5465 * 3. The locker is at an in_interrupt() context.
5466 */
5467bool read_lock_is_recursive(void)
5468{
5469 return force_read_lock_recursive ||
5470 !IS_ENABLED(CONFIG_QUEUED_RWLOCKS) ||
5471 in_interrupt();
5472}
5473EXPORT_SYMBOL_GPL(read_lock_is_recursive);
5474
5475/*
5476 * We are not always called with irqs disabled - do that here,
5477 * and also avoid lockdep recursion:
5478 */
5479void lock_acquire(struct lockdep_map *lock, unsigned int subclass,
5480 int trylock, int read, int check,
5481 struct lockdep_map *nest_lock, unsigned long ip)
5482{
5483 unsigned long flags;
5484
5485 trace_lock_acquire(lock, subclass, trylock, read, check, nest_lock, ip);
5486
5487 if (!debug_locks)
5488 return;
5489
5490 if (unlikely(!lockdep_enabled())) {
5491 /* XXX allow trylock from NMI ?!? */
5492 if (lockdep_nmi() && !trylock) {
5493 struct held_lock hlock;
5494
5495 hlock.acquire_ip = ip;
5496 hlock.instance = lock;
5497 hlock.nest_lock = nest_lock;
5498 hlock.irq_context = 2; // XXX
5499 hlock.trylock = trylock;
5500 hlock.read = read;
5501 hlock.check = check;
5502 hlock.hardirqs_off = true;
5503 hlock.references = 0;
5504
5505 verify_lock_unused(lock, &hlock, subclass);
5506 }
5507 return;
5508 }
5509
5510 raw_local_irq_save(flags);
5511 check_flags(flags);
5512
5513 lockdep_recursion_inc();
5514 __lock_acquire(lock, subclass, trylock, read, check,
5515 irqs_disabled_flags(flags), nest_lock, ip, 0, 0);
5516 lockdep_recursion_finish();
5517 raw_local_irq_restore(flags);
5518}
5519EXPORT_SYMBOL_GPL(lock_acquire);
5520
5521void lock_release(struct lockdep_map *lock, unsigned long ip)
5522{
5523 unsigned long flags;
5524
5525 trace_lock_release(lock, ip);
5526
5527 if (unlikely(!lockdep_enabled()))
5528 return;
5529
5530 raw_local_irq_save(flags);
5531 check_flags(flags);
5532
5533 lockdep_recursion_inc();
5534 if (__lock_release(lock, ip))
5535 check_chain_key(current);
5536 lockdep_recursion_finish();
5537 raw_local_irq_restore(flags);
5538}
5539EXPORT_SYMBOL_GPL(lock_release);
5540
5541noinstr int lock_is_held_type(const struct lockdep_map *lock, int read)
5542{
5543 unsigned long flags;
5544 int ret = LOCK_STATE_NOT_HELD;
5545
5546 /*
5547 * Avoid false negative lockdep_assert_held() and
5548 * lockdep_assert_not_held().
5549 */
5550 if (unlikely(!lockdep_enabled()))
5551 return LOCK_STATE_UNKNOWN;
5552
5553 raw_local_irq_save(flags);
5554 check_flags(flags);
5555
5556 lockdep_recursion_inc();
5557 ret = __lock_is_held(lock, read);
5558 lockdep_recursion_finish();
5559 raw_local_irq_restore(flags);
5560
5561 return ret;
5562}
5563EXPORT_SYMBOL_GPL(lock_is_held_type);
5564NOKPROBE_SYMBOL(lock_is_held_type);
5565
5566struct pin_cookie lock_pin_lock(struct lockdep_map *lock)
5567{
5568 struct pin_cookie cookie = NIL_COOKIE;
5569 unsigned long flags;
5570
5571 if (unlikely(!lockdep_enabled()))
5572 return cookie;
5573
5574 raw_local_irq_save(flags);
5575 check_flags(flags);
5576
5577 lockdep_recursion_inc();
5578 cookie = __lock_pin_lock(lock);
5579 lockdep_recursion_finish();
5580 raw_local_irq_restore(flags);
5581
5582 return cookie;
5583}
5584EXPORT_SYMBOL_GPL(lock_pin_lock);
5585
5586void lock_repin_lock(struct lockdep_map *lock, struct pin_cookie cookie)
5587{
5588 unsigned long flags;
5589
5590 if (unlikely(!lockdep_enabled()))
5591 return;
5592
5593 raw_local_irq_save(flags);
5594 check_flags(flags);
5595
5596 lockdep_recursion_inc();
5597 __lock_repin_lock(lock, cookie);
5598 lockdep_recursion_finish();
5599 raw_local_irq_restore(flags);
5600}
5601EXPORT_SYMBOL_GPL(lock_repin_lock);
5602
5603void lock_unpin_lock(struct lockdep_map *lock, struct pin_cookie cookie)
5604{
5605 unsigned long flags;
5606
5607 if (unlikely(!lockdep_enabled()))
5608 return;
5609
5610 raw_local_irq_save(flags);
5611 check_flags(flags);
5612
5613 lockdep_recursion_inc();
5614 __lock_unpin_lock(lock, cookie);
5615 lockdep_recursion_finish();
5616 raw_local_irq_restore(flags);
5617}
5618EXPORT_SYMBOL_GPL(lock_unpin_lock);
5619
5620#ifdef CONFIG_LOCK_STAT
5621static void print_lock_contention_bug(struct task_struct *curr,
5622 struct lockdep_map *lock,
5623 unsigned long ip)
5624{
5625 if (!debug_locks_off())
5626 return;
5627 if (debug_locks_silent)
5628 return;
5629
5630 pr_warn("\n");
5631 pr_warn("=================================\n");
5632 pr_warn("WARNING: bad contention detected!\n");
5633 print_kernel_ident();
5634 pr_warn("---------------------------------\n");
5635 pr_warn("%s/%d is trying to contend lock (",
5636 curr->comm, task_pid_nr(curr));
5637 print_lockdep_cache(lock);
5638 pr_cont(") at:\n");
5639 print_ip_sym(KERN_WARNING, ip);
5640 pr_warn("but there are no locks held!\n");
5641 pr_warn("\nother info that might help us debug this:\n");
5642 lockdep_print_held_locks(curr);
5643
5644 pr_warn("\nstack backtrace:\n");
5645 dump_stack();
5646}
5647
5648static void
5649__lock_contended(struct lockdep_map *lock, unsigned long ip)
5650{
5651 struct task_struct *curr = current;
5652 struct held_lock *hlock;
5653 struct lock_class_stats *stats;
5654 unsigned int depth;
5655 int i, contention_point, contending_point;
5656
5657 depth = curr->lockdep_depth;
5658 /*
5659 * Whee, we contended on this lock, except it seems we're not
5660 * actually trying to acquire anything much at all..
5661 */
5662 if (DEBUG_LOCKS_WARN_ON(!depth))
5663 return;
5664
5665 hlock = find_held_lock(curr, lock, depth, &i);
5666 if (!hlock) {
5667 print_lock_contention_bug(curr, lock, ip);
5668 return;
5669 }
5670
5671 if (hlock->instance != lock)
5672 return;
5673
5674 hlock->waittime_stamp = lockstat_clock();
5675
5676 contention_point = lock_point(hlock_class(hlock)->contention_point, ip);
5677 contending_point = lock_point(hlock_class(hlock)->contending_point,
5678 lock->ip);
5679
5680 stats = get_lock_stats(hlock_class(hlock));
5681 if (contention_point < LOCKSTAT_POINTS)
5682 stats->contention_point[contention_point]++;
5683 if (contending_point < LOCKSTAT_POINTS)
5684 stats->contending_point[contending_point]++;
5685 if (lock->cpu != smp_processor_id())
5686 stats->bounces[bounce_contended + !!hlock->read]++;
5687}
5688
5689static void
5690__lock_acquired(struct lockdep_map *lock, unsigned long ip)
5691{
5692 struct task_struct *curr = current;
5693 struct held_lock *hlock;
5694 struct lock_class_stats *stats;
5695 unsigned int depth;
5696 u64 now, waittime = 0;
5697 int i, cpu;
5698
5699 depth = curr->lockdep_depth;
5700 /*
5701 * Yay, we acquired ownership of this lock we didn't try to
5702 * acquire, how the heck did that happen?
5703 */
5704 if (DEBUG_LOCKS_WARN_ON(!depth))
5705 return;
5706
5707 hlock = find_held_lock(curr, lock, depth, &i);
5708 if (!hlock) {
5709 print_lock_contention_bug(curr, lock, _RET_IP_);
5710 return;
5711 }
5712
5713 if (hlock->instance != lock)
5714 return;
5715
5716 cpu = smp_processor_id();
5717 if (hlock->waittime_stamp) {
5718 now = lockstat_clock();
5719 waittime = now - hlock->waittime_stamp;
5720 hlock->holdtime_stamp = now;
5721 }
5722
5723 stats = get_lock_stats(hlock_class(hlock));
5724 if (waittime) {
5725 if (hlock->read)
5726 lock_time_inc(&stats->read_waittime, waittime);
5727 else
5728 lock_time_inc(&stats->write_waittime, waittime);
5729 }
5730 if (lock->cpu != cpu)
5731 stats->bounces[bounce_acquired + !!hlock->read]++;
5732
5733 lock->cpu = cpu;
5734 lock->ip = ip;
5735}
5736
5737void lock_contended(struct lockdep_map *lock, unsigned long ip)
5738{
5739 unsigned long flags;
5740
5741 trace_lock_contended(lock, ip);
5742
5743 if (unlikely(!lock_stat || !lockdep_enabled()))
5744 return;
5745
5746 raw_local_irq_save(flags);
5747 check_flags(flags);
5748 lockdep_recursion_inc();
5749 __lock_contended(lock, ip);
5750 lockdep_recursion_finish();
5751 raw_local_irq_restore(flags);
5752}
5753EXPORT_SYMBOL_GPL(lock_contended);
5754
5755void lock_acquired(struct lockdep_map *lock, unsigned long ip)
5756{
5757 unsigned long flags;
5758
5759 trace_lock_acquired(lock, ip);
5760
5761 if (unlikely(!lock_stat || !lockdep_enabled()))
5762 return;
5763
5764 raw_local_irq_save(flags);
5765 check_flags(flags);
5766 lockdep_recursion_inc();
5767 __lock_acquired(lock, ip);
5768 lockdep_recursion_finish();
5769 raw_local_irq_restore(flags);
5770}
5771EXPORT_SYMBOL_GPL(lock_acquired);
5772#endif
5773
5774/*
5775 * Used by the testsuite, sanitize the validator state
5776 * after a simulated failure:
5777 */
5778
5779void lockdep_reset(void)
5780{
5781 unsigned long flags;
5782 int i;
5783
5784 raw_local_irq_save(flags);
5785 lockdep_init_task(current);
5786 memset(current->held_locks, 0, MAX_LOCK_DEPTH*sizeof(struct held_lock));
5787 nr_hardirq_chains = 0;
5788 nr_softirq_chains = 0;
5789 nr_process_chains = 0;
5790 debug_locks = 1;
5791 for (i = 0; i < CHAINHASH_SIZE; i++)
5792 INIT_HLIST_HEAD(chainhash_table + i);
5793 raw_local_irq_restore(flags);
5794}
5795
5796/* Remove a class from a lock chain. Must be called with the graph lock held. */
5797static void remove_class_from_lock_chain(struct pending_free *pf,
5798 struct lock_chain *chain,
5799 struct lock_class *class)
5800{
5801#ifdef CONFIG_PROVE_LOCKING
5802 int i;
5803
5804 for (i = chain->base; i < chain->base + chain->depth; i++) {
5805 if (chain_hlock_class_idx(chain_hlocks[i]) != class - lock_classes)
5806 continue;
5807 /*
5808 * Each lock class occurs at most once in a lock chain so once
5809 * we found a match we can break out of this loop.
5810 */
5811 goto free_lock_chain;
5812 }
5813 /* Since the chain has not been modified, return. */
5814 return;
5815
5816free_lock_chain:
5817 free_chain_hlocks(chain->base, chain->depth);
5818 /* Overwrite the chain key for concurrent RCU readers. */
5819 WRITE_ONCE(chain->chain_key, INITIAL_CHAIN_KEY);
5820 dec_chains(chain->irq_context);
5821
5822 /*
5823 * Note: calling hlist_del_rcu() from inside a
5824 * hlist_for_each_entry_rcu() loop is safe.
5825 */
5826 hlist_del_rcu(&chain->entry);
5827 __set_bit(chain - lock_chains, pf->lock_chains_being_freed);
5828 nr_zapped_lock_chains++;
5829#endif
5830}
5831
5832/* Must be called with the graph lock held. */
5833static void remove_class_from_lock_chains(struct pending_free *pf,
5834 struct lock_class *class)
5835{
5836 struct lock_chain *chain;
5837 struct hlist_head *head;
5838 int i;
5839
5840 for (i = 0; i < ARRAY_SIZE(chainhash_table); i++) {
5841 head = chainhash_table + i;
5842 hlist_for_each_entry_rcu(chain, head, entry) {
5843 remove_class_from_lock_chain(pf, chain, class);
5844 }
5845 }
5846}
5847
5848/*
5849 * Remove all references to a lock class. The caller must hold the graph lock.
5850 */
5851static void zap_class(struct pending_free *pf, struct lock_class *class)
5852{
5853 struct lock_list *entry;
5854 int i;
5855
5856 WARN_ON_ONCE(!class->key);
5857
5858 /*
5859 * Remove all dependencies this lock is
5860 * involved in:
5861 */
5862 for_each_set_bit(i, list_entries_in_use, ARRAY_SIZE(list_entries)) {
5863 entry = list_entries + i;
5864 if (entry->class != class && entry->links_to != class)
5865 continue;
5866 __clear_bit(i, list_entries_in_use);
5867 nr_list_entries--;
5868 list_del_rcu(&entry->entry);
5869 }
5870 if (list_empty(&class->locks_after) &&
5871 list_empty(&class->locks_before)) {
5872 list_move_tail(&class->lock_entry, &pf->zapped);
5873 hlist_del_rcu(&class->hash_entry);
5874 WRITE_ONCE(class->key, NULL);
5875 WRITE_ONCE(class->name, NULL);
5876 nr_lock_classes--;
5877 __clear_bit(class - lock_classes, lock_classes_in_use);
5878 } else {
5879 WARN_ONCE(true, "%s() failed for class %s\n", __func__,
5880 class->name);
5881 }
5882
5883 remove_class_from_lock_chains(pf, class);
5884 nr_zapped_classes++;
5885}
5886
5887static void reinit_class(struct lock_class *class)
5888{
5889 void *const p = class;
5890 const unsigned int offset = offsetof(struct lock_class, key);
5891
5892 WARN_ON_ONCE(!class->lock_entry.next);
5893 WARN_ON_ONCE(!list_empty(&class->locks_after));
5894 WARN_ON_ONCE(!list_empty(&class->locks_before));
5895 memset(p + offset, 0, sizeof(*class) - offset);
5896 WARN_ON_ONCE(!class->lock_entry.next);
5897 WARN_ON_ONCE(!list_empty(&class->locks_after));
5898 WARN_ON_ONCE(!list_empty(&class->locks_before));
5899}
5900
5901static inline int within(const void *addr, void *start, unsigned long size)
5902{
5903 return addr >= start && addr < start + size;
5904}
5905
5906static bool inside_selftest(void)
5907{
5908 return current == lockdep_selftest_task_struct;
5909}
5910
5911/* The caller must hold the graph lock. */
5912static struct pending_free *get_pending_free(void)
5913{
5914 return delayed_free.pf + delayed_free.index;
5915}
5916
5917static void free_zapped_rcu(struct rcu_head *cb);
5918
5919/*
5920 * Schedule an RCU callback if no RCU callback is pending. Must be called with
5921 * the graph lock held.
5922 */
5923static void call_rcu_zapped(struct pending_free *pf)
5924{
5925 WARN_ON_ONCE(inside_selftest());
5926
5927 if (list_empty(&pf->zapped))
5928 return;
5929
5930 if (delayed_free.scheduled)
5931 return;
5932
5933 delayed_free.scheduled = true;
5934
5935 WARN_ON_ONCE(delayed_free.pf + delayed_free.index != pf);
5936 delayed_free.index ^= 1;
5937
5938 call_rcu(&delayed_free.rcu_head, free_zapped_rcu);
5939}
5940
5941/* The caller must hold the graph lock. May be called from RCU context. */
5942static void __free_zapped_classes(struct pending_free *pf)
5943{
5944 struct lock_class *class;
5945
5946 check_data_structures();
5947
5948 list_for_each_entry(class, &pf->zapped, lock_entry)
5949 reinit_class(class);
5950
5951 list_splice_init(&pf->zapped, &free_lock_classes);
5952
5953#ifdef CONFIG_PROVE_LOCKING
5954 bitmap_andnot(lock_chains_in_use, lock_chains_in_use,
5955 pf->lock_chains_being_freed, ARRAY_SIZE(lock_chains));
5956 bitmap_clear(pf->lock_chains_being_freed, 0, ARRAY_SIZE(lock_chains));
5957#endif
5958}
5959
5960static void free_zapped_rcu(struct rcu_head *ch)
5961{
5962 struct pending_free *pf;
5963 unsigned long flags;
5964
5965 if (WARN_ON_ONCE(ch != &delayed_free.rcu_head))
5966 return;
5967
5968 raw_local_irq_save(flags);
5969 lockdep_lock();
5970
5971 /* closed head */
5972 pf = delayed_free.pf + (delayed_free.index ^ 1);
5973 __free_zapped_classes(pf);
5974 delayed_free.scheduled = false;
5975
5976 /*
5977 * If there's anything on the open list, close and start a new callback.
5978 */
5979 call_rcu_zapped(delayed_free.pf + delayed_free.index);
5980
5981 lockdep_unlock();
5982 raw_local_irq_restore(flags);
5983}
5984
5985/*
5986 * Remove all lock classes from the class hash table and from the
5987 * all_lock_classes list whose key or name is in the address range [start,
5988 * start + size). Move these lock classes to the zapped_classes list. Must
5989 * be called with the graph lock held.
5990 */
5991static void __lockdep_free_key_range(struct pending_free *pf, void *start,
5992 unsigned long size)
5993{
5994 struct lock_class *class;
5995 struct hlist_head *head;
5996 int i;
5997
5998 /* Unhash all classes that were created by a module. */
5999 for (i = 0; i < CLASSHASH_SIZE; i++) {
6000 head = classhash_table + i;
6001 hlist_for_each_entry_rcu(class, head, hash_entry) {
6002 if (!within(class->key, start, size) &&
6003 !within(class->name, start, size))
6004 continue;
6005 zap_class(pf, class);
6006 }
6007 }
6008}
6009
6010/*
6011 * Used in module.c to remove lock classes from memory that is going to be
6012 * freed; and possibly re-used by other modules.
6013 *
6014 * We will have had one synchronize_rcu() before getting here, so we're
6015 * guaranteed nobody will look up these exact classes -- they're properly dead
6016 * but still allocated.
6017 */
6018static void lockdep_free_key_range_reg(void *start, unsigned long size)
6019{
6020 struct pending_free *pf;
6021 unsigned long flags;
6022
6023 init_data_structures_once();
6024
6025 raw_local_irq_save(flags);
6026 lockdep_lock();
6027 pf = get_pending_free();
6028 __lockdep_free_key_range(pf, start, size);
6029 call_rcu_zapped(pf);
6030 lockdep_unlock();
6031 raw_local_irq_restore(flags);
6032
6033 /*
6034 * Wait for any possible iterators from look_up_lock_class() to pass
6035 * before continuing to free the memory they refer to.
6036 */
6037 synchronize_rcu();
6038}
6039
6040/*
6041 * Free all lockdep keys in the range [start, start+size). Does not sleep.
6042 * Ignores debug_locks. Must only be used by the lockdep selftests.
6043 */
6044static void lockdep_free_key_range_imm(void *start, unsigned long size)
6045{
6046 struct pending_free *pf = delayed_free.pf;
6047 unsigned long flags;
6048
6049 init_data_structures_once();
6050
6051 raw_local_irq_save(flags);
6052 lockdep_lock();
6053 __lockdep_free_key_range(pf, start, size);
6054 __free_zapped_classes(pf);
6055 lockdep_unlock();
6056 raw_local_irq_restore(flags);
6057}
6058
6059void lockdep_free_key_range(void *start, unsigned long size)
6060{
6061 init_data_structures_once();
6062
6063 if (inside_selftest())
6064 lockdep_free_key_range_imm(start, size);
6065 else
6066 lockdep_free_key_range_reg(start, size);
6067}
6068
6069/*
6070 * Check whether any element of the @lock->class_cache[] array refers to a
6071 * registered lock class. The caller must hold either the graph lock or the
6072 * RCU read lock.
6073 */
6074static bool lock_class_cache_is_registered(struct lockdep_map *lock)
6075{
6076 struct lock_class *class;
6077 struct hlist_head *head;
6078 int i, j;
6079
6080 for (i = 0; i < CLASSHASH_SIZE; i++) {
6081 head = classhash_table + i;
6082 hlist_for_each_entry_rcu(class, head, hash_entry) {
6083 for (j = 0; j < NR_LOCKDEP_CACHING_CLASSES; j++)
6084 if (lock->class_cache[j] == class)
6085 return true;
6086 }
6087 }
6088 return false;
6089}
6090
6091/* The caller must hold the graph lock. Does not sleep. */
6092static void __lockdep_reset_lock(struct pending_free *pf,
6093 struct lockdep_map *lock)
6094{
6095 struct lock_class *class;
6096 int j;
6097
6098 /*
6099 * Remove all classes this lock might have:
6100 */
6101 for (j = 0; j < MAX_LOCKDEP_SUBCLASSES; j++) {
6102 /*
6103 * If the class exists we look it up and zap it:
6104 */
6105 class = look_up_lock_class(lock, j);
6106 if (class)
6107 zap_class(pf, class);
6108 }
6109 /*
6110 * Debug check: in the end all mapped classes should
6111 * be gone.
6112 */
6113 if (WARN_ON_ONCE(lock_class_cache_is_registered(lock)))
6114 debug_locks_off();
6115}
6116
6117/*
6118 * Remove all information lockdep has about a lock if debug_locks == 1. Free
6119 * released data structures from RCU context.
6120 */
6121static void lockdep_reset_lock_reg(struct lockdep_map *lock)
6122{
6123 struct pending_free *pf;
6124 unsigned long flags;
6125 int locked;
6126
6127 raw_local_irq_save(flags);
6128 locked = graph_lock();
6129 if (!locked)
6130 goto out_irq;
6131
6132 pf = get_pending_free();
6133 __lockdep_reset_lock(pf, lock);
6134 call_rcu_zapped(pf);
6135
6136 graph_unlock();
6137out_irq:
6138 raw_local_irq_restore(flags);
6139}
6140
6141/*
6142 * Reset a lock. Does not sleep. Ignores debug_locks. Must only be used by the
6143 * lockdep selftests.
6144 */
6145static void lockdep_reset_lock_imm(struct lockdep_map *lock)
6146{
6147 struct pending_free *pf = delayed_free.pf;
6148 unsigned long flags;
6149
6150 raw_local_irq_save(flags);
6151 lockdep_lock();
6152 __lockdep_reset_lock(pf, lock);
6153 __free_zapped_classes(pf);
6154 lockdep_unlock();
6155 raw_local_irq_restore(flags);
6156}
6157
6158void lockdep_reset_lock(struct lockdep_map *lock)
6159{
6160 init_data_structures_once();
6161
6162 if (inside_selftest())
6163 lockdep_reset_lock_imm(lock);
6164 else
6165 lockdep_reset_lock_reg(lock);
6166}
6167
6168/* Unregister a dynamically allocated key. */
6169void lockdep_unregister_key(struct lock_class_key *key)
6170{
6171 struct hlist_head *hash_head = keyhashentry(key);
6172 struct lock_class_key *k;
6173 struct pending_free *pf;
6174 unsigned long flags;
6175 bool found = false;
6176
6177 might_sleep();
6178
6179 if (WARN_ON_ONCE(static_obj(key)))
6180 return;
6181
6182 raw_local_irq_save(flags);
6183 if (!graph_lock())
6184 goto out_irq;
6185
6186 pf = get_pending_free();
6187 hlist_for_each_entry_rcu(k, hash_head, hash_entry) {
6188 if (k == key) {
6189 hlist_del_rcu(&k->hash_entry);
6190 found = true;
6191 break;
6192 }
6193 }
6194 WARN_ON_ONCE(!found);
6195 __lockdep_free_key_range(pf, key, 1);
6196 call_rcu_zapped(pf);
6197 graph_unlock();
6198out_irq:
6199 raw_local_irq_restore(flags);
6200
6201 /* Wait until is_dynamic_key() has finished accessing k->hash_entry. */
6202 synchronize_rcu();
6203}
6204EXPORT_SYMBOL_GPL(lockdep_unregister_key);
6205
6206void __init lockdep_init(void)
6207{
6208 printk("Lock dependency validator: Copyright (c) 2006 Red Hat, Inc., Ingo Molnar\n");
6209
6210 printk("... MAX_LOCKDEP_SUBCLASSES: %lu\n", MAX_LOCKDEP_SUBCLASSES);
6211 printk("... MAX_LOCK_DEPTH: %lu\n", MAX_LOCK_DEPTH);
6212 printk("... MAX_LOCKDEP_KEYS: %lu\n", MAX_LOCKDEP_KEYS);
6213 printk("... CLASSHASH_SIZE: %lu\n", CLASSHASH_SIZE);
6214 printk("... MAX_LOCKDEP_ENTRIES: %lu\n", MAX_LOCKDEP_ENTRIES);
6215 printk("... MAX_LOCKDEP_CHAINS: %lu\n", MAX_LOCKDEP_CHAINS);
6216 printk("... CHAINHASH_SIZE: %lu\n", CHAINHASH_SIZE);
6217
6218 printk(" memory used by lock dependency info: %zu kB\n",
6219 (sizeof(lock_classes) +
6220 sizeof(lock_classes_in_use) +
6221 sizeof(classhash_table) +
6222 sizeof(list_entries) +
6223 sizeof(list_entries_in_use) +
6224 sizeof(chainhash_table) +
6225 sizeof(delayed_free)
6226#ifdef CONFIG_PROVE_LOCKING
6227 + sizeof(lock_cq)
6228 + sizeof(lock_chains)
6229 + sizeof(lock_chains_in_use)
6230 + sizeof(chain_hlocks)
6231#endif
6232 ) / 1024
6233 );
6234
6235#if defined(CONFIG_TRACE_IRQFLAGS) && defined(CONFIG_PROVE_LOCKING)
6236 printk(" memory used for stack traces: %zu kB\n",
6237 (sizeof(stack_trace) + sizeof(stack_trace_hash)) / 1024
6238 );
6239#endif
6240
6241 printk(" per task-struct memory footprint: %zu bytes\n",
6242 sizeof(((struct task_struct *)NULL)->held_locks));
6243}
6244
6245static void
6246print_freed_lock_bug(struct task_struct *curr, const void *mem_from,
6247 const void *mem_to, struct held_lock *hlock)
6248{
6249 if (!debug_locks_off())
6250 return;
6251 if (debug_locks_silent)
6252 return;
6253
6254 pr_warn("\n");
6255 pr_warn("=========================\n");
6256 pr_warn("WARNING: held lock freed!\n");
6257 print_kernel_ident();
6258 pr_warn("-------------------------\n");
6259 pr_warn("%s/%d is freeing memory %px-%px, with a lock still held there!\n",
6260 curr->comm, task_pid_nr(curr), mem_from, mem_to-1);
6261 print_lock(hlock);
6262 lockdep_print_held_locks(curr);
6263
6264 pr_warn("\nstack backtrace:\n");
6265 dump_stack();
6266}
6267
6268static inline int not_in_range(const void* mem_from, unsigned long mem_len,
6269 const void* lock_from, unsigned long lock_len)
6270{
6271 return lock_from + lock_len <= mem_from ||
6272 mem_from + mem_len <= lock_from;
6273}
6274
6275/*
6276 * Called when kernel memory is freed (or unmapped), or if a lock
6277 * is destroyed or reinitialized - this code checks whether there is
6278 * any held lock in the memory range of <from> to <to>:
6279 */
6280void debug_check_no_locks_freed(const void *mem_from, unsigned long mem_len)
6281{
6282 struct task_struct *curr = current;
6283 struct held_lock *hlock;
6284 unsigned long flags;
6285 int i;
6286
6287 if (unlikely(!debug_locks))
6288 return;
6289
6290 raw_local_irq_save(flags);
6291 for (i = 0; i < curr->lockdep_depth; i++) {
6292 hlock = curr->held_locks + i;
6293
6294 if (not_in_range(mem_from, mem_len, hlock->instance,
6295 sizeof(*hlock->instance)))
6296 continue;
6297
6298 print_freed_lock_bug(curr, mem_from, mem_from + mem_len, hlock);
6299 break;
6300 }
6301 raw_local_irq_restore(flags);
6302}
6303EXPORT_SYMBOL_GPL(debug_check_no_locks_freed);
6304
6305static void print_held_locks_bug(void)
6306{
6307 if (!debug_locks_off())
6308 return;
6309 if (debug_locks_silent)
6310 return;
6311
6312 pr_warn("\n");
6313 pr_warn("====================================\n");
6314 pr_warn("WARNING: %s/%d still has locks held!\n",
6315 current->comm, task_pid_nr(current));
6316 print_kernel_ident();
6317 pr_warn("------------------------------------\n");
6318 lockdep_print_held_locks(current);
6319 pr_warn("\nstack backtrace:\n");
6320 dump_stack();
6321}
6322
6323void debug_check_no_locks_held(void)
6324{
6325 if (unlikely(current->lockdep_depth > 0))
6326 print_held_locks_bug();
6327}
6328EXPORT_SYMBOL_GPL(debug_check_no_locks_held);
6329
6330#ifdef __KERNEL__
6331void debug_show_all_locks(void)
6332{
6333 struct task_struct *g, *p;
6334
6335 if (unlikely(!debug_locks)) {
6336 pr_warn("INFO: lockdep is turned off.\n");
6337 return;
6338 }
6339 pr_warn("\nShowing all locks held in the system:\n");
6340
6341 rcu_read_lock();
6342 for_each_process_thread(g, p) {
6343 if (!p->lockdep_depth)
6344 continue;
6345 lockdep_print_held_locks(p);
6346 touch_nmi_watchdog();
6347 touch_all_softlockup_watchdogs();
6348 }
6349 rcu_read_unlock();
6350
6351 pr_warn("\n");
6352 pr_warn("=============================================\n\n");
6353}
6354EXPORT_SYMBOL_GPL(debug_show_all_locks);
6355#endif
6356
6357/*
6358 * Careful: only use this function if you are sure that
6359 * the task cannot run in parallel!
6360 */
6361void debug_show_held_locks(struct task_struct *task)
6362{
6363 if (unlikely(!debug_locks)) {
6364 printk("INFO: lockdep is turned off.\n");
6365 return;
6366 }
6367 lockdep_print_held_locks(task);
6368}
6369EXPORT_SYMBOL_GPL(debug_show_held_locks);
6370
6371asmlinkage __visible void lockdep_sys_exit(void)
6372{
6373 struct task_struct *curr = current;
6374
6375 if (unlikely(curr->lockdep_depth)) {
6376 if (!debug_locks_off())
6377 return;
6378 pr_warn("\n");
6379 pr_warn("================================================\n");
6380 pr_warn("WARNING: lock held when returning to user space!\n");
6381 print_kernel_ident();
6382 pr_warn("------------------------------------------------\n");
6383 pr_warn("%s/%d is leaving the kernel with locks still held!\n",
6384 curr->comm, curr->pid);
6385 lockdep_print_held_locks(curr);
6386 }
6387
6388 /*
6389 * The lock history for each syscall should be independent. So wipe the
6390 * slate clean on return to userspace.
6391 */
6392 lockdep_invariant_state(false);
6393}
6394
6395void lockdep_rcu_suspicious(const char *file, const int line, const char *s)
6396{
6397 struct task_struct *curr = current;
6398
6399 /* Note: the following can be executed concurrently, so be careful. */
6400 pr_warn("\n");
6401 pr_warn("=============================\n");
6402 pr_warn("WARNING: suspicious RCU usage\n");
6403 print_kernel_ident();
6404 pr_warn("-----------------------------\n");
6405 pr_warn("%s:%d %s!\n", file, line, s);
6406 pr_warn("\nother info that might help us debug this:\n\n");
6407 pr_warn("\n%srcu_scheduler_active = %d, debug_locks = %d\n",
6408 !rcu_lockdep_current_cpu_online()
6409 ? "RCU used illegally from offline CPU!\n"
6410 : "",
6411 rcu_scheduler_active, debug_locks);
6412
6413 /*
6414 * If a CPU is in the RCU-free window in idle (ie: in the section
6415 * between rcu_idle_enter() and rcu_idle_exit(), then RCU
6416 * considers that CPU to be in an "extended quiescent state",
6417 * which means that RCU will be completely ignoring that CPU.
6418 * Therefore, rcu_read_lock() and friends have absolutely no
6419 * effect on a CPU running in that state. In other words, even if
6420 * such an RCU-idle CPU has called rcu_read_lock(), RCU might well
6421 * delete data structures out from under it. RCU really has no
6422 * choice here: we need to keep an RCU-free window in idle where
6423 * the CPU may possibly enter into low power mode. This way we can
6424 * notice an extended quiescent state to other CPUs that started a grace
6425 * period. Otherwise we would delay any grace period as long as we run
6426 * in the idle task.
6427 *
6428 * So complain bitterly if someone does call rcu_read_lock(),
6429 * rcu_read_lock_bh() and so on from extended quiescent states.
6430 */
6431 if (!rcu_is_watching())
6432 pr_warn("RCU used illegally from extended quiescent state!\n");
6433
6434 lockdep_print_held_locks(curr);
6435 pr_warn("\nstack backtrace:\n");
6436 dump_stack();
6437}
6438EXPORT_SYMBOL_GPL(lockdep_rcu_suspicious);