projects/linux·blob master · refs

6976 lines · 178.5 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#include <linux/context_tracking.h>
59#include <linux/console.h>
60#include <linux/kasan.h>
61
62#include <asm/sections.h>
63
64#include "lockdep_internals.h"
65#include "lock_events.h"
66
67#include <trace/events/lock.h>
68
69#ifdef CONFIG_PROVE_LOCKING
70static int prove_locking = 1;
71module_param(prove_locking, int, 0644);
72#else
73#define prove_locking 0
74#endif
75
76#ifdef CONFIG_LOCK_STAT
77static int lock_stat = 1;
78module_param(lock_stat, int, 0644);
79#else
80#define lock_stat 0
81#endif
82
83#ifdef CONFIG_SYSCTL
84static const struct ctl_table kern_lockdep_table[] = {
85#ifdef CONFIG_PROVE_LOCKING
86 {
87 .procname = "prove_locking",
88 .data = &prove_locking,
89 .maxlen = sizeof(int),
90 .mode = 0644,
91 .proc_handler = proc_dointvec,
92 },
93#endif /* CONFIG_PROVE_LOCKING */
94#ifdef CONFIG_LOCK_STAT
95 {
96 .procname = "lock_stat",
97 .data = &lock_stat,
98 .maxlen = sizeof(int),
99 .mode = 0644,
100 .proc_handler = proc_dointvec,
101 },
102#endif /* CONFIG_LOCK_STAT */
103};
104
105static __init int kernel_lockdep_sysctls_init(void)
106{
107 register_sysctl_init("kernel", kern_lockdep_table);
108 return 0;
109}
110late_initcall(kernel_lockdep_sysctls_init);
111#endif /* CONFIG_SYSCTL */
112
113DEFINE_PER_CPU(unsigned int, lockdep_recursion);
114EXPORT_PER_CPU_SYMBOL_GPL(lockdep_recursion);
115
116static __always_inline bool lockdep_enabled(void)
117{
118 if (!debug_locks)
119 return false;
120
121 if (this_cpu_read(lockdep_recursion))
122 return false;
123
124 if (current->lockdep_recursion)
125 return false;
126
127 return true;
128}
129
130/*
131 * lockdep_lock: protects the lockdep graph, the hashes and the
132 * class/list/hash allocators.
133 *
134 * This is one of the rare exceptions where it's justified
135 * to use a raw spinlock - we really dont want the spinlock
136 * code to recurse back into the lockdep code...
137 */
138static arch_spinlock_t __lock = (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
139static struct task_struct *__owner;
140
141static inline void lockdep_lock(void)
142{
143 DEBUG_LOCKS_WARN_ON(!irqs_disabled());
144
145 __this_cpu_inc(lockdep_recursion);
146 arch_spin_lock(&__lock);
147 __owner = current;
148}
149
150static inline void lockdep_unlock(void)
151{
152 DEBUG_LOCKS_WARN_ON(!irqs_disabled());
153
154 if (debug_locks && DEBUG_LOCKS_WARN_ON(__owner != current))
155 return;
156
157 __owner = NULL;
158 arch_spin_unlock(&__lock);
159 __this_cpu_dec(lockdep_recursion);
160}
161
162#ifdef CONFIG_PROVE_LOCKING
163static inline bool lockdep_assert_locked(void)
164{
165 return DEBUG_LOCKS_WARN_ON(__owner != current);
166}
167#endif
168
169static struct task_struct *lockdep_selftest_task_struct;
170
171
172static int graph_lock(void)
173{
174 lockdep_lock();
175 lockevent_inc(lockdep_lock);
176 /*
177 * Make sure that if another CPU detected a bug while
178 * walking the graph we dont change it (while the other
179 * CPU is busy printing out stuff with the graph lock
180 * dropped already)
181 */
182 if (!debug_locks) {
183 lockdep_unlock();
184 return 0;
185 }
186 return 1;
187}
188
189static inline void graph_unlock(void)
190{
191 lockdep_unlock();
192}
193
194/*
195 * Turn lock debugging off and return with 0 if it was off already,
196 * and also release the graph lock:
197 */
198static inline int debug_locks_off_graph_unlock(void)
199{
200 int ret = debug_locks_off();
201
202 lockdep_unlock();
203
204 return ret;
205}
206
207unsigned long nr_list_entries;
208static struct lock_list list_entries[MAX_LOCKDEP_ENTRIES];
209static DECLARE_BITMAP(list_entries_in_use, MAX_LOCKDEP_ENTRIES);
210
211/*
212 * All data structures here are protected by the global debug_lock.
213 *
214 * nr_lock_classes is the number of elements of lock_classes[] that is
215 * in use.
216 */
217#define KEYHASH_BITS (MAX_LOCKDEP_KEYS_BITS - 1)
218#define KEYHASH_SIZE (1UL << KEYHASH_BITS)
219static struct hlist_head lock_keys_hash[KEYHASH_SIZE];
220unsigned long nr_lock_classes;
221unsigned long nr_zapped_classes;
222unsigned long nr_dynamic_keys;
223unsigned long max_lock_class_idx;
224struct lock_class lock_classes[MAX_LOCKDEP_KEYS];
225DECLARE_BITMAP(lock_classes_in_use, MAX_LOCKDEP_KEYS);
226
227static inline struct lock_class *hlock_class(struct held_lock *hlock)
228{
229 unsigned int class_idx = hlock->class_idx;
230
231 /* Don't re-read hlock->class_idx, can't use READ_ONCE() on bitfield */
232 barrier();
233
234 if (!test_bit(class_idx, lock_classes_in_use)) {
235 /*
236 * Someone passed in garbage, we give up.
237 */
238 DEBUG_LOCKS_WARN_ON(1);
239 return NULL;
240 }
241
242 /*
243 * At this point, if the passed hlock->class_idx is still garbage,
244 * we just have to live with it
245 */
246 return lock_classes + class_idx;
247}
248
249#ifdef CONFIG_LOCK_STAT
250static DEFINE_PER_CPU(struct lock_class_stats[MAX_LOCKDEP_KEYS], cpu_lock_stats);
251
252static inline u64 lockstat_clock(void)
253{
254 return local_clock();
255}
256
257static int lock_point(unsigned long points[], unsigned long ip)
258{
259 int i;
260
261 for (i = 0; i < LOCKSTAT_POINTS; i++) {
262 if (points[i] == 0) {
263 points[i] = ip;
264 break;
265 }
266 if (points[i] == ip)
267 break;
268 }
269
270 return i;
271}
272
273static void lock_time_inc(struct lock_time *lt, u64 time)
274{
275 if (time > lt->max)
276 lt->max = time;
277
278 if (time < lt->min || !lt->nr)
279 lt->min = time;
280
281 lt->total += time;
282 lt->nr++;
283}
284
285static inline void lock_time_add(struct lock_time *src, struct lock_time *dst)
286{
287 if (!src->nr)
288 return;
289
290 if (src->max > dst->max)
291 dst->max = src->max;
292
293 if (src->min < dst->min || !dst->nr)
294 dst->min = src->min;
295
296 dst->total += src->total;
297 dst->nr += src->nr;
298}
299
300void lock_stats(struct lock_class *class, struct lock_class_stats *stats)
301{
302 int cpu, i;
303
304 memset(stats, 0, sizeof(struct lock_class_stats));
305 for_each_possible_cpu(cpu) {
306 struct lock_class_stats *pcs =
307 &per_cpu(cpu_lock_stats, cpu)[class - lock_classes];
308
309 for (i = 0; i < ARRAY_SIZE(stats->contention_point); i++)
310 stats->contention_point[i] += pcs->contention_point[i];
311
312 for (i = 0; i < ARRAY_SIZE(stats->contending_point); i++)
313 stats->contending_point[i] += pcs->contending_point[i];
314
315 lock_time_add(&pcs->read_waittime, &stats->read_waittime);
316 lock_time_add(&pcs->write_waittime, &stats->write_waittime);
317
318 lock_time_add(&pcs->read_holdtime, &stats->read_holdtime);
319 lock_time_add(&pcs->write_holdtime, &stats->write_holdtime);
320
321 for (i = 0; i < ARRAY_SIZE(stats->bounces); i++)
322 stats->bounces[i] += pcs->bounces[i];
323 }
324}
325
326void clear_lock_stats(struct lock_class *class)
327{
328 int cpu;
329
330 for_each_possible_cpu(cpu) {
331 struct lock_class_stats *cpu_stats =
332 &per_cpu(cpu_lock_stats, cpu)[class - lock_classes];
333
334 memset(cpu_stats, 0, sizeof(struct lock_class_stats));
335 }
336 memset(class->contention_point, 0, sizeof(class->contention_point));
337 memset(class->contending_point, 0, sizeof(class->contending_point));
338}
339
340static struct lock_class_stats *get_lock_stats(struct lock_class *class)
341{
342 return &this_cpu_ptr(cpu_lock_stats)[class - lock_classes];
343}
344
345static void lock_release_holdtime(struct held_lock *hlock)
346{
347 struct lock_class_stats *stats;
348 u64 holdtime;
349
350 if (!lock_stat)
351 return;
352
353 holdtime = lockstat_clock() - hlock->holdtime_stamp;
354
355 stats = get_lock_stats(hlock_class(hlock));
356 if (hlock->read)
357 lock_time_inc(&stats->read_holdtime, holdtime);
358 else
359 lock_time_inc(&stats->write_holdtime, holdtime);
360}
361#else
362static inline void lock_release_holdtime(struct held_lock *hlock)
363{
364}
365#endif
366
367/*
368 * We keep a global list of all lock classes. The list is only accessed with
369 * the lockdep spinlock lock held. free_lock_classes is a list with free
370 * elements. These elements are linked together by the lock_entry member in
371 * struct lock_class.
372 */
373static LIST_HEAD(all_lock_classes);
374static LIST_HEAD(free_lock_classes);
375
376/**
377 * struct pending_free - information about data structures about to be freed
378 * @zapped: Head of a list with struct lock_class elements.
379 * @lock_chains_being_freed: Bitmap that indicates which lock_chains[] elements
380 * are about to be freed.
381 */
382struct pending_free {
383 struct list_head zapped;
384 DECLARE_BITMAP(lock_chains_being_freed, MAX_LOCKDEP_CHAINS);
385};
386
387/**
388 * struct delayed_free - data structures used for delayed freeing
389 *
390 * A data structure for delayed freeing of data structures that may be
391 * accessed by RCU readers at the time these were freed.
392 *
393 * @rcu_head: Used to schedule an RCU callback for freeing data structures.
394 * @index: Index of @pf to which freed data structures are added.
395 * @scheduled: Whether or not an RCU callback has been scheduled.
396 * @pf: Array with information about data structures about to be freed.
397 */
398static struct delayed_free {
399 struct rcu_head rcu_head;
400 int index;
401 int scheduled;
402 struct pending_free pf[2];
403} delayed_free;
404
405/*
406 * The lockdep classes are in a hash-table as well, for fast lookup:
407 */
408#define CLASSHASH_BITS (MAX_LOCKDEP_KEYS_BITS - 1)
409#define CLASSHASH_SIZE (1UL << CLASSHASH_BITS)
410#define __classhashfn(key) hash_long((unsigned long)key, CLASSHASH_BITS)
411#define classhashentry(key) (classhash_table + __classhashfn((key)))
412
413static struct hlist_head classhash_table[CLASSHASH_SIZE];
414
415/*
416 * We put the lock dependency chains into a hash-table as well, to cache
417 * their existence:
418 */
419#define CHAINHASH_BITS (MAX_LOCKDEP_CHAINS_BITS-1)
420#define CHAINHASH_SIZE (1UL << CHAINHASH_BITS)
421#define __chainhashfn(chain) hash_long(chain, CHAINHASH_BITS)
422#define chainhashentry(chain) (chainhash_table + __chainhashfn((chain)))
423
424static struct hlist_head chainhash_table[CHAINHASH_SIZE];
425
426/*
427 * the id of held_lock
428 */
429static inline u16 hlock_id(struct held_lock *hlock)
430{
431 BUILD_BUG_ON(MAX_LOCKDEP_KEYS_BITS + 2 > 16);
432
433 return (hlock->class_idx | (hlock->read << MAX_LOCKDEP_KEYS_BITS));
434}
435
436static inline __maybe_unused unsigned int chain_hlock_class_idx(u16 hlock_id)
437{
438 return hlock_id & (MAX_LOCKDEP_KEYS - 1);
439}
440
441/*
442 * The hash key of the lock dependency chains is a hash itself too:
443 * it's a hash of all locks taken up to that lock, including that lock.
444 * It's a 64-bit hash, because it's important for the keys to be
445 * unique.
446 */
447static inline u64 iterate_chain_key(u64 key, u32 idx)
448{
449 u32 k0 = key, k1 = key >> 32;
450
451 __jhash_mix(idx, k0, k1); /* Macro that modifies arguments! */
452
453 return k0 | (u64)k1 << 32;
454}
455
456void lockdep_init_task(struct task_struct *task)
457{
458 task->lockdep_depth = 0; /* no locks held yet */
459 task->curr_chain_key = INITIAL_CHAIN_KEY;
460 task->lockdep_recursion = 0;
461}
462
463static __always_inline void lockdep_recursion_inc(void)
464{
465 __this_cpu_inc(lockdep_recursion);
466}
467
468static __always_inline void lockdep_recursion_finish(void)
469{
470 if (WARN_ON_ONCE(__this_cpu_dec_return(lockdep_recursion)))
471 __this_cpu_write(lockdep_recursion, 0);
472}
473
474void lockdep_set_selftest_task(struct task_struct *task)
475{
476 lockdep_selftest_task_struct = task;
477}
478
479/*
480 * Debugging switches:
481 */
482
483#define VERBOSE 0
484#define VERY_VERBOSE 0
485
486#if VERBOSE
487# define HARDIRQ_VERBOSE 1
488# define SOFTIRQ_VERBOSE 1
489#else
490# define HARDIRQ_VERBOSE 0
491# define SOFTIRQ_VERBOSE 0
492#endif
493
494#if VERBOSE || HARDIRQ_VERBOSE || SOFTIRQ_VERBOSE
495/*
496 * Quick filtering for interesting events:
497 */
498static int class_filter(struct lock_class *class)
499{
500#if 0
501 /* Example */
502 if (class->name_version == 1 &&
503 !strcmp(class->name, "lockname"))
504 return 1;
505 if (class->name_version == 1 &&
506 !strcmp(class->name, "&struct->lockfield"))
507 return 1;
508#endif
509 /* Filter everything else. 1 would be to allow everything else */
510 return 0;
511}
512#endif
513
514static int verbose(struct lock_class *class)
515{
516#if VERBOSE
517 return class_filter(class);
518#endif
519 return 0;
520}
521
522static void print_lockdep_off(const char *bug_msg)
523{
524 printk(KERN_DEBUG "%s\n", bug_msg);
525 printk(KERN_DEBUG "turning off the locking correctness validator.\n");
526#ifdef CONFIG_LOCK_STAT
527 printk(KERN_DEBUG "Please attach the output of /proc/lock_stat to the bug report\n");
528#endif
529}
530
531unsigned long nr_stack_trace_entries;
532
533#ifdef CONFIG_PROVE_LOCKING
534/**
535 * struct lock_trace - single stack backtrace
536 * @hash_entry: Entry in a stack_trace_hash[] list.
537 * @hash: jhash() of @entries.
538 * @nr_entries: Number of entries in @entries.
539 * @entries: Actual stack backtrace.
540 */
541struct lock_trace {
542 struct hlist_node hash_entry;
543 u32 hash;
544 u32 nr_entries;
545 unsigned long entries[] __aligned(sizeof(unsigned long));
546};
547#define LOCK_TRACE_SIZE_IN_LONGS \
548 (sizeof(struct lock_trace) / sizeof(unsigned long))
549/*
550 * Stack-trace: sequence of lock_trace structures. Protected by the graph_lock.
551 */
552static unsigned long stack_trace[MAX_STACK_TRACE_ENTRIES];
553static struct hlist_head stack_trace_hash[STACK_TRACE_HASH_SIZE];
554
555static bool traces_identical(struct lock_trace *t1, struct lock_trace *t2)
556{
557 return t1->hash == t2->hash && t1->nr_entries == t2->nr_entries &&
558 memcmp(t1->entries, t2->entries,
559 t1->nr_entries * sizeof(t1->entries[0])) == 0;
560}
561
562static struct lock_trace *save_trace(void)
563{
564 struct lock_trace *trace, *t2;
565 struct hlist_head *hash_head;
566 u32 hash;
567 int max_entries;
568
569 BUILD_BUG_ON_NOT_POWER_OF_2(STACK_TRACE_HASH_SIZE);
570 BUILD_BUG_ON(LOCK_TRACE_SIZE_IN_LONGS >= MAX_STACK_TRACE_ENTRIES);
571
572 trace = (struct lock_trace *)(stack_trace + nr_stack_trace_entries);
573 max_entries = MAX_STACK_TRACE_ENTRIES - nr_stack_trace_entries -
574 LOCK_TRACE_SIZE_IN_LONGS;
575
576 if (max_entries <= 0) {
577 if (!debug_locks_off_graph_unlock())
578 return NULL;
579
580 nbcon_cpu_emergency_enter();
581 print_lockdep_off("BUG: MAX_STACK_TRACE_ENTRIES too low!");
582 dump_stack();
583 nbcon_cpu_emergency_exit();
584
585 return NULL;
586 }
587 trace->nr_entries = stack_trace_save(trace->entries, max_entries, 3);
588
589 hash = jhash(trace->entries, trace->nr_entries *
590 sizeof(trace->entries[0]), 0);
591 trace->hash = hash;
592 hash_head = stack_trace_hash + (hash & (STACK_TRACE_HASH_SIZE - 1));
593 hlist_for_each_entry(t2, hash_head, hash_entry) {
594 if (traces_identical(trace, t2))
595 return t2;
596 }
597 nr_stack_trace_entries += LOCK_TRACE_SIZE_IN_LONGS + trace->nr_entries;
598 hlist_add_head(&trace->hash_entry, hash_head);
599
600 return trace;
601}
602
603/* Return the number of stack traces in the stack_trace[] array. */
604u64 lockdep_stack_trace_count(void)
605{
606 struct lock_trace *trace;
607 u64 c = 0;
608 int i;
609
610 for (i = 0; i < ARRAY_SIZE(stack_trace_hash); i++) {
611 hlist_for_each_entry(trace, &stack_trace_hash[i], hash_entry) {
612 c++;
613 }
614 }
615
616 return c;
617}
618
619/* Return the number of stack hash chains that have at least one stack trace. */
620u64 lockdep_stack_hash_count(void)
621{
622 u64 c = 0;
623 int i;
624
625 for (i = 0; i < ARRAY_SIZE(stack_trace_hash); i++)
626 if (!hlist_empty(&stack_trace_hash[i]))
627 c++;
628
629 return c;
630}
631#endif
632
633unsigned int nr_hardirq_chains;
634unsigned int nr_softirq_chains;
635unsigned int nr_process_chains;
636unsigned int max_lockdep_depth;
637
638#ifdef CONFIG_DEBUG_LOCKDEP
639/*
640 * Various lockdep statistics:
641 */
642DEFINE_PER_CPU(struct lockdep_stats, lockdep_stats);
643#endif
644
645#ifdef CONFIG_PROVE_LOCKING
646/*
647 * Locking printouts:
648 */
649
650#define __USAGE(__STATE) \
651 [LOCK_USED_IN_##__STATE] = "IN-"__stringify(__STATE)"-W", \
652 [LOCK_ENABLED_##__STATE] = __stringify(__STATE)"-ON-W", \
653 [LOCK_USED_IN_##__STATE##_READ] = "IN-"__stringify(__STATE)"-R",\
654 [LOCK_ENABLED_##__STATE##_READ] = __stringify(__STATE)"-ON-R",
655
656static const char *usage_str[] =
657{
658#define LOCKDEP_STATE(__STATE) __USAGE(__STATE)
659#include "lockdep_states.h"
660#undef LOCKDEP_STATE
661 [LOCK_USED] = "INITIAL USE",
662 [LOCK_USED_READ] = "INITIAL READ USE",
663 /* abused as string storage for verify_lock_unused() */
664 [LOCK_USAGE_STATES] = "IN-NMI",
665};
666#endif
667
668const char *__get_key_name(const struct lockdep_subclass_key *key, char *str)
669{
670 return kallsyms_lookup((unsigned long)key, NULL, NULL, NULL, str);
671}
672
673static inline unsigned long lock_flag(enum lock_usage_bit bit)
674{
675 return 1UL << bit;
676}
677
678static char get_usage_char(struct lock_class *class, enum lock_usage_bit bit)
679{
680 /*
681 * The usage character defaults to '.' (i.e., irqs disabled and not in
682 * irq context), which is the safest usage category.
683 */
684 char c = '.';
685
686 /*
687 * The order of the following usage checks matters, which will
688 * result in the outcome character as follows:
689 *
690 * - '+': irq is enabled and not in irq context
691 * - '-': in irq context and irq is disabled
692 * - '?': in irq context and irq is enabled
693 */
694 if (class->usage_mask & lock_flag(bit + LOCK_USAGE_DIR_MASK)) {
695 c = '+';
696 if (class->usage_mask & lock_flag(bit))
697 c = '?';
698 } else if (class->usage_mask & lock_flag(bit))
699 c = '-';
700
701 return c;
702}
703
704void get_usage_chars(struct lock_class *class, char usage[LOCK_USAGE_CHARS])
705{
706 int i = 0;
707
708#define LOCKDEP_STATE(__STATE) \
709 usage[i++] = get_usage_char(class, LOCK_USED_IN_##__STATE); \
710 usage[i++] = get_usage_char(class, LOCK_USED_IN_##__STATE##_READ);
711#include "lockdep_states.h"
712#undef LOCKDEP_STATE
713
714 usage[i] = '\0';
715}
716
717static void __print_lock_name(struct held_lock *hlock, struct lock_class *class)
718{
719 char str[KSYM_NAME_LEN];
720 const char *name;
721
722 name = class->name;
723 if (!name) {
724 name = __get_key_name(class->key, str);
725 printk(KERN_CONT "%s", name);
726 } else {
727 printk(KERN_CONT "%s", name);
728 if (class->name_version > 1)
729 printk(KERN_CONT "#%d", class->name_version);
730 if (class->subclass)
731 printk(KERN_CONT "/%d", class->subclass);
732 if (hlock && class->print_fn)
733 class->print_fn(hlock->instance);
734 }
735}
736
737static void print_lock_name(struct held_lock *hlock, struct lock_class *class)
738{
739 char usage[LOCK_USAGE_CHARS];
740
741 get_usage_chars(class, usage);
742
743 printk(KERN_CONT " (");
744 __print_lock_name(hlock, class);
745 printk(KERN_CONT "){%s}-{%d:%d}", usage,
746 class->wait_type_outer ?: class->wait_type_inner,
747 class->wait_type_inner);
748}
749
750static void print_lockdep_cache(struct lockdep_map *lock)
751{
752 const char *name;
753 char str[KSYM_NAME_LEN];
754
755 name = lock->name;
756 if (!name)
757 name = __get_key_name(lock->key->subkeys, str);
758
759 printk(KERN_CONT "%s", name);
760}
761
762static void print_lock(struct held_lock *hlock)
763{
764 /*
765 * We can be called locklessly through debug_show_all_locks() so be
766 * extra careful, the hlock might have been released and cleared.
767 *
768 * If this indeed happens, lets pretend it does not hurt to continue
769 * to print the lock unless the hlock class_idx does not point to a
770 * registered class. The rationale here is: since we don't attempt
771 * to distinguish whether we are in this situation, if it just
772 * happened we can't count on class_idx to tell either.
773 */
774 struct lock_class *lock = hlock_class(hlock);
775
776 if (!lock) {
777 printk(KERN_CONT "<RELEASED>\n");
778 return;
779 }
780
781 printk(KERN_CONT "%px", hlock->instance);
782 print_lock_name(hlock, lock);
783 printk(KERN_CONT ", at: %pS\n", (void *)hlock->acquire_ip);
784}
785
786static void lockdep_print_held_locks(struct task_struct *p)
787{
788 int i, depth = READ_ONCE(p->lockdep_depth);
789
790 /*
791 * Note that it's always somewhat unreliable to print held locks
792 * of a task that is running on another CPU, but we cannot guarantee
793 * the stability of ->held_locks without actually stopping all active
794 * remote CPUs, which we absolutely do not want to do because it's
795 * very intrusive and thus slow.
796 *
797 * So we do the next best thing here: we print out the held lock
798 * array on a best-effort basis, without crashing even if the
799 * fields are being modified on another CPU. Note the careful
800 * construction of print_lock() so that it never crashes.
801 *
802 * We also print out the CPU the task is or was last running on, with
803 * the message saying 'on CPU...' if the task is running, and
804 * 'last CPU' if it's not.
805 *
806 * Also note that the task_is_running(p) information is fundamentally
807 * racy: even if the message says the task is 'on CPU', the task may
808 * have scheduled out already, or if it says 'last CPU', it may just
809 * have scheduled in on another CPU. But even with these limitations
810 * it's still useful debuggining information.
811 */
812 printk("locks held by %s/%d: %d, %s CPU#%d%s\n",
813 p->comm, task_pid_nr(p), depth,
814 task_is_running(p) ? "last" : "on", task_cpu(p),
815 depth > 0 ? ":" : "");
816
817 for (i = 0; i < depth; i++) {
818 printk(" #%d: ", i);
819 print_lock(p->held_locks + i);
820 }
821}
822
823static void print_kernel_ident(void)
824{
825 printk("%s %.*s %s\n", init_utsname()->release,
826 (int)strcspn(init_utsname()->version, " "),
827 init_utsname()->version,
828 print_tainted());
829}
830
831static int very_verbose(struct lock_class *class)
832{
833#if VERY_VERBOSE
834 return class_filter(class);
835#endif
836 return 0;
837}
838
839/*
840 * Is this the address of a static object:
841 */
842#ifdef __KERNEL__
843static int static_obj(const void *obj)
844{
845 unsigned long addr = (unsigned long) obj;
846
847 if (is_kernel_core_data(addr))
848 return 1;
849
850 /*
851 * keys are allowed in the __ro_after_init section.
852 */
853 if (is_kernel_rodata(addr))
854 return 1;
855
856 /*
857 * in initdata section and used during bootup only?
858 * NOTE: On some platforms the initdata section is
859 * outside of the _stext ... _end range.
860 */
861 if (system_state < SYSTEM_FREEING_INITMEM &&
862 init_section_contains((void *)addr, 1))
863 return 1;
864
865 /*
866 * in-kernel percpu var?
867 */
868 if (is_kernel_percpu_address(addr))
869 return 1;
870
871 /*
872 * module static or percpu var?
873 */
874 return is_module_address(addr) || is_module_percpu_address(addr);
875}
876#endif
877
878/*
879 * To make lock name printouts unique, we calculate a unique
880 * class->name_version generation counter. The caller must hold the graph
881 * lock.
882 */
883static int count_matching_names(struct lock_class *new_class)
884{
885 struct lock_class *class;
886 int count = 0;
887
888 if (!new_class->name)
889 return 0;
890
891 list_for_each_entry(class, &all_lock_classes, lock_entry) {
892 if (new_class->key - new_class->subclass == class->key)
893 return class->name_version;
894 if (class->name && !strcmp(class->name, new_class->name))
895 count = max(count, class->name_version);
896 }
897
898 return count + 1;
899}
900
901/* used from NMI context -- must be lockless */
902static noinstr struct lock_class *
903look_up_lock_class(const struct lockdep_map *lock, unsigned int subclass)
904{
905 struct lockdep_subclass_key *key;
906 struct hlist_head *hash_head;
907 struct lock_class *class;
908
909 if (unlikely(subclass >= MAX_LOCKDEP_SUBCLASSES)) {
910 instrumentation_begin();
911 debug_locks_off();
912 nbcon_cpu_emergency_enter();
913 printk(KERN_ERR
914 "BUG: looking up invalid subclass: %u\n", subclass);
915 printk(KERN_ERR
916 "turning off the locking correctness validator.\n");
917 dump_stack();
918 nbcon_cpu_emergency_exit();
919 instrumentation_end();
920 return NULL;
921 }
922
923 /*
924 * If it is not initialised then it has never been locked,
925 * so it won't be present in the hash table.
926 */
927 if (unlikely(!lock->key))
928 return NULL;
929
930 /*
931 * NOTE: the class-key must be unique. For dynamic locks, a static
932 * lock_class_key variable is passed in through the mutex_init()
933 * (or spin_lock_init()) call - which acts as the key. For static
934 * locks we use the lock object itself as the key.
935 */
936 BUILD_BUG_ON(sizeof(struct lock_class_key) >
937 sizeof(struct lockdep_map));
938
939 key = lock->key->subkeys + subclass;
940
941 hash_head = classhashentry(key);
942
943 /*
944 * We do an RCU walk of the hash, see lockdep_free_key_range().
945 */
946 if (DEBUG_LOCKS_WARN_ON(!irqs_disabled()))
947 return NULL;
948
949 hlist_for_each_entry_rcu_notrace(class, hash_head, hash_entry) {
950 if (class->key == key) {
951 /*
952 * Huh! same key, different name? Did someone trample
953 * on some memory? We're most confused.
954 */
955 WARN_ONCE(class->name != lock->name &&
956 lock->key != &__lockdep_no_validate__,
957 "Looking for class \"%s\" with key %ps, but found a different class \"%s\" with the same key\n",
958 lock->name, lock->key, class->name);
959 return class;
960 }
961 }
962
963 return NULL;
964}
965
966static __always_inline bool lock_class_cache_is_valid(const struct lockdep_map *lock,
967 const struct lock_class *class,
968 unsigned int subclass)
969{
970 unsigned int class_subclass;
971
972 if (!class)
973 return false;
974
975 if (unlikely(class < lock_classes || class >= lock_classes + MAX_LOCKDEP_KEYS))
976 return false;
977
978 if (unlikely(!arch_test_bit(class - lock_classes, lock_classes_in_use)))
979 return false;
980
981 if (unlikely(!lock->key))
982 return false;
983
984 class_subclass = subclass ? subclass : class->subclass;
985 if (unlikely(class_subclass >= MAX_LOCKDEP_SUBCLASSES))
986 return false;
987
988 if (unlikely(READ_ONCE(class->key) != lock->key->subkeys + class_subclass))
989 return false;
990
991 return true;
992}
993
994/*
995 * Static locks do not have their class-keys yet - for them the key is
996 * the lock object itself. If the lock is in the per cpu area, the
997 * canonical address of the lock (per cpu offset removed) is used.
998 */
999static bool assign_lock_key(struct lockdep_map *lock)
1000{
1001 unsigned long can_addr, addr = (unsigned long)lock;
1002
1003#ifdef __KERNEL__
1004 /*
1005 * lockdep_free_key_range() assumes that struct lock_class_key
1006 * objects do not overlap. Since we use the address of lock
1007 * objects as class key for static objects, check whether the
1008 * size of lock_class_key objects does not exceed the size of
1009 * the smallest lock object.
1010 */
1011 BUILD_BUG_ON(sizeof(struct lock_class_key) > sizeof(raw_spinlock_t));
1012#endif
1013
1014 if (__is_kernel_percpu_address(addr, &can_addr))
1015 lock->key = (void *)can_addr;
1016 else if (__is_module_percpu_address(addr, &can_addr))
1017 lock->key = (void *)can_addr;
1018 else if (static_obj(lock))
1019 lock->key = (void *)lock;
1020 else {
1021 /* Debug-check: all keys must be persistent! */
1022 debug_locks_off();
1023 nbcon_cpu_emergency_enter();
1024 pr_err("INFO: trying to register non-static key.\n");
1025 pr_err("The code is fine but needs lockdep annotation, or maybe\n");
1026 pr_err("you didn't initialize this object before use?\n");
1027 pr_err("turning off the locking correctness validator.\n");
1028 dump_stack();
1029 nbcon_cpu_emergency_exit();
1030 return false;
1031 }
1032
1033 return true;
1034}
1035
1036#ifdef CONFIG_DEBUG_LOCKDEP
1037
1038/* Check whether element @e occurs in list @h */
1039static bool in_list(struct list_head *e, struct list_head *h)
1040{
1041 struct list_head *f;
1042
1043 list_for_each(f, h) {
1044 if (e == f)
1045 return true;
1046 }
1047
1048 return false;
1049}
1050
1051/*
1052 * Check whether entry @e occurs in any of the locks_after or locks_before
1053 * lists.
1054 */
1055static bool in_any_class_list(struct list_head *e)
1056{
1057 struct lock_class *class;
1058 int i;
1059
1060 for (i = 0; i < ARRAY_SIZE(lock_classes); i++) {
1061 class = &lock_classes[i];
1062 if (in_list(e, &class->locks_after) ||
1063 in_list(e, &class->locks_before))
1064 return true;
1065 }
1066 return false;
1067}
1068
1069static bool class_lock_list_valid(struct lock_class *c, struct list_head *h)
1070{
1071 struct lock_list *e;
1072
1073 list_for_each_entry(e, h, entry) {
1074 if (e->links_to != c) {
1075 printk(KERN_INFO "class %s: mismatch for lock entry %ld; class %s <> %s",
1076 c->name ? : "(?)",
1077 (unsigned long)(e - list_entries),
1078 e->links_to && e->links_to->name ?
1079 e->links_to->name : "(?)",
1080 e->class && e->class->name ? e->class->name :
1081 "(?)");
1082 return false;
1083 }
1084 }
1085 return true;
1086}
1087
1088#ifdef CONFIG_PROVE_LOCKING
1089static u16 chain_hlocks[MAX_LOCKDEP_CHAIN_HLOCKS];
1090#endif
1091
1092static bool check_lock_chain_key(struct lock_chain *chain)
1093{
1094#ifdef CONFIG_PROVE_LOCKING
1095 u64 chain_key = INITIAL_CHAIN_KEY;
1096 int i;
1097
1098 for (i = chain->base; i < chain->base + chain->depth; i++)
1099 chain_key = iterate_chain_key(chain_key, chain_hlocks[i]);
1100 /*
1101 * The 'unsigned long long' casts avoid that a compiler warning
1102 * is reported when building tools/lib/lockdep.
1103 */
1104 if (chain->chain_key != chain_key) {
1105 printk(KERN_INFO "chain %lld: key %#llx <> %#llx\n",
1106 (unsigned long long)(chain - lock_chains),
1107 (unsigned long long)chain->chain_key,
1108 (unsigned long long)chain_key);
1109 return false;
1110 }
1111#endif
1112 return true;
1113}
1114
1115static bool in_any_zapped_class_list(struct lock_class *class)
1116{
1117 struct pending_free *pf;
1118 int i;
1119
1120 for (i = 0, pf = delayed_free.pf; i < ARRAY_SIZE(delayed_free.pf); i++, pf++) {
1121 if (in_list(&class->lock_entry, &pf->zapped))
1122 return true;
1123 }
1124
1125 return false;
1126}
1127
1128static bool __check_data_structures(void)
1129{
1130 struct lock_class *class;
1131 struct lock_chain *chain;
1132 struct hlist_head *head;
1133 struct lock_list *e;
1134 int i;
1135
1136 /* Check whether all classes occur in a lock list. */
1137 for (i = 0; i < ARRAY_SIZE(lock_classes); i++) {
1138 class = &lock_classes[i];
1139 if (!in_list(&class->lock_entry, &all_lock_classes) &&
1140 !in_list(&class->lock_entry, &free_lock_classes) &&
1141 !in_any_zapped_class_list(class)) {
1142 printk(KERN_INFO "class %px/%s is not in any class list\n",
1143 class, class->name ? : "(?)");
1144 return false;
1145 }
1146 }
1147
1148 /* Check whether all classes have valid lock lists. */
1149 for (i = 0; i < ARRAY_SIZE(lock_classes); i++) {
1150 class = &lock_classes[i];
1151 if (!class_lock_list_valid(class, &class->locks_before))
1152 return false;
1153 if (!class_lock_list_valid(class, &class->locks_after))
1154 return false;
1155 }
1156
1157 /* Check the chain_key of all lock chains. */
1158 for (i = 0; i < ARRAY_SIZE(chainhash_table); i++) {
1159 head = chainhash_table + i;
1160 hlist_for_each_entry_rcu(chain, head, entry) {
1161 if (!check_lock_chain_key(chain))
1162 return false;
1163 }
1164 }
1165
1166 /*
1167 * Check whether all list entries that are in use occur in a class
1168 * lock list.
1169 */
1170 for_each_set_bit(i, list_entries_in_use, ARRAY_SIZE(list_entries)) {
1171 e = list_entries + i;
1172 if (!in_any_class_list(&e->entry)) {
1173 printk(KERN_INFO "list entry %d is not in any class list; class %s <> %s\n",
1174 (unsigned int)(e - list_entries),
1175 e->class->name ? : "(?)",
1176 e->links_to->name ? : "(?)");
1177 return false;
1178 }
1179 }
1180
1181 /*
1182 * Check whether all list entries that are not in use do not occur in
1183 * a class lock list.
1184 */
1185 for_each_clear_bit(i, list_entries_in_use, ARRAY_SIZE(list_entries)) {
1186 e = list_entries + i;
1187 if (in_any_class_list(&e->entry)) {
1188 printk(KERN_INFO "list entry %d occurs in a class list; class %s <> %s\n",
1189 (unsigned int)(e - list_entries),
1190 e->class && e->class->name ? e->class->name :
1191 "(?)",
1192 e->links_to && e->links_to->name ?
1193 e->links_to->name : "(?)");
1194 return false;
1195 }
1196 }
1197
1198 return true;
1199}
1200
1201int check_consistency = 0;
1202module_param(check_consistency, int, 0644);
1203
1204static void check_data_structures(void)
1205{
1206 static bool once = false;
1207
1208 if (check_consistency && !once) {
1209 if (!__check_data_structures()) {
1210 once = true;
1211 WARN_ON(once);
1212 }
1213 }
1214}
1215
1216#else /* CONFIG_DEBUG_LOCKDEP */
1217
1218static inline void check_data_structures(void) { }
1219
1220#endif /* CONFIG_DEBUG_LOCKDEP */
1221
1222static void init_chain_block_buckets(void);
1223
1224/*
1225 * Initialize the lock_classes[] array elements, the free_lock_classes list
1226 * and also the delayed_free structure.
1227 */
1228static void init_data_structures_once(void)
1229{
1230 static bool __read_mostly ds_initialized, rcu_head_initialized;
1231 int i;
1232
1233 if (likely(rcu_head_initialized))
1234 return;
1235
1236 if (system_state >= SYSTEM_SCHEDULING) {
1237 init_rcu_head(&delayed_free.rcu_head);
1238 rcu_head_initialized = true;
1239 }
1240
1241 if (ds_initialized)
1242 return;
1243
1244 ds_initialized = true;
1245
1246 INIT_LIST_HEAD(&delayed_free.pf[0].zapped);
1247 INIT_LIST_HEAD(&delayed_free.pf[1].zapped);
1248
1249 for (i = 0; i < ARRAY_SIZE(lock_classes); i++) {
1250 list_add_tail(&lock_classes[i].lock_entry, &free_lock_classes);
1251 INIT_LIST_HEAD(&lock_classes[i].locks_after);
1252 INIT_LIST_HEAD(&lock_classes[i].locks_before);
1253 }
1254 init_chain_block_buckets();
1255}
1256
1257static inline struct hlist_head *keyhashentry(const struct lock_class_key *key)
1258{
1259 unsigned long hash = hash_long((uintptr_t)key, KEYHASH_BITS);
1260
1261 return lock_keys_hash + hash;
1262}
1263
1264/* Register a dynamically allocated key. */
1265void lockdep_register_key(struct lock_class_key *key)
1266{
1267 struct hlist_head *hash_head;
1268 struct lock_class_key *k;
1269 unsigned long flags;
1270
1271 if (WARN_ON_ONCE(static_obj(key)))
1272 return;
1273 hash_head = keyhashentry(key);
1274
1275 raw_local_irq_save(flags);
1276 if (!graph_lock())
1277 goto restore_irqs;
1278 hlist_for_each_entry_rcu(k, hash_head, hash_entry) {
1279 if (WARN_ON_ONCE(k == key))
1280 goto out_unlock;
1281 }
1282 hlist_add_head_rcu(&key->hash_entry, hash_head);
1283 nr_dynamic_keys++;
1284out_unlock:
1285 graph_unlock();
1286restore_irqs:
1287 raw_local_irq_restore(flags);
1288}
1289EXPORT_SYMBOL_GPL(lockdep_register_key);
1290
1291/* Check whether a key has been registered as a dynamic key. */
1292static bool is_dynamic_key(const struct lock_class_key *key)
1293{
1294 struct hlist_head *hash_head;
1295 struct lock_class_key *k;
1296 bool found = false;
1297
1298 if (WARN_ON_ONCE(static_obj(key)))
1299 return false;
1300
1301 /*
1302 * If lock debugging is disabled lock_keys_hash[] may contain
1303 * pointers to memory that has already been freed. Avoid triggering
1304 * a use-after-free in that case by returning early.
1305 */
1306 if (!debug_locks)
1307 return true;
1308
1309 hash_head = keyhashentry(key);
1310
1311 rcu_read_lock();
1312 hlist_for_each_entry_rcu(k, hash_head, hash_entry) {
1313 if (k == key) {
1314 found = true;
1315 break;
1316 }
1317 }
1318 rcu_read_unlock();
1319
1320 return found;
1321}
1322
1323/*
1324 * Register a lock's class in the hash-table, if the class is not present
1325 * yet. Otherwise we look it up. We cache the result in the lock object
1326 * itself, so actual lookup of the hash should be once per lock object.
1327 */
1328static struct lock_class *
1329register_lock_class(struct lockdep_map *lock, unsigned int subclass, int force)
1330{
1331 struct lockdep_subclass_key *key;
1332 struct hlist_head *hash_head;
1333 struct lock_class *class;
1334 int idx;
1335
1336 DEBUG_LOCKS_WARN_ON(!irqs_disabled());
1337
1338 class = look_up_lock_class(lock, subclass);
1339 if (likely(class))
1340 goto out_set_class_cache;
1341
1342 if (!lock->key) {
1343 if (!assign_lock_key(lock))
1344 return NULL;
1345 } else if (!static_obj(lock->key) && !is_dynamic_key(lock->key)) {
1346 return NULL;
1347 }
1348
1349 key = lock->key->subkeys + subclass;
1350 hash_head = classhashentry(key);
1351
1352 if (!graph_lock()) {
1353 return NULL;
1354 }
1355 /*
1356 * We have to do the hash-walk again, to avoid races
1357 * with another CPU:
1358 */
1359 hlist_for_each_entry_rcu(class, hash_head, hash_entry) {
1360 if (class->key == key)
1361 goto out_unlock_set;
1362 }
1363
1364 init_data_structures_once();
1365
1366 /* Allocate a new lock class and add it to the hash. */
1367 class = list_first_entry_or_null(&free_lock_classes, typeof(*class),
1368 lock_entry);
1369 if (!class) {
1370 if (!debug_locks_off_graph_unlock()) {
1371 return NULL;
1372 }
1373
1374 nbcon_cpu_emergency_enter();
1375 print_lockdep_off("BUG: MAX_LOCKDEP_KEYS too low!");
1376 dump_stack();
1377 nbcon_cpu_emergency_exit();
1378 return NULL;
1379 }
1380 nr_lock_classes++;
1381 __set_bit(class - lock_classes, lock_classes_in_use);
1382 debug_atomic_inc(nr_unused_locks);
1383 class->key = key;
1384 class->name = lock->name;
1385 class->subclass = subclass;
1386 WARN_ON_ONCE(!list_empty(&class->locks_before));
1387 WARN_ON_ONCE(!list_empty(&class->locks_after));
1388 class->name_version = count_matching_names(class);
1389 class->wait_type_inner = lock->wait_type_inner;
1390 class->wait_type_outer = lock->wait_type_outer;
1391 class->lock_type = lock->lock_type;
1392 /*
1393 * We use RCU's safe list-add method to make
1394 * parallel walking of the hash-list safe:
1395 */
1396 hlist_add_head_rcu(&class->hash_entry, hash_head);
1397 /*
1398 * Remove the class from the free list and add it to the global list
1399 * of classes.
1400 */
1401 list_move_tail(&class->lock_entry, &all_lock_classes);
1402 idx = class - lock_classes;
1403 if (idx > max_lock_class_idx)
1404 max_lock_class_idx = idx;
1405
1406 if (verbose(class)) {
1407 graph_unlock();
1408
1409 nbcon_cpu_emergency_enter();
1410 printk("\nnew class %px: %s", class->key, class->name);
1411 if (class->name_version > 1)
1412 printk(KERN_CONT "#%d", class->name_version);
1413 printk(KERN_CONT "\n");
1414 dump_stack();
1415 nbcon_cpu_emergency_exit();
1416
1417 if (!graph_lock()) {
1418 return NULL;
1419 }
1420 }
1421out_unlock_set:
1422 graph_unlock();
1423
1424out_set_class_cache:
1425 if (!subclass || force)
1426 WRITE_ONCE(lock->class_cache[0], class);
1427 else if (subclass < NR_LOCKDEP_CACHING_CLASSES)
1428 WRITE_ONCE(lock->class_cache[subclass], class);
1429
1430 /*
1431 * Hash collision, did we smoke some? We found a class with a matching
1432 * hash but the subclass -- which is hashed in -- didn't match.
1433 */
1434 if (DEBUG_LOCKS_WARN_ON(class->subclass != subclass))
1435 return NULL;
1436
1437 return class;
1438}
1439
1440#ifdef CONFIG_PROVE_LOCKING
1441/*
1442 * Allocate a lockdep entry. (assumes the graph_lock held, returns
1443 * with NULL on failure)
1444 */
1445static struct lock_list *alloc_list_entry(void)
1446{
1447 int idx = find_first_zero_bit(list_entries_in_use,
1448 ARRAY_SIZE(list_entries));
1449
1450 if (idx >= ARRAY_SIZE(list_entries)) {
1451 if (!debug_locks_off_graph_unlock())
1452 return NULL;
1453
1454 nbcon_cpu_emergency_enter();
1455 print_lockdep_off("BUG: MAX_LOCKDEP_ENTRIES too low!");
1456 dump_stack();
1457 nbcon_cpu_emergency_exit();
1458 return NULL;
1459 }
1460 nr_list_entries++;
1461 __set_bit(idx, list_entries_in_use);
1462 return list_entries + idx;
1463}
1464
1465/*
1466 * Add a new dependency to the head of the list:
1467 */
1468static int add_lock_to_list(struct lock_class *this,
1469 struct lock_class *links_to, struct list_head *head,
1470 u16 distance, u8 dep,
1471 const struct lock_trace *trace)
1472{
1473 struct lock_list *entry;
1474 /*
1475 * Lock not present yet - get a new dependency struct and
1476 * add it to the list:
1477 */
1478 entry = alloc_list_entry();
1479 if (!entry)
1480 return 0;
1481
1482 entry->class = this;
1483 entry->links_to = links_to;
1484 entry->dep = dep;
1485 entry->distance = distance;
1486 entry->trace = trace;
1487 /*
1488 * Both allocation and removal are done under the graph lock; but
1489 * iteration is under RCU-sched; see look_up_lock_class() and
1490 * lockdep_free_key_range().
1491 */
1492 list_add_tail_rcu(&entry->entry, head);
1493
1494 return 1;
1495}
1496
1497/*
1498 * For good efficiency of modular, we use power of 2
1499 */
1500#define MAX_CIRCULAR_QUEUE_SIZE (1UL << CONFIG_LOCKDEP_CIRCULAR_QUEUE_BITS)
1501#define CQ_MASK (MAX_CIRCULAR_QUEUE_SIZE-1)
1502
1503/*
1504 * The circular_queue and helpers are used to implement graph
1505 * breadth-first search (BFS) algorithm, by which we can determine
1506 * whether there is a path from a lock to another. In deadlock checks,
1507 * a path from the next lock to be acquired to a previous held lock
1508 * indicates that adding the <prev> -> <next> lock dependency will
1509 * produce a circle in the graph. Breadth-first search instead of
1510 * depth-first search is used in order to find the shortest (circular)
1511 * path.
1512 */
1513struct circular_queue {
1514 struct lock_list *element[MAX_CIRCULAR_QUEUE_SIZE];
1515 unsigned int front, rear;
1516};
1517
1518static struct circular_queue lock_cq;
1519
1520unsigned int max_bfs_queue_depth;
1521
1522static unsigned int lockdep_dependency_gen_id;
1523
1524static inline void __cq_init(struct circular_queue *cq)
1525{
1526 cq->front = cq->rear = 0;
1527 lockdep_dependency_gen_id++;
1528}
1529
1530static inline int __cq_empty(struct circular_queue *cq)
1531{
1532 return (cq->front == cq->rear);
1533}
1534
1535static inline int __cq_full(struct circular_queue *cq)
1536{
1537 return ((cq->rear + 1) & CQ_MASK) == cq->front;
1538}
1539
1540static inline int __cq_enqueue(struct circular_queue *cq, struct lock_list *elem)
1541{
1542 if (__cq_full(cq))
1543 return -1;
1544
1545 cq->element[cq->rear] = elem;
1546 cq->rear = (cq->rear + 1) & CQ_MASK;
1547 return 0;
1548}
1549
1550/*
1551 * Dequeue an element from the circular_queue, return a lock_list if
1552 * the queue is not empty, or NULL if otherwise.
1553 */
1554static inline struct lock_list * __cq_dequeue(struct circular_queue *cq)
1555{
1556 struct lock_list * lock;
1557
1558 if (__cq_empty(cq))
1559 return NULL;
1560
1561 lock = cq->element[cq->front];
1562 cq->front = (cq->front + 1) & CQ_MASK;
1563
1564 return lock;
1565}
1566
1567static inline unsigned int __cq_get_elem_count(struct circular_queue *cq)
1568{
1569 return (cq->rear - cq->front) & CQ_MASK;
1570}
1571
1572static inline void mark_lock_accessed(struct lock_list *lock)
1573{
1574 lock->class->dep_gen_id = lockdep_dependency_gen_id;
1575}
1576
1577static inline void visit_lock_entry(struct lock_list *lock,
1578 struct lock_list *parent)
1579{
1580 lock->parent = parent;
1581}
1582
1583static inline unsigned long lock_accessed(struct lock_list *lock)
1584{
1585 return lock->class->dep_gen_id == lockdep_dependency_gen_id;
1586}
1587
1588static inline struct lock_list *get_lock_parent(struct lock_list *child)
1589{
1590 return child->parent;
1591}
1592
1593static inline int get_lock_depth(struct lock_list *child)
1594{
1595 int depth = 0;
1596 struct lock_list *parent;
1597
1598 while ((parent = get_lock_parent(child))) {
1599 child = parent;
1600 depth++;
1601 }
1602 return depth;
1603}
1604
1605/*
1606 * Return the forward or backward dependency list.
1607 *
1608 * @lock: the lock_list to get its class's dependency list
1609 * @offset: the offset to struct lock_class to determine whether it is
1610 * locks_after or locks_before
1611 */
1612static inline struct list_head *get_dep_list(struct lock_list *lock, int offset)
1613{
1614 void *lock_class = lock->class;
1615
1616 return lock_class + offset;
1617}
1618/*
1619 * Return values of a bfs search:
1620 *
1621 * BFS_E* indicates an error
1622 * BFS_R* indicates a result (match or not)
1623 *
1624 * BFS_EINVALIDNODE: Find a invalid node in the graph.
1625 *
1626 * BFS_EQUEUEFULL: The queue is full while doing the bfs.
1627 *
1628 * BFS_RMATCH: Find the matched node in the graph, and put that node into
1629 * *@target_entry.
1630 *
1631 * BFS_RNOMATCH: Haven't found the matched node and keep *@target_entry
1632 * _unchanged_.
1633 */
1634enum bfs_result {
1635 BFS_EINVALIDNODE = -2,
1636 BFS_EQUEUEFULL = -1,
1637 BFS_RMATCH = 0,
1638 BFS_RNOMATCH = 1,
1639};
1640
1641/*
1642 * bfs_result < 0 means error
1643 */
1644static inline bool bfs_error(enum bfs_result res)
1645{
1646 return res < 0;
1647}
1648
1649/*
1650 * DEP_*_BIT in lock_list::dep
1651 *
1652 * For dependency @prev -> @next:
1653 *
1654 * SR: @prev is shared reader (->read != 0) and @next is recursive reader
1655 * (->read == 2)
1656 * ER: @prev is exclusive locker (->read == 0) and @next is recursive reader
1657 * SN: @prev is shared reader and @next is non-recursive locker (->read != 2)
1658 * EN: @prev is exclusive locker and @next is non-recursive locker
1659 *
1660 * Note that we define the value of DEP_*_BITs so that:
1661 * bit0 is prev->read == 0
1662 * bit1 is next->read != 2
1663 */
1664#define DEP_SR_BIT (0 + (0 << 1)) /* 0 */
1665#define DEP_ER_BIT (1 + (0 << 1)) /* 1 */
1666#define DEP_SN_BIT (0 + (1 << 1)) /* 2 */
1667#define DEP_EN_BIT (1 + (1 << 1)) /* 3 */
1668
1669#define DEP_SR_MASK (1U << (DEP_SR_BIT))
1670#define DEP_ER_MASK (1U << (DEP_ER_BIT))
1671#define DEP_SN_MASK (1U << (DEP_SN_BIT))
1672#define DEP_EN_MASK (1U << (DEP_EN_BIT))
1673
1674static inline unsigned int
1675__calc_dep_bit(struct held_lock *prev, struct held_lock *next)
1676{
1677 return (prev->read == 0) + ((next->read != 2) << 1);
1678}
1679
1680static inline u8 calc_dep(struct held_lock *prev, struct held_lock *next)
1681{
1682 return 1U << __calc_dep_bit(prev, next);
1683}
1684
1685/*
1686 * calculate the dep_bit for backwards edges. We care about whether @prev is
1687 * shared and whether @next is recursive.
1688 */
1689static inline unsigned int
1690__calc_dep_bitb(struct held_lock *prev, struct held_lock *next)
1691{
1692 return (next->read != 2) + ((prev->read == 0) << 1);
1693}
1694
1695static inline u8 calc_depb(struct held_lock *prev, struct held_lock *next)
1696{
1697 return 1U << __calc_dep_bitb(prev, next);
1698}
1699
1700/*
1701 * Initialize a lock_list entry @lock belonging to @class as the root for a BFS
1702 * search.
1703 */
1704static inline void __bfs_init_root(struct lock_list *lock,
1705 struct lock_class *class)
1706{
1707 lock->class = class;
1708 lock->parent = NULL;
1709 lock->only_xr = 0;
1710}
1711
1712/*
1713 * Initialize a lock_list entry @lock based on a lock acquisition @hlock as the
1714 * root for a BFS search.
1715 *
1716 * ->only_xr of the initial lock node is set to @hlock->read == 2, to make sure
1717 * that <prev> -> @hlock and @hlock -> <whatever __bfs() found> is not -(*R)->
1718 * and -(S*)->.
1719 */
1720static inline void bfs_init_root(struct lock_list *lock,
1721 struct held_lock *hlock)
1722{
1723 __bfs_init_root(lock, hlock_class(hlock));
1724 lock->only_xr = (hlock->read == 2);
1725}
1726
1727/*
1728 * Similar to bfs_init_root() but initialize the root for backwards BFS.
1729 *
1730 * ->only_xr of the initial lock node is set to @hlock->read != 0, to make sure
1731 * that <next> -> @hlock and @hlock -> <whatever backwards BFS found> is not
1732 * -(*S)-> and -(R*)-> (reverse order of -(*R)-> and -(S*)->).
1733 */
1734static inline void bfs_init_rootb(struct lock_list *lock,
1735 struct held_lock *hlock)
1736{
1737 __bfs_init_root(lock, hlock_class(hlock));
1738 lock->only_xr = (hlock->read != 0);
1739}
1740
1741static inline struct lock_list *__bfs_next(struct lock_list *lock, int offset)
1742{
1743 if (!lock || !lock->parent)
1744 return NULL;
1745
1746 return list_next_or_null_rcu(get_dep_list(lock->parent, offset),
1747 &lock->entry, struct lock_list, entry);
1748}
1749
1750/*
1751 * Breadth-First Search to find a strong path in the dependency graph.
1752 *
1753 * @source_entry: the source of the path we are searching for.
1754 * @data: data used for the second parameter of @match function
1755 * @match: match function for the search
1756 * @target_entry: pointer to the target of a matched path
1757 * @offset: the offset to struct lock_class to determine whether it is
1758 * locks_after or locks_before
1759 *
1760 * We may have multiple edges (considering different kinds of dependencies,
1761 * e.g. ER and SN) between two nodes in the dependency graph. But
1762 * only the strong dependency path in the graph is relevant to deadlocks. A
1763 * strong dependency path is a dependency path that doesn't have two adjacent
1764 * dependencies as -(*R)-> -(S*)->, please see:
1765 *
1766 * Documentation/locking/lockdep-design.rst
1767 *
1768 * for more explanation of the definition of strong dependency paths
1769 *
1770 * In __bfs(), we only traverse in the strong dependency path:
1771 *
1772 * In lock_list::only_xr, we record whether the previous dependency only
1773 * has -(*R)-> in the search, and if it does (prev only has -(*R)->), we
1774 * filter out any -(S*)-> in the current dependency and after that, the
1775 * ->only_xr is set according to whether we only have -(*R)-> left.
1776 */
1777static enum bfs_result __bfs(struct lock_list *source_entry,
1778 void *data,
1779 bool (*match)(struct lock_list *entry, void *data),
1780 bool (*skip)(struct lock_list *entry, void *data),
1781 struct lock_list **target_entry,
1782 int offset)
1783{
1784 struct circular_queue *cq = &lock_cq;
1785 struct lock_list *lock = NULL;
1786 struct lock_list *entry;
1787 struct list_head *head;
1788 unsigned int cq_depth;
1789 bool first;
1790
1791 lockdep_assert_locked();
1792
1793 __cq_init(cq);
1794 __cq_enqueue(cq, source_entry);
1795
1796 while ((lock = __bfs_next(lock, offset)) || (lock = __cq_dequeue(cq))) {
1797 if (!lock->class)
1798 return BFS_EINVALIDNODE;
1799
1800 /*
1801 * Step 1: check whether we already finish on this one.
1802 *
1803 * If we have visited all the dependencies from this @lock to
1804 * others (iow, if we have visited all lock_list entries in
1805 * @lock->class->locks_{after,before}) we skip, otherwise go
1806 * and visit all the dependencies in the list and mark this
1807 * list accessed.
1808 */
1809 if (lock_accessed(lock))
1810 continue;
1811 else
1812 mark_lock_accessed(lock);
1813
1814 /*
1815 * Step 2: check whether prev dependency and this form a strong
1816 * dependency path.
1817 */
1818 if (lock->parent) { /* Parent exists, check prev dependency */
1819 u8 dep = lock->dep;
1820 bool prev_only_xr = lock->parent->only_xr;
1821
1822 /*
1823 * Mask out all -(S*)-> if we only have *R in previous
1824 * step, because -(*R)-> -(S*)-> don't make up a strong
1825 * dependency.
1826 */
1827 if (prev_only_xr)
1828 dep &= ~(DEP_SR_MASK | DEP_SN_MASK);
1829
1830 /* If nothing left, we skip */
1831 if (!dep)
1832 continue;
1833
1834 /* If there are only -(*R)-> left, set that for the next step */
1835 lock->only_xr = !(dep & (DEP_SN_MASK | DEP_EN_MASK));
1836 }
1837
1838 /*
1839 * Step 3: we haven't visited this and there is a strong
1840 * dependency path to this, so check with @match.
1841 * If @skip is provide and returns true, we skip this
1842 * lock (and any path this lock is in).
1843 */
1844 if (skip && skip(lock, data))
1845 continue;
1846
1847 if (match(lock, data)) {
1848 *target_entry = lock;
1849 return BFS_RMATCH;
1850 }
1851
1852 /*
1853 * Step 4: if not match, expand the path by adding the
1854 * forward or backwards dependencies in the search
1855 *
1856 */
1857 first = true;
1858 head = get_dep_list(lock, offset);
1859 list_for_each_entry_rcu(entry, head, entry) {
1860 visit_lock_entry(entry, lock);
1861
1862 /*
1863 * Note we only enqueue the first of the list into the
1864 * queue, because we can always find a sibling
1865 * dependency from one (see __bfs_next()), as a result
1866 * the space of queue is saved.
1867 */
1868 if (!first)
1869 continue;
1870
1871 first = false;
1872
1873 if (__cq_enqueue(cq, entry))
1874 return BFS_EQUEUEFULL;
1875
1876 cq_depth = __cq_get_elem_count(cq);
1877 if (max_bfs_queue_depth < cq_depth)
1878 max_bfs_queue_depth = cq_depth;
1879 }
1880 }
1881
1882 return BFS_RNOMATCH;
1883}
1884
1885static inline enum bfs_result
1886__bfs_forwards(struct lock_list *src_entry,
1887 void *data,
1888 bool (*match)(struct lock_list *entry, void *data),
1889 bool (*skip)(struct lock_list *entry, void *data),
1890 struct lock_list **target_entry)
1891{
1892 return __bfs(src_entry, data, match, skip, target_entry,
1893 offsetof(struct lock_class, locks_after));
1894
1895}
1896
1897static inline enum bfs_result
1898__bfs_backwards(struct lock_list *src_entry,
1899 void *data,
1900 bool (*match)(struct lock_list *entry, void *data),
1901 bool (*skip)(struct lock_list *entry, void *data),
1902 struct lock_list **target_entry)
1903{
1904 return __bfs(src_entry, data, match, skip, target_entry,
1905 offsetof(struct lock_class, locks_before));
1906
1907}
1908
1909static void print_lock_trace(const struct lock_trace *trace,
1910 unsigned int spaces)
1911{
1912 stack_trace_print(trace->entries, trace->nr_entries, spaces);
1913}
1914
1915/*
1916 * Print a dependency chain entry (this is only done when a deadlock
1917 * has been detected):
1918 */
1919static noinline void
1920print_circular_bug_entry(struct lock_list *target, int depth)
1921{
1922 if (debug_locks_silent)
1923 return;
1924 printk("\n-> #%u", depth);
1925 print_lock_name(NULL, target->class);
1926 printk(KERN_CONT ":\n");
1927 print_lock_trace(target->trace, 6);
1928}
1929
1930static void
1931print_circular_lock_scenario(struct held_lock *src,
1932 struct held_lock *tgt,
1933 struct lock_list *prt)
1934{
1935 struct lock_class *source = hlock_class(src);
1936 struct lock_class *target = hlock_class(tgt);
1937 struct lock_class *parent = prt->class;
1938 int src_read = src->read;
1939 int tgt_read = tgt->read;
1940
1941 /*
1942 * A direct locking problem where unsafe_class lock is taken
1943 * directly by safe_class lock, then all we need to show
1944 * is the deadlock scenario, as it is obvious that the
1945 * unsafe lock is taken under the safe lock.
1946 *
1947 * But if there is a chain instead, where the safe lock takes
1948 * an intermediate lock (middle_class) where this lock is
1949 * not the same as the safe lock, then the lock chain is
1950 * used to describe the problem. Otherwise we would need
1951 * to show a different CPU case for each link in the chain
1952 * from the safe_class lock to the unsafe_class lock.
1953 */
1954 if (parent != source) {
1955 printk("Chain exists of:\n ");
1956 __print_lock_name(src, source);
1957 printk(KERN_CONT " --> ");
1958 __print_lock_name(NULL, parent);
1959 printk(KERN_CONT " --> ");
1960 __print_lock_name(tgt, target);
1961 printk(KERN_CONT "\n\n");
1962 }
1963
1964 printk(" Possible unsafe locking scenario:\n\n");
1965 printk(" CPU0 CPU1\n");
1966 printk(" ---- ----\n");
1967 if (tgt_read != 0)
1968 printk(" rlock(");
1969 else
1970 printk(" lock(");
1971 __print_lock_name(tgt, target);
1972 printk(KERN_CONT ");\n");
1973 printk(" lock(");
1974 __print_lock_name(NULL, parent);
1975 printk(KERN_CONT ");\n");
1976 printk(" lock(");
1977 __print_lock_name(tgt, target);
1978 printk(KERN_CONT ");\n");
1979 if (src_read != 0)
1980 printk(" rlock(");
1981 else if (src->sync)
1982 printk(" sync(");
1983 else
1984 printk(" lock(");
1985 __print_lock_name(src, source);
1986 printk(KERN_CONT ");\n");
1987 printk("\n *** DEADLOCK ***\n\n");
1988}
1989
1990/*
1991 * When a circular dependency is detected, print the
1992 * header first:
1993 */
1994static noinline void
1995print_circular_bug_header(struct lock_list *entry, unsigned int depth,
1996 struct held_lock *check_src,
1997 struct held_lock *check_tgt)
1998{
1999 struct task_struct *curr = current;
2000
2001 if (debug_locks_silent)
2002 return;
2003
2004 pr_warn("\n");
2005 pr_warn("======================================================\n");
2006 pr_warn("WARNING: possible circular locking dependency detected\n");
2007 print_kernel_ident();
2008 pr_warn("------------------------------------------------------\n");
2009 pr_warn("%s/%d is trying to acquire lock:\n",
2010 curr->comm, task_pid_nr(curr));
2011 print_lock(check_src);
2012
2013 pr_warn("\nbut task is already holding lock:\n");
2014
2015 print_lock(check_tgt);
2016 pr_warn("\nwhich lock already depends on the new lock.\n\n");
2017 pr_warn("\nthe existing dependency chain (in reverse order) is:\n");
2018
2019 print_circular_bug_entry(entry, depth);
2020}
2021
2022/*
2023 * We are about to add B -> A into the dependency graph, and in __bfs() a
2024 * strong dependency path A -> .. -> B is found: hlock_class equals
2025 * entry->class.
2026 *
2027 * We will have a deadlock case (conflict) if A -> .. -> B -> A is a strong
2028 * dependency cycle, that means:
2029 *
2030 * Either
2031 *
2032 * a) B -> A is -(E*)->
2033 *
2034 * or
2035 *
2036 * b) A -> .. -> B is -(*N)-> (i.e. A -> .. -(*N)-> B)
2037 *
2038 * as then we don't have -(*R)-> -(S*)-> in the cycle.
2039 */
2040static inline bool hlock_conflict(struct lock_list *entry, void *data)
2041{
2042 struct held_lock *hlock = (struct held_lock *)data;
2043
2044 return hlock_class(hlock) == entry->class && /* Found A -> .. -> B */
2045 (hlock->read == 0 || /* B -> A is -(E*)-> */
2046 !entry->only_xr); /* A -> .. -> B is -(*N)-> */
2047}
2048
2049static noinline void print_circular_bug(struct lock_list *this,
2050 struct lock_list *target,
2051 struct held_lock *check_src,
2052 struct held_lock *check_tgt)
2053{
2054 struct task_struct *curr = current;
2055 struct lock_list *parent;
2056 struct lock_list *first_parent;
2057 int depth;
2058
2059 if (!debug_locks_off_graph_unlock() || debug_locks_silent)
2060 return;
2061
2062 this->trace = save_trace();
2063 if (!this->trace)
2064 return;
2065
2066 depth = get_lock_depth(target);
2067
2068 nbcon_cpu_emergency_enter();
2069
2070 print_circular_bug_header(target, depth, check_src, check_tgt);
2071
2072 parent = get_lock_parent(target);
2073 first_parent = parent;
2074
2075 while (parent) {
2076 print_circular_bug_entry(parent, --depth);
2077 parent = get_lock_parent(parent);
2078 }
2079
2080 printk("\nother info that might help us debug this:\n\n");
2081 print_circular_lock_scenario(check_src, check_tgt,
2082 first_parent);
2083
2084 lockdep_print_held_locks(curr);
2085
2086 printk("\nstack backtrace:\n");
2087 dump_stack();
2088
2089 nbcon_cpu_emergency_exit();
2090}
2091
2092static noinline void print_bfs_bug(int ret)
2093{
2094 if (!debug_locks_off_graph_unlock())
2095 return;
2096
2097 /*
2098 * Breadth-first-search failed, graph got corrupted?
2099 */
2100 if (ret == BFS_EQUEUEFULL)
2101 pr_warn("Increase LOCKDEP_CIRCULAR_QUEUE_BITS to avoid this warning:\n");
2102
2103 WARN(1, "lockdep bfs error:%d\n", ret);
2104}
2105
2106static bool noop_count(struct lock_list *entry, void *data)
2107{
2108 (*(unsigned long *)data)++;
2109 return false;
2110}
2111
2112static unsigned long __lockdep_count_forward_deps(struct lock_list *this)
2113{
2114 unsigned long count = 0;
2115 struct lock_list *target_entry;
2116
2117 __bfs_forwards(this, (void *)&count, noop_count, NULL, &target_entry);
2118
2119 return count;
2120}
2121unsigned long lockdep_count_forward_deps(struct lock_class *class)
2122{
2123 unsigned long ret, flags;
2124 struct lock_list this;
2125
2126 __bfs_init_root(&this, class);
2127
2128 raw_local_irq_save(flags);
2129 lockdep_lock();
2130 ret = __lockdep_count_forward_deps(&this);
2131 lockdep_unlock();
2132 raw_local_irq_restore(flags);
2133
2134 return ret;
2135}
2136
2137static unsigned long __lockdep_count_backward_deps(struct lock_list *this)
2138{
2139 unsigned long count = 0;
2140 struct lock_list *target_entry;
2141
2142 __bfs_backwards(this, (void *)&count, noop_count, NULL, &target_entry);
2143
2144 return count;
2145}
2146
2147unsigned long lockdep_count_backward_deps(struct lock_class *class)
2148{
2149 unsigned long ret, flags;
2150 struct lock_list this;
2151
2152 __bfs_init_root(&this, class);
2153
2154 raw_local_irq_save(flags);
2155 lockdep_lock();
2156 ret = __lockdep_count_backward_deps(&this);
2157 lockdep_unlock();
2158 raw_local_irq_restore(flags);
2159
2160 return ret;
2161}
2162
2163/*
2164 * Check that the dependency graph starting at <src> can lead to
2165 * <target> or not.
2166 */
2167static noinline enum bfs_result
2168check_path(struct held_lock *target, struct lock_list *src_entry,
2169 bool (*match)(struct lock_list *entry, void *data),
2170 bool (*skip)(struct lock_list *entry, void *data),
2171 struct lock_list **target_entry)
2172{
2173 enum bfs_result ret;
2174
2175 ret = __bfs_forwards(src_entry, target, match, skip, target_entry);
2176
2177 if (unlikely(bfs_error(ret)))
2178 print_bfs_bug(ret);
2179
2180 return ret;
2181}
2182
2183static void print_deadlock_bug(struct task_struct *, struct held_lock *, struct held_lock *);
2184
2185/*
2186 * Prove that the dependency graph starting at <src> can not
2187 * lead to <target>. If it can, there is a circle when adding
2188 * <target> -> <src> dependency.
2189 *
2190 * Print an error and return BFS_RMATCH if it does.
2191 */
2192static noinline enum bfs_result
2193check_noncircular(struct held_lock *src, struct held_lock *target,
2194 struct lock_trace **const trace)
2195{
2196 enum bfs_result ret;
2197 struct lock_list *target_entry;
2198 struct lock_list src_entry;
2199
2200 bfs_init_root(&src_entry, src);
2201
2202 debug_atomic_inc(nr_cyclic_checks);
2203
2204 ret = check_path(target, &src_entry, hlock_conflict, NULL, &target_entry);
2205
2206 if (unlikely(ret == BFS_RMATCH)) {
2207 if (!*trace) {
2208 /*
2209 * If save_trace fails here, the printing might
2210 * trigger a WARN but because of the !nr_entries it
2211 * should not do bad things.
2212 */
2213 *trace = save_trace();
2214 }
2215
2216 if (src->class_idx == target->class_idx)
2217 print_deadlock_bug(current, src, target);
2218 else
2219 print_circular_bug(&src_entry, target_entry, src, target);
2220 }
2221
2222 return ret;
2223}
2224
2225#ifdef CONFIG_TRACE_IRQFLAGS
2226
2227/*
2228 * Forwards and backwards subgraph searching, for the purposes of
2229 * proving that two subgraphs can be connected by a new dependency
2230 * without creating any illegal irq-safe -> irq-unsafe lock dependency.
2231 *
2232 * A irq safe->unsafe deadlock happens with the following conditions:
2233 *
2234 * 1) We have a strong dependency path A -> ... -> B
2235 *
2236 * 2) and we have ENABLED_IRQ usage of B and USED_IN_IRQ usage of A, therefore
2237 * irq can create a new dependency B -> A (consider the case that a holder
2238 * of B gets interrupted by an irq whose handler will try to acquire A).
2239 *
2240 * 3) the dependency circle A -> ... -> B -> A we get from 1) and 2) is a
2241 * strong circle:
2242 *
2243 * For the usage bits of B:
2244 * a) if A -> B is -(*N)->, then B -> A could be any type, so any
2245 * ENABLED_IRQ usage suffices.
2246 * b) if A -> B is -(*R)->, then B -> A must be -(E*)->, so only
2247 * ENABLED_IRQ_*_READ usage suffices.
2248 *
2249 * For the usage bits of A:
2250 * c) if A -> B is -(E*)->, then B -> A could be any type, so any
2251 * USED_IN_IRQ usage suffices.
2252 * d) if A -> B is -(S*)->, then B -> A must be -(*N)->, so only
2253 * USED_IN_IRQ_*_READ usage suffices.
2254 */
2255
2256/*
2257 * There is a strong dependency path in the dependency graph: A -> B, and now
2258 * we need to decide which usage bit of A should be accumulated to detect
2259 * safe->unsafe bugs.
2260 *
2261 * Note that usage_accumulate() is used in backwards search, so ->only_xr
2262 * stands for whether A -> B only has -(S*)-> (in this case ->only_xr is true).
2263 *
2264 * As above, if only_xr is false, which means A -> B has -(E*)-> dependency
2265 * path, any usage of A should be considered. Otherwise, we should only
2266 * consider _READ usage.
2267 */
2268static inline bool usage_accumulate(struct lock_list *entry, void *mask)
2269{
2270 if (!entry->only_xr)
2271 *(unsigned long *)mask |= entry->class->usage_mask;
2272 else /* Mask out _READ usage bits */
2273 *(unsigned long *)mask |= (entry->class->usage_mask & LOCKF_IRQ);
2274
2275 return false;
2276}
2277
2278/*
2279 * There is a strong dependency path in the dependency graph: A -> B, and now
2280 * we need to decide which usage bit of B conflicts with the usage bits of A,
2281 * i.e. which usage bit of B may introduce safe->unsafe deadlocks.
2282 *
2283 * As above, if only_xr is false, which means A -> B has -(*N)-> dependency
2284 * path, any usage of B should be considered. Otherwise, we should only
2285 * consider _READ usage.
2286 */
2287static inline bool usage_match(struct lock_list *entry, void *mask)
2288{
2289 if (!entry->only_xr)
2290 return !!(entry->class->usage_mask & *(unsigned long *)mask);
2291 else /* Mask out _READ usage bits */
2292 return !!((entry->class->usage_mask & LOCKF_IRQ) & *(unsigned long *)mask);
2293}
2294
2295static inline bool usage_skip(struct lock_list *entry, void *mask)
2296{
2297 if (entry->class->lock_type == LD_LOCK_NORMAL)
2298 return false;
2299
2300 /*
2301 * Skip local_lock() for irq inversion detection.
2302 *
2303 * For !RT, local_lock() is not a real lock, so it won't carry any
2304 * dependency.
2305 *
2306 * For RT, an irq inversion happens when we have lock A and B, and on
2307 * some CPU we can have:
2308 *
2309 * lock(A);
2310 * <interrupted>
2311 * lock(B);
2312 *
2313 * where lock(B) cannot sleep, and we have a dependency B -> ... -> A.
2314 *
2315 * Now we prove local_lock() cannot exist in that dependency. First we
2316 * have the observation for any lock chain L1 -> ... -> Ln, for any
2317 * 1 <= i <= n, Li.inner_wait_type <= L1.inner_wait_type, otherwise
2318 * wait context check will complain. And since B is not a sleep lock,
2319 * therefore B.inner_wait_type >= 2, and since the inner_wait_type of
2320 * local_lock() is 3, which is greater than 2, therefore there is no
2321 * way the local_lock() exists in the dependency B -> ... -> A.
2322 *
2323 * As a result, we will skip local_lock(), when we search for irq
2324 * inversion bugs.
2325 */
2326 if (entry->class->lock_type == LD_LOCK_PERCPU &&
2327 DEBUG_LOCKS_WARN_ON(entry->class->wait_type_inner < LD_WAIT_CONFIG))
2328 return false;
2329
2330 /*
2331 * Skip WAIT_OVERRIDE for irq inversion detection -- it's not actually
2332 * a lock and only used to override the wait_type.
2333 */
2334
2335 return true;
2336}
2337
2338/*
2339 * Find a node in the forwards-direction dependency sub-graph starting
2340 * at @root->class that matches @bit.
2341 *
2342 * Return BFS_MATCH if such a node exists in the subgraph, and put that node
2343 * into *@target_entry.
2344 */
2345static enum bfs_result
2346find_usage_forwards(struct lock_list *root, unsigned long usage_mask,
2347 struct lock_list **target_entry)
2348{
2349 enum bfs_result result;
2350
2351 debug_atomic_inc(nr_find_usage_forwards_checks);
2352
2353 result = __bfs_forwards(root, &usage_mask, usage_match, usage_skip, target_entry);
2354
2355 return result;
2356}
2357
2358/*
2359 * Find a node in the backwards-direction dependency sub-graph starting
2360 * at @root->class that matches @bit.
2361 */
2362static enum bfs_result
2363find_usage_backwards(struct lock_list *root, unsigned long usage_mask,
2364 struct lock_list **target_entry)
2365{
2366 enum bfs_result result;
2367
2368 debug_atomic_inc(nr_find_usage_backwards_checks);
2369
2370 result = __bfs_backwards(root, &usage_mask, usage_match, usage_skip, target_entry);
2371
2372 return result;
2373}
2374
2375static void print_lock_class_header(struct lock_class *class, int depth)
2376{
2377 int bit;
2378
2379 printk("%*s->", depth, "");
2380 print_lock_name(NULL, class);
2381#ifdef CONFIG_DEBUG_LOCKDEP
2382 printk(KERN_CONT " ops: %lu", debug_class_ops_read(class));
2383#endif
2384 printk(KERN_CONT " {\n");
2385
2386 for (bit = 0; bit < LOCK_TRACE_STATES; bit++) {
2387 if (class->usage_mask & (1 << bit)) {
2388 int len = depth;
2389
2390 len += printk("%*s %s", depth, "", usage_str[bit]);
2391 len += printk(KERN_CONT " at:\n");
2392 print_lock_trace(class->usage_traces[bit], len);
2393 }
2394 }
2395 printk("%*s }\n", depth, "");
2396
2397 printk("%*s ... key at: [<%px>] %pS\n",
2398 depth, "", class->key, class->key);
2399}
2400
2401/*
2402 * Dependency path printing:
2403 *
2404 * After BFS we get a lock dependency path (linked via ->parent of lock_list),
2405 * printing out each lock in the dependency path will help on understanding how
2406 * the deadlock could happen. Here are some details about dependency path
2407 * printing:
2408 *
2409 * 1) A lock_list can be either forwards or backwards for a lock dependency,
2410 * for a lock dependency A -> B, there are two lock_lists:
2411 *
2412 * a) lock_list in the ->locks_after list of A, whose ->class is B and
2413 * ->links_to is A. In this case, we can say the lock_list is
2414 * "A -> B" (forwards case).
2415 *
2416 * b) lock_list in the ->locks_before list of B, whose ->class is A
2417 * and ->links_to is B. In this case, we can say the lock_list is
2418 * "B <- A" (bacwards case).
2419 *
2420 * The ->trace of both a) and b) point to the call trace where B was
2421 * acquired with A held.
2422 *
2423 * 2) A "helper" lock_list is introduced during BFS, this lock_list doesn't
2424 * represent a certain lock dependency, it only provides an initial entry
2425 * for BFS. For example, BFS may introduce a "helper" lock_list whose
2426 * ->class is A, as a result BFS will search all dependencies starting with
2427 * A, e.g. A -> B or A -> C.
2428 *
2429 * The notation of a forwards helper lock_list is like "-> A", which means
2430 * we should search the forwards dependencies starting with "A", e.g A -> B
2431 * or A -> C.
2432 *
2433 * The notation of a bacwards helper lock_list is like "<- B", which means
2434 * we should search the backwards dependencies ending with "B", e.g.
2435 * B <- A or B <- C.
2436 */
2437
2438/*
2439 * printk the shortest lock dependencies from @root to @leaf in reverse order.
2440 *
2441 * We have a lock dependency path as follow:
2442 *
2443 * @root @leaf
2444 * | |
2445 * V V
2446 * ->parent ->parent
2447 * | lock_list | <--------- | lock_list | ... | lock_list | <--------- | lock_list |
2448 * | -> L1 | | L1 -> L2 | ... |Ln-2 -> Ln-1| | Ln-1 -> Ln|
2449 *
2450 * , so it's natural that we start from @leaf and print every ->class and
2451 * ->trace until we reach the @root.
2452 */
2453static void __used
2454print_shortest_lock_dependencies(struct lock_list *leaf,
2455 struct lock_list *root)
2456{
2457 struct lock_list *entry = leaf;
2458 int depth;
2459
2460 /*compute depth from generated tree by BFS*/
2461 depth = get_lock_depth(leaf);
2462
2463 do {
2464 print_lock_class_header(entry->class, depth);
2465 printk("%*s ... acquired at:\n", depth, "");
2466 print_lock_trace(entry->trace, 2);
2467 printk("\n");
2468
2469 if (depth == 0 && (entry != root)) {
2470 printk("lockdep:%s bad path found in chain graph\n", __func__);
2471 break;
2472 }
2473
2474 entry = get_lock_parent(entry);
2475 depth--;
2476 } while (entry && (depth >= 0));
2477}
2478
2479/*
2480 * printk the shortest lock dependencies from @leaf to @root.
2481 *
2482 * We have a lock dependency path (from a backwards search) as follow:
2483 *
2484 * @leaf @root
2485 * | |
2486 * V V
2487 * ->parent ->parent
2488 * | lock_list | ---------> | lock_list | ... | lock_list | ---------> | lock_list |
2489 * | L2 <- L1 | | L3 <- L2 | ... | Ln <- Ln-1 | | <- Ln |
2490 *
2491 * , so when we iterate from @leaf to @root, we actually print the lock
2492 * dependency path L1 -> L2 -> .. -> Ln in the non-reverse order.
2493 *
2494 * Another thing to notice here is that ->class of L2 <- L1 is L1, while the
2495 * ->trace of L2 <- L1 is the call trace of L2, in fact we don't have the call
2496 * trace of L1 in the dependency path, which is alright, because most of the
2497 * time we can figure out where L1 is held from the call trace of L2.
2498 */
2499static void __used
2500print_shortest_lock_dependencies_backwards(struct lock_list *leaf,
2501 struct lock_list *root)
2502{
2503 struct lock_list *entry = leaf;
2504 const struct lock_trace *trace = NULL;
2505 int depth;
2506
2507 /*compute depth from generated tree by BFS*/
2508 depth = get_lock_depth(leaf);
2509
2510 do {
2511 print_lock_class_header(entry->class, depth);
2512 if (trace) {
2513 printk("%*s ... acquired at:\n", depth, "");
2514 print_lock_trace(trace, 2);
2515 printk("\n");
2516 }
2517
2518 /*
2519 * Record the pointer to the trace for the next lock_list
2520 * entry, see the comments for the function.
2521 */
2522 trace = entry->trace;
2523
2524 if (depth == 0 && (entry != root)) {
2525 printk("lockdep:%s bad path found in chain graph\n", __func__);
2526 break;
2527 }
2528
2529 entry = get_lock_parent(entry);
2530 depth--;
2531 } while (entry && (depth >= 0));
2532}
2533
2534static void
2535print_irq_lock_scenario(struct lock_list *safe_entry,
2536 struct lock_list *unsafe_entry,
2537 struct lock_class *prev_class,
2538 struct lock_class *next_class)
2539{
2540 struct lock_class *safe_class = safe_entry->class;
2541 struct lock_class *unsafe_class = unsafe_entry->class;
2542 struct lock_class *middle_class = prev_class;
2543
2544 if (middle_class == safe_class)
2545 middle_class = next_class;
2546
2547 /*
2548 * A direct locking problem where unsafe_class lock is taken
2549 * directly by safe_class lock, then all we need to show
2550 * is the deadlock scenario, as it is obvious that the
2551 * unsafe lock is taken under the safe lock.
2552 *
2553 * But if there is a chain instead, where the safe lock takes
2554 * an intermediate lock (middle_class) where this lock is
2555 * not the same as the safe lock, then the lock chain is
2556 * used to describe the problem. Otherwise we would need
2557 * to show a different CPU case for each link in the chain
2558 * from the safe_class lock to the unsafe_class lock.
2559 */
2560 if (middle_class != unsafe_class) {
2561 printk("Chain exists of:\n ");
2562 __print_lock_name(NULL, safe_class);
2563 printk(KERN_CONT " --> ");
2564 __print_lock_name(NULL, middle_class);
2565 printk(KERN_CONT " --> ");
2566 __print_lock_name(NULL, unsafe_class);
2567 printk(KERN_CONT "\n\n");
2568 }
2569
2570 printk(" Possible interrupt unsafe locking scenario:\n\n");
2571 printk(" CPU0 CPU1\n");
2572 printk(" ---- ----\n");
2573 printk(" lock(");
2574 __print_lock_name(NULL, unsafe_class);
2575 printk(KERN_CONT ");\n");
2576 printk(" local_irq_disable();\n");
2577 printk(" lock(");
2578 __print_lock_name(NULL, safe_class);
2579 printk(KERN_CONT ");\n");
2580 printk(" lock(");
2581 __print_lock_name(NULL, middle_class);
2582 printk(KERN_CONT ");\n");
2583 printk(" <Interrupt>\n");
2584 printk(" lock(");
2585 __print_lock_name(NULL, safe_class);
2586 printk(KERN_CONT ");\n");
2587 printk("\n *** DEADLOCK ***\n\n");
2588}
2589
2590static void
2591print_bad_irq_dependency(struct task_struct *curr,
2592 struct lock_list *prev_root,
2593 struct lock_list *next_root,
2594 struct lock_list *backwards_entry,
2595 struct lock_list *forwards_entry,
2596 struct held_lock *prev,
2597 struct held_lock *next,
2598 enum lock_usage_bit bit1,
2599 enum lock_usage_bit bit2,
2600 const char *irqclass)
2601{
2602 if (!debug_locks_off_graph_unlock() || debug_locks_silent)
2603 return;
2604
2605 nbcon_cpu_emergency_enter();
2606
2607 pr_warn("\n");
2608 pr_warn("=====================================================\n");
2609 pr_warn("WARNING: %s-safe -> %s-unsafe lock order detected\n",
2610 irqclass, irqclass);
2611 print_kernel_ident();
2612 pr_warn("-----------------------------------------------------\n");
2613 pr_warn("%s/%d [HC%u[%lu]:SC%u[%lu]:HE%u:SE%u] is trying to acquire:\n",
2614 curr->comm, task_pid_nr(curr),
2615 lockdep_hardirq_context(), hardirq_count() >> HARDIRQ_SHIFT,
2616 curr->softirq_context, softirq_count() >> SOFTIRQ_SHIFT,
2617 lockdep_hardirqs_enabled(),
2618 curr->softirqs_enabled);
2619 print_lock(next);
2620
2621 pr_warn("\nand this task is already holding:\n");
2622 print_lock(prev);
2623 pr_warn("which would create a new lock dependency:\n");
2624 print_lock_name(prev, hlock_class(prev));
2625 pr_cont(" ->");
2626 print_lock_name(next, hlock_class(next));
2627 pr_cont("\n");
2628
2629 pr_warn("\nbut this new dependency connects a %s-irq-safe lock:\n",
2630 irqclass);
2631 print_lock_name(NULL, backwards_entry->class);
2632 pr_warn("\n... which became %s-irq-safe at:\n", irqclass);
2633
2634 print_lock_trace(backwards_entry->class->usage_traces[bit1], 1);
2635
2636 pr_warn("\nto a %s-irq-unsafe lock:\n", irqclass);
2637 print_lock_name(NULL, forwards_entry->class);
2638 pr_warn("\n... which became %s-irq-unsafe at:\n", irqclass);
2639 pr_warn("...");
2640
2641 print_lock_trace(forwards_entry->class->usage_traces[bit2], 1);
2642
2643 pr_warn("\nother info that might help us debug this:\n\n");
2644 print_irq_lock_scenario(backwards_entry, forwards_entry,
2645 hlock_class(prev), hlock_class(next));
2646
2647 lockdep_print_held_locks(curr);
2648
2649 pr_warn("\nthe dependencies between %s-irq-safe lock and the holding lock:\n", irqclass);
2650 print_shortest_lock_dependencies_backwards(backwards_entry, prev_root);
2651
2652 pr_warn("\nthe dependencies between the lock to be acquired");
2653 pr_warn(" and %s-irq-unsafe lock:\n", irqclass);
2654 next_root->trace = save_trace();
2655 if (!next_root->trace)
2656 goto out;
2657 print_shortest_lock_dependencies(forwards_entry, next_root);
2658
2659 pr_warn("\nstack backtrace:\n");
2660 dump_stack();
2661out:
2662 nbcon_cpu_emergency_exit();
2663}
2664
2665static const char *state_names[] = {
2666#define LOCKDEP_STATE(__STATE) \
2667 __stringify(__STATE),
2668#include "lockdep_states.h"
2669#undef LOCKDEP_STATE
2670};
2671
2672static const char *state_rnames[] = {
2673#define LOCKDEP_STATE(__STATE) \
2674 __stringify(__STATE)"-READ",
2675#include "lockdep_states.h"
2676#undef LOCKDEP_STATE
2677};
2678
2679static inline const char *state_name(enum lock_usage_bit bit)
2680{
2681 if (bit & LOCK_USAGE_READ_MASK)
2682 return state_rnames[bit >> LOCK_USAGE_DIR_MASK];
2683 else
2684 return state_names[bit >> LOCK_USAGE_DIR_MASK];
2685}
2686
2687/*
2688 * The bit number is encoded like:
2689 *
2690 * bit0: 0 exclusive, 1 read lock
2691 * bit1: 0 used in irq, 1 irq enabled
2692 * bit2-n: state
2693 */
2694static int exclusive_bit(int new_bit)
2695{
2696 int state = new_bit & LOCK_USAGE_STATE_MASK;
2697 int dir = new_bit & LOCK_USAGE_DIR_MASK;
2698
2699 /*
2700 * keep state, bit flip the direction and strip read.
2701 */
2702 return state | (dir ^ LOCK_USAGE_DIR_MASK);
2703}
2704
2705/*
2706 * Observe that when given a bitmask where each bitnr is encoded as above, a
2707 * right shift of the mask transforms the individual bitnrs as -1 and
2708 * conversely, a left shift transforms into +1 for the individual bitnrs.
2709 *
2710 * So for all bits whose number have LOCK_ENABLED_* set (bitnr1 == 1), we can
2711 * create the mask with those bit numbers using LOCK_USED_IN_* (bitnr1 == 0)
2712 * instead by subtracting the bit number by 2, or shifting the mask right by 2.
2713 *
2714 * Similarly, bitnr1 == 0 becomes bitnr1 == 1 by adding 2, or shifting left 2.
2715 *
2716 * So split the mask (note that LOCKF_ENABLED_IRQ_ALL|LOCKF_USED_IN_IRQ_ALL is
2717 * all bits set) and recompose with bitnr1 flipped.
2718 */
2719static unsigned long invert_dir_mask(unsigned long mask)
2720{
2721 unsigned long excl = 0;
2722
2723 /* Invert dir */
2724 excl |= (mask & LOCKF_ENABLED_IRQ_ALL) >> LOCK_USAGE_DIR_MASK;
2725 excl |= (mask & LOCKF_USED_IN_IRQ_ALL) << LOCK_USAGE_DIR_MASK;
2726
2727 return excl;
2728}
2729
2730/*
2731 * Note that a LOCK_ENABLED_IRQ_*_READ usage and a LOCK_USED_IN_IRQ_*_READ
2732 * usage may cause deadlock too, for example:
2733 *
2734 * P1 P2
2735 * <irq disabled>
2736 * write_lock(l1); <irq enabled>
2737 * read_lock(l2);
2738 * write_lock(l2);
2739 * <in irq>
2740 * read_lock(l1);
2741 *
2742 * , in above case, l1 will be marked as LOCK_USED_IN_IRQ_HARDIRQ_READ and l2
2743 * will marked as LOCK_ENABLE_IRQ_HARDIRQ_READ, and this is a possible
2744 * deadlock.
2745 *
2746 * In fact, all of the following cases may cause deadlocks:
2747 *
2748 * LOCK_USED_IN_IRQ_* -> LOCK_ENABLED_IRQ_*
2749 * LOCK_USED_IN_IRQ_*_READ -> LOCK_ENABLED_IRQ_*
2750 * LOCK_USED_IN_IRQ_* -> LOCK_ENABLED_IRQ_*_READ
2751 * LOCK_USED_IN_IRQ_*_READ -> LOCK_ENABLED_IRQ_*_READ
2752 *
2753 * As a result, to calculate the "exclusive mask", first we invert the
2754 * direction (USED_IN/ENABLED) of the original mask, and 1) for all bits with
2755 * bitnr0 set (LOCK_*_READ), add those with bitnr0 cleared (LOCK_*). 2) for all
2756 * bits with bitnr0 cleared (LOCK_*_READ), add those with bitnr0 set (LOCK_*).
2757 */
2758static unsigned long exclusive_mask(unsigned long mask)
2759{
2760 unsigned long excl = invert_dir_mask(mask);
2761
2762 excl |= (excl & LOCKF_IRQ_READ) >> LOCK_USAGE_READ_MASK;
2763 excl |= (excl & LOCKF_IRQ) << LOCK_USAGE_READ_MASK;
2764
2765 return excl;
2766}
2767
2768/*
2769 * Retrieve the _possible_ original mask to which @mask is
2770 * exclusive. Ie: this is the opposite of exclusive_mask().
2771 * Note that 2 possible original bits can match an exclusive
2772 * bit: one has LOCK_USAGE_READ_MASK set, the other has it
2773 * cleared. So both are returned for each exclusive bit.
2774 */
2775static unsigned long original_mask(unsigned long mask)
2776{
2777 unsigned long excl = invert_dir_mask(mask);
2778
2779 /* Include read in existing usages */
2780 excl |= (excl & LOCKF_IRQ_READ) >> LOCK_USAGE_READ_MASK;
2781 excl |= (excl & LOCKF_IRQ) << LOCK_USAGE_READ_MASK;
2782
2783 return excl;
2784}
2785
2786/*
2787 * Find the first pair of bit match between an original
2788 * usage mask and an exclusive usage mask.
2789 */
2790static int find_exclusive_match(unsigned long mask,
2791 unsigned long excl_mask,
2792 enum lock_usage_bit *bitp,
2793 enum lock_usage_bit *excl_bitp)
2794{
2795 int bit, excl, excl_read;
2796
2797 for_each_set_bit(bit, &mask, LOCK_USED) {
2798 /*
2799 * exclusive_bit() strips the read bit, however,
2800 * LOCK_ENABLED_IRQ_*_READ may cause deadlocks too, so we need
2801 * to search excl | LOCK_USAGE_READ_MASK as well.
2802 */
2803 excl = exclusive_bit(bit);
2804 excl_read = excl | LOCK_USAGE_READ_MASK;
2805 if (excl_mask & lock_flag(excl)) {
2806 *bitp = bit;
2807 *excl_bitp = excl;
2808 return 0;
2809 } else if (excl_mask & lock_flag(excl_read)) {
2810 *bitp = bit;
2811 *excl_bitp = excl_read;
2812 return 0;
2813 }
2814 }
2815 return -1;
2816}
2817
2818/*
2819 * Prove that the new dependency does not connect a hardirq-safe(-read)
2820 * lock with a hardirq-unsafe lock - to achieve this we search
2821 * the backwards-subgraph starting at <prev>, and the
2822 * forwards-subgraph starting at <next>:
2823 */
2824static int check_irq_usage(struct task_struct *curr, struct held_lock *prev,
2825 struct held_lock *next)
2826{
2827 unsigned long usage_mask = 0, forward_mask, backward_mask;
2828 enum lock_usage_bit forward_bit = 0, backward_bit = 0;
2829 struct lock_list *target_entry1;
2830 struct lock_list *target_entry;
2831 struct lock_list this, that;
2832 enum bfs_result ret;
2833
2834 /*
2835 * Step 1: gather all hard/soft IRQs usages backward in an
2836 * accumulated usage mask.
2837 */
2838 bfs_init_rootb(&this, prev);
2839
2840 ret = __bfs_backwards(&this, &usage_mask, usage_accumulate, usage_skip, NULL);
2841 if (bfs_error(ret)) {
2842 print_bfs_bug(ret);
2843 return 0;
2844 }
2845
2846 usage_mask &= LOCKF_USED_IN_IRQ_ALL;
2847 if (!usage_mask)
2848 return 1;
2849
2850 /*
2851 * Step 2: find exclusive uses forward that match the previous
2852 * backward accumulated mask.
2853 */
2854 forward_mask = exclusive_mask(usage_mask);
2855
2856 bfs_init_root(&that, next);
2857
2858 ret = find_usage_forwards(&that, forward_mask, &target_entry1);
2859 if (bfs_error(ret)) {
2860 print_bfs_bug(ret);
2861 return 0;
2862 }
2863 if (ret == BFS_RNOMATCH)
2864 return 1;
2865
2866 /*
2867 * Step 3: we found a bad match! Now retrieve a lock from the backward
2868 * list whose usage mask matches the exclusive usage mask from the
2869 * lock found on the forward list.
2870 *
2871 * Note, we should only keep the LOCKF_ENABLED_IRQ_ALL bits, considering
2872 * the follow case:
2873 *
2874 * When trying to add A -> B to the graph, we find that there is a
2875 * hardirq-safe L, that L -> ... -> A, and another hardirq-unsafe M,
2876 * that B -> ... -> M. However M is **softirq-safe**, if we use exact
2877 * invert bits of M's usage_mask, we will find another lock N that is
2878 * **softirq-unsafe** and N -> ... -> A, however N -> .. -> M will not
2879 * cause a inversion deadlock.
2880 */
2881 backward_mask = original_mask(target_entry1->class->usage_mask & LOCKF_ENABLED_IRQ_ALL);
2882
2883 ret = find_usage_backwards(&this, backward_mask, &target_entry);
2884 if (bfs_error(ret)) {
2885 print_bfs_bug(ret);
2886 return 0;
2887 }
2888 if (DEBUG_LOCKS_WARN_ON(ret == BFS_RNOMATCH))
2889 return 1;
2890
2891 /*
2892 * Step 4: narrow down to a pair of incompatible usage bits
2893 * and report it.
2894 */
2895 ret = find_exclusive_match(target_entry->class->usage_mask,
2896 target_entry1->class->usage_mask,
2897 &backward_bit, &forward_bit);
2898 if (DEBUG_LOCKS_WARN_ON(ret == -1))
2899 return 1;
2900
2901 print_bad_irq_dependency(curr, &this, &that,
2902 target_entry, target_entry1,
2903 prev, next,
2904 backward_bit, forward_bit,
2905 state_name(backward_bit));
2906
2907 return 0;
2908}
2909
2910#else
2911
2912static inline int check_irq_usage(struct task_struct *curr,
2913 struct held_lock *prev, struct held_lock *next)
2914{
2915 return 1;
2916}
2917
2918static inline bool usage_skip(struct lock_list *entry, void *mask)
2919{
2920 return false;
2921}
2922
2923#endif /* CONFIG_TRACE_IRQFLAGS */
2924
2925#ifdef CONFIG_LOCKDEP_SMALL
2926/*
2927 * We are about to add A -> B into the dependency graph, and in __bfs() a
2928 * strong dependency path A -> .. -> B is found: hlock_class equals
2929 * entry->class.
2930 *
2931 * If A -> .. -> B can replace A -> B in any __bfs() search (means the former
2932 * is _stronger_ than or equal to the latter), we consider A -> B as redundant.
2933 * For example if A -> .. -> B is -(EN)-> (i.e. A -(E*)-> .. -(*N)-> B), and A
2934 * -> B is -(ER)-> or -(EN)->, then we don't need to add A -> B into the
2935 * dependency graph, as any strong path ..-> A -> B ->.. we can get with
2936 * having dependency A -> B, we could already get a equivalent path ..-> A ->
2937 * .. -> B -> .. with A -> .. -> B. Therefore A -> B is redundant.
2938 *
2939 * We need to make sure both the start and the end of A -> .. -> B is not
2940 * weaker than A -> B. For the start part, please see the comment in
2941 * check_redundant(). For the end part, we need:
2942 *
2943 * Either
2944 *
2945 * a) A -> B is -(*R)-> (everything is not weaker than that)
2946 *
2947 * or
2948 *
2949 * b) A -> .. -> B is -(*N)-> (nothing is stronger than this)
2950 *
2951 */
2952static inline bool hlock_equal(struct lock_list *entry, void *data)
2953{
2954 struct held_lock *hlock = (struct held_lock *)data;
2955
2956 return hlock_class(hlock) == entry->class && /* Found A -> .. -> B */
2957 (hlock->read == 2 || /* A -> B is -(*R)-> */
2958 !entry->only_xr); /* A -> .. -> B is -(*N)-> */
2959}
2960
2961/*
2962 * Check that the dependency graph starting at <src> can lead to
2963 * <target> or not. If it can, <src> -> <target> dependency is already
2964 * in the graph.
2965 *
2966 * Return BFS_RMATCH if it does, or BFS_RNOMATCH if it does not, return BFS_E* if
2967 * any error appears in the bfs search.
2968 */
2969static noinline enum bfs_result
2970check_redundant(struct held_lock *src, struct held_lock *target)
2971{
2972 enum bfs_result ret;
2973 struct lock_list *target_entry;
2974 struct lock_list src_entry;
2975
2976 bfs_init_root(&src_entry, src);
2977 /*
2978 * Special setup for check_redundant().
2979 *
2980 * To report redundant, we need to find a strong dependency path that
2981 * is equal to or stronger than <src> -> <target>. So if <src> is E,
2982 * we need to let __bfs() only search for a path starting at a -(E*)->,
2983 * we achieve this by setting the initial node's ->only_xr to true in
2984 * that case. And if <prev> is S, we set initial ->only_xr to false
2985 * because both -(S*)-> (equal) and -(E*)-> (stronger) are redundant.
2986 */
2987 src_entry.only_xr = src->read == 0;
2988
2989 debug_atomic_inc(nr_redundant_checks);
2990
2991 /*
2992 * Note: we skip local_lock() for redundant check, because as the
2993 * comment in usage_skip(), A -> local_lock() -> B and A -> B are not
2994 * the same.
2995 */
2996 ret = check_path(target, &src_entry, hlock_equal, usage_skip, &target_entry);
2997
2998 if (ret == BFS_RMATCH)
2999 debug_atomic_inc(nr_redundant);
3000
3001 return ret;
3002}
3003
3004#else
3005
3006static inline enum bfs_result
3007check_redundant(struct held_lock *src, struct held_lock *target)
3008{
3009 return BFS_RNOMATCH;
3010}
3011
3012#endif
3013
3014static void inc_chains(int irq_context)
3015{
3016 if (irq_context & LOCK_CHAIN_HARDIRQ_CONTEXT)
3017 nr_hardirq_chains++;
3018 else if (irq_context & LOCK_CHAIN_SOFTIRQ_CONTEXT)
3019 nr_softirq_chains++;
3020 else
3021 nr_process_chains++;
3022}
3023
3024static void dec_chains(int irq_context)
3025{
3026 if (irq_context & LOCK_CHAIN_HARDIRQ_CONTEXT)
3027 nr_hardirq_chains--;
3028 else if (irq_context & LOCK_CHAIN_SOFTIRQ_CONTEXT)
3029 nr_softirq_chains--;
3030 else
3031 nr_process_chains--;
3032}
3033
3034static void
3035print_deadlock_scenario(struct held_lock *nxt, struct held_lock *prv)
3036{
3037 struct lock_class *next = hlock_class(nxt);
3038 struct lock_class *prev = hlock_class(prv);
3039
3040 printk(" Possible unsafe locking scenario:\n\n");
3041 printk(" CPU0\n");
3042 printk(" ----\n");
3043 printk(" lock(");
3044 __print_lock_name(prv, prev);
3045 printk(KERN_CONT ");\n");
3046 printk(" lock(");
3047 __print_lock_name(nxt, next);
3048 printk(KERN_CONT ");\n");
3049 printk("\n *** DEADLOCK ***\n\n");
3050 printk(" May be due to missing lock nesting notation\n\n");
3051}
3052
3053static void
3054print_deadlock_bug(struct task_struct *curr, struct held_lock *prev,
3055 struct held_lock *next)
3056{
3057 struct lock_class *class = hlock_class(prev);
3058
3059 if (!debug_locks_off_graph_unlock() || debug_locks_silent)
3060 return;
3061
3062 nbcon_cpu_emergency_enter();
3063
3064 pr_warn("\n");
3065 pr_warn("============================================\n");
3066 pr_warn("WARNING: possible recursive locking detected\n");
3067 print_kernel_ident();
3068 pr_warn("--------------------------------------------\n");
3069 pr_warn("%s/%d is trying to acquire lock:\n",
3070 curr->comm, task_pid_nr(curr));
3071 print_lock(next);
3072 pr_warn("\nbut task is already holding lock:\n");
3073 print_lock(prev);
3074
3075 if (class->cmp_fn) {
3076 pr_warn("and the lock comparison function returns %i:\n",
3077 class->cmp_fn(prev->instance, next->instance));
3078 }
3079
3080 pr_warn("\nother info that might help us debug this:\n");
3081 print_deadlock_scenario(next, prev);
3082 lockdep_print_held_locks(curr);
3083
3084 pr_warn("\nstack backtrace:\n");
3085 dump_stack();
3086
3087 nbcon_cpu_emergency_exit();
3088}
3089
3090/*
3091 * Check whether we are holding such a class already.
3092 *
3093 * (Note that this has to be done separately, because the graph cannot
3094 * detect such classes of deadlocks.)
3095 *
3096 * Returns: 0 on deadlock detected, 1 on OK, 2 if another lock with the same
3097 * lock class is held but nest_lock is also held, i.e. we rely on the
3098 * nest_lock to avoid the deadlock.
3099 */
3100static int
3101check_deadlock(struct task_struct *curr, struct held_lock *next)
3102{
3103 struct lock_class *class;
3104 struct held_lock *prev;
3105 struct held_lock *nest = NULL;
3106 int i;
3107
3108 for (i = 0; i < curr->lockdep_depth; i++) {
3109 prev = curr->held_locks + i;
3110
3111 if (prev->instance == next->nest_lock)
3112 nest = prev;
3113
3114 if (hlock_class(prev) != hlock_class(next))
3115 continue;
3116
3117 /*
3118 * Allow read-after-read recursion of the same
3119 * lock class (i.e. read_lock(lock)+read_lock(lock)):
3120 */
3121 if ((next->read == 2) && prev->read)
3122 continue;
3123
3124 class = hlock_class(prev);
3125
3126 if (class->cmp_fn &&
3127 class->cmp_fn(prev->instance, next->instance) < 0)
3128 continue;
3129
3130 /*
3131 * We're holding the nest_lock, which serializes this lock's
3132 * nesting behaviour.
3133 */
3134 if (nest)
3135 return 2;
3136
3137 print_deadlock_bug(curr, prev, next);
3138 return 0;
3139 }
3140 return 1;
3141}
3142
3143/*
3144 * There was a chain-cache miss, and we are about to add a new dependency
3145 * to a previous lock. We validate the following rules:
3146 *
3147 * - would the adding of the <prev> -> <next> dependency create a
3148 * circular dependency in the graph? [== circular deadlock]
3149 *
3150 * - does the new prev->next dependency connect any hardirq-safe lock
3151 * (in the full backwards-subgraph starting at <prev>) with any
3152 * hardirq-unsafe lock (in the full forwards-subgraph starting at
3153 * <next>)? [== illegal lock inversion with hardirq contexts]
3154 *
3155 * - does the new prev->next dependency connect any softirq-safe lock
3156 * (in the full backwards-subgraph starting at <prev>) with any
3157 * softirq-unsafe lock (in the full forwards-subgraph starting at
3158 * <next>)? [== illegal lock inversion with softirq contexts]
3159 *
3160 * any of these scenarios could lead to a deadlock.
3161 *
3162 * Then if all the validations pass, we add the forwards and backwards
3163 * dependency.
3164 */
3165static int
3166check_prev_add(struct task_struct *curr, struct held_lock *prev,
3167 struct held_lock *next, u16 distance,
3168 struct lock_trace **const trace)
3169{
3170 struct lock_list *entry;
3171 enum bfs_result ret;
3172
3173 if (!hlock_class(prev)->key || !hlock_class(next)->key) {
3174 /*
3175 * The warning statements below may trigger a use-after-free
3176 * of the class name. It is better to trigger a use-after free
3177 * and to have the class name most of the time instead of not
3178 * having the class name available.
3179 */
3180 WARN_ONCE(!debug_locks_silent && !hlock_class(prev)->key,
3181 "Detected use-after-free of lock class %px/%s\n",
3182 hlock_class(prev),
3183 hlock_class(prev)->name);
3184 WARN_ONCE(!debug_locks_silent && !hlock_class(next)->key,
3185 "Detected use-after-free of lock class %px/%s\n",
3186 hlock_class(next),
3187 hlock_class(next)->name);
3188 return 2;
3189 }
3190
3191 if (prev->class_idx == next->class_idx) {
3192 struct lock_class *class = hlock_class(prev);
3193
3194 if (class->cmp_fn &&
3195 class->cmp_fn(prev->instance, next->instance) < 0)
3196 return 2;
3197 }
3198
3199 /*
3200 * Prove that the new <prev> -> <next> dependency would not
3201 * create a circular dependency in the graph. (We do this by
3202 * a breadth-first search into the graph starting at <next>,
3203 * and check whether we can reach <prev>.)
3204 *
3205 * The search is limited by the size of the circular queue (i.e.,
3206 * MAX_CIRCULAR_QUEUE_SIZE) which keeps track of a breadth of nodes
3207 * in the graph whose neighbours are to be checked.
3208 */
3209 ret = check_noncircular(next, prev, trace);
3210 if (unlikely(bfs_error(ret) || ret == BFS_RMATCH))
3211 return 0;
3212
3213 if (!check_irq_usage(curr, prev, next))
3214 return 0;
3215
3216 /*
3217 * Is the <prev> -> <next> dependency already present?
3218 *
3219 * (this may occur even though this is a new chain: consider
3220 * e.g. the L1 -> L2 -> L3 -> L4 and the L5 -> L1 -> L2 -> L3
3221 * chains - the second one will be new, but L1 already has
3222 * L2 added to its dependency list, due to the first chain.)
3223 */
3224 list_for_each_entry(entry, &hlock_class(prev)->locks_after, entry) {
3225 if (entry->class == hlock_class(next)) {
3226 if (distance == 1)
3227 entry->distance = 1;
3228 entry->dep |= calc_dep(prev, next);
3229
3230 /*
3231 * Also, update the reverse dependency in @next's
3232 * ->locks_before list.
3233 *
3234 * Here we reuse @entry as the cursor, which is fine
3235 * because we won't go to the next iteration of the
3236 * outer loop:
3237 *
3238 * For normal cases, we return in the inner loop.
3239 *
3240 * If we fail to return, we have inconsistency, i.e.
3241 * <prev>::locks_after contains <next> while
3242 * <next>::locks_before doesn't contain <prev>. In
3243 * that case, we return after the inner and indicate
3244 * something is wrong.
3245 */
3246 list_for_each_entry(entry, &hlock_class(next)->locks_before, entry) {
3247 if (entry->class == hlock_class(prev)) {
3248 if (distance == 1)
3249 entry->distance = 1;
3250 entry->dep |= calc_depb(prev, next);
3251 return 1;
3252 }
3253 }
3254
3255 /* <prev> is not found in <next>::locks_before */
3256 return 0;
3257 }
3258 }
3259
3260 /*
3261 * Is the <prev> -> <next> link redundant?
3262 */
3263 ret = check_redundant(prev, next);
3264 if (bfs_error(ret))
3265 return 0;
3266 else if (ret == BFS_RMATCH)
3267 return 2;
3268
3269 if (!*trace) {
3270 *trace = save_trace();
3271 if (!*trace)
3272 return 0;
3273 }
3274
3275 /*
3276 * Ok, all validations passed, add the new lock
3277 * to the previous lock's dependency list:
3278 */
3279 ret = add_lock_to_list(hlock_class(next), hlock_class(prev),
3280 &hlock_class(prev)->locks_after, distance,
3281 calc_dep(prev, next), *trace);
3282
3283 if (!ret)
3284 return 0;
3285
3286 ret = add_lock_to_list(hlock_class(prev), hlock_class(next),
3287 &hlock_class(next)->locks_before, distance,
3288 calc_depb(prev, next), *trace);
3289 if (!ret)
3290 return 0;
3291
3292 return 2;
3293}
3294
3295/*
3296 * Add the dependency to all directly-previous locks that are 'relevant'.
3297 * The ones that are relevant are (in increasing distance from curr):
3298 * all consecutive trylock entries and the final non-trylock entry - or
3299 * the end of this context's lock-chain - whichever comes first.
3300 */
3301static int
3302check_prevs_add(struct task_struct *curr, struct held_lock *next)
3303{
3304 struct lock_trace *trace = NULL;
3305 int depth = curr->lockdep_depth;
3306 struct held_lock *hlock;
3307
3308 /*
3309 * Debugging checks.
3310 *
3311 * Depth must not be zero for a non-head lock:
3312 */
3313 if (!depth)
3314 goto out_bug;
3315 /*
3316 * At least two relevant locks must exist for this
3317 * to be a head:
3318 */
3319 if (curr->held_locks[depth].irq_context !=
3320 curr->held_locks[depth-1].irq_context)
3321 goto out_bug;
3322
3323 for (;;) {
3324 u16 distance = curr->lockdep_depth - depth + 1;
3325 hlock = curr->held_locks + depth - 1;
3326
3327 if (hlock->check) {
3328 int ret = check_prev_add(curr, hlock, next, distance, &trace);
3329 if (!ret)
3330 return 0;
3331
3332 /*
3333 * Stop after the first non-trylock entry,
3334 * as non-trylock entries have added their
3335 * own direct dependencies already, so this
3336 * lock is connected to them indirectly:
3337 */
3338 if (!hlock->trylock)
3339 break;
3340 }
3341
3342 depth--;
3343 /*
3344 * End of lock-stack?
3345 */
3346 if (!depth)
3347 break;
3348 /*
3349 * Stop the search if we cross into another context:
3350 */
3351 if (curr->held_locks[depth].irq_context !=
3352 curr->held_locks[depth-1].irq_context)
3353 break;
3354 }
3355 return 1;
3356out_bug:
3357 if (!debug_locks_off_graph_unlock())
3358 return 0;
3359
3360 /*
3361 * Clearly we all shouldn't be here, but since we made it we
3362 * can reliable say we messed up our state. See the above two
3363 * gotos for reasons why we could possibly end up here.
3364 */
3365 WARN_ON(1);
3366
3367 return 0;
3368}
3369
3370struct lock_chain lock_chains[MAX_LOCKDEP_CHAINS];
3371static DECLARE_BITMAP(lock_chains_in_use, MAX_LOCKDEP_CHAINS);
3372static u16 chain_hlocks[MAX_LOCKDEP_CHAIN_HLOCKS];
3373unsigned long nr_zapped_lock_chains;
3374unsigned int nr_free_chain_hlocks; /* Free chain_hlocks in buckets */
3375unsigned int nr_lost_chain_hlocks; /* Lost chain_hlocks */
3376unsigned int nr_large_chain_blocks; /* size > MAX_CHAIN_BUCKETS */
3377
3378/*
3379 * The first 2 chain_hlocks entries in the chain block in the bucket
3380 * list contains the following meta data:
3381 *
3382 * entry[0]:
3383 * Bit 15 - always set to 1 (it is not a class index)
3384 * Bits 0-14 - upper 15 bits of the next block index
3385 * entry[1] - lower 16 bits of next block index
3386 *
3387 * A next block index of all 1 bits means it is the end of the list.
3388 *
3389 * On the unsized bucket (bucket-0), the 3rd and 4th entries contain
3390 * the chain block size:
3391 *
3392 * entry[2] - upper 16 bits of the chain block size
3393 * entry[3] - lower 16 bits of the chain block size
3394 */
3395#define MAX_CHAIN_BUCKETS 16
3396#define CHAIN_BLK_FLAG (1U << 15)
3397#define CHAIN_BLK_LIST_END 0xFFFFU
3398
3399static int chain_block_buckets[MAX_CHAIN_BUCKETS];
3400
3401static inline int size_to_bucket(int size)
3402{
3403 if (size > MAX_CHAIN_BUCKETS)
3404 return 0;
3405
3406 return size - 1;
3407}
3408
3409/*
3410 * Iterate all the chain blocks in a bucket.
3411 */
3412#define for_each_chain_block(bucket, prev, curr) \
3413 for ((prev) = -1, (curr) = chain_block_buckets[bucket]; \
3414 (curr) >= 0; \
3415 (prev) = (curr), (curr) = chain_block_next(curr))
3416
3417/*
3418 * next block or -1
3419 */
3420static inline int chain_block_next(int offset)
3421{
3422 int next = chain_hlocks[offset];
3423
3424 WARN_ON_ONCE(!(next & CHAIN_BLK_FLAG));
3425
3426 if (next == CHAIN_BLK_LIST_END)
3427 return -1;
3428
3429 next &= ~CHAIN_BLK_FLAG;
3430 next <<= 16;
3431 next |= chain_hlocks[offset + 1];
3432
3433 return next;
3434}
3435
3436/*
3437 * bucket-0 only
3438 */
3439static inline int chain_block_size(int offset)
3440{
3441 return (chain_hlocks[offset + 2] << 16) | chain_hlocks[offset + 3];
3442}
3443
3444static inline void init_chain_block(int offset, int next, int bucket, int size)
3445{
3446 chain_hlocks[offset] = (next >> 16) | CHAIN_BLK_FLAG;
3447 chain_hlocks[offset + 1] = (u16)next;
3448
3449 if (size && !bucket) {
3450 chain_hlocks[offset + 2] = size >> 16;
3451 chain_hlocks[offset + 3] = (u16)size;
3452 }
3453}
3454
3455static inline void add_chain_block(int offset, int size)
3456{
3457 int bucket = size_to_bucket(size);
3458 int next = chain_block_buckets[bucket];
3459 int prev, curr;
3460
3461 if (unlikely(size < 2)) {
3462 /*
3463 * We can't store single entries on the freelist. Leak them.
3464 *
3465 * One possible way out would be to uniquely mark them, other
3466 * than with CHAIN_BLK_FLAG, such that we can recover them when
3467 * the block before it is re-added.
3468 */
3469 if (size)
3470 nr_lost_chain_hlocks++;
3471 return;
3472 }
3473
3474 nr_free_chain_hlocks += size;
3475 if (!bucket) {
3476 nr_large_chain_blocks++;
3477
3478 /*
3479 * Variable sized, sort large to small.
3480 */
3481 for_each_chain_block(0, prev, curr) {
3482 if (size >= chain_block_size(curr))
3483 break;
3484 }
3485 init_chain_block(offset, curr, 0, size);
3486 if (prev < 0)
3487 chain_block_buckets[0] = offset;
3488 else
3489 init_chain_block(prev, offset, 0, 0);
3490 return;
3491 }
3492 /*
3493 * Fixed size, add to head.
3494 */
3495 init_chain_block(offset, next, bucket, size);
3496 chain_block_buckets[bucket] = offset;
3497}
3498
3499/*
3500 * Only the first block in the list can be deleted.
3501 *
3502 * For the variable size bucket[0], the first block (the largest one) is
3503 * returned, broken up and put back into the pool. So if a chain block of
3504 * length > MAX_CHAIN_BUCKETS is ever used and zapped, it will just be
3505 * queued up after the primordial chain block and never be used until the
3506 * hlock entries in the primordial chain block is almost used up. That
3507 * causes fragmentation and reduce allocation efficiency. That can be
3508 * monitored by looking at the "large chain blocks" number in lockdep_stats.
3509 */
3510static inline void del_chain_block(int bucket, int size, int next)
3511{
3512 nr_free_chain_hlocks -= size;
3513 chain_block_buckets[bucket] = next;
3514
3515 if (!bucket)
3516 nr_large_chain_blocks--;
3517}
3518
3519static void init_chain_block_buckets(void)
3520{
3521 int i;
3522
3523 for (i = 0; i < MAX_CHAIN_BUCKETS; i++)
3524 chain_block_buckets[i] = -1;
3525
3526 add_chain_block(0, ARRAY_SIZE(chain_hlocks));
3527}
3528
3529/*
3530 * Return offset of a chain block of the right size or -1 if not found.
3531 *
3532 * Fairly simple worst-fit allocator with the addition of a number of size
3533 * specific free lists.
3534 */
3535static int alloc_chain_hlocks(int req)
3536{
3537 int bucket, curr, size;
3538
3539 /*
3540 * We rely on the MSB to act as an escape bit to denote freelist
3541 * pointers. Make sure this bit isn't set in 'normal' class_idx usage.
3542 */
3543 BUILD_BUG_ON((MAX_LOCKDEP_KEYS-1) & CHAIN_BLK_FLAG);
3544
3545 init_data_structures_once();
3546
3547 if (nr_free_chain_hlocks < req)
3548 return -1;
3549
3550 /*
3551 * We require a minimum of 2 (u16) entries to encode a freelist
3552 * 'pointer'.
3553 */
3554 req = max(req, 2);
3555 bucket = size_to_bucket(req);
3556 curr = chain_block_buckets[bucket];
3557
3558 if (bucket) {
3559 if (curr >= 0) {
3560 del_chain_block(bucket, req, chain_block_next(curr));
3561 return curr;
3562 }
3563 /* Try bucket 0 */
3564 curr = chain_block_buckets[0];
3565 }
3566
3567 /*
3568 * The variable sized freelist is sorted by size; the first entry is
3569 * the largest. Use it if it fits.
3570 */
3571 if (curr >= 0) {
3572 size = chain_block_size(curr);
3573 if (likely(size >= req)) {
3574 del_chain_block(0, size, chain_block_next(curr));
3575 if (size > req)
3576 add_chain_block(curr + req, size - req);
3577 return curr;
3578 }
3579 }
3580
3581 /*
3582 * Last resort, split a block in a larger sized bucket.
3583 */
3584 for (size = MAX_CHAIN_BUCKETS; size > req; size--) {
3585 bucket = size_to_bucket(size);
3586 curr = chain_block_buckets[bucket];
3587 if (curr < 0)
3588 continue;
3589
3590 del_chain_block(bucket, size, chain_block_next(curr));
3591 add_chain_block(curr + req, size - req);
3592 return curr;
3593 }
3594
3595 return -1;
3596}
3597
3598static inline void free_chain_hlocks(int base, int size)
3599{
3600 add_chain_block(base, max(size, 2));
3601}
3602
3603struct lock_class *lock_chain_get_class(struct lock_chain *chain, int i)
3604{
3605 u16 chain_hlock = chain_hlocks[chain->base + i];
3606 unsigned int class_idx = chain_hlock_class_idx(chain_hlock);
3607
3608 return lock_classes + class_idx;
3609}
3610
3611/*
3612 * Returns the index of the first held_lock of the current chain
3613 */
3614static inline int get_first_held_lock(struct task_struct *curr,
3615 struct held_lock *hlock)
3616{
3617 int i;
3618 struct held_lock *hlock_curr;
3619
3620 for (i = curr->lockdep_depth - 1; i >= 0; i--) {
3621 hlock_curr = curr->held_locks + i;
3622 if (hlock_curr->irq_context != hlock->irq_context)
3623 break;
3624
3625 }
3626
3627 return ++i;
3628}
3629
3630#ifdef CONFIG_DEBUG_LOCKDEP
3631/*
3632 * Returns the next chain_key iteration
3633 */
3634static u64 print_chain_key_iteration(u16 hlock_id, u64 chain_key)
3635{
3636 u64 new_chain_key = iterate_chain_key(chain_key, hlock_id);
3637
3638 printk(" hlock_id:%d -> chain_key:%016Lx",
3639 (unsigned int)hlock_id,
3640 (unsigned long long)new_chain_key);
3641 return new_chain_key;
3642}
3643
3644static void
3645print_chain_keys_held_locks(struct task_struct *curr, struct held_lock *hlock_next)
3646{
3647 struct held_lock *hlock;
3648 u64 chain_key = INITIAL_CHAIN_KEY;
3649 int depth = curr->lockdep_depth;
3650 int i = get_first_held_lock(curr, hlock_next);
3651
3652 printk("depth: %u (irq_context %u)\n", depth - i + 1,
3653 hlock_next->irq_context);
3654 for (; i < depth; i++) {
3655 hlock = curr->held_locks + i;
3656 chain_key = print_chain_key_iteration(hlock_id(hlock), chain_key);
3657
3658 print_lock(hlock);
3659 }
3660
3661 print_chain_key_iteration(hlock_id(hlock_next), chain_key);
3662 print_lock(hlock_next);
3663}
3664
3665static void print_chain_keys_chain(struct lock_chain *chain)
3666{
3667 int i;
3668 u64 chain_key = INITIAL_CHAIN_KEY;
3669 u16 hlock_id;
3670
3671 printk("depth: %u\n", chain->depth);
3672 for (i = 0; i < chain->depth; i++) {
3673 hlock_id = chain_hlocks[chain->base + i];
3674 chain_key = print_chain_key_iteration(hlock_id, chain_key);
3675
3676 print_lock_name(NULL, lock_classes + chain_hlock_class_idx(hlock_id));
3677 printk("\n");
3678 }
3679}
3680
3681static void print_collision(struct task_struct *curr,
3682 struct held_lock *hlock_next,
3683 struct lock_chain *chain)
3684{
3685 nbcon_cpu_emergency_enter();
3686
3687 pr_warn("\n");
3688 pr_warn("============================\n");
3689 pr_warn("WARNING: chain_key collision\n");
3690 print_kernel_ident();
3691 pr_warn("----------------------------\n");
3692 pr_warn("%s/%d: ", current->comm, task_pid_nr(current));
3693 pr_warn("Hash chain already cached but the contents don't match!\n");
3694
3695 pr_warn("Held locks:");
3696 print_chain_keys_held_locks(curr, hlock_next);
3697
3698 pr_warn("Locks in cached chain:");
3699 print_chain_keys_chain(chain);
3700
3701 pr_warn("\nstack backtrace:\n");
3702 dump_stack();
3703
3704 nbcon_cpu_emergency_exit();
3705}
3706#endif
3707
3708/*
3709 * Checks whether the chain and the current held locks are consistent
3710 * in depth and also in content. If they are not it most likely means
3711 * that there was a collision during the calculation of the chain_key.
3712 * Returns: 0 not passed, 1 passed
3713 */
3714static int check_no_collision(struct task_struct *curr,
3715 struct held_lock *hlock,
3716 struct lock_chain *chain)
3717{
3718#ifdef CONFIG_DEBUG_LOCKDEP
3719 int i, j, id;
3720
3721 i = get_first_held_lock(curr, hlock);
3722
3723 if (DEBUG_LOCKS_WARN_ON(chain->depth != curr->lockdep_depth - (i - 1))) {
3724 print_collision(curr, hlock, chain);
3725 return 0;
3726 }
3727
3728 for (j = 0; j < chain->depth - 1; j++, i++) {
3729 id = hlock_id(&curr->held_locks[i]);
3730
3731 if (DEBUG_LOCKS_WARN_ON(chain_hlocks[chain->base + j] != id)) {
3732 print_collision(curr, hlock, chain);
3733 return 0;
3734 }
3735 }
3736#endif
3737 return 1;
3738}
3739
3740/*
3741 * Given an index that is >= -1, return the index of the next lock chain.
3742 * Return -2 if there is no next lock chain.
3743 */
3744long lockdep_next_lockchain(long i)
3745{
3746 i = find_next_bit(lock_chains_in_use, ARRAY_SIZE(lock_chains), i + 1);
3747 return i < ARRAY_SIZE(lock_chains) ? i : -2;
3748}
3749
3750unsigned long lock_chain_count(void)
3751{
3752 return bitmap_weight(lock_chains_in_use, ARRAY_SIZE(lock_chains));
3753}
3754
3755/* Must be called with the graph lock held. */
3756static struct lock_chain *alloc_lock_chain(void)
3757{
3758 int idx = find_first_zero_bit(lock_chains_in_use,
3759 ARRAY_SIZE(lock_chains));
3760
3761 if (unlikely(idx >= ARRAY_SIZE(lock_chains)))
3762 return NULL;
3763 __set_bit(idx, lock_chains_in_use);
3764 return lock_chains + idx;
3765}
3766
3767/*
3768 * Adds a dependency chain into chain hashtable. And must be called with
3769 * graph_lock held.
3770 *
3771 * Return 0 if fail, and graph_lock is released.
3772 * Return 1 if succeed, with graph_lock held.
3773 */
3774static inline int add_chain_cache(struct task_struct *curr,
3775 struct held_lock *hlock,
3776 u64 chain_key)
3777{
3778 struct hlist_head *hash_head = chainhashentry(chain_key);
3779 struct lock_chain *chain;
3780 int i, j;
3781
3782 /*
3783 * The caller must hold the graph lock, ensure we've got IRQs
3784 * disabled to make this an IRQ-safe lock.. for recursion reasons
3785 * lockdep won't complain about its own locking errors.
3786 */
3787 if (lockdep_assert_locked())
3788 return 0;
3789
3790 chain = alloc_lock_chain();
3791 if (!chain) {
3792 if (!debug_locks_off_graph_unlock())
3793 return 0;
3794
3795 nbcon_cpu_emergency_enter();
3796 print_lockdep_off("BUG: MAX_LOCKDEP_CHAINS too low!");
3797 dump_stack();
3798 nbcon_cpu_emergency_exit();
3799 return 0;
3800 }
3801 chain->chain_key = chain_key;
3802 chain->irq_context = hlock->irq_context;
3803 i = get_first_held_lock(curr, hlock);
3804 chain->depth = curr->lockdep_depth + 1 - i;
3805
3806 BUILD_BUG_ON((1UL << 24) <= ARRAY_SIZE(chain_hlocks));
3807 BUILD_BUG_ON((1UL << 6) <= ARRAY_SIZE(curr->held_locks));
3808 BUILD_BUG_ON((1UL << 8*sizeof(chain_hlocks[0])) <= ARRAY_SIZE(lock_classes));
3809
3810 j = alloc_chain_hlocks(chain->depth);
3811 if (j < 0) {
3812 if (!debug_locks_off_graph_unlock())
3813 return 0;
3814
3815 nbcon_cpu_emergency_enter();
3816 print_lockdep_off("BUG: MAX_LOCKDEP_CHAIN_HLOCKS too low!");
3817 dump_stack();
3818 nbcon_cpu_emergency_exit();
3819 return 0;
3820 }
3821
3822 chain->base = j;
3823 for (j = 0; j < chain->depth - 1; j++, i++) {
3824 int lock_id = hlock_id(curr->held_locks + i);
3825
3826 chain_hlocks[chain->base + j] = lock_id;
3827 }
3828 chain_hlocks[chain->base + j] = hlock_id(hlock);
3829 hlist_add_head_rcu(&chain->entry, hash_head);
3830 debug_atomic_inc(chain_lookup_misses);
3831 inc_chains(chain->irq_context);
3832
3833 return 1;
3834}
3835
3836/*
3837 * Look up a dependency chain. Must be called with either the graph lock or
3838 * the RCU read lock held.
3839 */
3840static inline struct lock_chain *lookup_chain_cache(u64 chain_key)
3841{
3842 struct hlist_head *hash_head = chainhashentry(chain_key);
3843 struct lock_chain *chain;
3844
3845 hlist_for_each_entry_rcu(chain, hash_head, entry) {
3846 if (READ_ONCE(chain->chain_key) == chain_key) {
3847 debug_atomic_inc(chain_lookup_hits);
3848 return chain;
3849 }
3850 }
3851 return NULL;
3852}
3853
3854/*
3855 * If the key is not present yet in dependency chain cache then
3856 * add it and return 1 - in this case the new dependency chain is
3857 * validated. If the key is already hashed, return 0.
3858 * (On return with 1 graph_lock is held.)
3859 */
3860static inline int lookup_chain_cache_add(struct task_struct *curr,
3861 struct held_lock *hlock,
3862 u64 chain_key)
3863{
3864 struct lock_class *class = hlock_class(hlock);
3865 struct lock_chain *chain = lookup_chain_cache(chain_key);
3866
3867 if (chain) {
3868cache_hit:
3869 if (!check_no_collision(curr, hlock, chain))
3870 return 0;
3871
3872 if (very_verbose(class)) {
3873 printk("\nhash chain already cached, key: "
3874 "%016Lx tail class: [%px] %s\n",
3875 (unsigned long long)chain_key,
3876 class->key, class->name);
3877 }
3878
3879 return 0;
3880 }
3881
3882 if (very_verbose(class)) {
3883 printk("\nnew hash chain, key: %016Lx tail class: [%px] %s\n",
3884 (unsigned long long)chain_key, class->key, class->name);
3885 }
3886
3887 if (!graph_lock())
3888 return 0;
3889
3890 /*
3891 * We have to walk the chain again locked - to avoid duplicates:
3892 */
3893 chain = lookup_chain_cache(chain_key);
3894 if (chain) {
3895 graph_unlock();
3896 goto cache_hit;
3897 }
3898
3899 if (!add_chain_cache(curr, hlock, chain_key))
3900 return 0;
3901
3902 return 1;
3903}
3904
3905static int validate_chain(struct task_struct *curr,
3906 struct held_lock *hlock,
3907 int chain_head, u64 chain_key)
3908{
3909 /*
3910 * Trylock needs to maintain the stack of held locks, but it
3911 * does not add new dependencies, because trylock can be done
3912 * in any order.
3913 *
3914 * We look up the chain_key and do the O(N^2) check and update of
3915 * the dependencies only if this is a new dependency chain.
3916 * (If lookup_chain_cache_add() return with 1 it acquires
3917 * graph_lock for us)
3918 */
3919 if (!hlock->trylock && hlock->check &&
3920 lookup_chain_cache_add(curr, hlock, chain_key)) {
3921 /*
3922 * Check whether last held lock:
3923 *
3924 * - is irq-safe, if this lock is irq-unsafe
3925 * - is softirq-safe, if this lock is hardirq-unsafe
3926 *
3927 * And check whether the new lock's dependency graph
3928 * could lead back to the previous lock:
3929 *
3930 * - within the current held-lock stack
3931 * - across our accumulated lock dependency records
3932 *
3933 * any of these scenarios could lead to a deadlock.
3934 */
3935 /*
3936 * The simple case: does the current hold the same lock
3937 * already?
3938 */
3939 int ret = check_deadlock(curr, hlock);
3940
3941 if (!ret)
3942 return 0;
3943 /*
3944 * Add dependency only if this lock is not the head
3945 * of the chain, and if the new lock introduces no more
3946 * lock dependency (because we already hold a lock with the
3947 * same lock class) nor deadlock (because the nest_lock
3948 * serializes nesting locks), see the comments for
3949 * check_deadlock().
3950 */
3951 if (!chain_head && ret != 2) {
3952 if (!check_prevs_add(curr, hlock))
3953 return 0;
3954 }
3955
3956 graph_unlock();
3957 } else {
3958 /* after lookup_chain_cache_add(): */
3959 if (unlikely(!debug_locks))
3960 return 0;
3961 }
3962
3963 return 1;
3964}
3965#else
3966static inline int validate_chain(struct task_struct *curr,
3967 struct held_lock *hlock,
3968 int chain_head, u64 chain_key)
3969{
3970 return 1;
3971}
3972
3973static void init_chain_block_buckets(void) { }
3974#endif /* CONFIG_PROVE_LOCKING */
3975
3976/*
3977 * We are building curr_chain_key incrementally, so double-check
3978 * it from scratch, to make sure that it's done correctly:
3979 */
3980static void check_chain_key(struct task_struct *curr)
3981{
3982#ifdef CONFIG_DEBUG_LOCKDEP
3983 struct held_lock *hlock, *prev_hlock = NULL;
3984 unsigned int i;
3985 u64 chain_key = INITIAL_CHAIN_KEY;
3986
3987 for (i = 0; i < curr->lockdep_depth; i++) {
3988 hlock = curr->held_locks + i;
3989 if (chain_key != hlock->prev_chain_key) {
3990 debug_locks_off();
3991 /*
3992 * We got mighty confused, our chain keys don't match
3993 * with what we expect, someone trample on our task state?
3994 */
3995 WARN(1, "hm#1, depth: %u [%u], %016Lx != %016Lx\n",
3996 curr->lockdep_depth, i,
3997 (unsigned long long)chain_key,
3998 (unsigned long long)hlock->prev_chain_key);
3999 return;
4000 }
4001
4002 /*
4003 * hlock->class_idx can't go beyond MAX_LOCKDEP_KEYS, but is
4004 * it registered lock class index?
4005 */
4006 if (DEBUG_LOCKS_WARN_ON(!test_bit(hlock->class_idx, lock_classes_in_use)))
4007 return;
4008
4009 if (prev_hlock && (prev_hlock->irq_context !=
4010 hlock->irq_context))
4011 chain_key = INITIAL_CHAIN_KEY;
4012 chain_key = iterate_chain_key(chain_key, hlock_id(hlock));
4013 prev_hlock = hlock;
4014 }
4015 if (chain_key != curr->curr_chain_key) {
4016 debug_locks_off();
4017 /*
4018 * More smoking hash instead of calculating it, damn see these
4019 * numbers float.. I bet that a pink elephant stepped on my memory.
4020 */
4021 WARN(1, "hm#2, depth: %u [%u], %016Lx != %016Lx\n",
4022 curr->lockdep_depth, i,
4023 (unsigned long long)chain_key,
4024 (unsigned long long)curr->curr_chain_key);
4025 }
4026#endif
4027}
4028
4029#ifdef CONFIG_PROVE_LOCKING
4030static int mark_lock(struct task_struct *curr, struct held_lock *this,
4031 enum lock_usage_bit new_bit);
4032
4033static void print_usage_bug_scenario(struct held_lock *lock)
4034{
4035 struct lock_class *class = hlock_class(lock);
4036
4037 printk(" Possible unsafe locking scenario:\n\n");
4038 printk(" CPU0\n");
4039 printk(" ----\n");
4040 printk(" lock(");
4041 __print_lock_name(lock, class);
4042 printk(KERN_CONT ");\n");
4043 printk(" <Interrupt>\n");
4044 printk(" lock(");
4045 __print_lock_name(lock, class);
4046 printk(KERN_CONT ");\n");
4047 printk("\n *** DEADLOCK ***\n\n");
4048}
4049
4050static void
4051print_usage_bug(struct task_struct *curr, struct held_lock *this,
4052 enum lock_usage_bit prev_bit, enum lock_usage_bit new_bit)
4053{
4054 if (!debug_locks_off() || debug_locks_silent)
4055 return;
4056
4057 nbcon_cpu_emergency_enter();
4058
4059 pr_warn("\n");
4060 pr_warn("================================\n");
4061 pr_warn("WARNING: inconsistent lock state\n");
4062 print_kernel_ident();
4063 pr_warn("--------------------------------\n");
4064
4065 pr_warn("inconsistent {%s} -> {%s} usage.\n",
4066 usage_str[prev_bit], usage_str[new_bit]);
4067
4068 pr_warn("%s/%d [HC%u[%lu]:SC%u[%lu]:HE%u:SE%u] takes:\n",
4069 curr->comm, task_pid_nr(curr),
4070 lockdep_hardirq_context(), hardirq_count() >> HARDIRQ_SHIFT,
4071 lockdep_softirq_context(curr), softirq_count() >> SOFTIRQ_SHIFT,
4072 lockdep_hardirqs_enabled(),
4073 lockdep_softirqs_enabled(curr));
4074 print_lock(this);
4075
4076 pr_warn("{%s} state was registered at:\n", usage_str[prev_bit]);
4077 print_lock_trace(hlock_class(this)->usage_traces[prev_bit], 1);
4078
4079 print_irqtrace_events(curr);
4080 pr_warn("\nother info that might help us debug this:\n");
4081 print_usage_bug_scenario(this);
4082
4083 lockdep_print_held_locks(curr);
4084
4085 pr_warn("\nstack backtrace:\n");
4086 dump_stack();
4087
4088 nbcon_cpu_emergency_exit();
4089}
4090
4091/*
4092 * Print out an error if an invalid bit is set:
4093 */
4094static inline int
4095valid_state(struct task_struct *curr, struct held_lock *this,
4096 enum lock_usage_bit new_bit, enum lock_usage_bit bad_bit)
4097{
4098 if (unlikely(hlock_class(this)->usage_mask & (1 << bad_bit))) {
4099 graph_unlock();
4100 print_usage_bug(curr, this, bad_bit, new_bit);
4101 return 0;
4102 }
4103 return 1;
4104}
4105
4106
4107/*
4108 * print irq inversion bug:
4109 */
4110static void
4111print_irq_inversion_bug(struct task_struct *curr,
4112 struct lock_list *root, struct lock_list *other,
4113 struct held_lock *this, int forwards,
4114 const char *irqclass)
4115{
4116 struct lock_list *entry = other;
4117 struct lock_list *middle = NULL;
4118 int depth;
4119
4120 if (!debug_locks_off_graph_unlock() || debug_locks_silent)
4121 return;
4122
4123 nbcon_cpu_emergency_enter();
4124
4125 pr_warn("\n");
4126 pr_warn("========================================================\n");
4127 pr_warn("WARNING: possible irq lock inversion dependency detected\n");
4128 print_kernel_ident();
4129 pr_warn("--------------------------------------------------------\n");
4130 pr_warn("%s/%d just changed the state of lock:\n",
4131 curr->comm, task_pid_nr(curr));
4132 print_lock(this);
4133 if (forwards)
4134 pr_warn("but this lock took another, %s-unsafe lock in the past:\n", irqclass);
4135 else
4136 pr_warn("but this lock was taken by another, %s-safe lock in the past:\n", irqclass);
4137 print_lock_name(NULL, other->class);
4138 pr_warn("\n\nand interrupts could create inverse lock ordering between them.\n\n");
4139
4140 pr_warn("\nother info that might help us debug this:\n");
4141
4142 /* Find a middle lock (if one exists) */
4143 depth = get_lock_depth(other);
4144 do {
4145 if (depth == 0 && (entry != root)) {
4146 pr_warn("lockdep:%s bad path found in chain graph\n", __func__);
4147 break;
4148 }
4149 middle = entry;
4150 entry = get_lock_parent(entry);
4151 depth--;
4152 } while (entry && entry != root && (depth >= 0));
4153 if (forwards)
4154 print_irq_lock_scenario(root, other,
4155 middle ? middle->class : root->class, other->class);
4156 else
4157 print_irq_lock_scenario(other, root,
4158 middle ? middle->class : other->class, root->class);
4159
4160 lockdep_print_held_locks(curr);
4161
4162 pr_warn("\nthe shortest dependencies between 2nd lock and 1st lock:\n");
4163 root->trace = save_trace();
4164 if (!root->trace)
4165 goto out;
4166 print_shortest_lock_dependencies(other, root);
4167
4168 pr_warn("\nstack backtrace:\n");
4169 dump_stack();
4170out:
4171 nbcon_cpu_emergency_exit();
4172}
4173
4174/*
4175 * Prove that in the forwards-direction subgraph starting at <this>
4176 * there is no lock matching <mask>:
4177 */
4178static int
4179check_usage_forwards(struct task_struct *curr, struct held_lock *this,
4180 enum lock_usage_bit bit)
4181{
4182 enum bfs_result ret;
4183 struct lock_list root;
4184 struct lock_list *target_entry;
4185 enum lock_usage_bit read_bit = bit + LOCK_USAGE_READ_MASK;
4186 unsigned usage_mask = lock_flag(bit) | lock_flag(read_bit);
4187
4188 bfs_init_root(&root, this);
4189 ret = find_usage_forwards(&root, usage_mask, &target_entry);
4190 if (bfs_error(ret)) {
4191 print_bfs_bug(ret);
4192 return 0;
4193 }
4194 if (ret == BFS_RNOMATCH)
4195 return 1;
4196
4197 /* Check whether write or read usage is the match */
4198 if (target_entry->class->usage_mask & lock_flag(bit)) {
4199 print_irq_inversion_bug(curr, &root, target_entry,
4200 this, 1, state_name(bit));
4201 } else {
4202 print_irq_inversion_bug(curr, &root, target_entry,
4203 this, 1, state_name(read_bit));
4204 }
4205
4206 return 0;
4207}
4208
4209/*
4210 * Prove that in the backwards-direction subgraph starting at <this>
4211 * there is no lock matching <mask>:
4212 */
4213static int
4214check_usage_backwards(struct task_struct *curr, struct held_lock *this,
4215 enum lock_usage_bit bit)
4216{
4217 enum bfs_result ret;
4218 struct lock_list root;
4219 struct lock_list *target_entry;
4220 enum lock_usage_bit read_bit = bit + LOCK_USAGE_READ_MASK;
4221 unsigned usage_mask = lock_flag(bit) | lock_flag(read_bit);
4222
4223 bfs_init_rootb(&root, this);
4224 ret = find_usage_backwards(&root, usage_mask, &target_entry);
4225 if (bfs_error(ret)) {
4226 print_bfs_bug(ret);
4227 return 0;
4228 }
4229 if (ret == BFS_RNOMATCH)
4230 return 1;
4231
4232 /* Check whether write or read usage is the match */
4233 if (target_entry->class->usage_mask & lock_flag(bit)) {
4234 print_irq_inversion_bug(curr, &root, target_entry,
4235 this, 0, state_name(bit));
4236 } else {
4237 print_irq_inversion_bug(curr, &root, target_entry,
4238 this, 0, state_name(read_bit));
4239 }
4240
4241 return 0;
4242}
4243
4244void print_irqtrace_events(struct task_struct *curr)
4245{
4246 const struct irqtrace_events *trace = &curr->irqtrace;
4247
4248 nbcon_cpu_emergency_enter();
4249
4250 printk("irq event stamp: %u\n", trace->irq_events);
4251 printk("hardirqs last enabled at (%u): [<%px>] %pS\n",
4252 trace->hardirq_enable_event, (void *)trace->hardirq_enable_ip,
4253 (void *)trace->hardirq_enable_ip);
4254 printk("hardirqs last disabled at (%u): [<%px>] %pS\n",
4255 trace->hardirq_disable_event, (void *)trace->hardirq_disable_ip,
4256 (void *)trace->hardirq_disable_ip);
4257 printk("softirqs last enabled at (%u): [<%px>] %pS\n",
4258 trace->softirq_enable_event, (void *)trace->softirq_enable_ip,
4259 (void *)trace->softirq_enable_ip);
4260 printk("softirqs last disabled at (%u): [<%px>] %pS\n",
4261 trace->softirq_disable_event, (void *)trace->softirq_disable_ip,
4262 (void *)trace->softirq_disable_ip);
4263
4264 nbcon_cpu_emergency_exit();
4265}
4266
4267static int HARDIRQ_verbose(struct lock_class *class)
4268{
4269#if HARDIRQ_VERBOSE
4270 return class_filter(class);
4271#endif
4272 return 0;
4273}
4274
4275static int SOFTIRQ_verbose(struct lock_class *class)
4276{
4277#if SOFTIRQ_VERBOSE
4278 return class_filter(class);
4279#endif
4280 return 0;
4281}
4282
4283static int (*state_verbose_f[])(struct lock_class *class) = {
4284#define LOCKDEP_STATE(__STATE) \
4285 __STATE##_verbose,
4286#include "lockdep_states.h"
4287#undef LOCKDEP_STATE
4288};
4289
4290static inline int state_verbose(enum lock_usage_bit bit,
4291 struct lock_class *class)
4292{
4293 return state_verbose_f[bit >> LOCK_USAGE_DIR_MASK](class);
4294}
4295
4296typedef int (*check_usage_f)(struct task_struct *, struct held_lock *,
4297 enum lock_usage_bit bit, const char *name);
4298
4299static int
4300mark_lock_irq(struct task_struct *curr, struct held_lock *this,
4301 enum lock_usage_bit new_bit)
4302{
4303 int excl_bit = exclusive_bit(new_bit);
4304 int read = new_bit & LOCK_USAGE_READ_MASK;
4305 int dir = new_bit & LOCK_USAGE_DIR_MASK;
4306
4307 /*
4308 * Validate that this particular lock does not have conflicting
4309 * usage states.
4310 */
4311 if (!valid_state(curr, this, new_bit, excl_bit))
4312 return 0;
4313
4314 /*
4315 * Check for read in write conflicts
4316 */
4317 if (!read && !valid_state(curr, this, new_bit,
4318 excl_bit + LOCK_USAGE_READ_MASK))
4319 return 0;
4320
4321
4322 /*
4323 * Validate that the lock dependencies don't have conflicting usage
4324 * states.
4325 */
4326 if (dir) {
4327 /*
4328 * mark ENABLED has to look backwards -- to ensure no dependee
4329 * has USED_IN state, which, again, would allow recursion deadlocks.
4330 */
4331 if (!check_usage_backwards(curr, this, excl_bit))
4332 return 0;
4333 } else {
4334 /*
4335 * mark USED_IN has to look forwards -- to ensure no dependency
4336 * has ENABLED state, which would allow recursion deadlocks.
4337 */
4338 if (!check_usage_forwards(curr, this, excl_bit))
4339 return 0;
4340 }
4341
4342 if (state_verbose(new_bit, hlock_class(this)))
4343 return 2;
4344
4345 return 1;
4346}
4347
4348/*
4349 * Mark all held locks with a usage bit:
4350 */
4351static int
4352mark_held_locks(struct task_struct *curr, enum lock_usage_bit base_bit)
4353{
4354 struct held_lock *hlock;
4355 int i;
4356
4357 for (i = 0; i < curr->lockdep_depth; i++) {
4358 enum lock_usage_bit hlock_bit = base_bit;
4359 hlock = curr->held_locks + i;
4360
4361 if (hlock->read)
4362 hlock_bit += LOCK_USAGE_READ_MASK;
4363
4364 BUG_ON(hlock_bit >= LOCK_USAGE_STATES);
4365
4366 if (!hlock->check)
4367 continue;
4368
4369 if (!mark_lock(curr, hlock, hlock_bit))
4370 return 0;
4371 }
4372
4373 return 1;
4374}
4375
4376/*
4377 * Hardirqs will be enabled:
4378 */
4379static void __trace_hardirqs_on_caller(void)
4380{
4381 struct task_struct *curr = current;
4382
4383 /*
4384 * We are going to turn hardirqs on, so set the
4385 * usage bit for all held locks:
4386 */
4387 if (!mark_held_locks(curr, LOCK_ENABLED_HARDIRQ))
4388 return;
4389 /*
4390 * If we have softirqs enabled, then set the usage
4391 * bit for all held locks. (disabled hardirqs prevented
4392 * this bit from being set before)
4393 */
4394 if (curr->softirqs_enabled)
4395 mark_held_locks(curr, LOCK_ENABLED_SOFTIRQ);
4396}
4397
4398/**
4399 * lockdep_hardirqs_on_prepare - Prepare for enabling interrupts
4400 *
4401 * Invoked before a possible transition to RCU idle from exit to user or
4402 * guest mode. This ensures that all RCU operations are done before RCU
4403 * stops watching. After the RCU transition lockdep_hardirqs_on() has to be
4404 * invoked to set the final state.
4405 */
4406void lockdep_hardirqs_on_prepare(void)
4407{
4408 if (unlikely(!debug_locks))
4409 return;
4410
4411 /*
4412 * NMIs do not (and cannot) track lock dependencies, nothing to do.
4413 */
4414 if (unlikely(in_nmi()))
4415 return;
4416
4417 if (unlikely(this_cpu_read(lockdep_recursion)))
4418 return;
4419
4420 if (unlikely(lockdep_hardirqs_enabled())) {
4421 /*
4422 * Neither irq nor preemption are disabled here
4423 * so this is racy by nature but losing one hit
4424 * in a stat is not a big deal.
4425 */
4426 __debug_atomic_inc(redundant_hardirqs_on);
4427 return;
4428 }
4429
4430 /*
4431 * We're enabling irqs and according to our state above irqs weren't
4432 * already enabled, yet we find the hardware thinks they are in fact
4433 * enabled.. someone messed up their IRQ state tracing.
4434 */
4435 if (DEBUG_LOCKS_WARN_ON(!irqs_disabled()))
4436 return;
4437
4438 /*
4439 * See the fine text that goes along with this variable definition.
4440 */
4441 if (DEBUG_LOCKS_WARN_ON(early_boot_irqs_disabled))
4442 return;
4443
4444 /*
4445 * Can't allow enabling interrupts while in an interrupt handler,
4446 * that's general bad form and such. Recursion, limited stack etc..
4447 */
4448 if (DEBUG_LOCKS_WARN_ON(lockdep_hardirq_context()))
4449 return;
4450
4451 current->hardirq_chain_key = current->curr_chain_key;
4452
4453 lockdep_recursion_inc();
4454 __trace_hardirqs_on_caller();
4455 lockdep_recursion_finish();
4456}
4457EXPORT_SYMBOL_GPL(lockdep_hardirqs_on_prepare);
4458
4459void noinstr lockdep_hardirqs_on(unsigned long ip)
4460{
4461 struct irqtrace_events *trace = &current->irqtrace;
4462
4463 if (unlikely(!debug_locks))
4464 return;
4465
4466 /*
4467 * NMIs can happen in the middle of local_irq_{en,dis}able() where the
4468 * tracking state and hardware state are out of sync.
4469 *
4470 * NMIs must save lockdep_hardirqs_enabled() to restore IRQ state from,
4471 * and not rely on hardware state like normal interrupts.
4472 */
4473 if (unlikely(in_nmi())) {
4474 if (!IS_ENABLED(CONFIG_TRACE_IRQFLAGS_NMI))
4475 return;
4476
4477 /*
4478 * Skip:
4479 * - recursion check, because NMI can hit lockdep;
4480 * - hardware state check, because above;
4481 * - chain_key check, see lockdep_hardirqs_on_prepare().
4482 */
4483 goto skip_checks;
4484 }
4485
4486 if (unlikely(this_cpu_read(lockdep_recursion)))
4487 return;
4488
4489 if (lockdep_hardirqs_enabled()) {
4490 /*
4491 * Neither irq nor preemption are disabled here
4492 * so this is racy by nature but losing one hit
4493 * in a stat is not a big deal.
4494 */
4495 __debug_atomic_inc(redundant_hardirqs_on);
4496 return;
4497 }
4498
4499 /*
4500 * We're enabling irqs and according to our state above irqs weren't
4501 * already enabled, yet we find the hardware thinks they are in fact
4502 * enabled.. someone messed up their IRQ state tracing.
4503 */
4504 if (DEBUG_LOCKS_WARN_ON(!irqs_disabled()))
4505 return;
4506
4507 /*
4508 * Ensure the lock stack remained unchanged between
4509 * lockdep_hardirqs_on_prepare() and lockdep_hardirqs_on().
4510 */
4511 DEBUG_LOCKS_WARN_ON(current->hardirq_chain_key !=
4512 current->curr_chain_key);
4513
4514skip_checks:
4515 /* we'll do an OFF -> ON transition: */
4516 __this_cpu_write(hardirqs_enabled, 1);
4517 trace->hardirq_enable_ip = ip;
4518 trace->hardirq_enable_event = ++trace->irq_events;
4519 debug_atomic_inc(hardirqs_on_events);
4520}
4521EXPORT_SYMBOL_GPL(lockdep_hardirqs_on);
4522
4523/*
4524 * Hardirqs were disabled:
4525 */
4526void noinstr lockdep_hardirqs_off(unsigned long ip)
4527{
4528 if (unlikely(!debug_locks))
4529 return;
4530
4531 /*
4532 * Matching lockdep_hardirqs_on(), allow NMIs in the middle of lockdep;
4533 * they will restore the software state. This ensures the software
4534 * state is consistent inside NMIs as well.
4535 */
4536 if (in_nmi()) {
4537 if (!IS_ENABLED(CONFIG_TRACE_IRQFLAGS_NMI))
4538 return;
4539 } else if (__this_cpu_read(lockdep_recursion))
4540 return;
4541
4542 /*
4543 * So we're supposed to get called after you mask local IRQs, but for
4544 * some reason the hardware doesn't quite think you did a proper job.
4545 */
4546 if (DEBUG_LOCKS_WARN_ON(!irqs_disabled()))
4547 return;
4548
4549 if (lockdep_hardirqs_enabled()) {
4550 struct irqtrace_events *trace = &current->irqtrace;
4551
4552 /*
4553 * We have done an ON -> OFF transition:
4554 */
4555 __this_cpu_write(hardirqs_enabled, 0);
4556 trace->hardirq_disable_ip = ip;
4557 trace->hardirq_disable_event = ++trace->irq_events;
4558 debug_atomic_inc(hardirqs_off_events);
4559 } else {
4560 debug_atomic_inc(redundant_hardirqs_off);
4561 }
4562}
4563EXPORT_SYMBOL_GPL(lockdep_hardirqs_off);
4564
4565/*
4566 * Softirqs will be enabled:
4567 */
4568void lockdep_softirqs_on(unsigned long ip)
4569{
4570 struct irqtrace_events *trace = &current->irqtrace;
4571
4572 if (unlikely(!lockdep_enabled()))
4573 return;
4574
4575 /*
4576 * We fancy IRQs being disabled here, see softirq.c, avoids
4577 * funny state and nesting things.
4578 */
4579 if (DEBUG_LOCKS_WARN_ON(!irqs_disabled()))
4580 return;
4581
4582 if (current->softirqs_enabled) {
4583 debug_atomic_inc(redundant_softirqs_on);
4584 return;
4585 }
4586
4587 lockdep_recursion_inc();
4588 /*
4589 * We'll do an OFF -> ON transition:
4590 */
4591 current->softirqs_enabled = 1;
4592 trace->softirq_enable_ip = ip;
4593 trace->softirq_enable_event = ++trace->irq_events;
4594 debug_atomic_inc(softirqs_on_events);
4595 /*
4596 * We are going to turn softirqs on, so set the
4597 * usage bit for all held locks, if hardirqs are
4598 * enabled too:
4599 */
4600 if (lockdep_hardirqs_enabled())
4601 mark_held_locks(current, LOCK_ENABLED_SOFTIRQ);
4602 lockdep_recursion_finish();
4603}
4604
4605/*
4606 * Softirqs were disabled:
4607 */
4608void lockdep_softirqs_off(unsigned long ip)
4609{
4610 if (unlikely(!lockdep_enabled()))
4611 return;
4612
4613 /*
4614 * We fancy IRQs being disabled here, see softirq.c
4615 */
4616 if (DEBUG_LOCKS_WARN_ON(!irqs_disabled()))
4617 return;
4618
4619 if (current->softirqs_enabled) {
4620 struct irqtrace_events *trace = &current->irqtrace;
4621
4622 /*
4623 * We have done an ON -> OFF transition:
4624 */
4625 current->softirqs_enabled = 0;
4626 trace->softirq_disable_ip = ip;
4627 trace->softirq_disable_event = ++trace->irq_events;
4628 debug_atomic_inc(softirqs_off_events);
4629 /*
4630 * Whoops, we wanted softirqs off, so why aren't they?
4631 */
4632 DEBUG_LOCKS_WARN_ON(!softirq_count());
4633 } else
4634 debug_atomic_inc(redundant_softirqs_off);
4635}
4636
4637/**
4638 * lockdep_cleanup_dead_cpu - Ensure CPU lockdep state is cleanly stopped
4639 *
4640 * @cpu: index of offlined CPU
4641 * @idle: task pointer for offlined CPU's idle thread
4642 *
4643 * Invoked after the CPU is dead. Ensures that the tracing infrastructure
4644 * is left in a suitable state for the CPU to be subsequently brought
4645 * online again.
4646 */
4647void lockdep_cleanup_dead_cpu(unsigned int cpu, struct task_struct *idle)
4648{
4649 if (unlikely(!debug_locks))
4650 return;
4651
4652 if (unlikely(per_cpu(hardirqs_enabled, cpu))) {
4653 pr_warn("CPU %u left hardirqs enabled!", cpu);
4654 if (idle)
4655 print_irqtrace_events(idle);
4656 /* Clean it up for when the CPU comes online again. */
4657 per_cpu(hardirqs_enabled, cpu) = 0;
4658 }
4659}
4660
4661static int
4662mark_usage(struct task_struct *curr, struct held_lock *hlock, int check)
4663{
4664 if (!check)
4665 goto lock_used;
4666
4667 /*
4668 * If non-trylock use in a hardirq or softirq context, then
4669 * mark the lock as used in these contexts:
4670 */
4671 if (!hlock->trylock) {
4672 if (hlock->read) {
4673 if (lockdep_hardirq_context())
4674 if (!mark_lock(curr, hlock,
4675 LOCK_USED_IN_HARDIRQ_READ))
4676 return 0;
4677 if (curr->softirq_context)
4678 if (!mark_lock(curr, hlock,
4679 LOCK_USED_IN_SOFTIRQ_READ))
4680 return 0;
4681 } else {
4682 if (lockdep_hardirq_context())
4683 if (!mark_lock(curr, hlock, LOCK_USED_IN_HARDIRQ))
4684 return 0;
4685 if (curr->softirq_context)
4686 if (!mark_lock(curr, hlock, LOCK_USED_IN_SOFTIRQ))
4687 return 0;
4688 }
4689 }
4690
4691 /*
4692 * For lock_sync(), don't mark the ENABLED usage, since lock_sync()
4693 * creates no critical section and no extra dependency can be introduced
4694 * by interrupts
4695 */
4696 if (!hlock->hardirqs_off && !hlock->sync) {
4697 if (hlock->read) {
4698 if (!mark_lock(curr, hlock,
4699 LOCK_ENABLED_HARDIRQ_READ))
4700 return 0;
4701 if (curr->softirqs_enabled)
4702 if (!mark_lock(curr, hlock,
4703 LOCK_ENABLED_SOFTIRQ_READ))
4704 return 0;
4705 } else {
4706 if (!mark_lock(curr, hlock,
4707 LOCK_ENABLED_HARDIRQ))
4708 return 0;
4709 if (curr->softirqs_enabled)
4710 if (!mark_lock(curr, hlock,
4711 LOCK_ENABLED_SOFTIRQ))
4712 return 0;
4713 }
4714 }
4715
4716lock_used:
4717 /* mark it as used: */
4718 if (!mark_lock(curr, hlock, LOCK_USED))
4719 return 0;
4720
4721 return 1;
4722}
4723
4724static inline unsigned int task_irq_context(struct task_struct *task)
4725{
4726 return LOCK_CHAIN_HARDIRQ_CONTEXT * !!lockdep_hardirq_context() +
4727 LOCK_CHAIN_SOFTIRQ_CONTEXT * !!task->softirq_context;
4728}
4729
4730static int separate_irq_context(struct task_struct *curr,
4731 struct held_lock *hlock)
4732{
4733 unsigned int depth = curr->lockdep_depth;
4734
4735 /*
4736 * Keep track of points where we cross into an interrupt context:
4737 */
4738 if (depth) {
4739 struct held_lock *prev_hlock;
4740
4741 prev_hlock = curr->held_locks + depth-1;
4742 /*
4743 * If we cross into another context, reset the
4744 * hash key (this also prevents the checking and the
4745 * adding of the dependency to 'prev'):
4746 */
4747 if (prev_hlock->irq_context != hlock->irq_context)
4748 return 1;
4749 }
4750 return 0;
4751}
4752
4753/*
4754 * Mark a lock with a usage bit, and validate the state transition:
4755 */
4756static int mark_lock(struct task_struct *curr, struct held_lock *this,
4757 enum lock_usage_bit new_bit)
4758{
4759 unsigned int new_mask, ret = 1;
4760
4761 if (new_bit >= LOCK_USAGE_STATES) {
4762 DEBUG_LOCKS_WARN_ON(1);
4763 return 0;
4764 }
4765
4766 if (new_bit == LOCK_USED && this->read)
4767 new_bit = LOCK_USED_READ;
4768
4769 new_mask = 1 << new_bit;
4770
4771 /*
4772 * If already set then do not dirty the cacheline,
4773 * nor do any checks:
4774 */
4775 if (likely(hlock_class(this)->usage_mask & new_mask))
4776 return 1;
4777
4778 if (!graph_lock())
4779 return 0;
4780 /*
4781 * Make sure we didn't race:
4782 */
4783 if (unlikely(hlock_class(this)->usage_mask & new_mask))
4784 goto unlock;
4785
4786 if (!hlock_class(this)->usage_mask)
4787 debug_atomic_dec(nr_unused_locks);
4788
4789 hlock_class(this)->usage_mask |= new_mask;
4790
4791 if (new_bit < LOCK_TRACE_STATES) {
4792 if (!(hlock_class(this)->usage_traces[new_bit] = save_trace()))
4793 return 0;
4794 }
4795
4796 if (new_bit < LOCK_USED) {
4797 ret = mark_lock_irq(curr, this, new_bit);
4798 if (!ret)
4799 return 0;
4800 }
4801
4802unlock:
4803 graph_unlock();
4804
4805 /*
4806 * We must printk outside of the graph_lock:
4807 */
4808 if (ret == 2) {
4809 nbcon_cpu_emergency_enter();
4810 printk("\nmarked lock as {%s}:\n", usage_str[new_bit]);
4811 print_lock(this);
4812 print_irqtrace_events(curr);
4813 dump_stack();
4814 nbcon_cpu_emergency_exit();
4815 }
4816
4817 return ret;
4818}
4819
4820static inline short task_wait_context(struct task_struct *curr)
4821{
4822 /*
4823 * Set appropriate wait type for the context; for IRQs we have to take
4824 * into account force_irqthread as that is implied by PREEMPT_RT.
4825 */
4826 if (lockdep_hardirq_context()) {
4827 /*
4828 * Check if force_irqthreads will run us threaded.
4829 */
4830 if (curr->hardirq_threaded || curr->irq_config)
4831 return LD_WAIT_CONFIG;
4832
4833 return LD_WAIT_SPIN;
4834 } else if (curr->softirq_context) {
4835 /*
4836 * Softirqs are always threaded.
4837 */
4838 return LD_WAIT_CONFIG;
4839 }
4840
4841 return LD_WAIT_MAX;
4842}
4843
4844static int
4845print_lock_invalid_wait_context(struct task_struct *curr,
4846 struct held_lock *hlock)
4847{
4848 short curr_inner;
4849
4850 if (!debug_locks_off())
4851 return 0;
4852 if (debug_locks_silent)
4853 return 0;
4854
4855 nbcon_cpu_emergency_enter();
4856
4857 pr_warn("\n");
4858 pr_warn("=============================\n");
4859 pr_warn("[ BUG: Invalid wait context ]\n");
4860 print_kernel_ident();
4861 pr_warn("-----------------------------\n");
4862
4863 pr_warn("%s/%d is trying to lock:\n", curr->comm, task_pid_nr(curr));
4864 print_lock(hlock);
4865
4866 pr_warn("other info that might help us debug this:\n");
4867
4868 curr_inner = task_wait_context(curr);
4869 pr_warn("context-{%d:%d}\n", curr_inner, curr_inner);
4870
4871 lockdep_print_held_locks(curr);
4872
4873 pr_warn("stack backtrace:\n");
4874 dump_stack();
4875
4876 nbcon_cpu_emergency_exit();
4877
4878 return 0;
4879}
4880
4881/*
4882 * Verify the wait_type context.
4883 *
4884 * This check validates we take locks in the right wait-type order; that is it
4885 * ensures that we do not take mutexes inside spinlocks and do not attempt to
4886 * acquire spinlocks inside raw_spinlocks and the sort.
4887 *
4888 * The entire thing is slightly more complex because of RCU, RCU is a lock that
4889 * can be taken from (pretty much) any context but also has constraints.
4890 * However when taken in a stricter environment the RCU lock does not loosen
4891 * the constraints.
4892 *
4893 * Therefore we must look for the strictest environment in the lock stack and
4894 * compare that to the lock we're trying to acquire.
4895 */
4896static int check_wait_context(struct task_struct *curr, struct held_lock *next)
4897{
4898 u8 next_inner = hlock_class(next)->wait_type_inner;
4899 u8 next_outer = hlock_class(next)->wait_type_outer;
4900 u8 curr_inner;
4901 int depth;
4902
4903 if (!next_inner || next->trylock)
4904 return 0;
4905
4906 if (!next_outer)
4907 next_outer = next_inner;
4908
4909 /*
4910 * Find start of current irq_context..
4911 */
4912 for (depth = curr->lockdep_depth - 1; depth >= 0; depth--) {
4913 struct held_lock *prev = curr->held_locks + depth;
4914 if (prev->irq_context != next->irq_context)
4915 break;
4916 }
4917 depth++;
4918
4919 curr_inner = task_wait_context(curr);
4920
4921 for (; depth < curr->lockdep_depth; depth++) {
4922 struct held_lock *prev = curr->held_locks + depth;
4923 struct lock_class *class = hlock_class(prev);
4924 u8 prev_inner = class->wait_type_inner;
4925
4926 if (prev_inner) {
4927 /*
4928 * We can have a bigger inner than a previous one
4929 * when outer is smaller than inner, as with RCU.
4930 *
4931 * Also due to trylocks.
4932 */
4933 curr_inner = min(curr_inner, prev_inner);
4934
4935 /*
4936 * Allow override for annotations -- this is typically
4937 * only valid/needed for code that only exists when
4938 * CONFIG_PREEMPT_RT=n.
4939 */
4940 if (unlikely(class->lock_type == LD_LOCK_WAIT_OVERRIDE))
4941 curr_inner = prev_inner;
4942 }
4943 }
4944
4945 if (next_outer > curr_inner)
4946 return print_lock_invalid_wait_context(curr, next);
4947
4948 return 0;
4949}
4950
4951#else /* CONFIG_PROVE_LOCKING */
4952
4953static inline int
4954mark_usage(struct task_struct *curr, struct held_lock *hlock, int check)
4955{
4956 return 1;
4957}
4958
4959static inline unsigned int task_irq_context(struct task_struct *task)
4960{
4961 return 0;
4962}
4963
4964static inline int separate_irq_context(struct task_struct *curr,
4965 struct held_lock *hlock)
4966{
4967 return 0;
4968}
4969
4970static inline int check_wait_context(struct task_struct *curr,
4971 struct held_lock *next)
4972{
4973 return 0;
4974}
4975
4976#endif /* CONFIG_PROVE_LOCKING */
4977
4978/*
4979 * Initialize a lock instance's lock-class mapping info:
4980 */
4981void lockdep_init_map_type(struct lockdep_map *lock, const char *name,
4982 struct lock_class_key *key, int subclass,
4983 u8 inner, u8 outer, u8 lock_type)
4984{
4985 int i;
4986
4987 for (i = 0; i < NR_LOCKDEP_CACHING_CLASSES; i++)
4988 WRITE_ONCE(lock->class_cache[i], NULL);
4989
4990#ifdef CONFIG_LOCK_STAT
4991 lock->cpu = raw_smp_processor_id();
4992#endif
4993
4994 /*
4995 * Can't be having no nameless bastards around this place!
4996 */
4997 if (DEBUG_LOCKS_WARN_ON(!name)) {
4998 lock->name = "NULL";
4999 return;
5000 }
5001
5002 lock->name = name;
5003
5004 lock->wait_type_outer = outer;
5005 lock->wait_type_inner = inner;
5006 lock->lock_type = lock_type;
5007
5008 /*
5009 * No key, no joy, we need to hash something.
5010 */
5011 if (DEBUG_LOCKS_WARN_ON(!key))
5012 return;
5013 /*
5014 * Sanity check, the lock-class key must either have been allocated
5015 * statically or must have been registered as a dynamic key.
5016 */
5017 if (!static_obj(key) && !is_dynamic_key(key)) {
5018 if (debug_locks)
5019 printk(KERN_ERR "BUG: key %px has not been registered!\n", key);
5020 DEBUG_LOCKS_WARN_ON(1);
5021 return;
5022 }
5023 lock->key = key;
5024
5025 if (unlikely(!debug_locks))
5026 return;
5027
5028 if (subclass) {
5029 unsigned long flags;
5030
5031 if (DEBUG_LOCKS_WARN_ON(!lockdep_enabled()))
5032 return;
5033
5034 raw_local_irq_save(flags);
5035 lockdep_recursion_inc();
5036 register_lock_class(lock, subclass, 1);
5037 lockdep_recursion_finish();
5038 raw_local_irq_restore(flags);
5039 }
5040}
5041EXPORT_SYMBOL_GPL(lockdep_init_map_type);
5042
5043struct lock_class_key __lockdep_no_validate__;
5044EXPORT_SYMBOL_GPL(__lockdep_no_validate__);
5045
5046struct lock_class_key __lockdep_no_track__;
5047EXPORT_SYMBOL_GPL(__lockdep_no_track__);
5048
5049#ifdef CONFIG_PROVE_LOCKING
5050void lockdep_set_lock_cmp_fn(struct lockdep_map *lock, lock_cmp_fn cmp_fn,
5051 lock_print_fn print_fn)
5052{
5053 struct lock_class *class = READ_ONCE(lock->class_cache[0]);
5054 unsigned long flags;
5055
5056 raw_local_irq_save(flags);
5057 lockdep_recursion_inc();
5058
5059 if (!lock_class_cache_is_valid(lock, class, 0))
5060 class = NULL;
5061
5062 if (!class)
5063 class = register_lock_class(lock, 0, 0);
5064
5065 if (class) {
5066 WARN_ON(class->cmp_fn && class->cmp_fn != cmp_fn);
5067 WARN_ON(class->print_fn && class->print_fn != print_fn);
5068
5069 class->cmp_fn = cmp_fn;
5070 class->print_fn = print_fn;
5071 }
5072
5073 lockdep_recursion_finish();
5074 raw_local_irq_restore(flags);
5075}
5076EXPORT_SYMBOL_GPL(lockdep_set_lock_cmp_fn);
5077#endif
5078
5079static void
5080print_lock_nested_lock_not_held(struct task_struct *curr,
5081 struct held_lock *hlock)
5082{
5083 if (!debug_locks_off())
5084 return;
5085 if (debug_locks_silent)
5086 return;
5087
5088 nbcon_cpu_emergency_enter();
5089
5090 pr_warn("\n");
5091 pr_warn("==================================\n");
5092 pr_warn("WARNING: Nested lock was not taken\n");
5093 print_kernel_ident();
5094 pr_warn("----------------------------------\n");
5095
5096 pr_warn("%s/%d is trying to lock:\n", curr->comm, task_pid_nr(curr));
5097 print_lock(hlock);
5098
5099 pr_warn("\nbut this task is not holding:\n");
5100 pr_warn("%s\n", hlock->nest_lock->name);
5101
5102 pr_warn("\nstack backtrace:\n");
5103 dump_stack();
5104
5105 pr_warn("\nother info that might help us debug this:\n");
5106 lockdep_print_held_locks(curr);
5107
5108 pr_warn("\nstack backtrace:\n");
5109 dump_stack();
5110
5111 nbcon_cpu_emergency_exit();
5112}
5113
5114static int __lock_is_held(const struct lockdep_map *lock, int read);
5115
5116/*
5117 * This gets called for every mutex_lock*()/spin_lock*() operation.
5118 * We maintain the dependency maps and validate the locking attempt:
5119 *
5120 * The callers must make sure that IRQs are disabled before calling it,
5121 * otherwise we could get an interrupt which would want to take locks,
5122 * which would end up in lockdep again.
5123 */
5124static int __lock_acquire(struct lockdep_map *lock, unsigned int subclass,
5125 int trylock, int read, int check, int hardirqs_off,
5126 struct lockdep_map *nest_lock, unsigned long ip,
5127 int references, int pin_count, int sync, int seq)
5128{
5129 struct task_struct *curr = current;
5130 struct lock_class *class = NULL;
5131 struct held_lock *hlock;
5132 unsigned int depth;
5133 int chain_head = 0;
5134 int class_idx;
5135 u64 chain_key;
5136
5137 if (unlikely(!debug_locks))
5138 return 0;
5139
5140 if (unlikely(lock->key == &__lockdep_no_track__))
5141 return 0;
5142
5143 lockevent_inc(lockdep_acquire);
5144
5145 if (!prove_locking || lock->key == &__lockdep_no_validate__) {
5146 check = 0;
5147 lockevent_inc(lockdep_nocheck);
5148 }
5149
5150 if (DEBUG_LOCKS_WARN_ON(subclass >= MAX_LOCKDEP_SUBCLASSES))
5151 return 0;
5152
5153 if (subclass < NR_LOCKDEP_CACHING_CLASSES) {
5154 class = READ_ONCE(lock->class_cache[subclass]);
5155 if (!lock_class_cache_is_valid(lock, class, subclass))
5156 class = NULL;
5157 }
5158 /*
5159 * Not cached?
5160 */
5161 if (unlikely(!class)) {
5162 class = register_lock_class(lock, subclass, 0);
5163 if (!class)
5164 return 0;
5165 }
5166
5167 debug_class_ops_inc(class);
5168
5169 if (very_verbose(class)) {
5170 nbcon_cpu_emergency_enter();
5171 printk("\nacquire class [%px] %s", class->key, class->name);
5172 if (class->name_version > 1)
5173 printk(KERN_CONT "#%d", class->name_version);
5174 printk(KERN_CONT "\n");
5175 dump_stack();
5176 nbcon_cpu_emergency_exit();
5177 }
5178
5179 /*
5180 * Add the lock to the list of currently held locks.
5181 * (we dont increase the depth just yet, up until the
5182 * dependency checks are done)
5183 */
5184 depth = curr->lockdep_depth;
5185 /*
5186 * Ran out of static storage for our per-task lock stack again have we?
5187 */
5188 if (DEBUG_LOCKS_WARN_ON(depth >= MAX_LOCK_DEPTH))
5189 return 0;
5190
5191 class_idx = class - lock_classes;
5192
5193 if (depth && !sync) {
5194 /* we're holding locks and the new held lock is not a sync */
5195 hlock = curr->held_locks + depth - 1;
5196 if (hlock->class_idx == class_idx && nest_lock) {
5197 if (!references)
5198 references++;
5199
5200 if (!hlock->references)
5201 hlock->references++;
5202
5203 hlock->references += references;
5204
5205 /* Overflow */
5206 if (DEBUG_LOCKS_WARN_ON(hlock->references < references))
5207 return 0;
5208
5209 return 2;
5210 }
5211 }
5212
5213 hlock = curr->held_locks + depth;
5214 /*
5215 * Plain impossible, we just registered it and checked it weren't no
5216 * NULL like.. I bet this mushroom I ate was good!
5217 */
5218 if (DEBUG_LOCKS_WARN_ON(!class))
5219 return 0;
5220 hlock->class_idx = class_idx;
5221 hlock->acquire_ip = ip;
5222 hlock->instance = lock;
5223 hlock->nest_lock = nest_lock;
5224 hlock->irq_context = task_irq_context(curr);
5225 hlock->trylock = trylock;
5226 hlock->read = read;
5227 hlock->check = check;
5228 hlock->sync = !!sync;
5229 hlock->hardirqs_off = !!hardirqs_off;
5230 hlock->references = references;
5231#ifdef CONFIG_LOCK_STAT
5232 hlock->waittime_stamp = 0;
5233 hlock->holdtime_stamp = lockstat_clock();
5234#endif
5235 hlock->pin_count = pin_count;
5236 hlock->seq_count = seq;
5237
5238 if (check_wait_context(curr, hlock))
5239 return 0;
5240
5241 /* Initialize the lock usage bit */
5242 if (!mark_usage(curr, hlock, check))
5243 return 0;
5244
5245 /*
5246 * Calculate the chain hash: it's the combined hash of all the
5247 * lock keys along the dependency chain. We save the hash value
5248 * at every step so that we can get the current hash easily
5249 * after unlock. The chain hash is then used to cache dependency
5250 * results.
5251 *
5252 * The 'key ID' is what is the most compact key value to drive
5253 * the hash, not class->key.
5254 */
5255 /*
5256 * Whoops, we did it again.. class_idx is invalid.
5257 */
5258 if (DEBUG_LOCKS_WARN_ON(!test_bit(class_idx, lock_classes_in_use)))
5259 return 0;
5260
5261 chain_key = curr->curr_chain_key;
5262 if (!depth) {
5263 /*
5264 * How can we have a chain hash when we ain't got no keys?!
5265 */
5266 if (DEBUG_LOCKS_WARN_ON(chain_key != INITIAL_CHAIN_KEY))
5267 return 0;
5268 chain_head = 1;
5269 }
5270
5271 hlock->prev_chain_key = chain_key;
5272 if (separate_irq_context(curr, hlock)) {
5273 chain_key = INITIAL_CHAIN_KEY;
5274 chain_head = 1;
5275 }
5276 chain_key = iterate_chain_key(chain_key, hlock_id(hlock));
5277
5278 if (nest_lock && !__lock_is_held(nest_lock, -1)) {
5279 print_lock_nested_lock_not_held(curr, hlock);
5280 return 0;
5281 }
5282
5283 if (!debug_locks_silent) {
5284 WARN_ON_ONCE(depth && !hlock_class(hlock - 1)->key);
5285 WARN_ON_ONCE(!hlock_class(hlock)->key);
5286 }
5287
5288 if (!validate_chain(curr, hlock, chain_head, chain_key))
5289 return 0;
5290
5291 /* For lock_sync(), we are done here since no actual critical section */
5292 if (hlock->sync)
5293 return 1;
5294
5295 curr->curr_chain_key = chain_key;
5296 curr->lockdep_depth++;
5297 check_chain_key(curr);
5298#ifdef CONFIG_DEBUG_LOCKDEP
5299 if (unlikely(!debug_locks))
5300 return 0;
5301#endif
5302 if (unlikely(curr->lockdep_depth >= MAX_LOCK_DEPTH)) {
5303 debug_locks_off();
5304 nbcon_cpu_emergency_enter();
5305 print_lockdep_off("BUG: MAX_LOCK_DEPTH too low!");
5306 printk(KERN_DEBUG "depth: %i max: %lu!\n",
5307 curr->lockdep_depth, MAX_LOCK_DEPTH);
5308
5309 lockdep_print_held_locks(current);
5310 debug_show_all_locks();
5311 dump_stack();
5312 nbcon_cpu_emergency_exit();
5313
5314 return 0;
5315 }
5316
5317 if (unlikely(curr->lockdep_depth > max_lockdep_depth))
5318 max_lockdep_depth = curr->lockdep_depth;
5319
5320 return 1;
5321}
5322
5323static void print_unlock_imbalance_bug(struct task_struct *curr,
5324 struct lockdep_map *lock,
5325 unsigned long ip)
5326{
5327 if (!debug_locks_off())
5328 return;
5329 if (debug_locks_silent)
5330 return;
5331
5332 nbcon_cpu_emergency_enter();
5333
5334 pr_warn("\n");
5335 pr_warn("=====================================\n");
5336 pr_warn("WARNING: bad unlock balance detected!\n");
5337 print_kernel_ident();
5338 pr_warn("-------------------------------------\n");
5339 pr_warn("%s/%d is trying to release lock (",
5340 curr->comm, task_pid_nr(curr));
5341 print_lockdep_cache(lock);
5342 pr_cont(") at:\n");
5343 print_ip_sym(KERN_WARNING, ip);
5344 pr_warn("but there are no more locks to release!\n");
5345 pr_warn("\nother info that might help us debug this:\n");
5346 lockdep_print_held_locks(curr);
5347
5348 pr_warn("\nstack backtrace:\n");
5349 dump_stack();
5350
5351 nbcon_cpu_emergency_exit();
5352}
5353
5354static noinstr int match_held_lock(const struct held_lock *hlock,
5355 const struct lockdep_map *lock)
5356{
5357 if (hlock->instance == lock)
5358 return 1;
5359
5360 if (hlock->references) {
5361 const struct lock_class *class = READ_ONCE(lock->class_cache[0]);
5362
5363 if (!lock_class_cache_is_valid(lock, class, 0))
5364 class = look_up_lock_class(lock, 0);
5365
5366 /*
5367 * If look_up_lock_class() failed to find a class, we're trying
5368 * to test if we hold a lock that has never yet been acquired.
5369 * Clearly if the lock hasn't been acquired _ever_, we're not
5370 * holding it either, so report failure.
5371 */
5372 if (!class)
5373 return 0;
5374
5375 /*
5376 * References, but not a lock we're actually ref-counting?
5377 * State got messed up, follow the sites that change ->references
5378 * and try to make sense of it.
5379 */
5380 if (DEBUG_LOCKS_WARN_ON(!hlock->nest_lock))
5381 return 0;
5382
5383 if (hlock->class_idx == class - lock_classes)
5384 return 1;
5385 }
5386
5387 return 0;
5388}
5389
5390/* @depth must not be zero */
5391static struct held_lock *find_held_lock(struct task_struct *curr,
5392 struct lockdep_map *lock,
5393 unsigned int depth, int *idx)
5394{
5395 struct held_lock *ret, *hlock, *prev_hlock;
5396 int i;
5397
5398 i = depth - 1;
5399 hlock = curr->held_locks + i;
5400 ret = hlock;
5401 if (match_held_lock(hlock, lock))
5402 goto out;
5403
5404 ret = NULL;
5405 for (i--, prev_hlock = hlock--;
5406 i >= 0;
5407 i--, prev_hlock = hlock--) {
5408 /*
5409 * We must not cross into another context:
5410 */
5411 if (prev_hlock->irq_context != hlock->irq_context) {
5412 ret = NULL;
5413 break;
5414 }
5415 if (match_held_lock(hlock, lock)) {
5416 ret = hlock;
5417 break;
5418 }
5419 }
5420
5421out:
5422 *idx = i;
5423 return ret;
5424}
5425
5426static int reacquire_held_locks(struct task_struct *curr, unsigned int depth,
5427 int idx, unsigned int *merged)
5428{
5429 struct held_lock *hlock;
5430 int first_idx = idx;
5431
5432 if (DEBUG_LOCKS_WARN_ON(!irqs_disabled()))
5433 return 0;
5434
5435 for (hlock = curr->held_locks + idx; idx < depth; idx++, hlock++) {
5436 switch (__lock_acquire(hlock->instance,
5437 hlock_class(hlock)->subclass,
5438 hlock->trylock,
5439 hlock->read, hlock->check,
5440 hlock->hardirqs_off,
5441 hlock->nest_lock, hlock->acquire_ip,
5442 hlock->references, hlock->pin_count, 0, hlock->seq_count)) {
5443 case 0:
5444 return 1;
5445 case 1:
5446 break;
5447 case 2:
5448 *merged += (idx == first_idx);
5449 break;
5450 default:
5451 WARN_ON(1);
5452 return 0;
5453 }
5454 }
5455 return 0;
5456}
5457
5458static int
5459__lock_set_class(struct lockdep_map *lock, const char *name,
5460 struct lock_class_key *key, unsigned int subclass,
5461 unsigned long ip)
5462{
5463 struct task_struct *curr = current;
5464 unsigned int depth, merged = 0;
5465 struct held_lock *hlock;
5466 struct lock_class *class;
5467 int i;
5468
5469 if (unlikely(!debug_locks))
5470 return 0;
5471
5472 depth = curr->lockdep_depth;
5473 /*
5474 * This function is about (re)setting the class of a held lock,
5475 * yet we're not actually holding any locks. Naughty user!
5476 */
5477 if (DEBUG_LOCKS_WARN_ON(!depth))
5478 return 0;
5479
5480 hlock = find_held_lock(curr, lock, depth, &i);
5481 if (!hlock) {
5482 print_unlock_imbalance_bug(curr, lock, ip);
5483 return 0;
5484 }
5485
5486 lockdep_init_map_type(lock, name, key, 0,
5487 lock->wait_type_inner,
5488 lock->wait_type_outer,
5489 lock->lock_type);
5490 class = register_lock_class(lock, subclass, 0);
5491 if (!class)
5492 return 0;
5493 hlock->class_idx = class - lock_classes;
5494
5495 curr->lockdep_depth = i;
5496 curr->curr_chain_key = hlock->prev_chain_key;
5497
5498 if (reacquire_held_locks(curr, depth, i, &merged))
5499 return 0;
5500
5501 /*
5502 * I took it apart and put it back together again, except now I have
5503 * these 'spare' parts.. where shall I put them.
5504 */
5505 if (DEBUG_LOCKS_WARN_ON(curr->lockdep_depth != depth - merged))
5506 return 0;
5507 return 1;
5508}
5509
5510static int __lock_downgrade(struct lockdep_map *lock, unsigned long ip)
5511{
5512 struct task_struct *curr = current;
5513 unsigned int depth, merged = 0;
5514 struct held_lock *hlock;
5515 int i;
5516
5517 if (unlikely(!debug_locks))
5518 return 0;
5519
5520 depth = curr->lockdep_depth;
5521 /*
5522 * This function is about (re)setting the class of a held lock,
5523 * yet we're not actually holding any locks. Naughty user!
5524 */
5525 if (DEBUG_LOCKS_WARN_ON(!depth))
5526 return 0;
5527
5528 hlock = find_held_lock(curr, lock, depth, &i);
5529 if (!hlock) {
5530 print_unlock_imbalance_bug(curr, lock, ip);
5531 return 0;
5532 }
5533
5534 curr->lockdep_depth = i;
5535 curr->curr_chain_key = hlock->prev_chain_key;
5536
5537 WARN(hlock->read, "downgrading a read lock");
5538 hlock->read = 1;
5539 hlock->acquire_ip = ip;
5540
5541 if (reacquire_held_locks(curr, depth, i, &merged))
5542 return 0;
5543
5544 /* Merging can't happen with unchanged classes.. */
5545 if (DEBUG_LOCKS_WARN_ON(merged))
5546 return 0;
5547
5548 /*
5549 * I took it apart and put it back together again, except now I have
5550 * these 'spare' parts.. where shall I put them.
5551 */
5552 if (DEBUG_LOCKS_WARN_ON(curr->lockdep_depth != depth))
5553 return 0;
5554
5555 return 1;
5556}
5557
5558/*
5559 * Remove the lock from the list of currently held locks - this gets
5560 * called on mutex_unlock()/spin_unlock*() (or on a failed
5561 * mutex_lock_interruptible()).
5562 */
5563static int
5564__lock_release(struct lockdep_map *lock, unsigned long ip)
5565{
5566 struct task_struct *curr = current;
5567 unsigned int depth, merged = 1;
5568 struct held_lock *hlock;
5569 int i;
5570
5571 if (unlikely(!debug_locks))
5572 return 0;
5573
5574 depth = curr->lockdep_depth;
5575 /*
5576 * So we're all set to release this lock.. wait what lock? We don't
5577 * own any locks, you've been drinking again?
5578 */
5579 if (depth <= 0) {
5580 print_unlock_imbalance_bug(curr, lock, ip);
5581 return 0;
5582 }
5583
5584 /*
5585 * Check whether the lock exists in the current stack
5586 * of held locks:
5587 */
5588 hlock = find_held_lock(curr, lock, depth, &i);
5589 if (!hlock) {
5590 print_unlock_imbalance_bug(curr, lock, ip);
5591 return 0;
5592 }
5593
5594 if (hlock->instance == lock)
5595 lock_release_holdtime(hlock);
5596
5597 WARN(hlock->pin_count, "releasing a pinned lock\n");
5598
5599 if (hlock->references) {
5600 hlock->references--;
5601 if (hlock->references) {
5602 /*
5603 * We had, and after removing one, still have
5604 * references, the current lock stack is still
5605 * valid. We're done!
5606 */
5607 return 1;
5608 }
5609 }
5610
5611 /*
5612 * We have the right lock to unlock, 'hlock' points to it.
5613 * Now we remove it from the stack, and add back the other
5614 * entries (if any), recalculating the hash along the way:
5615 */
5616
5617 curr->lockdep_depth = i;
5618 curr->curr_chain_key = hlock->prev_chain_key;
5619
5620 /*
5621 * The most likely case is when the unlock is on the innermost
5622 * lock. In this case, we are done!
5623 */
5624 if (i == depth-1)
5625 return 1;
5626
5627 if (reacquire_held_locks(curr, depth, i + 1, &merged))
5628 return 0;
5629
5630 /*
5631 * We had N bottles of beer on the wall, we drank one, but now
5632 * there's not N-1 bottles of beer left on the wall...
5633 * Pouring two of the bottles together is acceptable.
5634 */
5635 DEBUG_LOCKS_WARN_ON(curr->lockdep_depth != depth - merged);
5636
5637 /*
5638 * Since reacquire_held_locks() would have called check_chain_key()
5639 * indirectly via __lock_acquire(), we don't need to do it again
5640 * on return.
5641 */
5642 return 0;
5643}
5644
5645static __always_inline
5646int __lock_is_held(const struct lockdep_map *lock, int read)
5647{
5648 struct task_struct *curr = current;
5649 int i;
5650
5651 for (i = 0; i < curr->lockdep_depth; i++) {
5652 struct held_lock *hlock = curr->held_locks + i;
5653
5654 if (match_held_lock(hlock, lock)) {
5655 if (read == -1 || !!hlock->read == read)
5656 return LOCK_STATE_HELD;
5657
5658 return LOCK_STATE_NOT_HELD;
5659 }
5660 }
5661
5662 return LOCK_STATE_NOT_HELD;
5663}
5664
5665static struct pin_cookie __lock_pin_lock(struct lockdep_map *lock)
5666{
5667 struct pin_cookie cookie = NIL_COOKIE;
5668 struct task_struct *curr = current;
5669 int i;
5670
5671 if (unlikely(!debug_locks))
5672 return cookie;
5673
5674 for (i = 0; i < curr->lockdep_depth; i++) {
5675 struct held_lock *hlock = curr->held_locks + i;
5676
5677 if (match_held_lock(hlock, lock)) {
5678 /*
5679 * Grab 16bits of randomness; this is sufficient to not
5680 * be guessable and still allows some pin nesting in
5681 * our u32 pin_count.
5682 */
5683 cookie.val = 1 + (sched_clock() & 0xffff);
5684 hlock->pin_count += cookie.val;
5685 return cookie;
5686 }
5687 }
5688
5689 WARN(1, "pinning an unheld lock\n");
5690 return cookie;
5691}
5692
5693static void __lock_repin_lock(struct lockdep_map *lock, struct pin_cookie cookie)
5694{
5695 struct task_struct *curr = current;
5696 int i;
5697
5698 if (unlikely(!debug_locks))
5699 return;
5700
5701 for (i = 0; i < curr->lockdep_depth; i++) {
5702 struct held_lock *hlock = curr->held_locks + i;
5703
5704 if (match_held_lock(hlock, lock)) {
5705 hlock->pin_count += cookie.val;
5706 return;
5707 }
5708 }
5709
5710 WARN(1, "pinning an unheld lock\n");
5711}
5712
5713static void __lock_unpin_lock(struct lockdep_map *lock, struct pin_cookie cookie)
5714{
5715 struct task_struct *curr = current;
5716 int i;
5717
5718 if (unlikely(!debug_locks))
5719 return;
5720
5721 for (i = 0; i < curr->lockdep_depth; i++) {
5722 struct held_lock *hlock = curr->held_locks + i;
5723
5724 if (match_held_lock(hlock, lock)) {
5725 int pin_count;
5726
5727 if (WARN(!hlock->pin_count, "unpinning an unpinned lock\n"))
5728 return;
5729
5730 pin_count = hlock->pin_count - cookie.val;
5731
5732 if (WARN(pin_count < 0, "pin count corrupted\n"))
5733 pin_count = 0;
5734
5735 hlock->pin_count = pin_count;
5736 return;
5737 }
5738 }
5739
5740 WARN(1, "unpinning an unheld lock\n");
5741}
5742
5743static u32 __lock_sequence(struct lockdep_map *lock)
5744{
5745 struct task_struct *curr = current;
5746 int i;
5747
5748 if (unlikely(!debug_locks))
5749 return ~0;
5750
5751 for (i = 0; i < curr->lockdep_depth; i++) {
5752 struct held_lock *hlock = curr->held_locks + i;
5753
5754 if (match_held_lock(hlock, lock))
5755 return hlock->seq_count;
5756 }
5757
5758 return ~0;
5759}
5760
5761/*
5762 * Check whether we follow the irq-flags state precisely:
5763 */
5764static noinstr void check_flags(unsigned long flags)
5765{
5766#if defined(CONFIG_PROVE_LOCKING) && defined(CONFIG_DEBUG_LOCKDEP)
5767 if (!debug_locks)
5768 return;
5769
5770 /* Get the warning out.. */
5771 instrumentation_begin();
5772
5773 if (irqs_disabled_flags(flags)) {
5774 if (DEBUG_LOCKS_WARN_ON(lockdep_hardirqs_enabled())) {
5775 printk("possible reason: unannotated irqs-off.\n");
5776 }
5777 } else {
5778 if (DEBUG_LOCKS_WARN_ON(!lockdep_hardirqs_enabled())) {
5779 printk("possible reason: unannotated irqs-on.\n");
5780 }
5781 }
5782
5783#ifndef CONFIG_PREEMPT_RT
5784 /*
5785 * We dont accurately track softirq state in e.g.
5786 * hardirq contexts (such as on 4KSTACKS), so only
5787 * check if not in hardirq contexts:
5788 */
5789 if (!hardirq_count()) {
5790 if (softirq_count()) {
5791 /* like the above, but with softirqs */
5792 DEBUG_LOCKS_WARN_ON(current->softirqs_enabled);
5793 } else {
5794 /* lick the above, does it taste good? */
5795 DEBUG_LOCKS_WARN_ON(!current->softirqs_enabled);
5796 }
5797 }
5798#endif
5799
5800 if (!debug_locks)
5801 print_irqtrace_events(current);
5802
5803 instrumentation_end();
5804#endif
5805}
5806
5807void lock_set_class(struct lockdep_map *lock, const char *name,
5808 struct lock_class_key *key, unsigned int subclass,
5809 unsigned long ip)
5810{
5811 unsigned long flags;
5812
5813 if (unlikely(!lockdep_enabled()))
5814 return;
5815
5816 raw_local_irq_save(flags);
5817 lockdep_recursion_inc();
5818 check_flags(flags);
5819 if (__lock_set_class(lock, name, key, subclass, ip))
5820 check_chain_key(current);
5821 lockdep_recursion_finish();
5822 raw_local_irq_restore(flags);
5823}
5824EXPORT_SYMBOL_GPL(lock_set_class);
5825
5826void lock_downgrade(struct lockdep_map *lock, unsigned long ip)
5827{
5828 unsigned long flags;
5829
5830 if (unlikely(!lockdep_enabled()))
5831 return;
5832
5833 raw_local_irq_save(flags);
5834 lockdep_recursion_inc();
5835 check_flags(flags);
5836 if (__lock_downgrade(lock, ip))
5837 check_chain_key(current);
5838 lockdep_recursion_finish();
5839 raw_local_irq_restore(flags);
5840}
5841EXPORT_SYMBOL_GPL(lock_downgrade);
5842
5843/* NMI context !!! */
5844static void verify_lock_unused(struct lockdep_map *lock, struct held_lock *hlock, int subclass)
5845{
5846#ifdef CONFIG_PROVE_LOCKING
5847 struct lock_class *class = look_up_lock_class(lock, subclass);
5848 unsigned long mask = LOCKF_USED;
5849
5850 /* if it doesn't have a class (yet), it certainly hasn't been used yet */
5851 if (!class)
5852 return;
5853
5854 /*
5855 * READ locks only conflict with USED, such that if we only ever use
5856 * READ locks, there is no deadlock possible -- RCU.
5857 */
5858 if (!hlock->read)
5859 mask |= LOCKF_USED_READ;
5860
5861 if (!(class->usage_mask & mask))
5862 return;
5863
5864 hlock->class_idx = class - lock_classes;
5865
5866 print_usage_bug(current, hlock, LOCK_USED, LOCK_USAGE_STATES);
5867#endif
5868}
5869
5870static bool lockdep_nmi(void)
5871{
5872 if (raw_cpu_read(lockdep_recursion))
5873 return false;
5874
5875 if (!in_nmi())
5876 return false;
5877
5878 return true;
5879}
5880
5881/*
5882 * read_lock() is recursive if:
5883 * 1. We force lockdep think this way in selftests or
5884 * 2. The implementation is not queued read/write lock or
5885 * 3. The locker is at an in_interrupt() context.
5886 */
5887bool read_lock_is_recursive(void)
5888{
5889 return force_read_lock_recursive ||
5890 !IS_ENABLED(CONFIG_QUEUED_RWLOCKS) ||
5891 in_interrupt();
5892}
5893EXPORT_SYMBOL_GPL(read_lock_is_recursive);
5894
5895/*
5896 * We are not always called with irqs disabled - do that here,
5897 * and also avoid lockdep recursion:
5898 */
5899void lock_acquire(struct lockdep_map *lock, unsigned int subclass,
5900 int trylock, int read, int check,
5901 struct lockdep_map *nest_lock, unsigned long ip)
5902{
5903 unsigned long flags;
5904
5905 trace_lock_acquire(lock, subclass, trylock, read, check, nest_lock, ip);
5906
5907 if (!debug_locks)
5908 return;
5909
5910 /*
5911 * As KASAN instrumentation is disabled and lock_acquire() is usually
5912 * the first lockdep call when a task tries to acquire a lock, add
5913 * kasan_check_byte() here to check for use-after-free and other
5914 * memory errors.
5915 */
5916 kasan_check_byte(lock);
5917
5918 if (unlikely(!lockdep_enabled())) {
5919 /* XXX allow trylock from NMI ?!? */
5920 if (lockdep_nmi() && !trylock) {
5921 struct held_lock hlock;
5922
5923 hlock.acquire_ip = ip;
5924 hlock.instance = lock;
5925 hlock.nest_lock = nest_lock;
5926 hlock.irq_context = 2; // XXX
5927 hlock.trylock = trylock;
5928 hlock.read = read;
5929 hlock.check = check;
5930 hlock.hardirqs_off = true;
5931 hlock.references = 0;
5932
5933 verify_lock_unused(lock, &hlock, subclass);
5934 }
5935 return;
5936 }
5937
5938 raw_local_irq_save(flags);
5939 check_flags(flags);
5940
5941 lockdep_recursion_inc();
5942 __lock_acquire(lock, subclass, trylock, read, check,
5943 irqs_disabled_flags(flags), nest_lock, ip, 0, 0, 0,
5944 ++current->lockdep_seq);
5945 lockdep_recursion_finish();
5946 raw_local_irq_restore(flags);
5947}
5948EXPORT_SYMBOL_GPL(lock_acquire);
5949
5950void lock_release(struct lockdep_map *lock, unsigned long ip)
5951{
5952 unsigned long flags;
5953
5954 trace_lock_release(lock, ip);
5955
5956 if (unlikely(!lockdep_enabled() ||
5957 lock->key == &__lockdep_no_track__))
5958 return;
5959
5960 raw_local_irq_save(flags);
5961 check_flags(flags);
5962
5963 lockdep_recursion_inc();
5964 if (__lock_release(lock, ip))
5965 check_chain_key(current);
5966 lockdep_recursion_finish();
5967 raw_local_irq_restore(flags);
5968}
5969EXPORT_SYMBOL_GPL(lock_release);
5970
5971/*
5972 * lock_sync() - A special annotation for synchronize_{s,}rcu()-like API.
5973 *
5974 * No actual critical section is created by the APIs annotated with this: these
5975 * APIs are used to wait for one or multiple critical sections (on other CPUs
5976 * or threads), and it means that calling these APIs inside these critical
5977 * sections is potential deadlock.
5978 */
5979void lock_sync(struct lockdep_map *lock, unsigned subclass, int read,
5980 int check, struct lockdep_map *nest_lock, unsigned long ip)
5981{
5982 unsigned long flags;
5983
5984 if (unlikely(!lockdep_enabled()))
5985 return;
5986
5987 raw_local_irq_save(flags);
5988 check_flags(flags);
5989
5990 lockdep_recursion_inc();
5991 __lock_acquire(lock, subclass, 0, read, check,
5992 irqs_disabled_flags(flags), nest_lock, ip, 0, 0, 1,
5993 ++current->lockdep_seq);
5994 check_chain_key(current);
5995 lockdep_recursion_finish();
5996 raw_local_irq_restore(flags);
5997}
5998EXPORT_SYMBOL_GPL(lock_sync);
5999
6000noinstr int lock_is_held_type(const struct lockdep_map *lock, int read)
6001{
6002 unsigned long flags;
6003 int ret = LOCK_STATE_NOT_HELD;
6004
6005 /*
6006 * Avoid false negative lockdep_assert_held() and
6007 * lockdep_assert_not_held().
6008 */
6009 if (unlikely(!lockdep_enabled()))
6010 return LOCK_STATE_UNKNOWN;
6011
6012 raw_local_irq_save(flags);
6013 check_flags(flags);
6014
6015 lockdep_recursion_inc();
6016 ret = __lock_is_held(lock, read);
6017 lockdep_recursion_finish();
6018 raw_local_irq_restore(flags);
6019
6020 return ret;
6021}
6022EXPORT_SYMBOL_GPL(lock_is_held_type);
6023NOKPROBE_SYMBOL(lock_is_held_type);
6024
6025struct pin_cookie lock_pin_lock(struct lockdep_map *lock)
6026{
6027 struct pin_cookie cookie = NIL_COOKIE;
6028 unsigned long flags;
6029
6030 if (unlikely(!lockdep_enabled()))
6031 return cookie;
6032
6033 raw_local_irq_save(flags);
6034 check_flags(flags);
6035
6036 lockdep_recursion_inc();
6037 cookie = __lock_pin_lock(lock);
6038 lockdep_recursion_finish();
6039 raw_local_irq_restore(flags);
6040
6041 return cookie;
6042}
6043EXPORT_SYMBOL_GPL(lock_pin_lock);
6044
6045void lock_repin_lock(struct lockdep_map *lock, struct pin_cookie cookie)
6046{
6047 unsigned long flags;
6048
6049 if (unlikely(!lockdep_enabled()))
6050 return;
6051
6052 raw_local_irq_save(flags);
6053 check_flags(flags);
6054
6055 lockdep_recursion_inc();
6056 __lock_repin_lock(lock, cookie);
6057 lockdep_recursion_finish();
6058 raw_local_irq_restore(flags);
6059}
6060EXPORT_SYMBOL_GPL(lock_repin_lock);
6061
6062void lock_unpin_lock(struct lockdep_map *lock, struct pin_cookie cookie)
6063{
6064 unsigned long flags;
6065
6066 if (unlikely(!lockdep_enabled()))
6067 return;
6068
6069 raw_local_irq_save(flags);
6070 check_flags(flags);
6071
6072 lockdep_recursion_inc();
6073 __lock_unpin_lock(lock, cookie);
6074 lockdep_recursion_finish();
6075 raw_local_irq_restore(flags);
6076}
6077EXPORT_SYMBOL_GPL(lock_unpin_lock);
6078
6079u32 lock_sequence(struct lockdep_map *lock)
6080{
6081 unsigned long flags;
6082 u32 seq = ~0;
6083
6084 if (unlikely(!lockdep_enabled()))
6085 return seq;
6086
6087 raw_local_irq_save(flags);
6088 check_flags(flags);
6089
6090 lockdep_recursion_inc();
6091 seq = __lock_sequence(lock);
6092 lockdep_recursion_finish();
6093 raw_local_irq_restore(flags);
6094
6095 return seq;
6096}
6097EXPORT_SYMBOL_GPL(lock_sequence);
6098
6099#ifdef CONFIG_LOCK_STAT
6100static void print_lock_contention_bug(struct task_struct *curr,
6101 struct lockdep_map *lock,
6102 unsigned long ip)
6103{
6104 if (!debug_locks_off())
6105 return;
6106 if (debug_locks_silent)
6107 return;
6108
6109 nbcon_cpu_emergency_enter();
6110
6111 pr_warn("\n");
6112 pr_warn("=================================\n");
6113 pr_warn("WARNING: bad contention detected!\n");
6114 print_kernel_ident();
6115 pr_warn("---------------------------------\n");
6116 pr_warn("%s/%d is trying to contend lock (",
6117 curr->comm, task_pid_nr(curr));
6118 print_lockdep_cache(lock);
6119 pr_cont(") at:\n");
6120 print_ip_sym(KERN_WARNING, ip);
6121 pr_warn("but there are no locks held!\n");
6122 pr_warn("\nother info that might help us debug this:\n");
6123 lockdep_print_held_locks(curr);
6124
6125 pr_warn("\nstack backtrace:\n");
6126 dump_stack();
6127
6128 nbcon_cpu_emergency_exit();
6129}
6130
6131static void
6132__lock_contended(struct lockdep_map *lock, unsigned long ip)
6133{
6134 struct task_struct *curr = current;
6135 struct held_lock *hlock;
6136 struct lock_class_stats *stats;
6137 unsigned int depth;
6138 int i, contention_point, contending_point;
6139
6140 depth = curr->lockdep_depth;
6141 /*
6142 * Whee, we contended on this lock, except it seems we're not
6143 * actually trying to acquire anything much at all..
6144 */
6145 if (DEBUG_LOCKS_WARN_ON(!depth))
6146 return;
6147
6148 if (unlikely(lock->key == &__lockdep_no_track__))
6149 return;
6150
6151 hlock = find_held_lock(curr, lock, depth, &i);
6152 if (!hlock) {
6153 print_lock_contention_bug(curr, lock, ip);
6154 return;
6155 }
6156
6157 if (hlock->instance != lock)
6158 return;
6159
6160 hlock->waittime_stamp = lockstat_clock();
6161
6162 contention_point = lock_point(hlock_class(hlock)->contention_point, ip);
6163 contending_point = lock_point(hlock_class(hlock)->contending_point,
6164 lock->ip);
6165
6166 stats = get_lock_stats(hlock_class(hlock));
6167 if (contention_point < LOCKSTAT_POINTS)
6168 stats->contention_point[contention_point]++;
6169 if (contending_point < LOCKSTAT_POINTS)
6170 stats->contending_point[contending_point]++;
6171 if (lock->cpu != smp_processor_id())
6172 stats->bounces[bounce_contended + !!hlock->read]++;
6173}
6174
6175static void
6176__lock_acquired(struct lockdep_map *lock, unsigned long ip)
6177{
6178 struct task_struct *curr = current;
6179 struct held_lock *hlock;
6180 struct lock_class_stats *stats;
6181 unsigned int depth;
6182 u64 now, waittime = 0;
6183 int i, cpu;
6184
6185 depth = curr->lockdep_depth;
6186 /*
6187 * Yay, we acquired ownership of this lock we didn't try to
6188 * acquire, how the heck did that happen?
6189 */
6190 if (DEBUG_LOCKS_WARN_ON(!depth))
6191 return;
6192
6193 if (unlikely(lock->key == &__lockdep_no_track__))
6194 return;
6195
6196 hlock = find_held_lock(curr, lock, depth, &i);
6197 if (!hlock) {
6198 print_lock_contention_bug(curr, lock, _RET_IP_);
6199 return;
6200 }
6201
6202 if (hlock->instance != lock)
6203 return;
6204
6205 cpu = smp_processor_id();
6206 if (hlock->waittime_stamp) {
6207 now = lockstat_clock();
6208 waittime = now - hlock->waittime_stamp;
6209 hlock->holdtime_stamp = now;
6210 }
6211
6212 stats = get_lock_stats(hlock_class(hlock));
6213 if (waittime) {
6214 if (hlock->read)
6215 lock_time_inc(&stats->read_waittime, waittime);
6216 else
6217 lock_time_inc(&stats->write_waittime, waittime);
6218 }
6219 if (lock->cpu != cpu)
6220 stats->bounces[bounce_acquired + !!hlock->read]++;
6221
6222 lock->cpu = cpu;
6223 lock->ip = ip;
6224}
6225
6226void lock_contended(struct lockdep_map *lock, unsigned long ip)
6227{
6228 unsigned long flags;
6229
6230 trace_lock_contended(lock, ip);
6231
6232 if (unlikely(!lock_stat || !lockdep_enabled()))
6233 return;
6234
6235 raw_local_irq_save(flags);
6236 check_flags(flags);
6237 lockdep_recursion_inc();
6238 __lock_contended(lock, ip);
6239 lockdep_recursion_finish();
6240 raw_local_irq_restore(flags);
6241}
6242EXPORT_SYMBOL_GPL(lock_contended);
6243
6244void lock_acquired(struct lockdep_map *lock, unsigned long ip)
6245{
6246 unsigned long flags;
6247
6248 trace_lock_acquired(lock, ip);
6249
6250 if (unlikely(!lock_stat || !lockdep_enabled()))
6251 return;
6252
6253 raw_local_irq_save(flags);
6254 check_flags(flags);
6255 lockdep_recursion_inc();
6256 __lock_acquired(lock, ip);
6257 lockdep_recursion_finish();
6258 raw_local_irq_restore(flags);
6259}
6260EXPORT_SYMBOL_GPL(lock_acquired);
6261#endif
6262
6263/*
6264 * Used by the testsuite, sanitize the validator state
6265 * after a simulated failure:
6266 */
6267
6268void lockdep_reset(void)
6269{
6270 unsigned long flags;
6271 int i;
6272
6273 raw_local_irq_save(flags);
6274 lockdep_init_task(current);
6275 memset(current->held_locks, 0, MAX_LOCK_DEPTH*sizeof(struct held_lock));
6276 nr_hardirq_chains = 0;
6277 nr_softirq_chains = 0;
6278 nr_process_chains = 0;
6279 debug_locks = 1;
6280 for (i = 0; i < CHAINHASH_SIZE; i++)
6281 INIT_HLIST_HEAD(chainhash_table + i);
6282 raw_local_irq_restore(flags);
6283}
6284
6285/* Remove a class from a lock chain. Must be called with the graph lock held. */
6286static void remove_class_from_lock_chain(struct pending_free *pf,
6287 struct lock_chain *chain,
6288 struct lock_class *class)
6289{
6290#ifdef CONFIG_PROVE_LOCKING
6291 int i;
6292
6293 for (i = chain->base; i < chain->base + chain->depth; i++) {
6294 if (chain_hlock_class_idx(chain_hlocks[i]) != class - lock_classes)
6295 continue;
6296 /*
6297 * Each lock class occurs at most once in a lock chain so once
6298 * we found a match we can break out of this loop.
6299 */
6300 goto free_lock_chain;
6301 }
6302 /* Since the chain has not been modified, return. */
6303 return;
6304
6305free_lock_chain:
6306 free_chain_hlocks(chain->base, chain->depth);
6307 /* Overwrite the chain key for concurrent RCU readers. */
6308 WRITE_ONCE(chain->chain_key, INITIAL_CHAIN_KEY);
6309 dec_chains(chain->irq_context);
6310
6311 /*
6312 * Note: calling hlist_del_rcu() from inside a
6313 * hlist_for_each_entry_rcu() loop is safe.
6314 */
6315 hlist_del_rcu(&chain->entry);
6316 __set_bit(chain - lock_chains, pf->lock_chains_being_freed);
6317 nr_zapped_lock_chains++;
6318#endif
6319}
6320
6321/* Must be called with the graph lock held. */
6322static void remove_class_from_lock_chains(struct pending_free *pf,
6323 struct lock_class *class)
6324{
6325 struct lock_chain *chain;
6326 struct hlist_head *head;
6327 int i;
6328
6329 for (i = 0; i < ARRAY_SIZE(chainhash_table); i++) {
6330 head = chainhash_table + i;
6331 hlist_for_each_entry_rcu(chain, head, entry) {
6332 remove_class_from_lock_chain(pf, chain, class);
6333 }
6334 }
6335}
6336
6337/*
6338 * Remove all references to a lock class. The caller must hold the graph lock.
6339 */
6340static void zap_class(struct pending_free *pf, struct lock_class *class)
6341{
6342 struct lock_list *entry;
6343 int i;
6344
6345 WARN_ON_ONCE(!class->key);
6346
6347 /*
6348 * Remove all dependencies this lock is
6349 * involved in:
6350 */
6351 for_each_set_bit(i, list_entries_in_use, ARRAY_SIZE(list_entries)) {
6352 entry = list_entries + i;
6353 if (entry->class != class && entry->links_to != class)
6354 continue;
6355 __clear_bit(i, list_entries_in_use);
6356 nr_list_entries--;
6357 list_del_rcu(&entry->entry);
6358 }
6359 if (list_empty(&class->locks_after) &&
6360 list_empty(&class->locks_before)) {
6361 list_move_tail(&class->lock_entry, &pf->zapped);
6362 hlist_del_rcu(&class->hash_entry);
6363 WRITE_ONCE(class->key, NULL);
6364 WRITE_ONCE(class->name, NULL);
6365 /* Class allocated but not used, -1 in nr_unused_locks */
6366 if (class->usage_mask == 0)
6367 debug_atomic_dec(nr_unused_locks);
6368 nr_lock_classes--;
6369 __clear_bit(class - lock_classes, lock_classes_in_use);
6370 if (class - lock_classes == max_lock_class_idx)
6371 max_lock_class_idx--;
6372 } else {
6373 WARN_ONCE(true, "%s() failed for class %s\n", __func__,
6374 class->name);
6375 }
6376
6377 remove_class_from_lock_chains(pf, class);
6378 nr_zapped_classes++;
6379}
6380
6381static void reinit_class(struct lock_class *class)
6382{
6383 WARN_ON_ONCE(!class->lock_entry.next);
6384 WARN_ON_ONCE(!list_empty(&class->locks_after));
6385 WARN_ON_ONCE(!list_empty(&class->locks_before));
6386 memset_startat(class, 0, key);
6387 WARN_ON_ONCE(!class->lock_entry.next);
6388 WARN_ON_ONCE(!list_empty(&class->locks_after));
6389 WARN_ON_ONCE(!list_empty(&class->locks_before));
6390}
6391
6392static inline int within(const void *addr, void *start, unsigned long size)
6393{
6394 return addr >= start && addr < start + size;
6395}
6396
6397static bool inside_selftest(void)
6398{
6399 return current == lockdep_selftest_task_struct;
6400}
6401
6402/* The caller must hold the graph lock. */
6403static struct pending_free *get_pending_free(void)
6404{
6405 return delayed_free.pf + delayed_free.index;
6406}
6407
6408static void free_zapped_rcu(struct rcu_head *cb);
6409
6410/*
6411* See if we need to queue an RCU callback, must called with
6412* the lockdep lock held, returns false if either we don't have
6413* any pending free or the callback is already scheduled.
6414* Otherwise, a call_rcu() must follow this function call.
6415*/
6416static bool prepare_call_rcu_zapped(struct pending_free *pf)
6417{
6418 WARN_ON_ONCE(inside_selftest());
6419
6420 if (list_empty(&pf->zapped))
6421 return false;
6422
6423 if (delayed_free.scheduled)
6424 return false;
6425
6426 delayed_free.scheduled = true;
6427
6428 WARN_ON_ONCE(delayed_free.pf + delayed_free.index != pf);
6429 delayed_free.index ^= 1;
6430
6431 return true;
6432}
6433
6434/* The caller must hold the graph lock. May be called from RCU context. */
6435static void __free_zapped_classes(struct pending_free *pf)
6436{
6437 struct lock_class *class;
6438
6439 check_data_structures();
6440
6441 list_for_each_entry(class, &pf->zapped, lock_entry)
6442 reinit_class(class);
6443
6444 list_splice_init(&pf->zapped, &free_lock_classes);
6445
6446#ifdef CONFIG_PROVE_LOCKING
6447 bitmap_andnot(lock_chains_in_use, lock_chains_in_use,
6448 pf->lock_chains_being_freed, ARRAY_SIZE(lock_chains));
6449 bitmap_clear(pf->lock_chains_being_freed, 0, ARRAY_SIZE(lock_chains));
6450#endif
6451}
6452
6453static void free_zapped_rcu(struct rcu_head *ch)
6454{
6455 struct pending_free *pf;
6456 unsigned long flags;
6457 bool need_callback;
6458
6459 if (WARN_ON_ONCE(ch != &delayed_free.rcu_head))
6460 return;
6461
6462 raw_local_irq_save(flags);
6463 lockdep_lock();
6464
6465 /* closed head */
6466 pf = delayed_free.pf + (delayed_free.index ^ 1);
6467 __free_zapped_classes(pf);
6468 delayed_free.scheduled = false;
6469 need_callback =
6470 prepare_call_rcu_zapped(delayed_free.pf + delayed_free.index);
6471 lockdep_unlock();
6472 raw_local_irq_restore(flags);
6473
6474 /*
6475 * If there's pending free and its callback has not been scheduled,
6476 * queue an RCU callback.
6477 */
6478 if (need_callback)
6479 call_rcu(&delayed_free.rcu_head, free_zapped_rcu);
6480
6481}
6482
6483/*
6484 * Remove all lock classes from the class hash table and from the
6485 * all_lock_classes list whose key or name is in the address range [start,
6486 * start + size). Move these lock classes to the zapped_classes list. Must
6487 * be called with the graph lock held.
6488 */
6489static void __lockdep_free_key_range(struct pending_free *pf, void *start,
6490 unsigned long size)
6491{
6492 struct lock_class *class;
6493 struct hlist_head *head;
6494 int i;
6495
6496 /* Unhash all classes that were created by a module. */
6497 for (i = 0; i < CLASSHASH_SIZE; i++) {
6498 head = classhash_table + i;
6499 hlist_for_each_entry_rcu(class, head, hash_entry) {
6500 if (!within(class->key, start, size) &&
6501 !within(class->name, start, size))
6502 continue;
6503 zap_class(pf, class);
6504 }
6505 }
6506}
6507
6508/*
6509 * Used in module.c to remove lock classes from memory that is going to be
6510 * freed; and possibly re-used by other modules.
6511 *
6512 * We will have had one synchronize_rcu() before getting here, so we're
6513 * guaranteed nobody will look up these exact classes -- they're properly dead
6514 * but still allocated.
6515 */
6516static void lockdep_free_key_range_reg(void *start, unsigned long size)
6517{
6518 struct pending_free *pf;
6519 unsigned long flags;
6520 bool need_callback;
6521
6522 init_data_structures_once();
6523
6524 raw_local_irq_save(flags);
6525 lockdep_lock();
6526 pf = get_pending_free();
6527 __lockdep_free_key_range(pf, start, size);
6528 need_callback = prepare_call_rcu_zapped(pf);
6529 lockdep_unlock();
6530 raw_local_irq_restore(flags);
6531 if (need_callback)
6532 call_rcu(&delayed_free.rcu_head, free_zapped_rcu);
6533 /*
6534 * Wait for any possible iterators from look_up_lock_class() to pass
6535 * before continuing to free the memory they refer to.
6536 */
6537 synchronize_rcu();
6538}
6539
6540/*
6541 * Free all lockdep keys in the range [start, start+size). Does not sleep.
6542 * Ignores debug_locks. Must only be used by the lockdep selftests.
6543 */
6544static void lockdep_free_key_range_imm(void *start, unsigned long size)
6545{
6546 struct pending_free *pf = delayed_free.pf;
6547 unsigned long flags;
6548
6549 init_data_structures_once();
6550
6551 raw_local_irq_save(flags);
6552 lockdep_lock();
6553 __lockdep_free_key_range(pf, start, size);
6554 __free_zapped_classes(pf);
6555 lockdep_unlock();
6556 raw_local_irq_restore(flags);
6557}
6558
6559void lockdep_free_key_range(void *start, unsigned long size)
6560{
6561 init_data_structures_once();
6562
6563 if (inside_selftest())
6564 lockdep_free_key_range_imm(start, size);
6565 else
6566 lockdep_free_key_range_reg(start, size);
6567}
6568
6569/*
6570 * Check whether any element of the @lock->class_cache[] array refers to a
6571 * registered lock class. The caller must hold either the graph lock or the
6572 * RCU read lock.
6573 */
6574static bool lock_class_cache_is_registered(struct lockdep_map *lock)
6575{
6576 struct lock_class *class;
6577 struct hlist_head *head;
6578 int i, j;
6579
6580 for (i = 0; i < CLASSHASH_SIZE; i++) {
6581 head = classhash_table + i;
6582 hlist_for_each_entry_rcu(class, head, hash_entry) {
6583 for (j = 0; j < NR_LOCKDEP_CACHING_CLASSES; j++)
6584 if (lock->class_cache[j] == class)
6585 return true;
6586 }
6587 }
6588 return false;
6589}
6590
6591/* The caller must hold the graph lock. Does not sleep. */
6592static void __lockdep_reset_lock(struct pending_free *pf,
6593 struct lockdep_map *lock)
6594{
6595 struct lock_class *class;
6596 int j;
6597
6598 /*
6599 * Remove all classes this lock might have:
6600 */
6601 for (j = 0; j < MAX_LOCKDEP_SUBCLASSES; j++) {
6602 /*
6603 * If the class exists we look it up and zap it:
6604 */
6605 class = look_up_lock_class(lock, j);
6606 if (class)
6607 zap_class(pf, class);
6608 }
6609 /*
6610 * Debug check: in the end all mapped classes should
6611 * be gone.
6612 */
6613 if (WARN_ON_ONCE(lock_class_cache_is_registered(lock)))
6614 debug_locks_off();
6615}
6616
6617/*
6618 * Remove all information lockdep has about a lock if debug_locks == 1. Free
6619 * released data structures from RCU context.
6620 */
6621static void lockdep_reset_lock_reg(struct lockdep_map *lock)
6622{
6623 struct pending_free *pf;
6624 unsigned long flags;
6625 int locked;
6626 bool need_callback = false;
6627
6628 raw_local_irq_save(flags);
6629 locked = graph_lock();
6630 if (!locked)
6631 goto out_irq;
6632
6633 pf = get_pending_free();
6634 __lockdep_reset_lock(pf, lock);
6635 need_callback = prepare_call_rcu_zapped(pf);
6636
6637 graph_unlock();
6638out_irq:
6639 raw_local_irq_restore(flags);
6640 if (need_callback)
6641 call_rcu(&delayed_free.rcu_head, free_zapped_rcu);
6642}
6643
6644/*
6645 * Reset a lock. Does not sleep. Ignores debug_locks. Must only be used by the
6646 * lockdep selftests.
6647 */
6648static void lockdep_reset_lock_imm(struct lockdep_map *lock)
6649{
6650 struct pending_free *pf = delayed_free.pf;
6651 unsigned long flags;
6652
6653 raw_local_irq_save(flags);
6654 lockdep_lock();
6655 __lockdep_reset_lock(pf, lock);
6656 __free_zapped_classes(pf);
6657 lockdep_unlock();
6658 raw_local_irq_restore(flags);
6659}
6660
6661void lockdep_reset_lock(struct lockdep_map *lock)
6662{
6663 init_data_structures_once();
6664
6665 if (inside_selftest())
6666 lockdep_reset_lock_imm(lock);
6667 else
6668 lockdep_reset_lock_reg(lock);
6669}
6670
6671/*
6672 * Unregister a dynamically allocated key.
6673 *
6674 * Unlike lockdep_register_key(), a search is always done to find a matching
6675 * key irrespective of debug_locks to avoid potential invalid access to freed
6676 * memory in lock_class entry.
6677 */
6678void lockdep_unregister_key(struct lock_class_key *key)
6679{
6680 struct hlist_head *hash_head = keyhashentry(key);
6681 struct lock_class_key *k;
6682 struct pending_free *pf;
6683 unsigned long flags;
6684 bool found = false;
6685 bool need_callback = false;
6686
6687 might_sleep();
6688
6689 if (WARN_ON_ONCE(static_obj(key)))
6690 return;
6691
6692 raw_local_irq_save(flags);
6693 lockdep_lock();
6694
6695 hlist_for_each_entry_rcu(k, hash_head, hash_entry) {
6696 if (k == key) {
6697 hlist_del_rcu(&k->hash_entry);
6698 found = true;
6699 break;
6700 }
6701 }
6702 WARN_ON_ONCE(!found && debug_locks);
6703 if (found) {
6704 pf = get_pending_free();
6705 __lockdep_free_key_range(pf, key, 1);
6706 need_callback = prepare_call_rcu_zapped(pf);
6707 nr_dynamic_keys--;
6708 }
6709 lockdep_unlock();
6710 raw_local_irq_restore(flags);
6711
6712 if (need_callback)
6713 call_rcu(&delayed_free.rcu_head, free_zapped_rcu);
6714
6715 /*
6716 * Wait until is_dynamic_key() has finished accessing k->hash_entry.
6717 *
6718 * Some operations like __qdisc_destroy() will call this in a debug
6719 * kernel, and the network traffic is disabled while waiting, hence
6720 * the delay of the wait matters in debugging cases. Currently use a
6721 * synchronize_rcu_expedited() to speed up the wait at the cost of
6722 * system IPIs. TODO: Replace RCU with hazptr for this.
6723 */
6724 synchronize_rcu_expedited();
6725}
6726EXPORT_SYMBOL_GPL(lockdep_unregister_key);
6727
6728void __init lockdep_init(void)
6729{
6730 pr_info("Lock dependency validator: Copyright (c) 2006 Red Hat, Inc., Ingo Molnar\n");
6731
6732 pr_info("... MAX_LOCKDEP_SUBCLASSES: %lu\n", MAX_LOCKDEP_SUBCLASSES);
6733 pr_info("... MAX_LOCK_DEPTH: %lu\n", MAX_LOCK_DEPTH);
6734 pr_info("... MAX_LOCKDEP_KEYS: %lu\n", MAX_LOCKDEP_KEYS);
6735 pr_info("... CLASSHASH_SIZE: %lu\n", CLASSHASH_SIZE);
6736 pr_info("... MAX_LOCKDEP_ENTRIES: %lu\n", MAX_LOCKDEP_ENTRIES);
6737 pr_info("... MAX_LOCKDEP_CHAINS: %lu\n", MAX_LOCKDEP_CHAINS);
6738 pr_info("... CHAINHASH_SIZE: %lu\n", CHAINHASH_SIZE);
6739
6740 pr_info(" memory used by lock dependency info: %zu kB\n",
6741 (sizeof(lock_classes) +
6742 sizeof(lock_classes_in_use) +
6743 sizeof(classhash_table) +
6744 sizeof(list_entries) +
6745 sizeof(list_entries_in_use) +
6746 sizeof(chainhash_table) +
6747 sizeof(delayed_free)
6748#ifdef CONFIG_PROVE_LOCKING
6749 + sizeof(lock_cq)
6750 + sizeof(lock_chains)
6751 + sizeof(lock_chains_in_use)
6752 + sizeof(chain_hlocks)
6753#endif
6754 ) / 1024
6755 );
6756
6757#if defined(CONFIG_TRACE_IRQFLAGS) && defined(CONFIG_PROVE_LOCKING)
6758 pr_info(" memory used for stack traces: %zu kB\n",
6759 (sizeof(stack_trace) + sizeof(stack_trace_hash)) / 1024
6760 );
6761#endif
6762
6763 pr_info(" per task-struct memory footprint: %zu bytes\n",
6764 sizeof(((struct task_struct *)NULL)->held_locks));
6765}
6766
6767static void
6768print_freed_lock_bug(struct task_struct *curr, const void *mem_from,
6769 const void *mem_to, struct held_lock *hlock)
6770{
6771 if (!debug_locks_off())
6772 return;
6773 if (debug_locks_silent)
6774 return;
6775
6776 nbcon_cpu_emergency_enter();
6777
6778 pr_warn("\n");
6779 pr_warn("=========================\n");
6780 pr_warn("WARNING: held lock freed!\n");
6781 print_kernel_ident();
6782 pr_warn("-------------------------\n");
6783 pr_warn("%s/%d is freeing memory %px-%px, with a lock still held there!\n",
6784 curr->comm, task_pid_nr(curr), mem_from, mem_to-1);
6785 print_lock(hlock);
6786 lockdep_print_held_locks(curr);
6787
6788 pr_warn("\nstack backtrace:\n");
6789 dump_stack();
6790
6791 nbcon_cpu_emergency_exit();
6792}
6793
6794static inline int not_in_range(const void* mem_from, unsigned long mem_len,
6795 const void* lock_from, unsigned long lock_len)
6796{
6797 return lock_from + lock_len <= mem_from ||
6798 mem_from + mem_len <= lock_from;
6799}
6800
6801/*
6802 * Called when kernel memory is freed (or unmapped), or if a lock
6803 * is destroyed or reinitialized - this code checks whether there is
6804 * any held lock in the memory range of <from> to <to>:
6805 */
6806void debug_check_no_locks_freed(const void *mem_from, unsigned long mem_len)
6807{
6808 struct task_struct *curr = current;
6809 struct held_lock *hlock;
6810 unsigned long flags;
6811 int i;
6812
6813 if (unlikely(!debug_locks))
6814 return;
6815
6816 raw_local_irq_save(flags);
6817 for (i = 0; i < curr->lockdep_depth; i++) {
6818 hlock = curr->held_locks + i;
6819
6820 if (not_in_range(mem_from, mem_len, hlock->instance,
6821 sizeof(*hlock->instance)))
6822 continue;
6823
6824 print_freed_lock_bug(curr, mem_from, mem_from + mem_len, hlock);
6825 break;
6826 }
6827 raw_local_irq_restore(flags);
6828}
6829EXPORT_SYMBOL_GPL(debug_check_no_locks_freed);
6830
6831static void print_held_locks_bug(void)
6832{
6833 if (!debug_locks_off())
6834 return;
6835 if (debug_locks_silent)
6836 return;
6837
6838 nbcon_cpu_emergency_enter();
6839
6840 pr_warn("\n");
6841 pr_warn("====================================\n");
6842 pr_warn("WARNING: %s/%d still has locks held!\n",
6843 current->comm, task_pid_nr(current));
6844 print_kernel_ident();
6845 pr_warn("------------------------------------\n");
6846 lockdep_print_held_locks(current);
6847 pr_warn("\nstack backtrace:\n");
6848 dump_stack();
6849
6850 nbcon_cpu_emergency_exit();
6851}
6852
6853void debug_check_no_locks_held(void)
6854{
6855 if (unlikely(current->lockdep_depth > 0))
6856 print_held_locks_bug();
6857}
6858EXPORT_SYMBOL_GPL(debug_check_no_locks_held);
6859
6860#ifdef __KERNEL__
6861void debug_show_all_locks(void)
6862{
6863 struct task_struct *g, *p;
6864
6865 if (unlikely(!debug_locks)) {
6866 pr_warn("INFO: lockdep is turned off.\n");
6867 return;
6868 }
6869 pr_warn("\nShowing all locks held in the system:\n");
6870
6871 rcu_read_lock();
6872 for_each_process_thread(g, p) {
6873 if (!p->lockdep_depth)
6874 continue;
6875 lockdep_print_held_locks(p);
6876 touch_nmi_watchdog();
6877 touch_all_softlockup_watchdogs();
6878 }
6879 rcu_read_unlock();
6880
6881 pr_warn("\n");
6882 pr_warn("=============================================\n\n");
6883}
6884EXPORT_SYMBOL_GPL(debug_show_all_locks);
6885#endif
6886
6887/*
6888 * Careful: only use this function if you are sure that
6889 * the task cannot run in parallel!
6890 */
6891void debug_show_held_locks(struct task_struct *task)
6892{
6893 if (unlikely(!debug_locks)) {
6894 printk("INFO: lockdep is turned off.\n");
6895 return;
6896 }
6897 lockdep_print_held_locks(task);
6898}
6899EXPORT_SYMBOL_GPL(debug_show_held_locks);
6900
6901asmlinkage __visible void lockdep_sys_exit(void)
6902{
6903 struct task_struct *curr = current;
6904
6905 if (unlikely(curr->lockdep_depth)) {
6906 if (!debug_locks_off())
6907 return;
6908 nbcon_cpu_emergency_enter();
6909 pr_warn("\n");
6910 pr_warn("================================================\n");
6911 pr_warn("WARNING: lock held when returning to user space!\n");
6912 print_kernel_ident();
6913 pr_warn("------------------------------------------------\n");
6914 pr_warn("%s/%d is leaving the kernel with locks still held!\n",
6915 curr->comm, curr->pid);
6916 lockdep_print_held_locks(curr);
6917 nbcon_cpu_emergency_exit();
6918 }
6919
6920 /*
6921 * The lock history for each syscall should be independent. So wipe the
6922 * slate clean on return to userspace.
6923 */
6924 lockdep_invariant_state(false);
6925}
6926
6927void lockdep_rcu_suspicious(const char *file, const int line, const char *s)
6928{
6929 struct task_struct *curr = current;
6930 int dl = READ_ONCE(debug_locks);
6931 bool rcu = warn_rcu_enter();
6932
6933 /* Note: the following can be executed concurrently, so be careful. */
6934 nbcon_cpu_emergency_enter();
6935 pr_warn("\n");
6936 pr_warn("=============================\n");
6937 pr_warn("WARNING: suspicious RCU usage\n");
6938 print_kernel_ident();
6939 pr_warn("-----------------------------\n");
6940 pr_warn("%s:%d %s!\n", file, line, s);
6941 pr_warn("\nother info that might help us debug this:\n\n");
6942 pr_warn("\n%srcu_scheduler_active = %d, debug_locks = %d\n%s",
6943 !rcu_lockdep_current_cpu_online()
6944 ? "RCU used illegally from offline CPU!\n"
6945 : "",
6946 rcu_scheduler_active, dl,
6947 dl ? "" : "Possible false positive due to lockdep disabling via debug_locks = 0\n");
6948
6949 /*
6950 * If a CPU is in the RCU-free window in idle (ie: in the section
6951 * between ct_idle_enter() and ct_idle_exit(), then RCU
6952 * considers that CPU to be in an "extended quiescent state",
6953 * which means that RCU will be completely ignoring that CPU.
6954 * Therefore, rcu_read_lock() and friends have absolutely no
6955 * effect on a CPU running in that state. In other words, even if
6956 * such an RCU-idle CPU has called rcu_read_lock(), RCU might well
6957 * delete data structures out from under it. RCU really has no
6958 * choice here: we need to keep an RCU-free window in idle where
6959 * the CPU may possibly enter into low power mode. This way we can
6960 * notice an extended quiescent state to other CPUs that started a grace
6961 * period. Otherwise we would delay any grace period as long as we run
6962 * in the idle task.
6963 *
6964 * So complain bitterly if someone does call rcu_read_lock(),
6965 * rcu_read_lock_bh() and so on from extended quiescent states.
6966 */
6967 if (!rcu_is_watching())
6968 pr_warn("RCU used illegally from extended quiescent state!\n");
6969
6970 lockdep_print_held_locks(curr);
6971 pr_warn("\nstack backtrace:\n");
6972 dump_stack();
6973 nbcon_cpu_emergency_exit();
6974 warn_rcu_exit(rcu);
6975}
6976EXPORT_SYMBOL_GPL(lockdep_rcu_suspicious);