xref: /openbmc/linux/kernel/locking/lockdep.c (revision b8d312aa)
1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * kernel/lockdep.c
4  *
5  * Runtime locking correctness validator
6  *
7  * Started by Ingo Molnar:
8  *
9  *  Copyright (C) 2006,2007 Red Hat, Inc., Ingo Molnar <mingo@redhat.com>
10  *  Copyright (C) 2007 Red Hat, Inc., Peter Zijlstra
11  *
12  * this code maps all the lock dependencies as they occur in a live kernel
13  * and will warn about the following classes of locking bugs:
14  *
15  * - lock inversion scenarios
16  * - circular lock dependencies
17  * - hardirq/softirq safe/unsafe locking bugs
18  *
19  * Bugs are reported even if the current locking scenario does not cause
20  * any deadlock at this point.
21  *
22  * I.e. if anytime in the past two locks were taken in a different order,
23  * even if it happened for another task, even if those were different
24  * locks (but of the same class as this lock), this code will detect it.
25  *
26  * Thanks to Arjan van de Ven for coming up with the initial idea of
27  * mapping lock dependencies runtime.
28  */
29 #define DISABLE_BRANCH_PROFILING
30 #include <linux/mutex.h>
31 #include <linux/sched.h>
32 #include <linux/sched/clock.h>
33 #include <linux/sched/task.h>
34 #include <linux/sched/mm.h>
35 #include <linux/delay.h>
36 #include <linux/module.h>
37 #include <linux/proc_fs.h>
38 #include <linux/seq_file.h>
39 #include <linux/spinlock.h>
40 #include <linux/kallsyms.h>
41 #include <linux/interrupt.h>
42 #include <linux/stacktrace.h>
43 #include <linux/debug_locks.h>
44 #include <linux/irqflags.h>
45 #include <linux/utsname.h>
46 #include <linux/hash.h>
47 #include <linux/ftrace.h>
48 #include <linux/stringify.h>
49 #include <linux/bitmap.h>
50 #include <linux/bitops.h>
51 #include <linux/gfp.h>
52 #include <linux/random.h>
53 #include <linux/jhash.h>
54 #include <linux/nmi.h>
55 #include <linux/rcupdate.h>
56 #include <linux/kprobes.h>
57 
58 #include <asm/sections.h>
59 
60 #include "lockdep_internals.h"
61 
62 #define CREATE_TRACE_POINTS
63 #include <trace/events/lock.h>
64 
65 #ifdef CONFIG_PROVE_LOCKING
66 int prove_locking = 1;
67 module_param(prove_locking, int, 0644);
68 #else
69 #define prove_locking 0
70 #endif
71 
72 #ifdef CONFIG_LOCK_STAT
73 int lock_stat = 1;
74 module_param(lock_stat, int, 0644);
75 #else
76 #define lock_stat 0
77 #endif
78 
79 /*
80  * lockdep_lock: protects the lockdep graph, the hashes and the
81  *               class/list/hash allocators.
82  *
83  * This is one of the rare exceptions where it's justified
84  * to use a raw spinlock - we really dont want the spinlock
85  * code to recurse back into the lockdep code...
86  */
87 static arch_spinlock_t lockdep_lock = (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
88 static struct task_struct *lockdep_selftest_task_struct;
89 
90 static int graph_lock(void)
91 {
92 	arch_spin_lock(&lockdep_lock);
93 	/*
94 	 * Make sure that if another CPU detected a bug while
95 	 * walking the graph we dont change it (while the other
96 	 * CPU is busy printing out stuff with the graph lock
97 	 * dropped already)
98 	 */
99 	if (!debug_locks) {
100 		arch_spin_unlock(&lockdep_lock);
101 		return 0;
102 	}
103 	/* prevent any recursions within lockdep from causing deadlocks */
104 	current->lockdep_recursion++;
105 	return 1;
106 }
107 
108 static inline int graph_unlock(void)
109 {
110 	if (debug_locks && !arch_spin_is_locked(&lockdep_lock)) {
111 		/*
112 		 * The lockdep graph lock isn't locked while we expect it to
113 		 * be, we're confused now, bye!
114 		 */
115 		return DEBUG_LOCKS_WARN_ON(1);
116 	}
117 
118 	current->lockdep_recursion--;
119 	arch_spin_unlock(&lockdep_lock);
120 	return 0;
121 }
122 
123 /*
124  * Turn lock debugging off and return with 0 if it was off already,
125  * and also release the graph lock:
126  */
127 static inline int debug_locks_off_graph_unlock(void)
128 {
129 	int ret = debug_locks_off();
130 
131 	arch_spin_unlock(&lockdep_lock);
132 
133 	return ret;
134 }
135 
136 unsigned long nr_list_entries;
137 static struct lock_list list_entries[MAX_LOCKDEP_ENTRIES];
138 static DECLARE_BITMAP(list_entries_in_use, MAX_LOCKDEP_ENTRIES);
139 
140 /*
141  * All data structures here are protected by the global debug_lock.
142  *
143  * nr_lock_classes is the number of elements of lock_classes[] that is
144  * in use.
145  */
146 #define KEYHASH_BITS		(MAX_LOCKDEP_KEYS_BITS - 1)
147 #define KEYHASH_SIZE		(1UL << KEYHASH_BITS)
148 static struct hlist_head lock_keys_hash[KEYHASH_SIZE];
149 unsigned long nr_lock_classes;
150 #ifndef CONFIG_DEBUG_LOCKDEP
151 static
152 #endif
153 struct lock_class lock_classes[MAX_LOCKDEP_KEYS];
154 static DECLARE_BITMAP(lock_classes_in_use, MAX_LOCKDEP_KEYS);
155 
156 static inline struct lock_class *hlock_class(struct held_lock *hlock)
157 {
158 	unsigned int class_idx = hlock->class_idx;
159 
160 	/* Don't re-read hlock->class_idx, can't use READ_ONCE() on bitfield */
161 	barrier();
162 
163 	if (!test_bit(class_idx, lock_classes_in_use)) {
164 		/*
165 		 * Someone passed in garbage, we give up.
166 		 */
167 		DEBUG_LOCKS_WARN_ON(1);
168 		return NULL;
169 	}
170 
171 	/*
172 	 * At this point, if the passed hlock->class_idx is still garbage,
173 	 * we just have to live with it
174 	 */
175 	return lock_classes + class_idx;
176 }
177 
178 #ifdef CONFIG_LOCK_STAT
179 static DEFINE_PER_CPU(struct lock_class_stats[MAX_LOCKDEP_KEYS], cpu_lock_stats);
180 
181 static inline u64 lockstat_clock(void)
182 {
183 	return local_clock();
184 }
185 
186 static int lock_point(unsigned long points[], unsigned long ip)
187 {
188 	int i;
189 
190 	for (i = 0; i < LOCKSTAT_POINTS; i++) {
191 		if (points[i] == 0) {
192 			points[i] = ip;
193 			break;
194 		}
195 		if (points[i] == ip)
196 			break;
197 	}
198 
199 	return i;
200 }
201 
202 static void lock_time_inc(struct lock_time *lt, u64 time)
203 {
204 	if (time > lt->max)
205 		lt->max = time;
206 
207 	if (time < lt->min || !lt->nr)
208 		lt->min = time;
209 
210 	lt->total += time;
211 	lt->nr++;
212 }
213 
214 static inline void lock_time_add(struct lock_time *src, struct lock_time *dst)
215 {
216 	if (!src->nr)
217 		return;
218 
219 	if (src->max > dst->max)
220 		dst->max = src->max;
221 
222 	if (src->min < dst->min || !dst->nr)
223 		dst->min = src->min;
224 
225 	dst->total += src->total;
226 	dst->nr += src->nr;
227 }
228 
229 struct lock_class_stats lock_stats(struct lock_class *class)
230 {
231 	struct lock_class_stats stats;
232 	int cpu, i;
233 
234 	memset(&stats, 0, sizeof(struct lock_class_stats));
235 	for_each_possible_cpu(cpu) {
236 		struct lock_class_stats *pcs =
237 			&per_cpu(cpu_lock_stats, cpu)[class - lock_classes];
238 
239 		for (i = 0; i < ARRAY_SIZE(stats.contention_point); i++)
240 			stats.contention_point[i] += pcs->contention_point[i];
241 
242 		for (i = 0; i < ARRAY_SIZE(stats.contending_point); i++)
243 			stats.contending_point[i] += pcs->contending_point[i];
244 
245 		lock_time_add(&pcs->read_waittime, &stats.read_waittime);
246 		lock_time_add(&pcs->write_waittime, &stats.write_waittime);
247 
248 		lock_time_add(&pcs->read_holdtime, &stats.read_holdtime);
249 		lock_time_add(&pcs->write_holdtime, &stats.write_holdtime);
250 
251 		for (i = 0; i < ARRAY_SIZE(stats.bounces); i++)
252 			stats.bounces[i] += pcs->bounces[i];
253 	}
254 
255 	return stats;
256 }
257 
258 void clear_lock_stats(struct lock_class *class)
259 {
260 	int cpu;
261 
262 	for_each_possible_cpu(cpu) {
263 		struct lock_class_stats *cpu_stats =
264 			&per_cpu(cpu_lock_stats, cpu)[class - lock_classes];
265 
266 		memset(cpu_stats, 0, sizeof(struct lock_class_stats));
267 	}
268 	memset(class->contention_point, 0, sizeof(class->contention_point));
269 	memset(class->contending_point, 0, sizeof(class->contending_point));
270 }
271 
272 static struct lock_class_stats *get_lock_stats(struct lock_class *class)
273 {
274 	return &this_cpu_ptr(cpu_lock_stats)[class - lock_classes];
275 }
276 
277 static void lock_release_holdtime(struct held_lock *hlock)
278 {
279 	struct lock_class_stats *stats;
280 	u64 holdtime;
281 
282 	if (!lock_stat)
283 		return;
284 
285 	holdtime = lockstat_clock() - hlock->holdtime_stamp;
286 
287 	stats = get_lock_stats(hlock_class(hlock));
288 	if (hlock->read)
289 		lock_time_inc(&stats->read_holdtime, holdtime);
290 	else
291 		lock_time_inc(&stats->write_holdtime, holdtime);
292 }
293 #else
294 static inline void lock_release_holdtime(struct held_lock *hlock)
295 {
296 }
297 #endif
298 
299 /*
300  * We keep a global list of all lock classes. The list is only accessed with
301  * the lockdep spinlock lock held. free_lock_classes is a list with free
302  * elements. These elements are linked together by the lock_entry member in
303  * struct lock_class.
304  */
305 LIST_HEAD(all_lock_classes);
306 static LIST_HEAD(free_lock_classes);
307 
308 /**
309  * struct pending_free - information about data structures about to be freed
310  * @zapped: Head of a list with struct lock_class elements.
311  * @lock_chains_being_freed: Bitmap that indicates which lock_chains[] elements
312  *	are about to be freed.
313  */
314 struct pending_free {
315 	struct list_head zapped;
316 	DECLARE_BITMAP(lock_chains_being_freed, MAX_LOCKDEP_CHAINS);
317 };
318 
319 /**
320  * struct delayed_free - data structures used for delayed freeing
321  *
322  * A data structure for delayed freeing of data structures that may be
323  * accessed by RCU readers at the time these were freed.
324  *
325  * @rcu_head:  Used to schedule an RCU callback for freeing data structures.
326  * @index:     Index of @pf to which freed data structures are added.
327  * @scheduled: Whether or not an RCU callback has been scheduled.
328  * @pf:        Array with information about data structures about to be freed.
329  */
330 static struct delayed_free {
331 	struct rcu_head		rcu_head;
332 	int			index;
333 	int			scheduled;
334 	struct pending_free	pf[2];
335 } delayed_free;
336 
337 /*
338  * The lockdep classes are in a hash-table as well, for fast lookup:
339  */
340 #define CLASSHASH_BITS		(MAX_LOCKDEP_KEYS_BITS - 1)
341 #define CLASSHASH_SIZE		(1UL << CLASSHASH_BITS)
342 #define __classhashfn(key)	hash_long((unsigned long)key, CLASSHASH_BITS)
343 #define classhashentry(key)	(classhash_table + __classhashfn((key)))
344 
345 static struct hlist_head classhash_table[CLASSHASH_SIZE];
346 
347 /*
348  * We put the lock dependency chains into a hash-table as well, to cache
349  * their existence:
350  */
351 #define CHAINHASH_BITS		(MAX_LOCKDEP_CHAINS_BITS-1)
352 #define CHAINHASH_SIZE		(1UL << CHAINHASH_BITS)
353 #define __chainhashfn(chain)	hash_long(chain, CHAINHASH_BITS)
354 #define chainhashentry(chain)	(chainhash_table + __chainhashfn((chain)))
355 
356 static struct hlist_head chainhash_table[CHAINHASH_SIZE];
357 
358 /*
359  * The hash key of the lock dependency chains is a hash itself too:
360  * it's a hash of all locks taken up to that lock, including that lock.
361  * It's a 64-bit hash, because it's important for the keys to be
362  * unique.
363  */
364 static inline u64 iterate_chain_key(u64 key, u32 idx)
365 {
366 	u32 k0 = key, k1 = key >> 32;
367 
368 	__jhash_mix(idx, k0, k1); /* Macro that modifies arguments! */
369 
370 	return k0 | (u64)k1 << 32;
371 }
372 
373 void lockdep_init_task(struct task_struct *task)
374 {
375 	task->lockdep_depth = 0; /* no locks held yet */
376 	task->curr_chain_key = INITIAL_CHAIN_KEY;
377 	task->lockdep_recursion = 0;
378 }
379 
380 void lockdep_off(void)
381 {
382 	current->lockdep_recursion++;
383 }
384 EXPORT_SYMBOL(lockdep_off);
385 
386 void lockdep_on(void)
387 {
388 	current->lockdep_recursion--;
389 }
390 EXPORT_SYMBOL(lockdep_on);
391 
392 void lockdep_set_selftest_task(struct task_struct *task)
393 {
394 	lockdep_selftest_task_struct = task;
395 }
396 
397 /*
398  * Debugging switches:
399  */
400 
401 #define VERBOSE			0
402 #define VERY_VERBOSE		0
403 
404 #if VERBOSE
405 # define HARDIRQ_VERBOSE	1
406 # define SOFTIRQ_VERBOSE	1
407 #else
408 # define HARDIRQ_VERBOSE	0
409 # define SOFTIRQ_VERBOSE	0
410 #endif
411 
412 #if VERBOSE || HARDIRQ_VERBOSE || SOFTIRQ_VERBOSE
413 /*
414  * Quick filtering for interesting events:
415  */
416 static int class_filter(struct lock_class *class)
417 {
418 #if 0
419 	/* Example */
420 	if (class->name_version == 1 &&
421 			!strcmp(class->name, "lockname"))
422 		return 1;
423 	if (class->name_version == 1 &&
424 			!strcmp(class->name, "&struct->lockfield"))
425 		return 1;
426 #endif
427 	/* Filter everything else. 1 would be to allow everything else */
428 	return 0;
429 }
430 #endif
431 
432 static int verbose(struct lock_class *class)
433 {
434 #if VERBOSE
435 	return class_filter(class);
436 #endif
437 	return 0;
438 }
439 
440 static void print_lockdep_off(const char *bug_msg)
441 {
442 	printk(KERN_DEBUG "%s\n", bug_msg);
443 	printk(KERN_DEBUG "turning off the locking correctness validator.\n");
444 #ifdef CONFIG_LOCK_STAT
445 	printk(KERN_DEBUG "Please attach the output of /proc/lock_stat to the bug report\n");
446 #endif
447 }
448 
449 unsigned long nr_stack_trace_entries;
450 
451 #ifdef CONFIG_PROVE_LOCKING
452 /*
453  * Stack-trace: tightly packed array of stack backtrace
454  * addresses. Protected by the graph_lock.
455  */
456 static unsigned long stack_trace[MAX_STACK_TRACE_ENTRIES];
457 
458 static int save_trace(struct lock_trace *trace)
459 {
460 	unsigned long *entries = stack_trace + nr_stack_trace_entries;
461 	unsigned int max_entries;
462 
463 	trace->offset = nr_stack_trace_entries;
464 	max_entries = MAX_STACK_TRACE_ENTRIES - nr_stack_trace_entries;
465 	trace->nr_entries = stack_trace_save(entries, max_entries, 3);
466 	nr_stack_trace_entries += trace->nr_entries;
467 
468 	if (nr_stack_trace_entries >= MAX_STACK_TRACE_ENTRIES-1) {
469 		if (!debug_locks_off_graph_unlock())
470 			return 0;
471 
472 		print_lockdep_off("BUG: MAX_STACK_TRACE_ENTRIES too low!");
473 		dump_stack();
474 
475 		return 0;
476 	}
477 
478 	return 1;
479 }
480 #endif
481 
482 unsigned int nr_hardirq_chains;
483 unsigned int nr_softirq_chains;
484 unsigned int nr_process_chains;
485 unsigned int max_lockdep_depth;
486 
487 #ifdef CONFIG_DEBUG_LOCKDEP
488 /*
489  * Various lockdep statistics:
490  */
491 DEFINE_PER_CPU(struct lockdep_stats, lockdep_stats);
492 #endif
493 
494 #ifdef CONFIG_PROVE_LOCKING
495 /*
496  * Locking printouts:
497  */
498 
499 #define __USAGE(__STATE)						\
500 	[LOCK_USED_IN_##__STATE] = "IN-"__stringify(__STATE)"-W",	\
501 	[LOCK_ENABLED_##__STATE] = __stringify(__STATE)"-ON-W",		\
502 	[LOCK_USED_IN_##__STATE##_READ] = "IN-"__stringify(__STATE)"-R",\
503 	[LOCK_ENABLED_##__STATE##_READ] = __stringify(__STATE)"-ON-R",
504 
505 static const char *usage_str[] =
506 {
507 #define LOCKDEP_STATE(__STATE) __USAGE(__STATE)
508 #include "lockdep_states.h"
509 #undef LOCKDEP_STATE
510 	[LOCK_USED] = "INITIAL USE",
511 };
512 #endif
513 
514 const char * __get_key_name(struct lockdep_subclass_key *key, char *str)
515 {
516 	return kallsyms_lookup((unsigned long)key, NULL, NULL, NULL, str);
517 }
518 
519 static inline unsigned long lock_flag(enum lock_usage_bit bit)
520 {
521 	return 1UL << bit;
522 }
523 
524 static char get_usage_char(struct lock_class *class, enum lock_usage_bit bit)
525 {
526 	/*
527 	 * The usage character defaults to '.' (i.e., irqs disabled and not in
528 	 * irq context), which is the safest usage category.
529 	 */
530 	char c = '.';
531 
532 	/*
533 	 * The order of the following usage checks matters, which will
534 	 * result in the outcome character as follows:
535 	 *
536 	 * - '+': irq is enabled and not in irq context
537 	 * - '-': in irq context and irq is disabled
538 	 * - '?': in irq context and irq is enabled
539 	 */
540 	if (class->usage_mask & lock_flag(bit + LOCK_USAGE_DIR_MASK)) {
541 		c = '+';
542 		if (class->usage_mask & lock_flag(bit))
543 			c = '?';
544 	} else if (class->usage_mask & lock_flag(bit))
545 		c = '-';
546 
547 	return c;
548 }
549 
550 void get_usage_chars(struct lock_class *class, char usage[LOCK_USAGE_CHARS])
551 {
552 	int i = 0;
553 
554 #define LOCKDEP_STATE(__STATE) 						\
555 	usage[i++] = get_usage_char(class, LOCK_USED_IN_##__STATE);	\
556 	usage[i++] = get_usage_char(class, LOCK_USED_IN_##__STATE##_READ);
557 #include "lockdep_states.h"
558 #undef LOCKDEP_STATE
559 
560 	usage[i] = '\0';
561 }
562 
563 static void __print_lock_name(struct lock_class *class)
564 {
565 	char str[KSYM_NAME_LEN];
566 	const char *name;
567 
568 	name = class->name;
569 	if (!name) {
570 		name = __get_key_name(class->key, str);
571 		printk(KERN_CONT "%s", name);
572 	} else {
573 		printk(KERN_CONT "%s", name);
574 		if (class->name_version > 1)
575 			printk(KERN_CONT "#%d", class->name_version);
576 		if (class->subclass)
577 			printk(KERN_CONT "/%d", class->subclass);
578 	}
579 }
580 
581 static void print_lock_name(struct lock_class *class)
582 {
583 	char usage[LOCK_USAGE_CHARS];
584 
585 	get_usage_chars(class, usage);
586 
587 	printk(KERN_CONT " (");
588 	__print_lock_name(class);
589 	printk(KERN_CONT "){%s}", usage);
590 }
591 
592 static void print_lockdep_cache(struct lockdep_map *lock)
593 {
594 	const char *name;
595 	char str[KSYM_NAME_LEN];
596 
597 	name = lock->name;
598 	if (!name)
599 		name = __get_key_name(lock->key->subkeys, str);
600 
601 	printk(KERN_CONT "%s", name);
602 }
603 
604 static void print_lock(struct held_lock *hlock)
605 {
606 	/*
607 	 * We can be called locklessly through debug_show_all_locks() so be
608 	 * extra careful, the hlock might have been released and cleared.
609 	 *
610 	 * If this indeed happens, lets pretend it does not hurt to continue
611 	 * to print the lock unless the hlock class_idx does not point to a
612 	 * registered class. The rationale here is: since we don't attempt
613 	 * to distinguish whether we are in this situation, if it just
614 	 * happened we can't count on class_idx to tell either.
615 	 */
616 	struct lock_class *lock = hlock_class(hlock);
617 
618 	if (!lock) {
619 		printk(KERN_CONT "<RELEASED>\n");
620 		return;
621 	}
622 
623 	printk(KERN_CONT "%p", hlock->instance);
624 	print_lock_name(lock);
625 	printk(KERN_CONT ", at: %pS\n", (void *)hlock->acquire_ip);
626 }
627 
628 static void lockdep_print_held_locks(struct task_struct *p)
629 {
630 	int i, depth = READ_ONCE(p->lockdep_depth);
631 
632 	if (!depth)
633 		printk("no locks held by %s/%d.\n", p->comm, task_pid_nr(p));
634 	else
635 		printk("%d lock%s held by %s/%d:\n", depth,
636 		       depth > 1 ? "s" : "", p->comm, task_pid_nr(p));
637 	/*
638 	 * It's not reliable to print a task's held locks if it's not sleeping
639 	 * and it's not the current task.
640 	 */
641 	if (p->state == TASK_RUNNING && p != current)
642 		return;
643 	for (i = 0; i < depth; i++) {
644 		printk(" #%d: ", i);
645 		print_lock(p->held_locks + i);
646 	}
647 }
648 
649 static void print_kernel_ident(void)
650 {
651 	printk("%s %.*s %s\n", init_utsname()->release,
652 		(int)strcspn(init_utsname()->version, " "),
653 		init_utsname()->version,
654 		print_tainted());
655 }
656 
657 static int very_verbose(struct lock_class *class)
658 {
659 #if VERY_VERBOSE
660 	return class_filter(class);
661 #endif
662 	return 0;
663 }
664 
665 /*
666  * Is this the address of a static object:
667  */
668 #ifdef __KERNEL__
669 static int static_obj(const void *obj)
670 {
671 	unsigned long start = (unsigned long) &_stext,
672 		      end   = (unsigned long) &_end,
673 		      addr  = (unsigned long) obj;
674 
675 	if (arch_is_kernel_initmem_freed(addr))
676 		return 0;
677 
678 	/*
679 	 * static variable?
680 	 */
681 	if ((addr >= start) && (addr < end))
682 		return 1;
683 
684 	if (arch_is_kernel_data(addr))
685 		return 1;
686 
687 	/*
688 	 * in-kernel percpu var?
689 	 */
690 	if (is_kernel_percpu_address(addr))
691 		return 1;
692 
693 	/*
694 	 * module static or percpu var?
695 	 */
696 	return is_module_address(addr) || is_module_percpu_address(addr);
697 }
698 #endif
699 
700 /*
701  * To make lock name printouts unique, we calculate a unique
702  * class->name_version generation counter. The caller must hold the graph
703  * lock.
704  */
705 static int count_matching_names(struct lock_class *new_class)
706 {
707 	struct lock_class *class;
708 	int count = 0;
709 
710 	if (!new_class->name)
711 		return 0;
712 
713 	list_for_each_entry(class, &all_lock_classes, lock_entry) {
714 		if (new_class->key - new_class->subclass == class->key)
715 			return class->name_version;
716 		if (class->name && !strcmp(class->name, new_class->name))
717 			count = max(count, class->name_version);
718 	}
719 
720 	return count + 1;
721 }
722 
723 static inline struct lock_class *
724 look_up_lock_class(const struct lockdep_map *lock, unsigned int subclass)
725 {
726 	struct lockdep_subclass_key *key;
727 	struct hlist_head *hash_head;
728 	struct lock_class *class;
729 
730 	if (unlikely(subclass >= MAX_LOCKDEP_SUBCLASSES)) {
731 		debug_locks_off();
732 		printk(KERN_ERR
733 			"BUG: looking up invalid subclass: %u\n", subclass);
734 		printk(KERN_ERR
735 			"turning off the locking correctness validator.\n");
736 		dump_stack();
737 		return NULL;
738 	}
739 
740 	/*
741 	 * If it is not initialised then it has never been locked,
742 	 * so it won't be present in the hash table.
743 	 */
744 	if (unlikely(!lock->key))
745 		return NULL;
746 
747 	/*
748 	 * NOTE: the class-key must be unique. For dynamic locks, a static
749 	 * lock_class_key variable is passed in through the mutex_init()
750 	 * (or spin_lock_init()) call - which acts as the key. For static
751 	 * locks we use the lock object itself as the key.
752 	 */
753 	BUILD_BUG_ON(sizeof(struct lock_class_key) >
754 			sizeof(struct lockdep_map));
755 
756 	key = lock->key->subkeys + subclass;
757 
758 	hash_head = classhashentry(key);
759 
760 	/*
761 	 * We do an RCU walk of the hash, see lockdep_free_key_range().
762 	 */
763 	if (DEBUG_LOCKS_WARN_ON(!irqs_disabled()))
764 		return NULL;
765 
766 	hlist_for_each_entry_rcu(class, hash_head, hash_entry) {
767 		if (class->key == key) {
768 			/*
769 			 * Huh! same key, different name? Did someone trample
770 			 * on some memory? We're most confused.
771 			 */
772 			WARN_ON_ONCE(class->name != lock->name &&
773 				     lock->key != &__lockdep_no_validate__);
774 			return class;
775 		}
776 	}
777 
778 	return NULL;
779 }
780 
781 /*
782  * Static locks do not have their class-keys yet - for them the key is
783  * the lock object itself. If the lock is in the per cpu area, the
784  * canonical address of the lock (per cpu offset removed) is used.
785  */
786 static bool assign_lock_key(struct lockdep_map *lock)
787 {
788 	unsigned long can_addr, addr = (unsigned long)lock;
789 
790 #ifdef __KERNEL__
791 	/*
792 	 * lockdep_free_key_range() assumes that struct lock_class_key
793 	 * objects do not overlap. Since we use the address of lock
794 	 * objects as class key for static objects, check whether the
795 	 * size of lock_class_key objects does not exceed the size of
796 	 * the smallest lock object.
797 	 */
798 	BUILD_BUG_ON(sizeof(struct lock_class_key) > sizeof(raw_spinlock_t));
799 #endif
800 
801 	if (__is_kernel_percpu_address(addr, &can_addr))
802 		lock->key = (void *)can_addr;
803 	else if (__is_module_percpu_address(addr, &can_addr))
804 		lock->key = (void *)can_addr;
805 	else if (static_obj(lock))
806 		lock->key = (void *)lock;
807 	else {
808 		/* Debug-check: all keys must be persistent! */
809 		debug_locks_off();
810 		pr_err("INFO: trying to register non-static key.\n");
811 		pr_err("the code is fine but needs lockdep annotation.\n");
812 		pr_err("turning off the locking correctness validator.\n");
813 		dump_stack();
814 		return false;
815 	}
816 
817 	return true;
818 }
819 
820 #ifdef CONFIG_DEBUG_LOCKDEP
821 
822 /* Check whether element @e occurs in list @h */
823 static bool in_list(struct list_head *e, struct list_head *h)
824 {
825 	struct list_head *f;
826 
827 	list_for_each(f, h) {
828 		if (e == f)
829 			return true;
830 	}
831 
832 	return false;
833 }
834 
835 /*
836  * Check whether entry @e occurs in any of the locks_after or locks_before
837  * lists.
838  */
839 static bool in_any_class_list(struct list_head *e)
840 {
841 	struct lock_class *class;
842 	int i;
843 
844 	for (i = 0; i < ARRAY_SIZE(lock_classes); i++) {
845 		class = &lock_classes[i];
846 		if (in_list(e, &class->locks_after) ||
847 		    in_list(e, &class->locks_before))
848 			return true;
849 	}
850 	return false;
851 }
852 
853 static bool class_lock_list_valid(struct lock_class *c, struct list_head *h)
854 {
855 	struct lock_list *e;
856 
857 	list_for_each_entry(e, h, entry) {
858 		if (e->links_to != c) {
859 			printk(KERN_INFO "class %s: mismatch for lock entry %ld; class %s <> %s",
860 			       c->name ? : "(?)",
861 			       (unsigned long)(e - list_entries),
862 			       e->links_to && e->links_to->name ?
863 			       e->links_to->name : "(?)",
864 			       e->class && e->class->name ? e->class->name :
865 			       "(?)");
866 			return false;
867 		}
868 	}
869 	return true;
870 }
871 
872 #ifdef CONFIG_PROVE_LOCKING
873 static u16 chain_hlocks[MAX_LOCKDEP_CHAIN_HLOCKS];
874 #endif
875 
876 static bool check_lock_chain_key(struct lock_chain *chain)
877 {
878 #ifdef CONFIG_PROVE_LOCKING
879 	u64 chain_key = INITIAL_CHAIN_KEY;
880 	int i;
881 
882 	for (i = chain->base; i < chain->base + chain->depth; i++)
883 		chain_key = iterate_chain_key(chain_key, chain_hlocks[i]);
884 	/*
885 	 * The 'unsigned long long' casts avoid that a compiler warning
886 	 * is reported when building tools/lib/lockdep.
887 	 */
888 	if (chain->chain_key != chain_key) {
889 		printk(KERN_INFO "chain %lld: key %#llx <> %#llx\n",
890 		       (unsigned long long)(chain - lock_chains),
891 		       (unsigned long long)chain->chain_key,
892 		       (unsigned long long)chain_key);
893 		return false;
894 	}
895 #endif
896 	return true;
897 }
898 
899 static bool in_any_zapped_class_list(struct lock_class *class)
900 {
901 	struct pending_free *pf;
902 	int i;
903 
904 	for (i = 0, pf = delayed_free.pf; i < ARRAY_SIZE(delayed_free.pf); i++, pf++) {
905 		if (in_list(&class->lock_entry, &pf->zapped))
906 			return true;
907 	}
908 
909 	return false;
910 }
911 
912 static bool __check_data_structures(void)
913 {
914 	struct lock_class *class;
915 	struct lock_chain *chain;
916 	struct hlist_head *head;
917 	struct lock_list *e;
918 	int i;
919 
920 	/* Check whether all classes occur in a lock list. */
921 	for (i = 0; i < ARRAY_SIZE(lock_classes); i++) {
922 		class = &lock_classes[i];
923 		if (!in_list(&class->lock_entry, &all_lock_classes) &&
924 		    !in_list(&class->lock_entry, &free_lock_classes) &&
925 		    !in_any_zapped_class_list(class)) {
926 			printk(KERN_INFO "class %px/%s is not in any class list\n",
927 			       class, class->name ? : "(?)");
928 			return false;
929 		}
930 	}
931 
932 	/* Check whether all classes have valid lock lists. */
933 	for (i = 0; i < ARRAY_SIZE(lock_classes); i++) {
934 		class = &lock_classes[i];
935 		if (!class_lock_list_valid(class, &class->locks_before))
936 			return false;
937 		if (!class_lock_list_valid(class, &class->locks_after))
938 			return false;
939 	}
940 
941 	/* Check the chain_key of all lock chains. */
942 	for (i = 0; i < ARRAY_SIZE(chainhash_table); i++) {
943 		head = chainhash_table + i;
944 		hlist_for_each_entry_rcu(chain, head, entry) {
945 			if (!check_lock_chain_key(chain))
946 				return false;
947 		}
948 	}
949 
950 	/*
951 	 * Check whether all list entries that are in use occur in a class
952 	 * lock list.
953 	 */
954 	for_each_set_bit(i, list_entries_in_use, ARRAY_SIZE(list_entries)) {
955 		e = list_entries + i;
956 		if (!in_any_class_list(&e->entry)) {
957 			printk(KERN_INFO "list entry %d is not in any class list; class %s <> %s\n",
958 			       (unsigned int)(e - list_entries),
959 			       e->class->name ? : "(?)",
960 			       e->links_to->name ? : "(?)");
961 			return false;
962 		}
963 	}
964 
965 	/*
966 	 * Check whether all list entries that are not in use do not occur in
967 	 * a class lock list.
968 	 */
969 	for_each_clear_bit(i, list_entries_in_use, ARRAY_SIZE(list_entries)) {
970 		e = list_entries + i;
971 		if (in_any_class_list(&e->entry)) {
972 			printk(KERN_INFO "list entry %d occurs in a class list; class %s <> %s\n",
973 			       (unsigned int)(e - list_entries),
974 			       e->class && e->class->name ? e->class->name :
975 			       "(?)",
976 			       e->links_to && e->links_to->name ?
977 			       e->links_to->name : "(?)");
978 			return false;
979 		}
980 	}
981 
982 	return true;
983 }
984 
985 int check_consistency = 0;
986 module_param(check_consistency, int, 0644);
987 
988 static void check_data_structures(void)
989 {
990 	static bool once = false;
991 
992 	if (check_consistency && !once) {
993 		if (!__check_data_structures()) {
994 			once = true;
995 			WARN_ON(once);
996 		}
997 	}
998 }
999 
1000 #else /* CONFIG_DEBUG_LOCKDEP */
1001 
1002 static inline void check_data_structures(void) { }
1003 
1004 #endif /* CONFIG_DEBUG_LOCKDEP */
1005 
1006 /*
1007  * Initialize the lock_classes[] array elements, the free_lock_classes list
1008  * and also the delayed_free structure.
1009  */
1010 static void init_data_structures_once(void)
1011 {
1012 	static bool ds_initialized, rcu_head_initialized;
1013 	int i;
1014 
1015 	if (likely(rcu_head_initialized))
1016 		return;
1017 
1018 	if (system_state >= SYSTEM_SCHEDULING) {
1019 		init_rcu_head(&delayed_free.rcu_head);
1020 		rcu_head_initialized = true;
1021 	}
1022 
1023 	if (ds_initialized)
1024 		return;
1025 
1026 	ds_initialized = true;
1027 
1028 	INIT_LIST_HEAD(&delayed_free.pf[0].zapped);
1029 	INIT_LIST_HEAD(&delayed_free.pf[1].zapped);
1030 
1031 	for (i = 0; i < ARRAY_SIZE(lock_classes); i++) {
1032 		list_add_tail(&lock_classes[i].lock_entry, &free_lock_classes);
1033 		INIT_LIST_HEAD(&lock_classes[i].locks_after);
1034 		INIT_LIST_HEAD(&lock_classes[i].locks_before);
1035 	}
1036 }
1037 
1038 static inline struct hlist_head *keyhashentry(const struct lock_class_key *key)
1039 {
1040 	unsigned long hash = hash_long((uintptr_t)key, KEYHASH_BITS);
1041 
1042 	return lock_keys_hash + hash;
1043 }
1044 
1045 /* Register a dynamically allocated key. */
1046 void lockdep_register_key(struct lock_class_key *key)
1047 {
1048 	struct hlist_head *hash_head;
1049 	struct lock_class_key *k;
1050 	unsigned long flags;
1051 
1052 	if (WARN_ON_ONCE(static_obj(key)))
1053 		return;
1054 	hash_head = keyhashentry(key);
1055 
1056 	raw_local_irq_save(flags);
1057 	if (!graph_lock())
1058 		goto restore_irqs;
1059 	hlist_for_each_entry_rcu(k, hash_head, hash_entry) {
1060 		if (WARN_ON_ONCE(k == key))
1061 			goto out_unlock;
1062 	}
1063 	hlist_add_head_rcu(&key->hash_entry, hash_head);
1064 out_unlock:
1065 	graph_unlock();
1066 restore_irqs:
1067 	raw_local_irq_restore(flags);
1068 }
1069 EXPORT_SYMBOL_GPL(lockdep_register_key);
1070 
1071 /* Check whether a key has been registered as a dynamic key. */
1072 static bool is_dynamic_key(const struct lock_class_key *key)
1073 {
1074 	struct hlist_head *hash_head;
1075 	struct lock_class_key *k;
1076 	bool found = false;
1077 
1078 	if (WARN_ON_ONCE(static_obj(key)))
1079 		return false;
1080 
1081 	/*
1082 	 * If lock debugging is disabled lock_keys_hash[] may contain
1083 	 * pointers to memory that has already been freed. Avoid triggering
1084 	 * a use-after-free in that case by returning early.
1085 	 */
1086 	if (!debug_locks)
1087 		return true;
1088 
1089 	hash_head = keyhashentry(key);
1090 
1091 	rcu_read_lock();
1092 	hlist_for_each_entry_rcu(k, hash_head, hash_entry) {
1093 		if (k == key) {
1094 			found = true;
1095 			break;
1096 		}
1097 	}
1098 	rcu_read_unlock();
1099 
1100 	return found;
1101 }
1102 
1103 /*
1104  * Register a lock's class in the hash-table, if the class is not present
1105  * yet. Otherwise we look it up. We cache the result in the lock object
1106  * itself, so actual lookup of the hash should be once per lock object.
1107  */
1108 static struct lock_class *
1109 register_lock_class(struct lockdep_map *lock, unsigned int subclass, int force)
1110 {
1111 	struct lockdep_subclass_key *key;
1112 	struct hlist_head *hash_head;
1113 	struct lock_class *class;
1114 
1115 	DEBUG_LOCKS_WARN_ON(!irqs_disabled());
1116 
1117 	class = look_up_lock_class(lock, subclass);
1118 	if (likely(class))
1119 		goto out_set_class_cache;
1120 
1121 	if (!lock->key) {
1122 		if (!assign_lock_key(lock))
1123 			return NULL;
1124 	} else if (!static_obj(lock->key) && !is_dynamic_key(lock->key)) {
1125 		return NULL;
1126 	}
1127 
1128 	key = lock->key->subkeys + subclass;
1129 	hash_head = classhashentry(key);
1130 
1131 	if (!graph_lock()) {
1132 		return NULL;
1133 	}
1134 	/*
1135 	 * We have to do the hash-walk again, to avoid races
1136 	 * with another CPU:
1137 	 */
1138 	hlist_for_each_entry_rcu(class, hash_head, hash_entry) {
1139 		if (class->key == key)
1140 			goto out_unlock_set;
1141 	}
1142 
1143 	init_data_structures_once();
1144 
1145 	/* Allocate a new lock class and add it to the hash. */
1146 	class = list_first_entry_or_null(&free_lock_classes, typeof(*class),
1147 					 lock_entry);
1148 	if (!class) {
1149 		if (!debug_locks_off_graph_unlock()) {
1150 			return NULL;
1151 		}
1152 
1153 		print_lockdep_off("BUG: MAX_LOCKDEP_KEYS too low!");
1154 		dump_stack();
1155 		return NULL;
1156 	}
1157 	nr_lock_classes++;
1158 	__set_bit(class - lock_classes, lock_classes_in_use);
1159 	debug_atomic_inc(nr_unused_locks);
1160 	class->key = key;
1161 	class->name = lock->name;
1162 	class->subclass = subclass;
1163 	WARN_ON_ONCE(!list_empty(&class->locks_before));
1164 	WARN_ON_ONCE(!list_empty(&class->locks_after));
1165 	class->name_version = count_matching_names(class);
1166 	/*
1167 	 * We use RCU's safe list-add method to make
1168 	 * parallel walking of the hash-list safe:
1169 	 */
1170 	hlist_add_head_rcu(&class->hash_entry, hash_head);
1171 	/*
1172 	 * Remove the class from the free list and add it to the global list
1173 	 * of classes.
1174 	 */
1175 	list_move_tail(&class->lock_entry, &all_lock_classes);
1176 
1177 	if (verbose(class)) {
1178 		graph_unlock();
1179 
1180 		printk("\nnew class %px: %s", class->key, class->name);
1181 		if (class->name_version > 1)
1182 			printk(KERN_CONT "#%d", class->name_version);
1183 		printk(KERN_CONT "\n");
1184 		dump_stack();
1185 
1186 		if (!graph_lock()) {
1187 			return NULL;
1188 		}
1189 	}
1190 out_unlock_set:
1191 	graph_unlock();
1192 
1193 out_set_class_cache:
1194 	if (!subclass || force)
1195 		lock->class_cache[0] = class;
1196 	else if (subclass < NR_LOCKDEP_CACHING_CLASSES)
1197 		lock->class_cache[subclass] = class;
1198 
1199 	/*
1200 	 * Hash collision, did we smoke some? We found a class with a matching
1201 	 * hash but the subclass -- which is hashed in -- didn't match.
1202 	 */
1203 	if (DEBUG_LOCKS_WARN_ON(class->subclass != subclass))
1204 		return NULL;
1205 
1206 	return class;
1207 }
1208 
1209 #ifdef CONFIG_PROVE_LOCKING
1210 /*
1211  * Allocate a lockdep entry. (assumes the graph_lock held, returns
1212  * with NULL on failure)
1213  */
1214 static struct lock_list *alloc_list_entry(void)
1215 {
1216 	int idx = find_first_zero_bit(list_entries_in_use,
1217 				      ARRAY_SIZE(list_entries));
1218 
1219 	if (idx >= ARRAY_SIZE(list_entries)) {
1220 		if (!debug_locks_off_graph_unlock())
1221 			return NULL;
1222 
1223 		print_lockdep_off("BUG: MAX_LOCKDEP_ENTRIES too low!");
1224 		dump_stack();
1225 		return NULL;
1226 	}
1227 	nr_list_entries++;
1228 	__set_bit(idx, list_entries_in_use);
1229 	return list_entries + idx;
1230 }
1231 
1232 /*
1233  * Add a new dependency to the head of the list:
1234  */
1235 static int add_lock_to_list(struct lock_class *this,
1236 			    struct lock_class *links_to, struct list_head *head,
1237 			    unsigned long ip, int distance,
1238 			    struct lock_trace *trace)
1239 {
1240 	struct lock_list *entry;
1241 	/*
1242 	 * Lock not present yet - get a new dependency struct and
1243 	 * add it to the list:
1244 	 */
1245 	entry = alloc_list_entry();
1246 	if (!entry)
1247 		return 0;
1248 
1249 	entry->class = this;
1250 	entry->links_to = links_to;
1251 	entry->distance = distance;
1252 	entry->trace = *trace;
1253 	/*
1254 	 * Both allocation and removal are done under the graph lock; but
1255 	 * iteration is under RCU-sched; see look_up_lock_class() and
1256 	 * lockdep_free_key_range().
1257 	 */
1258 	list_add_tail_rcu(&entry->entry, head);
1259 
1260 	return 1;
1261 }
1262 
1263 /*
1264  * For good efficiency of modular, we use power of 2
1265  */
1266 #define MAX_CIRCULAR_QUEUE_SIZE		4096UL
1267 #define CQ_MASK				(MAX_CIRCULAR_QUEUE_SIZE-1)
1268 
1269 /*
1270  * The circular_queue and helpers are used to implement graph
1271  * breadth-first search (BFS) algorithm, by which we can determine
1272  * whether there is a path from a lock to another. In deadlock checks,
1273  * a path from the next lock to be acquired to a previous held lock
1274  * indicates that adding the <prev> -> <next> lock dependency will
1275  * produce a circle in the graph. Breadth-first search instead of
1276  * depth-first search is used in order to find the shortest (circular)
1277  * path.
1278  */
1279 struct circular_queue {
1280 	struct lock_list *element[MAX_CIRCULAR_QUEUE_SIZE];
1281 	unsigned int  front, rear;
1282 };
1283 
1284 static struct circular_queue lock_cq;
1285 
1286 unsigned int max_bfs_queue_depth;
1287 
1288 static unsigned int lockdep_dependency_gen_id;
1289 
1290 static inline void __cq_init(struct circular_queue *cq)
1291 {
1292 	cq->front = cq->rear = 0;
1293 	lockdep_dependency_gen_id++;
1294 }
1295 
1296 static inline int __cq_empty(struct circular_queue *cq)
1297 {
1298 	return (cq->front == cq->rear);
1299 }
1300 
1301 static inline int __cq_full(struct circular_queue *cq)
1302 {
1303 	return ((cq->rear + 1) & CQ_MASK) == cq->front;
1304 }
1305 
1306 static inline int __cq_enqueue(struct circular_queue *cq, struct lock_list *elem)
1307 {
1308 	if (__cq_full(cq))
1309 		return -1;
1310 
1311 	cq->element[cq->rear] = elem;
1312 	cq->rear = (cq->rear + 1) & CQ_MASK;
1313 	return 0;
1314 }
1315 
1316 /*
1317  * Dequeue an element from the circular_queue, return a lock_list if
1318  * the queue is not empty, or NULL if otherwise.
1319  */
1320 static inline struct lock_list * __cq_dequeue(struct circular_queue *cq)
1321 {
1322 	struct lock_list * lock;
1323 
1324 	if (__cq_empty(cq))
1325 		return NULL;
1326 
1327 	lock = cq->element[cq->front];
1328 	cq->front = (cq->front + 1) & CQ_MASK;
1329 
1330 	return lock;
1331 }
1332 
1333 static inline unsigned int  __cq_get_elem_count(struct circular_queue *cq)
1334 {
1335 	return (cq->rear - cq->front) & CQ_MASK;
1336 }
1337 
1338 static inline void mark_lock_accessed(struct lock_list *lock,
1339 					struct lock_list *parent)
1340 {
1341 	unsigned long nr;
1342 
1343 	nr = lock - list_entries;
1344 	WARN_ON(nr >= ARRAY_SIZE(list_entries)); /* Out-of-bounds, input fail */
1345 	lock->parent = parent;
1346 	lock->class->dep_gen_id = lockdep_dependency_gen_id;
1347 }
1348 
1349 static inline unsigned long lock_accessed(struct lock_list *lock)
1350 {
1351 	unsigned long nr;
1352 
1353 	nr = lock - list_entries;
1354 	WARN_ON(nr >= ARRAY_SIZE(list_entries)); /* Out-of-bounds, input fail */
1355 	return lock->class->dep_gen_id == lockdep_dependency_gen_id;
1356 }
1357 
1358 static inline struct lock_list *get_lock_parent(struct lock_list *child)
1359 {
1360 	return child->parent;
1361 }
1362 
1363 static inline int get_lock_depth(struct lock_list *child)
1364 {
1365 	int depth = 0;
1366 	struct lock_list *parent;
1367 
1368 	while ((parent = get_lock_parent(child))) {
1369 		child = parent;
1370 		depth++;
1371 	}
1372 	return depth;
1373 }
1374 
1375 /*
1376  * Return the forward or backward dependency list.
1377  *
1378  * @lock:   the lock_list to get its class's dependency list
1379  * @offset: the offset to struct lock_class to determine whether it is
1380  *          locks_after or locks_before
1381  */
1382 static inline struct list_head *get_dep_list(struct lock_list *lock, int offset)
1383 {
1384 	void *lock_class = lock->class;
1385 
1386 	return lock_class + offset;
1387 }
1388 
1389 /*
1390  * Forward- or backward-dependency search, used for both circular dependency
1391  * checking and hardirq-unsafe/softirq-unsafe checking.
1392  */
1393 static int __bfs(struct lock_list *source_entry,
1394 		 void *data,
1395 		 int (*match)(struct lock_list *entry, void *data),
1396 		 struct lock_list **target_entry,
1397 		 int offset)
1398 {
1399 	struct lock_list *entry;
1400 	struct lock_list *lock;
1401 	struct list_head *head;
1402 	struct circular_queue *cq = &lock_cq;
1403 	int ret = 1;
1404 
1405 	if (match(source_entry, data)) {
1406 		*target_entry = source_entry;
1407 		ret = 0;
1408 		goto exit;
1409 	}
1410 
1411 	head = get_dep_list(source_entry, offset);
1412 	if (list_empty(head))
1413 		goto exit;
1414 
1415 	__cq_init(cq);
1416 	__cq_enqueue(cq, source_entry);
1417 
1418 	while ((lock = __cq_dequeue(cq))) {
1419 
1420 		if (!lock->class) {
1421 			ret = -2;
1422 			goto exit;
1423 		}
1424 
1425 		head = get_dep_list(lock, offset);
1426 
1427 		DEBUG_LOCKS_WARN_ON(!irqs_disabled());
1428 
1429 		list_for_each_entry_rcu(entry, head, entry) {
1430 			if (!lock_accessed(entry)) {
1431 				unsigned int cq_depth;
1432 				mark_lock_accessed(entry, lock);
1433 				if (match(entry, data)) {
1434 					*target_entry = entry;
1435 					ret = 0;
1436 					goto exit;
1437 				}
1438 
1439 				if (__cq_enqueue(cq, entry)) {
1440 					ret = -1;
1441 					goto exit;
1442 				}
1443 				cq_depth = __cq_get_elem_count(cq);
1444 				if (max_bfs_queue_depth < cq_depth)
1445 					max_bfs_queue_depth = cq_depth;
1446 			}
1447 		}
1448 	}
1449 exit:
1450 	return ret;
1451 }
1452 
1453 static inline int __bfs_forwards(struct lock_list *src_entry,
1454 			void *data,
1455 			int (*match)(struct lock_list *entry, void *data),
1456 			struct lock_list **target_entry)
1457 {
1458 	return __bfs(src_entry, data, match, target_entry,
1459 		     offsetof(struct lock_class, locks_after));
1460 
1461 }
1462 
1463 static inline int __bfs_backwards(struct lock_list *src_entry,
1464 			void *data,
1465 			int (*match)(struct lock_list *entry, void *data),
1466 			struct lock_list **target_entry)
1467 {
1468 	return __bfs(src_entry, data, match, target_entry,
1469 		     offsetof(struct lock_class, locks_before));
1470 
1471 }
1472 
1473 static void print_lock_trace(struct lock_trace *trace, unsigned int spaces)
1474 {
1475 	unsigned long *entries = stack_trace + trace->offset;
1476 
1477 	stack_trace_print(entries, trace->nr_entries, spaces);
1478 }
1479 
1480 /*
1481  * Print a dependency chain entry (this is only done when a deadlock
1482  * has been detected):
1483  */
1484 static noinline void
1485 print_circular_bug_entry(struct lock_list *target, int depth)
1486 {
1487 	if (debug_locks_silent)
1488 		return;
1489 	printk("\n-> #%u", depth);
1490 	print_lock_name(target->class);
1491 	printk(KERN_CONT ":\n");
1492 	print_lock_trace(&target->trace, 6);
1493 }
1494 
1495 static void
1496 print_circular_lock_scenario(struct held_lock *src,
1497 			     struct held_lock *tgt,
1498 			     struct lock_list *prt)
1499 {
1500 	struct lock_class *source = hlock_class(src);
1501 	struct lock_class *target = hlock_class(tgt);
1502 	struct lock_class *parent = prt->class;
1503 
1504 	/*
1505 	 * A direct locking problem where unsafe_class lock is taken
1506 	 * directly by safe_class lock, then all we need to show
1507 	 * is the deadlock scenario, as it is obvious that the
1508 	 * unsafe lock is taken under the safe lock.
1509 	 *
1510 	 * But if there is a chain instead, where the safe lock takes
1511 	 * an intermediate lock (middle_class) where this lock is
1512 	 * not the same as the safe lock, then the lock chain is
1513 	 * used to describe the problem. Otherwise we would need
1514 	 * to show a different CPU case for each link in the chain
1515 	 * from the safe_class lock to the unsafe_class lock.
1516 	 */
1517 	if (parent != source) {
1518 		printk("Chain exists of:\n  ");
1519 		__print_lock_name(source);
1520 		printk(KERN_CONT " --> ");
1521 		__print_lock_name(parent);
1522 		printk(KERN_CONT " --> ");
1523 		__print_lock_name(target);
1524 		printk(KERN_CONT "\n\n");
1525 	}
1526 
1527 	printk(" Possible unsafe locking scenario:\n\n");
1528 	printk("       CPU0                    CPU1\n");
1529 	printk("       ----                    ----\n");
1530 	printk("  lock(");
1531 	__print_lock_name(target);
1532 	printk(KERN_CONT ");\n");
1533 	printk("                               lock(");
1534 	__print_lock_name(parent);
1535 	printk(KERN_CONT ");\n");
1536 	printk("                               lock(");
1537 	__print_lock_name(target);
1538 	printk(KERN_CONT ");\n");
1539 	printk("  lock(");
1540 	__print_lock_name(source);
1541 	printk(KERN_CONT ");\n");
1542 	printk("\n *** DEADLOCK ***\n\n");
1543 }
1544 
1545 /*
1546  * When a circular dependency is detected, print the
1547  * header first:
1548  */
1549 static noinline void
1550 print_circular_bug_header(struct lock_list *entry, unsigned int depth,
1551 			struct held_lock *check_src,
1552 			struct held_lock *check_tgt)
1553 {
1554 	struct task_struct *curr = current;
1555 
1556 	if (debug_locks_silent)
1557 		return;
1558 
1559 	pr_warn("\n");
1560 	pr_warn("======================================================\n");
1561 	pr_warn("WARNING: possible circular locking dependency detected\n");
1562 	print_kernel_ident();
1563 	pr_warn("------------------------------------------------------\n");
1564 	pr_warn("%s/%d is trying to acquire lock:\n",
1565 		curr->comm, task_pid_nr(curr));
1566 	print_lock(check_src);
1567 
1568 	pr_warn("\nbut task is already holding lock:\n");
1569 
1570 	print_lock(check_tgt);
1571 	pr_warn("\nwhich lock already depends on the new lock.\n\n");
1572 	pr_warn("\nthe existing dependency chain (in reverse order) is:\n");
1573 
1574 	print_circular_bug_entry(entry, depth);
1575 }
1576 
1577 static inline int class_equal(struct lock_list *entry, void *data)
1578 {
1579 	return entry->class == data;
1580 }
1581 
1582 static noinline void print_circular_bug(struct lock_list *this,
1583 					struct lock_list *target,
1584 					struct held_lock *check_src,
1585 					struct held_lock *check_tgt)
1586 {
1587 	struct task_struct *curr = current;
1588 	struct lock_list *parent;
1589 	struct lock_list *first_parent;
1590 	int depth;
1591 
1592 	if (!debug_locks_off_graph_unlock() || debug_locks_silent)
1593 		return;
1594 
1595 	if (!save_trace(&this->trace))
1596 		return;
1597 
1598 	depth = get_lock_depth(target);
1599 
1600 	print_circular_bug_header(target, depth, check_src, check_tgt);
1601 
1602 	parent = get_lock_parent(target);
1603 	first_parent = parent;
1604 
1605 	while (parent) {
1606 		print_circular_bug_entry(parent, --depth);
1607 		parent = get_lock_parent(parent);
1608 	}
1609 
1610 	printk("\nother info that might help us debug this:\n\n");
1611 	print_circular_lock_scenario(check_src, check_tgt,
1612 				     first_parent);
1613 
1614 	lockdep_print_held_locks(curr);
1615 
1616 	printk("\nstack backtrace:\n");
1617 	dump_stack();
1618 }
1619 
1620 static noinline void print_bfs_bug(int ret)
1621 {
1622 	if (!debug_locks_off_graph_unlock())
1623 		return;
1624 
1625 	/*
1626 	 * Breadth-first-search failed, graph got corrupted?
1627 	 */
1628 	WARN(1, "lockdep bfs error:%d\n", ret);
1629 }
1630 
1631 static int noop_count(struct lock_list *entry, void *data)
1632 {
1633 	(*(unsigned long *)data)++;
1634 	return 0;
1635 }
1636 
1637 static unsigned long __lockdep_count_forward_deps(struct lock_list *this)
1638 {
1639 	unsigned long  count = 0;
1640 	struct lock_list *uninitialized_var(target_entry);
1641 
1642 	__bfs_forwards(this, (void *)&count, noop_count, &target_entry);
1643 
1644 	return count;
1645 }
1646 unsigned long lockdep_count_forward_deps(struct lock_class *class)
1647 {
1648 	unsigned long ret, flags;
1649 	struct lock_list this;
1650 
1651 	this.parent = NULL;
1652 	this.class = class;
1653 
1654 	raw_local_irq_save(flags);
1655 	arch_spin_lock(&lockdep_lock);
1656 	ret = __lockdep_count_forward_deps(&this);
1657 	arch_spin_unlock(&lockdep_lock);
1658 	raw_local_irq_restore(flags);
1659 
1660 	return ret;
1661 }
1662 
1663 static unsigned long __lockdep_count_backward_deps(struct lock_list *this)
1664 {
1665 	unsigned long  count = 0;
1666 	struct lock_list *uninitialized_var(target_entry);
1667 
1668 	__bfs_backwards(this, (void *)&count, noop_count, &target_entry);
1669 
1670 	return count;
1671 }
1672 
1673 unsigned long lockdep_count_backward_deps(struct lock_class *class)
1674 {
1675 	unsigned long ret, flags;
1676 	struct lock_list this;
1677 
1678 	this.parent = NULL;
1679 	this.class = class;
1680 
1681 	raw_local_irq_save(flags);
1682 	arch_spin_lock(&lockdep_lock);
1683 	ret = __lockdep_count_backward_deps(&this);
1684 	arch_spin_unlock(&lockdep_lock);
1685 	raw_local_irq_restore(flags);
1686 
1687 	return ret;
1688 }
1689 
1690 /*
1691  * Check that the dependency graph starting at <src> can lead to
1692  * <target> or not. Print an error and return 0 if it does.
1693  */
1694 static noinline int
1695 check_path(struct lock_class *target, struct lock_list *src_entry,
1696 	   struct lock_list **target_entry)
1697 {
1698 	int ret;
1699 
1700 	ret = __bfs_forwards(src_entry, (void *)target, class_equal,
1701 			     target_entry);
1702 
1703 	if (unlikely(ret < 0))
1704 		print_bfs_bug(ret);
1705 
1706 	return ret;
1707 }
1708 
1709 /*
1710  * Prove that the dependency graph starting at <src> can not
1711  * lead to <target>. If it can, there is a circle when adding
1712  * <target> -> <src> dependency.
1713  *
1714  * Print an error and return 0 if it does.
1715  */
1716 static noinline int
1717 check_noncircular(struct held_lock *src, struct held_lock *target,
1718 		  struct lock_trace *trace)
1719 {
1720 	int ret;
1721 	struct lock_list *uninitialized_var(target_entry);
1722 	struct lock_list src_entry = {
1723 		.class = hlock_class(src),
1724 		.parent = NULL,
1725 	};
1726 
1727 	debug_atomic_inc(nr_cyclic_checks);
1728 
1729 	ret = check_path(hlock_class(target), &src_entry, &target_entry);
1730 
1731 	if (unlikely(!ret)) {
1732 		if (!trace->nr_entries) {
1733 			/*
1734 			 * If save_trace fails here, the printing might
1735 			 * trigger a WARN but because of the !nr_entries it
1736 			 * should not do bad things.
1737 			 */
1738 			save_trace(trace);
1739 		}
1740 
1741 		print_circular_bug(&src_entry, target_entry, src, target);
1742 	}
1743 
1744 	return ret;
1745 }
1746 
1747 #ifdef CONFIG_LOCKDEP_SMALL
1748 /*
1749  * Check that the dependency graph starting at <src> can lead to
1750  * <target> or not. If it can, <src> -> <target> dependency is already
1751  * in the graph.
1752  *
1753  * Print an error and return 2 if it does or 1 if it does not.
1754  */
1755 static noinline int
1756 check_redundant(struct held_lock *src, struct held_lock *target)
1757 {
1758 	int ret;
1759 	struct lock_list *uninitialized_var(target_entry);
1760 	struct lock_list src_entry = {
1761 		.class = hlock_class(src),
1762 		.parent = NULL,
1763 	};
1764 
1765 	debug_atomic_inc(nr_redundant_checks);
1766 
1767 	ret = check_path(hlock_class(target), &src_entry, &target_entry);
1768 
1769 	if (!ret) {
1770 		debug_atomic_inc(nr_redundant);
1771 		ret = 2;
1772 	} else if (ret < 0)
1773 		ret = 0;
1774 
1775 	return ret;
1776 }
1777 #endif
1778 
1779 #ifdef CONFIG_TRACE_IRQFLAGS
1780 
1781 static inline int usage_accumulate(struct lock_list *entry, void *mask)
1782 {
1783 	*(unsigned long *)mask |= entry->class->usage_mask;
1784 
1785 	return 0;
1786 }
1787 
1788 /*
1789  * Forwards and backwards subgraph searching, for the purposes of
1790  * proving that two subgraphs can be connected by a new dependency
1791  * without creating any illegal irq-safe -> irq-unsafe lock dependency.
1792  */
1793 
1794 static inline int usage_match(struct lock_list *entry, void *mask)
1795 {
1796 	return entry->class->usage_mask & *(unsigned long *)mask;
1797 }
1798 
1799 /*
1800  * Find a node in the forwards-direction dependency sub-graph starting
1801  * at @root->class that matches @bit.
1802  *
1803  * Return 0 if such a node exists in the subgraph, and put that node
1804  * into *@target_entry.
1805  *
1806  * Return 1 otherwise and keep *@target_entry unchanged.
1807  * Return <0 on error.
1808  */
1809 static int
1810 find_usage_forwards(struct lock_list *root, unsigned long usage_mask,
1811 			struct lock_list **target_entry)
1812 {
1813 	int result;
1814 
1815 	debug_atomic_inc(nr_find_usage_forwards_checks);
1816 
1817 	result = __bfs_forwards(root, &usage_mask, usage_match, target_entry);
1818 
1819 	return result;
1820 }
1821 
1822 /*
1823  * Find a node in the backwards-direction dependency sub-graph starting
1824  * at @root->class that matches @bit.
1825  *
1826  * Return 0 if such a node exists in the subgraph, and put that node
1827  * into *@target_entry.
1828  *
1829  * Return 1 otherwise and keep *@target_entry unchanged.
1830  * Return <0 on error.
1831  */
1832 static int
1833 find_usage_backwards(struct lock_list *root, unsigned long usage_mask,
1834 			struct lock_list **target_entry)
1835 {
1836 	int result;
1837 
1838 	debug_atomic_inc(nr_find_usage_backwards_checks);
1839 
1840 	result = __bfs_backwards(root, &usage_mask, usage_match, target_entry);
1841 
1842 	return result;
1843 }
1844 
1845 static void print_lock_class_header(struct lock_class *class, int depth)
1846 {
1847 	int bit;
1848 
1849 	printk("%*s->", depth, "");
1850 	print_lock_name(class);
1851 #ifdef CONFIG_DEBUG_LOCKDEP
1852 	printk(KERN_CONT " ops: %lu", debug_class_ops_read(class));
1853 #endif
1854 	printk(KERN_CONT " {\n");
1855 
1856 	for (bit = 0; bit < LOCK_USAGE_STATES; bit++) {
1857 		if (class->usage_mask & (1 << bit)) {
1858 			int len = depth;
1859 
1860 			len += printk("%*s   %s", depth, "", usage_str[bit]);
1861 			len += printk(KERN_CONT " at:\n");
1862 			print_lock_trace(class->usage_traces + bit, len);
1863 		}
1864 	}
1865 	printk("%*s }\n", depth, "");
1866 
1867 	printk("%*s ... key      at: [<%px>] %pS\n",
1868 		depth, "", class->key, class->key);
1869 }
1870 
1871 /*
1872  * printk the shortest lock dependencies from @start to @end in reverse order:
1873  */
1874 static void __used
1875 print_shortest_lock_dependencies(struct lock_list *leaf,
1876 				 struct lock_list *root)
1877 {
1878 	struct lock_list *entry = leaf;
1879 	int depth;
1880 
1881 	/*compute depth from generated tree by BFS*/
1882 	depth = get_lock_depth(leaf);
1883 
1884 	do {
1885 		print_lock_class_header(entry->class, depth);
1886 		printk("%*s ... acquired at:\n", depth, "");
1887 		print_lock_trace(&entry->trace, 2);
1888 		printk("\n");
1889 
1890 		if (depth == 0 && (entry != root)) {
1891 			printk("lockdep:%s bad path found in chain graph\n", __func__);
1892 			break;
1893 		}
1894 
1895 		entry = get_lock_parent(entry);
1896 		depth--;
1897 	} while (entry && (depth >= 0));
1898 }
1899 
1900 static void
1901 print_irq_lock_scenario(struct lock_list *safe_entry,
1902 			struct lock_list *unsafe_entry,
1903 			struct lock_class *prev_class,
1904 			struct lock_class *next_class)
1905 {
1906 	struct lock_class *safe_class = safe_entry->class;
1907 	struct lock_class *unsafe_class = unsafe_entry->class;
1908 	struct lock_class *middle_class = prev_class;
1909 
1910 	if (middle_class == safe_class)
1911 		middle_class = next_class;
1912 
1913 	/*
1914 	 * A direct locking problem where unsafe_class lock is taken
1915 	 * directly by safe_class lock, then all we need to show
1916 	 * is the deadlock scenario, as it is obvious that the
1917 	 * unsafe lock is taken under the safe lock.
1918 	 *
1919 	 * But if there is a chain instead, where the safe lock takes
1920 	 * an intermediate lock (middle_class) where this lock is
1921 	 * not the same as the safe lock, then the lock chain is
1922 	 * used to describe the problem. Otherwise we would need
1923 	 * to show a different CPU case for each link in the chain
1924 	 * from the safe_class lock to the unsafe_class lock.
1925 	 */
1926 	if (middle_class != unsafe_class) {
1927 		printk("Chain exists of:\n  ");
1928 		__print_lock_name(safe_class);
1929 		printk(KERN_CONT " --> ");
1930 		__print_lock_name(middle_class);
1931 		printk(KERN_CONT " --> ");
1932 		__print_lock_name(unsafe_class);
1933 		printk(KERN_CONT "\n\n");
1934 	}
1935 
1936 	printk(" Possible interrupt unsafe locking scenario:\n\n");
1937 	printk("       CPU0                    CPU1\n");
1938 	printk("       ----                    ----\n");
1939 	printk("  lock(");
1940 	__print_lock_name(unsafe_class);
1941 	printk(KERN_CONT ");\n");
1942 	printk("                               local_irq_disable();\n");
1943 	printk("                               lock(");
1944 	__print_lock_name(safe_class);
1945 	printk(KERN_CONT ");\n");
1946 	printk("                               lock(");
1947 	__print_lock_name(middle_class);
1948 	printk(KERN_CONT ");\n");
1949 	printk("  <Interrupt>\n");
1950 	printk("    lock(");
1951 	__print_lock_name(safe_class);
1952 	printk(KERN_CONT ");\n");
1953 	printk("\n *** DEADLOCK ***\n\n");
1954 }
1955 
1956 static void
1957 print_bad_irq_dependency(struct task_struct *curr,
1958 			 struct lock_list *prev_root,
1959 			 struct lock_list *next_root,
1960 			 struct lock_list *backwards_entry,
1961 			 struct lock_list *forwards_entry,
1962 			 struct held_lock *prev,
1963 			 struct held_lock *next,
1964 			 enum lock_usage_bit bit1,
1965 			 enum lock_usage_bit bit2,
1966 			 const char *irqclass)
1967 {
1968 	if (!debug_locks_off_graph_unlock() || debug_locks_silent)
1969 		return;
1970 
1971 	pr_warn("\n");
1972 	pr_warn("=====================================================\n");
1973 	pr_warn("WARNING: %s-safe -> %s-unsafe lock order detected\n",
1974 		irqclass, irqclass);
1975 	print_kernel_ident();
1976 	pr_warn("-----------------------------------------------------\n");
1977 	pr_warn("%s/%d [HC%u[%lu]:SC%u[%lu]:HE%u:SE%u] is trying to acquire:\n",
1978 		curr->comm, task_pid_nr(curr),
1979 		curr->hardirq_context, hardirq_count() >> HARDIRQ_SHIFT,
1980 		curr->softirq_context, softirq_count() >> SOFTIRQ_SHIFT,
1981 		curr->hardirqs_enabled,
1982 		curr->softirqs_enabled);
1983 	print_lock(next);
1984 
1985 	pr_warn("\nand this task is already holding:\n");
1986 	print_lock(prev);
1987 	pr_warn("which would create a new lock dependency:\n");
1988 	print_lock_name(hlock_class(prev));
1989 	pr_cont(" ->");
1990 	print_lock_name(hlock_class(next));
1991 	pr_cont("\n");
1992 
1993 	pr_warn("\nbut this new dependency connects a %s-irq-safe lock:\n",
1994 		irqclass);
1995 	print_lock_name(backwards_entry->class);
1996 	pr_warn("\n... which became %s-irq-safe at:\n", irqclass);
1997 
1998 	print_lock_trace(backwards_entry->class->usage_traces + bit1, 1);
1999 
2000 	pr_warn("\nto a %s-irq-unsafe lock:\n", irqclass);
2001 	print_lock_name(forwards_entry->class);
2002 	pr_warn("\n... which became %s-irq-unsafe at:\n", irqclass);
2003 	pr_warn("...");
2004 
2005 	print_lock_trace(forwards_entry->class->usage_traces + bit2, 1);
2006 
2007 	pr_warn("\nother info that might help us debug this:\n\n");
2008 	print_irq_lock_scenario(backwards_entry, forwards_entry,
2009 				hlock_class(prev), hlock_class(next));
2010 
2011 	lockdep_print_held_locks(curr);
2012 
2013 	pr_warn("\nthe dependencies between %s-irq-safe lock and the holding lock:\n", irqclass);
2014 	if (!save_trace(&prev_root->trace))
2015 		return;
2016 	print_shortest_lock_dependencies(backwards_entry, prev_root);
2017 
2018 	pr_warn("\nthe dependencies between the lock to be acquired");
2019 	pr_warn(" and %s-irq-unsafe lock:\n", irqclass);
2020 	if (!save_trace(&next_root->trace))
2021 		return;
2022 	print_shortest_lock_dependencies(forwards_entry, next_root);
2023 
2024 	pr_warn("\nstack backtrace:\n");
2025 	dump_stack();
2026 }
2027 
2028 static const char *state_names[] = {
2029 #define LOCKDEP_STATE(__STATE) \
2030 	__stringify(__STATE),
2031 #include "lockdep_states.h"
2032 #undef LOCKDEP_STATE
2033 };
2034 
2035 static const char *state_rnames[] = {
2036 #define LOCKDEP_STATE(__STATE) \
2037 	__stringify(__STATE)"-READ",
2038 #include "lockdep_states.h"
2039 #undef LOCKDEP_STATE
2040 };
2041 
2042 static inline const char *state_name(enum lock_usage_bit bit)
2043 {
2044 	if (bit & LOCK_USAGE_READ_MASK)
2045 		return state_rnames[bit >> LOCK_USAGE_DIR_MASK];
2046 	else
2047 		return state_names[bit >> LOCK_USAGE_DIR_MASK];
2048 }
2049 
2050 /*
2051  * The bit number is encoded like:
2052  *
2053  *  bit0: 0 exclusive, 1 read lock
2054  *  bit1: 0 used in irq, 1 irq enabled
2055  *  bit2-n: state
2056  */
2057 static int exclusive_bit(int new_bit)
2058 {
2059 	int state = new_bit & LOCK_USAGE_STATE_MASK;
2060 	int dir = new_bit & LOCK_USAGE_DIR_MASK;
2061 
2062 	/*
2063 	 * keep state, bit flip the direction and strip read.
2064 	 */
2065 	return state | (dir ^ LOCK_USAGE_DIR_MASK);
2066 }
2067 
2068 /*
2069  * Observe that when given a bitmask where each bitnr is encoded as above, a
2070  * right shift of the mask transforms the individual bitnrs as -1 and
2071  * conversely, a left shift transforms into +1 for the individual bitnrs.
2072  *
2073  * So for all bits whose number have LOCK_ENABLED_* set (bitnr1 == 1), we can
2074  * create the mask with those bit numbers using LOCK_USED_IN_* (bitnr1 == 0)
2075  * instead by subtracting the bit number by 2, or shifting the mask right by 2.
2076  *
2077  * Similarly, bitnr1 == 0 becomes bitnr1 == 1 by adding 2, or shifting left 2.
2078  *
2079  * So split the mask (note that LOCKF_ENABLED_IRQ_ALL|LOCKF_USED_IN_IRQ_ALL is
2080  * all bits set) and recompose with bitnr1 flipped.
2081  */
2082 static unsigned long invert_dir_mask(unsigned long mask)
2083 {
2084 	unsigned long excl = 0;
2085 
2086 	/* Invert dir */
2087 	excl |= (mask & LOCKF_ENABLED_IRQ_ALL) >> LOCK_USAGE_DIR_MASK;
2088 	excl |= (mask & LOCKF_USED_IN_IRQ_ALL) << LOCK_USAGE_DIR_MASK;
2089 
2090 	return excl;
2091 }
2092 
2093 /*
2094  * As above, we clear bitnr0 (LOCK_*_READ off) with bitmask ops. First, for all
2095  * bits with bitnr0 set (LOCK_*_READ), add those with bitnr0 cleared (LOCK_*).
2096  * And then mask out all bitnr0.
2097  */
2098 static unsigned long exclusive_mask(unsigned long mask)
2099 {
2100 	unsigned long excl = invert_dir_mask(mask);
2101 
2102 	/* Strip read */
2103 	excl |= (excl & LOCKF_IRQ_READ) >> LOCK_USAGE_READ_MASK;
2104 	excl &= ~LOCKF_IRQ_READ;
2105 
2106 	return excl;
2107 }
2108 
2109 /*
2110  * Retrieve the _possible_ original mask to which @mask is
2111  * exclusive. Ie: this is the opposite of exclusive_mask().
2112  * Note that 2 possible original bits can match an exclusive
2113  * bit: one has LOCK_USAGE_READ_MASK set, the other has it
2114  * cleared. So both are returned for each exclusive bit.
2115  */
2116 static unsigned long original_mask(unsigned long mask)
2117 {
2118 	unsigned long excl = invert_dir_mask(mask);
2119 
2120 	/* Include read in existing usages */
2121 	excl |= (excl & LOCKF_IRQ) << LOCK_USAGE_READ_MASK;
2122 
2123 	return excl;
2124 }
2125 
2126 /*
2127  * Find the first pair of bit match between an original
2128  * usage mask and an exclusive usage mask.
2129  */
2130 static int find_exclusive_match(unsigned long mask,
2131 				unsigned long excl_mask,
2132 				enum lock_usage_bit *bitp,
2133 				enum lock_usage_bit *excl_bitp)
2134 {
2135 	int bit, excl;
2136 
2137 	for_each_set_bit(bit, &mask, LOCK_USED) {
2138 		excl = exclusive_bit(bit);
2139 		if (excl_mask & lock_flag(excl)) {
2140 			*bitp = bit;
2141 			*excl_bitp = excl;
2142 			return 0;
2143 		}
2144 	}
2145 	return -1;
2146 }
2147 
2148 /*
2149  * Prove that the new dependency does not connect a hardirq-safe(-read)
2150  * lock with a hardirq-unsafe lock - to achieve this we search
2151  * the backwards-subgraph starting at <prev>, and the
2152  * forwards-subgraph starting at <next>:
2153  */
2154 static int check_irq_usage(struct task_struct *curr, struct held_lock *prev,
2155 			   struct held_lock *next)
2156 {
2157 	unsigned long usage_mask = 0, forward_mask, backward_mask;
2158 	enum lock_usage_bit forward_bit = 0, backward_bit = 0;
2159 	struct lock_list *uninitialized_var(target_entry1);
2160 	struct lock_list *uninitialized_var(target_entry);
2161 	struct lock_list this, that;
2162 	int ret;
2163 
2164 	/*
2165 	 * Step 1: gather all hard/soft IRQs usages backward in an
2166 	 * accumulated usage mask.
2167 	 */
2168 	this.parent = NULL;
2169 	this.class = hlock_class(prev);
2170 
2171 	ret = __bfs_backwards(&this, &usage_mask, usage_accumulate, NULL);
2172 	if (ret < 0) {
2173 		print_bfs_bug(ret);
2174 		return 0;
2175 	}
2176 
2177 	usage_mask &= LOCKF_USED_IN_IRQ_ALL;
2178 	if (!usage_mask)
2179 		return 1;
2180 
2181 	/*
2182 	 * Step 2: find exclusive uses forward that match the previous
2183 	 * backward accumulated mask.
2184 	 */
2185 	forward_mask = exclusive_mask(usage_mask);
2186 
2187 	that.parent = NULL;
2188 	that.class = hlock_class(next);
2189 
2190 	ret = find_usage_forwards(&that, forward_mask, &target_entry1);
2191 	if (ret < 0) {
2192 		print_bfs_bug(ret);
2193 		return 0;
2194 	}
2195 	if (ret == 1)
2196 		return ret;
2197 
2198 	/*
2199 	 * Step 3: we found a bad match! Now retrieve a lock from the backward
2200 	 * list whose usage mask matches the exclusive usage mask from the
2201 	 * lock found on the forward list.
2202 	 */
2203 	backward_mask = original_mask(target_entry1->class->usage_mask);
2204 
2205 	ret = find_usage_backwards(&this, backward_mask, &target_entry);
2206 	if (ret < 0) {
2207 		print_bfs_bug(ret);
2208 		return 0;
2209 	}
2210 	if (DEBUG_LOCKS_WARN_ON(ret == 1))
2211 		return 1;
2212 
2213 	/*
2214 	 * Step 4: narrow down to a pair of incompatible usage bits
2215 	 * and report it.
2216 	 */
2217 	ret = find_exclusive_match(target_entry->class->usage_mask,
2218 				   target_entry1->class->usage_mask,
2219 				   &backward_bit, &forward_bit);
2220 	if (DEBUG_LOCKS_WARN_ON(ret == -1))
2221 		return 1;
2222 
2223 	print_bad_irq_dependency(curr, &this, &that,
2224 				 target_entry, target_entry1,
2225 				 prev, next,
2226 				 backward_bit, forward_bit,
2227 				 state_name(backward_bit));
2228 
2229 	return 0;
2230 }
2231 
2232 static void inc_chains(void)
2233 {
2234 	if (current->hardirq_context)
2235 		nr_hardirq_chains++;
2236 	else {
2237 		if (current->softirq_context)
2238 			nr_softirq_chains++;
2239 		else
2240 			nr_process_chains++;
2241 	}
2242 }
2243 
2244 #else
2245 
2246 static inline int check_irq_usage(struct task_struct *curr,
2247 				  struct held_lock *prev, struct held_lock *next)
2248 {
2249 	return 1;
2250 }
2251 
2252 static inline void inc_chains(void)
2253 {
2254 	nr_process_chains++;
2255 }
2256 
2257 #endif /* CONFIG_TRACE_IRQFLAGS */
2258 
2259 static void
2260 print_deadlock_scenario(struct held_lock *nxt, struct held_lock *prv)
2261 {
2262 	struct lock_class *next = hlock_class(nxt);
2263 	struct lock_class *prev = hlock_class(prv);
2264 
2265 	printk(" Possible unsafe locking scenario:\n\n");
2266 	printk("       CPU0\n");
2267 	printk("       ----\n");
2268 	printk("  lock(");
2269 	__print_lock_name(prev);
2270 	printk(KERN_CONT ");\n");
2271 	printk("  lock(");
2272 	__print_lock_name(next);
2273 	printk(KERN_CONT ");\n");
2274 	printk("\n *** DEADLOCK ***\n\n");
2275 	printk(" May be due to missing lock nesting notation\n\n");
2276 }
2277 
2278 static void
2279 print_deadlock_bug(struct task_struct *curr, struct held_lock *prev,
2280 		   struct held_lock *next)
2281 {
2282 	if (!debug_locks_off_graph_unlock() || debug_locks_silent)
2283 		return;
2284 
2285 	pr_warn("\n");
2286 	pr_warn("============================================\n");
2287 	pr_warn("WARNING: possible recursive locking detected\n");
2288 	print_kernel_ident();
2289 	pr_warn("--------------------------------------------\n");
2290 	pr_warn("%s/%d is trying to acquire lock:\n",
2291 		curr->comm, task_pid_nr(curr));
2292 	print_lock(next);
2293 	pr_warn("\nbut task is already holding lock:\n");
2294 	print_lock(prev);
2295 
2296 	pr_warn("\nother info that might help us debug this:\n");
2297 	print_deadlock_scenario(next, prev);
2298 	lockdep_print_held_locks(curr);
2299 
2300 	pr_warn("\nstack backtrace:\n");
2301 	dump_stack();
2302 }
2303 
2304 /*
2305  * Check whether we are holding such a class already.
2306  *
2307  * (Note that this has to be done separately, because the graph cannot
2308  * detect such classes of deadlocks.)
2309  *
2310  * Returns: 0 on deadlock detected, 1 on OK, 2 on recursive read
2311  */
2312 static int
2313 check_deadlock(struct task_struct *curr, struct held_lock *next)
2314 {
2315 	struct held_lock *prev;
2316 	struct held_lock *nest = NULL;
2317 	int i;
2318 
2319 	for (i = 0; i < curr->lockdep_depth; i++) {
2320 		prev = curr->held_locks + i;
2321 
2322 		if (prev->instance == next->nest_lock)
2323 			nest = prev;
2324 
2325 		if (hlock_class(prev) != hlock_class(next))
2326 			continue;
2327 
2328 		/*
2329 		 * Allow read-after-read recursion of the same
2330 		 * lock class (i.e. read_lock(lock)+read_lock(lock)):
2331 		 */
2332 		if ((next->read == 2) && prev->read)
2333 			return 2;
2334 
2335 		/*
2336 		 * We're holding the nest_lock, which serializes this lock's
2337 		 * nesting behaviour.
2338 		 */
2339 		if (nest)
2340 			return 2;
2341 
2342 		print_deadlock_bug(curr, prev, next);
2343 		return 0;
2344 	}
2345 	return 1;
2346 }
2347 
2348 /*
2349  * There was a chain-cache miss, and we are about to add a new dependency
2350  * to a previous lock. We validate the following rules:
2351  *
2352  *  - would the adding of the <prev> -> <next> dependency create a
2353  *    circular dependency in the graph? [== circular deadlock]
2354  *
2355  *  - does the new prev->next dependency connect any hardirq-safe lock
2356  *    (in the full backwards-subgraph starting at <prev>) with any
2357  *    hardirq-unsafe lock (in the full forwards-subgraph starting at
2358  *    <next>)? [== illegal lock inversion with hardirq contexts]
2359  *
2360  *  - does the new prev->next dependency connect any softirq-safe lock
2361  *    (in the full backwards-subgraph starting at <prev>) with any
2362  *    softirq-unsafe lock (in the full forwards-subgraph starting at
2363  *    <next>)? [== illegal lock inversion with softirq contexts]
2364  *
2365  * any of these scenarios could lead to a deadlock.
2366  *
2367  * Then if all the validations pass, we add the forwards and backwards
2368  * dependency.
2369  */
2370 static int
2371 check_prev_add(struct task_struct *curr, struct held_lock *prev,
2372 	       struct held_lock *next, int distance, struct lock_trace *trace)
2373 {
2374 	struct lock_list *entry;
2375 	int ret;
2376 
2377 	if (!hlock_class(prev)->key || !hlock_class(next)->key) {
2378 		/*
2379 		 * The warning statements below may trigger a use-after-free
2380 		 * of the class name. It is better to trigger a use-after free
2381 		 * and to have the class name most of the time instead of not
2382 		 * having the class name available.
2383 		 */
2384 		WARN_ONCE(!debug_locks_silent && !hlock_class(prev)->key,
2385 			  "Detected use-after-free of lock class %px/%s\n",
2386 			  hlock_class(prev),
2387 			  hlock_class(prev)->name);
2388 		WARN_ONCE(!debug_locks_silent && !hlock_class(next)->key,
2389 			  "Detected use-after-free of lock class %px/%s\n",
2390 			  hlock_class(next),
2391 			  hlock_class(next)->name);
2392 		return 2;
2393 	}
2394 
2395 	/*
2396 	 * Prove that the new <prev> -> <next> dependency would not
2397 	 * create a circular dependency in the graph. (We do this by
2398 	 * a breadth-first search into the graph starting at <next>,
2399 	 * and check whether we can reach <prev>.)
2400 	 *
2401 	 * The search is limited by the size of the circular queue (i.e.,
2402 	 * MAX_CIRCULAR_QUEUE_SIZE) which keeps track of a breadth of nodes
2403 	 * in the graph whose neighbours are to be checked.
2404 	 */
2405 	ret = check_noncircular(next, prev, trace);
2406 	if (unlikely(ret <= 0))
2407 		return 0;
2408 
2409 	if (!check_irq_usage(curr, prev, next))
2410 		return 0;
2411 
2412 	/*
2413 	 * For recursive read-locks we do all the dependency checks,
2414 	 * but we dont store read-triggered dependencies (only
2415 	 * write-triggered dependencies). This ensures that only the
2416 	 * write-side dependencies matter, and that if for example a
2417 	 * write-lock never takes any other locks, then the reads are
2418 	 * equivalent to a NOP.
2419 	 */
2420 	if (next->read == 2 || prev->read == 2)
2421 		return 1;
2422 	/*
2423 	 * Is the <prev> -> <next> dependency already present?
2424 	 *
2425 	 * (this may occur even though this is a new chain: consider
2426 	 *  e.g. the L1 -> L2 -> L3 -> L4 and the L5 -> L1 -> L2 -> L3
2427 	 *  chains - the second one will be new, but L1 already has
2428 	 *  L2 added to its dependency list, due to the first chain.)
2429 	 */
2430 	list_for_each_entry(entry, &hlock_class(prev)->locks_after, entry) {
2431 		if (entry->class == hlock_class(next)) {
2432 			if (distance == 1)
2433 				entry->distance = 1;
2434 			return 1;
2435 		}
2436 	}
2437 
2438 #ifdef CONFIG_LOCKDEP_SMALL
2439 	/*
2440 	 * Is the <prev> -> <next> link redundant?
2441 	 */
2442 	ret = check_redundant(prev, next);
2443 	if (ret != 1)
2444 		return ret;
2445 #endif
2446 
2447 	if (!trace->nr_entries && !save_trace(trace))
2448 		return 0;
2449 
2450 	/*
2451 	 * Ok, all validations passed, add the new lock
2452 	 * to the previous lock's dependency list:
2453 	 */
2454 	ret = add_lock_to_list(hlock_class(next), hlock_class(prev),
2455 			       &hlock_class(prev)->locks_after,
2456 			       next->acquire_ip, distance, trace);
2457 
2458 	if (!ret)
2459 		return 0;
2460 
2461 	ret = add_lock_to_list(hlock_class(prev), hlock_class(next),
2462 			       &hlock_class(next)->locks_before,
2463 			       next->acquire_ip, distance, trace);
2464 	if (!ret)
2465 		return 0;
2466 
2467 	return 2;
2468 }
2469 
2470 /*
2471  * Add the dependency to all directly-previous locks that are 'relevant'.
2472  * The ones that are relevant are (in increasing distance from curr):
2473  * all consecutive trylock entries and the final non-trylock entry - or
2474  * the end of this context's lock-chain - whichever comes first.
2475  */
2476 static int
2477 check_prevs_add(struct task_struct *curr, struct held_lock *next)
2478 {
2479 	struct lock_trace trace = { .nr_entries = 0 };
2480 	int depth = curr->lockdep_depth;
2481 	struct held_lock *hlock;
2482 
2483 	/*
2484 	 * Debugging checks.
2485 	 *
2486 	 * Depth must not be zero for a non-head lock:
2487 	 */
2488 	if (!depth)
2489 		goto out_bug;
2490 	/*
2491 	 * At least two relevant locks must exist for this
2492 	 * to be a head:
2493 	 */
2494 	if (curr->held_locks[depth].irq_context !=
2495 			curr->held_locks[depth-1].irq_context)
2496 		goto out_bug;
2497 
2498 	for (;;) {
2499 		int distance = curr->lockdep_depth - depth + 1;
2500 		hlock = curr->held_locks + depth - 1;
2501 
2502 		/*
2503 		 * Only non-recursive-read entries get new dependencies
2504 		 * added:
2505 		 */
2506 		if (hlock->read != 2 && hlock->check) {
2507 			int ret = check_prev_add(curr, hlock, next, distance,
2508 						 &trace);
2509 			if (!ret)
2510 				return 0;
2511 
2512 			/*
2513 			 * Stop after the first non-trylock entry,
2514 			 * as non-trylock entries have added their
2515 			 * own direct dependencies already, so this
2516 			 * lock is connected to them indirectly:
2517 			 */
2518 			if (!hlock->trylock)
2519 				break;
2520 		}
2521 
2522 		depth--;
2523 		/*
2524 		 * End of lock-stack?
2525 		 */
2526 		if (!depth)
2527 			break;
2528 		/*
2529 		 * Stop the search if we cross into another context:
2530 		 */
2531 		if (curr->held_locks[depth].irq_context !=
2532 				curr->held_locks[depth-1].irq_context)
2533 			break;
2534 	}
2535 	return 1;
2536 out_bug:
2537 	if (!debug_locks_off_graph_unlock())
2538 		return 0;
2539 
2540 	/*
2541 	 * Clearly we all shouldn't be here, but since we made it we
2542 	 * can reliable say we messed up our state. See the above two
2543 	 * gotos for reasons why we could possibly end up here.
2544 	 */
2545 	WARN_ON(1);
2546 
2547 	return 0;
2548 }
2549 
2550 struct lock_chain lock_chains[MAX_LOCKDEP_CHAINS];
2551 static DECLARE_BITMAP(lock_chains_in_use, MAX_LOCKDEP_CHAINS);
2552 int nr_chain_hlocks;
2553 static u16 chain_hlocks[MAX_LOCKDEP_CHAIN_HLOCKS];
2554 
2555 struct lock_class *lock_chain_get_class(struct lock_chain *chain, int i)
2556 {
2557 	return lock_classes + chain_hlocks[chain->base + i];
2558 }
2559 
2560 /*
2561  * Returns the index of the first held_lock of the current chain
2562  */
2563 static inline int get_first_held_lock(struct task_struct *curr,
2564 					struct held_lock *hlock)
2565 {
2566 	int i;
2567 	struct held_lock *hlock_curr;
2568 
2569 	for (i = curr->lockdep_depth - 1; i >= 0; i--) {
2570 		hlock_curr = curr->held_locks + i;
2571 		if (hlock_curr->irq_context != hlock->irq_context)
2572 			break;
2573 
2574 	}
2575 
2576 	return ++i;
2577 }
2578 
2579 #ifdef CONFIG_DEBUG_LOCKDEP
2580 /*
2581  * Returns the next chain_key iteration
2582  */
2583 static u64 print_chain_key_iteration(int class_idx, u64 chain_key)
2584 {
2585 	u64 new_chain_key = iterate_chain_key(chain_key, class_idx);
2586 
2587 	printk(" class_idx:%d -> chain_key:%016Lx",
2588 		class_idx,
2589 		(unsigned long long)new_chain_key);
2590 	return new_chain_key;
2591 }
2592 
2593 static void
2594 print_chain_keys_held_locks(struct task_struct *curr, struct held_lock *hlock_next)
2595 {
2596 	struct held_lock *hlock;
2597 	u64 chain_key = INITIAL_CHAIN_KEY;
2598 	int depth = curr->lockdep_depth;
2599 	int i = get_first_held_lock(curr, hlock_next);
2600 
2601 	printk("depth: %u (irq_context %u)\n", depth - i + 1,
2602 		hlock_next->irq_context);
2603 	for (; i < depth; i++) {
2604 		hlock = curr->held_locks + i;
2605 		chain_key = print_chain_key_iteration(hlock->class_idx, chain_key);
2606 
2607 		print_lock(hlock);
2608 	}
2609 
2610 	print_chain_key_iteration(hlock_next->class_idx, chain_key);
2611 	print_lock(hlock_next);
2612 }
2613 
2614 static void print_chain_keys_chain(struct lock_chain *chain)
2615 {
2616 	int i;
2617 	u64 chain_key = INITIAL_CHAIN_KEY;
2618 	int class_id;
2619 
2620 	printk("depth: %u\n", chain->depth);
2621 	for (i = 0; i < chain->depth; i++) {
2622 		class_id = chain_hlocks[chain->base + i];
2623 		chain_key = print_chain_key_iteration(class_id, chain_key);
2624 
2625 		print_lock_name(lock_classes + class_id);
2626 		printk("\n");
2627 	}
2628 }
2629 
2630 static void print_collision(struct task_struct *curr,
2631 			struct held_lock *hlock_next,
2632 			struct lock_chain *chain)
2633 {
2634 	pr_warn("\n");
2635 	pr_warn("============================\n");
2636 	pr_warn("WARNING: chain_key collision\n");
2637 	print_kernel_ident();
2638 	pr_warn("----------------------------\n");
2639 	pr_warn("%s/%d: ", current->comm, task_pid_nr(current));
2640 	pr_warn("Hash chain already cached but the contents don't match!\n");
2641 
2642 	pr_warn("Held locks:");
2643 	print_chain_keys_held_locks(curr, hlock_next);
2644 
2645 	pr_warn("Locks in cached chain:");
2646 	print_chain_keys_chain(chain);
2647 
2648 	pr_warn("\nstack backtrace:\n");
2649 	dump_stack();
2650 }
2651 #endif
2652 
2653 /*
2654  * Checks whether the chain and the current held locks are consistent
2655  * in depth and also in content. If they are not it most likely means
2656  * that there was a collision during the calculation of the chain_key.
2657  * Returns: 0 not passed, 1 passed
2658  */
2659 static int check_no_collision(struct task_struct *curr,
2660 			struct held_lock *hlock,
2661 			struct lock_chain *chain)
2662 {
2663 #ifdef CONFIG_DEBUG_LOCKDEP
2664 	int i, j, id;
2665 
2666 	i = get_first_held_lock(curr, hlock);
2667 
2668 	if (DEBUG_LOCKS_WARN_ON(chain->depth != curr->lockdep_depth - (i - 1))) {
2669 		print_collision(curr, hlock, chain);
2670 		return 0;
2671 	}
2672 
2673 	for (j = 0; j < chain->depth - 1; j++, i++) {
2674 		id = curr->held_locks[i].class_idx;
2675 
2676 		if (DEBUG_LOCKS_WARN_ON(chain_hlocks[chain->base + j] != id)) {
2677 			print_collision(curr, hlock, chain);
2678 			return 0;
2679 		}
2680 	}
2681 #endif
2682 	return 1;
2683 }
2684 
2685 /*
2686  * Given an index that is >= -1, return the index of the next lock chain.
2687  * Return -2 if there is no next lock chain.
2688  */
2689 long lockdep_next_lockchain(long i)
2690 {
2691 	i = find_next_bit(lock_chains_in_use, ARRAY_SIZE(lock_chains), i + 1);
2692 	return i < ARRAY_SIZE(lock_chains) ? i : -2;
2693 }
2694 
2695 unsigned long lock_chain_count(void)
2696 {
2697 	return bitmap_weight(lock_chains_in_use, ARRAY_SIZE(lock_chains));
2698 }
2699 
2700 /* Must be called with the graph lock held. */
2701 static struct lock_chain *alloc_lock_chain(void)
2702 {
2703 	int idx = find_first_zero_bit(lock_chains_in_use,
2704 				      ARRAY_SIZE(lock_chains));
2705 
2706 	if (unlikely(idx >= ARRAY_SIZE(lock_chains)))
2707 		return NULL;
2708 	__set_bit(idx, lock_chains_in_use);
2709 	return lock_chains + idx;
2710 }
2711 
2712 /*
2713  * Adds a dependency chain into chain hashtable. And must be called with
2714  * graph_lock held.
2715  *
2716  * Return 0 if fail, and graph_lock is released.
2717  * Return 1 if succeed, with graph_lock held.
2718  */
2719 static inline int add_chain_cache(struct task_struct *curr,
2720 				  struct held_lock *hlock,
2721 				  u64 chain_key)
2722 {
2723 	struct lock_class *class = hlock_class(hlock);
2724 	struct hlist_head *hash_head = chainhashentry(chain_key);
2725 	struct lock_chain *chain;
2726 	int i, j;
2727 
2728 	/*
2729 	 * The caller must hold the graph lock, ensure we've got IRQs
2730 	 * disabled to make this an IRQ-safe lock.. for recursion reasons
2731 	 * lockdep won't complain about its own locking errors.
2732 	 */
2733 	if (DEBUG_LOCKS_WARN_ON(!irqs_disabled()))
2734 		return 0;
2735 
2736 	chain = alloc_lock_chain();
2737 	if (!chain) {
2738 		if (!debug_locks_off_graph_unlock())
2739 			return 0;
2740 
2741 		print_lockdep_off("BUG: MAX_LOCKDEP_CHAINS too low!");
2742 		dump_stack();
2743 		return 0;
2744 	}
2745 	chain->chain_key = chain_key;
2746 	chain->irq_context = hlock->irq_context;
2747 	i = get_first_held_lock(curr, hlock);
2748 	chain->depth = curr->lockdep_depth + 1 - i;
2749 
2750 	BUILD_BUG_ON((1UL << 24) <= ARRAY_SIZE(chain_hlocks));
2751 	BUILD_BUG_ON((1UL << 6)  <= ARRAY_SIZE(curr->held_locks));
2752 	BUILD_BUG_ON((1UL << 8*sizeof(chain_hlocks[0])) <= ARRAY_SIZE(lock_classes));
2753 
2754 	if (likely(nr_chain_hlocks + chain->depth <= MAX_LOCKDEP_CHAIN_HLOCKS)) {
2755 		chain->base = nr_chain_hlocks;
2756 		for (j = 0; j < chain->depth - 1; j++, i++) {
2757 			int lock_id = curr->held_locks[i].class_idx;
2758 			chain_hlocks[chain->base + j] = lock_id;
2759 		}
2760 		chain_hlocks[chain->base + j] = class - lock_classes;
2761 		nr_chain_hlocks += chain->depth;
2762 	} else {
2763 		if (!debug_locks_off_graph_unlock())
2764 			return 0;
2765 
2766 		print_lockdep_off("BUG: MAX_LOCKDEP_CHAIN_HLOCKS too low!");
2767 		dump_stack();
2768 		return 0;
2769 	}
2770 
2771 	hlist_add_head_rcu(&chain->entry, hash_head);
2772 	debug_atomic_inc(chain_lookup_misses);
2773 	inc_chains();
2774 
2775 	return 1;
2776 }
2777 
2778 /*
2779  * Look up a dependency chain. Must be called with either the graph lock or
2780  * the RCU read lock held.
2781  */
2782 static inline struct lock_chain *lookup_chain_cache(u64 chain_key)
2783 {
2784 	struct hlist_head *hash_head = chainhashentry(chain_key);
2785 	struct lock_chain *chain;
2786 
2787 	hlist_for_each_entry_rcu(chain, hash_head, entry) {
2788 		if (READ_ONCE(chain->chain_key) == chain_key) {
2789 			debug_atomic_inc(chain_lookup_hits);
2790 			return chain;
2791 		}
2792 	}
2793 	return NULL;
2794 }
2795 
2796 /*
2797  * If the key is not present yet in dependency chain cache then
2798  * add it and return 1 - in this case the new dependency chain is
2799  * validated. If the key is already hashed, return 0.
2800  * (On return with 1 graph_lock is held.)
2801  */
2802 static inline int lookup_chain_cache_add(struct task_struct *curr,
2803 					 struct held_lock *hlock,
2804 					 u64 chain_key)
2805 {
2806 	struct lock_class *class = hlock_class(hlock);
2807 	struct lock_chain *chain = lookup_chain_cache(chain_key);
2808 
2809 	if (chain) {
2810 cache_hit:
2811 		if (!check_no_collision(curr, hlock, chain))
2812 			return 0;
2813 
2814 		if (very_verbose(class)) {
2815 			printk("\nhash chain already cached, key: "
2816 					"%016Lx tail class: [%px] %s\n",
2817 					(unsigned long long)chain_key,
2818 					class->key, class->name);
2819 		}
2820 
2821 		return 0;
2822 	}
2823 
2824 	if (very_verbose(class)) {
2825 		printk("\nnew hash chain, key: %016Lx tail class: [%px] %s\n",
2826 			(unsigned long long)chain_key, class->key, class->name);
2827 	}
2828 
2829 	if (!graph_lock())
2830 		return 0;
2831 
2832 	/*
2833 	 * We have to walk the chain again locked - to avoid duplicates:
2834 	 */
2835 	chain = lookup_chain_cache(chain_key);
2836 	if (chain) {
2837 		graph_unlock();
2838 		goto cache_hit;
2839 	}
2840 
2841 	if (!add_chain_cache(curr, hlock, chain_key))
2842 		return 0;
2843 
2844 	return 1;
2845 }
2846 
2847 static int validate_chain(struct task_struct *curr,
2848 			  struct held_lock *hlock,
2849 			  int chain_head, u64 chain_key)
2850 {
2851 	/*
2852 	 * Trylock needs to maintain the stack of held locks, but it
2853 	 * does not add new dependencies, because trylock can be done
2854 	 * in any order.
2855 	 *
2856 	 * We look up the chain_key and do the O(N^2) check and update of
2857 	 * the dependencies only if this is a new dependency chain.
2858 	 * (If lookup_chain_cache_add() return with 1 it acquires
2859 	 * graph_lock for us)
2860 	 */
2861 	if (!hlock->trylock && hlock->check &&
2862 	    lookup_chain_cache_add(curr, hlock, chain_key)) {
2863 		/*
2864 		 * Check whether last held lock:
2865 		 *
2866 		 * - is irq-safe, if this lock is irq-unsafe
2867 		 * - is softirq-safe, if this lock is hardirq-unsafe
2868 		 *
2869 		 * And check whether the new lock's dependency graph
2870 		 * could lead back to the previous lock:
2871 		 *
2872 		 * - within the current held-lock stack
2873 		 * - across our accumulated lock dependency records
2874 		 *
2875 		 * any of these scenarios could lead to a deadlock.
2876 		 */
2877 		/*
2878 		 * The simple case: does the current hold the same lock
2879 		 * already?
2880 		 */
2881 		int ret = check_deadlock(curr, hlock);
2882 
2883 		if (!ret)
2884 			return 0;
2885 		/*
2886 		 * Mark recursive read, as we jump over it when
2887 		 * building dependencies (just like we jump over
2888 		 * trylock entries):
2889 		 */
2890 		if (ret == 2)
2891 			hlock->read = 2;
2892 		/*
2893 		 * Add dependency only if this lock is not the head
2894 		 * of the chain, and if it's not a secondary read-lock:
2895 		 */
2896 		if (!chain_head && ret != 2) {
2897 			if (!check_prevs_add(curr, hlock))
2898 				return 0;
2899 		}
2900 
2901 		graph_unlock();
2902 	} else {
2903 		/* after lookup_chain_cache_add(): */
2904 		if (unlikely(!debug_locks))
2905 			return 0;
2906 	}
2907 
2908 	return 1;
2909 }
2910 #else
2911 static inline int validate_chain(struct task_struct *curr,
2912 				 struct held_lock *hlock,
2913 				 int chain_head, u64 chain_key)
2914 {
2915 	return 1;
2916 }
2917 #endif /* CONFIG_PROVE_LOCKING */
2918 
2919 /*
2920  * We are building curr_chain_key incrementally, so double-check
2921  * it from scratch, to make sure that it's done correctly:
2922  */
2923 static void check_chain_key(struct task_struct *curr)
2924 {
2925 #ifdef CONFIG_DEBUG_LOCKDEP
2926 	struct held_lock *hlock, *prev_hlock = NULL;
2927 	unsigned int i;
2928 	u64 chain_key = INITIAL_CHAIN_KEY;
2929 
2930 	for (i = 0; i < curr->lockdep_depth; i++) {
2931 		hlock = curr->held_locks + i;
2932 		if (chain_key != hlock->prev_chain_key) {
2933 			debug_locks_off();
2934 			/*
2935 			 * We got mighty confused, our chain keys don't match
2936 			 * with what we expect, someone trample on our task state?
2937 			 */
2938 			WARN(1, "hm#1, depth: %u [%u], %016Lx != %016Lx\n",
2939 				curr->lockdep_depth, i,
2940 				(unsigned long long)chain_key,
2941 				(unsigned long long)hlock->prev_chain_key);
2942 			return;
2943 		}
2944 
2945 		/*
2946 		 * hlock->class_idx can't go beyond MAX_LOCKDEP_KEYS, but is
2947 		 * it registered lock class index?
2948 		 */
2949 		if (DEBUG_LOCKS_WARN_ON(!test_bit(hlock->class_idx, lock_classes_in_use)))
2950 			return;
2951 
2952 		if (prev_hlock && (prev_hlock->irq_context !=
2953 							hlock->irq_context))
2954 			chain_key = INITIAL_CHAIN_KEY;
2955 		chain_key = iterate_chain_key(chain_key, hlock->class_idx);
2956 		prev_hlock = hlock;
2957 	}
2958 	if (chain_key != curr->curr_chain_key) {
2959 		debug_locks_off();
2960 		/*
2961 		 * More smoking hash instead of calculating it, damn see these
2962 		 * numbers float.. I bet that a pink elephant stepped on my memory.
2963 		 */
2964 		WARN(1, "hm#2, depth: %u [%u], %016Lx != %016Lx\n",
2965 			curr->lockdep_depth, i,
2966 			(unsigned long long)chain_key,
2967 			(unsigned long long)curr->curr_chain_key);
2968 	}
2969 #endif
2970 }
2971 
2972 #ifdef CONFIG_PROVE_LOCKING
2973 static int mark_lock(struct task_struct *curr, struct held_lock *this,
2974 		     enum lock_usage_bit new_bit);
2975 
2976 static void print_usage_bug_scenario(struct held_lock *lock)
2977 {
2978 	struct lock_class *class = hlock_class(lock);
2979 
2980 	printk(" Possible unsafe locking scenario:\n\n");
2981 	printk("       CPU0\n");
2982 	printk("       ----\n");
2983 	printk("  lock(");
2984 	__print_lock_name(class);
2985 	printk(KERN_CONT ");\n");
2986 	printk("  <Interrupt>\n");
2987 	printk("    lock(");
2988 	__print_lock_name(class);
2989 	printk(KERN_CONT ");\n");
2990 	printk("\n *** DEADLOCK ***\n\n");
2991 }
2992 
2993 static void
2994 print_usage_bug(struct task_struct *curr, struct held_lock *this,
2995 		enum lock_usage_bit prev_bit, enum lock_usage_bit new_bit)
2996 {
2997 	if (!debug_locks_off_graph_unlock() || debug_locks_silent)
2998 		return;
2999 
3000 	pr_warn("\n");
3001 	pr_warn("================================\n");
3002 	pr_warn("WARNING: inconsistent lock state\n");
3003 	print_kernel_ident();
3004 	pr_warn("--------------------------------\n");
3005 
3006 	pr_warn("inconsistent {%s} -> {%s} usage.\n",
3007 		usage_str[prev_bit], usage_str[new_bit]);
3008 
3009 	pr_warn("%s/%d [HC%u[%lu]:SC%u[%lu]:HE%u:SE%u] takes:\n",
3010 		curr->comm, task_pid_nr(curr),
3011 		trace_hardirq_context(curr), hardirq_count() >> HARDIRQ_SHIFT,
3012 		trace_softirq_context(curr), softirq_count() >> SOFTIRQ_SHIFT,
3013 		trace_hardirqs_enabled(curr),
3014 		trace_softirqs_enabled(curr));
3015 	print_lock(this);
3016 
3017 	pr_warn("{%s} state was registered at:\n", usage_str[prev_bit]);
3018 	print_lock_trace(hlock_class(this)->usage_traces + prev_bit, 1);
3019 
3020 	print_irqtrace_events(curr);
3021 	pr_warn("\nother info that might help us debug this:\n");
3022 	print_usage_bug_scenario(this);
3023 
3024 	lockdep_print_held_locks(curr);
3025 
3026 	pr_warn("\nstack backtrace:\n");
3027 	dump_stack();
3028 }
3029 
3030 /*
3031  * Print out an error if an invalid bit is set:
3032  */
3033 static inline int
3034 valid_state(struct task_struct *curr, struct held_lock *this,
3035 	    enum lock_usage_bit new_bit, enum lock_usage_bit bad_bit)
3036 {
3037 	if (unlikely(hlock_class(this)->usage_mask & (1 << bad_bit))) {
3038 		print_usage_bug(curr, this, bad_bit, new_bit);
3039 		return 0;
3040 	}
3041 	return 1;
3042 }
3043 
3044 
3045 /*
3046  * print irq inversion bug:
3047  */
3048 static void
3049 print_irq_inversion_bug(struct task_struct *curr,
3050 			struct lock_list *root, struct lock_list *other,
3051 			struct held_lock *this, int forwards,
3052 			const char *irqclass)
3053 {
3054 	struct lock_list *entry = other;
3055 	struct lock_list *middle = NULL;
3056 	int depth;
3057 
3058 	if (!debug_locks_off_graph_unlock() || debug_locks_silent)
3059 		return;
3060 
3061 	pr_warn("\n");
3062 	pr_warn("========================================================\n");
3063 	pr_warn("WARNING: possible irq lock inversion dependency detected\n");
3064 	print_kernel_ident();
3065 	pr_warn("--------------------------------------------------------\n");
3066 	pr_warn("%s/%d just changed the state of lock:\n",
3067 		curr->comm, task_pid_nr(curr));
3068 	print_lock(this);
3069 	if (forwards)
3070 		pr_warn("but this lock took another, %s-unsafe lock in the past:\n", irqclass);
3071 	else
3072 		pr_warn("but this lock was taken by another, %s-safe lock in the past:\n", irqclass);
3073 	print_lock_name(other->class);
3074 	pr_warn("\n\nand interrupts could create inverse lock ordering between them.\n\n");
3075 
3076 	pr_warn("\nother info that might help us debug this:\n");
3077 
3078 	/* Find a middle lock (if one exists) */
3079 	depth = get_lock_depth(other);
3080 	do {
3081 		if (depth == 0 && (entry != root)) {
3082 			pr_warn("lockdep:%s bad path found in chain graph\n", __func__);
3083 			break;
3084 		}
3085 		middle = entry;
3086 		entry = get_lock_parent(entry);
3087 		depth--;
3088 	} while (entry && entry != root && (depth >= 0));
3089 	if (forwards)
3090 		print_irq_lock_scenario(root, other,
3091 			middle ? middle->class : root->class, other->class);
3092 	else
3093 		print_irq_lock_scenario(other, root,
3094 			middle ? middle->class : other->class, root->class);
3095 
3096 	lockdep_print_held_locks(curr);
3097 
3098 	pr_warn("\nthe shortest dependencies between 2nd lock and 1st lock:\n");
3099 	if (!save_trace(&root->trace))
3100 		return;
3101 	print_shortest_lock_dependencies(other, root);
3102 
3103 	pr_warn("\nstack backtrace:\n");
3104 	dump_stack();
3105 }
3106 
3107 /*
3108  * Prove that in the forwards-direction subgraph starting at <this>
3109  * there is no lock matching <mask>:
3110  */
3111 static int
3112 check_usage_forwards(struct task_struct *curr, struct held_lock *this,
3113 		     enum lock_usage_bit bit, const char *irqclass)
3114 {
3115 	int ret;
3116 	struct lock_list root;
3117 	struct lock_list *uninitialized_var(target_entry);
3118 
3119 	root.parent = NULL;
3120 	root.class = hlock_class(this);
3121 	ret = find_usage_forwards(&root, lock_flag(bit), &target_entry);
3122 	if (ret < 0) {
3123 		print_bfs_bug(ret);
3124 		return 0;
3125 	}
3126 	if (ret == 1)
3127 		return ret;
3128 
3129 	print_irq_inversion_bug(curr, &root, target_entry,
3130 				this, 1, irqclass);
3131 	return 0;
3132 }
3133 
3134 /*
3135  * Prove that in the backwards-direction subgraph starting at <this>
3136  * there is no lock matching <mask>:
3137  */
3138 static int
3139 check_usage_backwards(struct task_struct *curr, struct held_lock *this,
3140 		      enum lock_usage_bit bit, const char *irqclass)
3141 {
3142 	int ret;
3143 	struct lock_list root;
3144 	struct lock_list *uninitialized_var(target_entry);
3145 
3146 	root.parent = NULL;
3147 	root.class = hlock_class(this);
3148 	ret = find_usage_backwards(&root, lock_flag(bit), &target_entry);
3149 	if (ret < 0) {
3150 		print_bfs_bug(ret);
3151 		return 0;
3152 	}
3153 	if (ret == 1)
3154 		return ret;
3155 
3156 	print_irq_inversion_bug(curr, &root, target_entry,
3157 				this, 0, irqclass);
3158 	return 0;
3159 }
3160 
3161 void print_irqtrace_events(struct task_struct *curr)
3162 {
3163 	printk("irq event stamp: %u\n", curr->irq_events);
3164 	printk("hardirqs last  enabled at (%u): [<%px>] %pS\n",
3165 		curr->hardirq_enable_event, (void *)curr->hardirq_enable_ip,
3166 		(void *)curr->hardirq_enable_ip);
3167 	printk("hardirqs last disabled at (%u): [<%px>] %pS\n",
3168 		curr->hardirq_disable_event, (void *)curr->hardirq_disable_ip,
3169 		(void *)curr->hardirq_disable_ip);
3170 	printk("softirqs last  enabled at (%u): [<%px>] %pS\n",
3171 		curr->softirq_enable_event, (void *)curr->softirq_enable_ip,
3172 		(void *)curr->softirq_enable_ip);
3173 	printk("softirqs last disabled at (%u): [<%px>] %pS\n",
3174 		curr->softirq_disable_event, (void *)curr->softirq_disable_ip,
3175 		(void *)curr->softirq_disable_ip);
3176 }
3177 
3178 static int HARDIRQ_verbose(struct lock_class *class)
3179 {
3180 #if HARDIRQ_VERBOSE
3181 	return class_filter(class);
3182 #endif
3183 	return 0;
3184 }
3185 
3186 static int SOFTIRQ_verbose(struct lock_class *class)
3187 {
3188 #if SOFTIRQ_VERBOSE
3189 	return class_filter(class);
3190 #endif
3191 	return 0;
3192 }
3193 
3194 #define STRICT_READ_CHECKS	1
3195 
3196 static int (*state_verbose_f[])(struct lock_class *class) = {
3197 #define LOCKDEP_STATE(__STATE) \
3198 	__STATE##_verbose,
3199 #include "lockdep_states.h"
3200 #undef LOCKDEP_STATE
3201 };
3202 
3203 static inline int state_verbose(enum lock_usage_bit bit,
3204 				struct lock_class *class)
3205 {
3206 	return state_verbose_f[bit >> LOCK_USAGE_DIR_MASK](class);
3207 }
3208 
3209 typedef int (*check_usage_f)(struct task_struct *, struct held_lock *,
3210 			     enum lock_usage_bit bit, const char *name);
3211 
3212 static int
3213 mark_lock_irq(struct task_struct *curr, struct held_lock *this,
3214 		enum lock_usage_bit new_bit)
3215 {
3216 	int excl_bit = exclusive_bit(new_bit);
3217 	int read = new_bit & LOCK_USAGE_READ_MASK;
3218 	int dir = new_bit & LOCK_USAGE_DIR_MASK;
3219 
3220 	/*
3221 	 * mark USED_IN has to look forwards -- to ensure no dependency
3222 	 * has ENABLED state, which would allow recursion deadlocks.
3223 	 *
3224 	 * mark ENABLED has to look backwards -- to ensure no dependee
3225 	 * has USED_IN state, which, again, would allow  recursion deadlocks.
3226 	 */
3227 	check_usage_f usage = dir ?
3228 		check_usage_backwards : check_usage_forwards;
3229 
3230 	/*
3231 	 * Validate that this particular lock does not have conflicting
3232 	 * usage states.
3233 	 */
3234 	if (!valid_state(curr, this, new_bit, excl_bit))
3235 		return 0;
3236 
3237 	/*
3238 	 * Validate that the lock dependencies don't have conflicting usage
3239 	 * states.
3240 	 */
3241 	if ((!read || STRICT_READ_CHECKS) &&
3242 			!usage(curr, this, excl_bit, state_name(new_bit & ~LOCK_USAGE_READ_MASK)))
3243 		return 0;
3244 
3245 	/*
3246 	 * Check for read in write conflicts
3247 	 */
3248 	if (!read) {
3249 		if (!valid_state(curr, this, new_bit, excl_bit + LOCK_USAGE_READ_MASK))
3250 			return 0;
3251 
3252 		if (STRICT_READ_CHECKS &&
3253 			!usage(curr, this, excl_bit + LOCK_USAGE_READ_MASK,
3254 				state_name(new_bit + LOCK_USAGE_READ_MASK)))
3255 			return 0;
3256 	}
3257 
3258 	if (state_verbose(new_bit, hlock_class(this)))
3259 		return 2;
3260 
3261 	return 1;
3262 }
3263 
3264 /*
3265  * Mark all held locks with a usage bit:
3266  */
3267 static int
3268 mark_held_locks(struct task_struct *curr, enum lock_usage_bit base_bit)
3269 {
3270 	struct held_lock *hlock;
3271 	int i;
3272 
3273 	for (i = 0; i < curr->lockdep_depth; i++) {
3274 		enum lock_usage_bit hlock_bit = base_bit;
3275 		hlock = curr->held_locks + i;
3276 
3277 		if (hlock->read)
3278 			hlock_bit += LOCK_USAGE_READ_MASK;
3279 
3280 		BUG_ON(hlock_bit >= LOCK_USAGE_STATES);
3281 
3282 		if (!hlock->check)
3283 			continue;
3284 
3285 		if (!mark_lock(curr, hlock, hlock_bit))
3286 			return 0;
3287 	}
3288 
3289 	return 1;
3290 }
3291 
3292 /*
3293  * Hardirqs will be enabled:
3294  */
3295 static void __trace_hardirqs_on_caller(unsigned long ip)
3296 {
3297 	struct task_struct *curr = current;
3298 
3299 	/* we'll do an OFF -> ON transition: */
3300 	curr->hardirqs_enabled = 1;
3301 
3302 	/*
3303 	 * We are going to turn hardirqs on, so set the
3304 	 * usage bit for all held locks:
3305 	 */
3306 	if (!mark_held_locks(curr, LOCK_ENABLED_HARDIRQ))
3307 		return;
3308 	/*
3309 	 * If we have softirqs enabled, then set the usage
3310 	 * bit for all held locks. (disabled hardirqs prevented
3311 	 * this bit from being set before)
3312 	 */
3313 	if (curr->softirqs_enabled)
3314 		if (!mark_held_locks(curr, LOCK_ENABLED_SOFTIRQ))
3315 			return;
3316 
3317 	curr->hardirq_enable_ip = ip;
3318 	curr->hardirq_enable_event = ++curr->irq_events;
3319 	debug_atomic_inc(hardirqs_on_events);
3320 }
3321 
3322 void lockdep_hardirqs_on(unsigned long ip)
3323 {
3324 	if (unlikely(!debug_locks || current->lockdep_recursion))
3325 		return;
3326 
3327 	if (unlikely(current->hardirqs_enabled)) {
3328 		/*
3329 		 * Neither irq nor preemption are disabled here
3330 		 * so this is racy by nature but losing one hit
3331 		 * in a stat is not a big deal.
3332 		 */
3333 		__debug_atomic_inc(redundant_hardirqs_on);
3334 		return;
3335 	}
3336 
3337 	/*
3338 	 * We're enabling irqs and according to our state above irqs weren't
3339 	 * already enabled, yet we find the hardware thinks they are in fact
3340 	 * enabled.. someone messed up their IRQ state tracing.
3341 	 */
3342 	if (DEBUG_LOCKS_WARN_ON(!irqs_disabled()))
3343 		return;
3344 
3345 	/*
3346 	 * See the fine text that goes along with this variable definition.
3347 	 */
3348 	if (DEBUG_LOCKS_WARN_ON(early_boot_irqs_disabled))
3349 		return;
3350 
3351 	/*
3352 	 * Can't allow enabling interrupts while in an interrupt handler,
3353 	 * that's general bad form and such. Recursion, limited stack etc..
3354 	 */
3355 	if (DEBUG_LOCKS_WARN_ON(current->hardirq_context))
3356 		return;
3357 
3358 	current->lockdep_recursion = 1;
3359 	__trace_hardirqs_on_caller(ip);
3360 	current->lockdep_recursion = 0;
3361 }
3362 NOKPROBE_SYMBOL(lockdep_hardirqs_on);
3363 
3364 /*
3365  * Hardirqs were disabled:
3366  */
3367 void lockdep_hardirqs_off(unsigned long ip)
3368 {
3369 	struct task_struct *curr = current;
3370 
3371 	if (unlikely(!debug_locks || current->lockdep_recursion))
3372 		return;
3373 
3374 	/*
3375 	 * So we're supposed to get called after you mask local IRQs, but for
3376 	 * some reason the hardware doesn't quite think you did a proper job.
3377 	 */
3378 	if (DEBUG_LOCKS_WARN_ON(!irqs_disabled()))
3379 		return;
3380 
3381 	if (curr->hardirqs_enabled) {
3382 		/*
3383 		 * We have done an ON -> OFF transition:
3384 		 */
3385 		curr->hardirqs_enabled = 0;
3386 		curr->hardirq_disable_ip = ip;
3387 		curr->hardirq_disable_event = ++curr->irq_events;
3388 		debug_atomic_inc(hardirqs_off_events);
3389 	} else
3390 		debug_atomic_inc(redundant_hardirqs_off);
3391 }
3392 NOKPROBE_SYMBOL(lockdep_hardirqs_off);
3393 
3394 /*
3395  * Softirqs will be enabled:
3396  */
3397 void trace_softirqs_on(unsigned long ip)
3398 {
3399 	struct task_struct *curr = current;
3400 
3401 	if (unlikely(!debug_locks || current->lockdep_recursion))
3402 		return;
3403 
3404 	/*
3405 	 * We fancy IRQs being disabled here, see softirq.c, avoids
3406 	 * funny state and nesting things.
3407 	 */
3408 	if (DEBUG_LOCKS_WARN_ON(!irqs_disabled()))
3409 		return;
3410 
3411 	if (curr->softirqs_enabled) {
3412 		debug_atomic_inc(redundant_softirqs_on);
3413 		return;
3414 	}
3415 
3416 	current->lockdep_recursion = 1;
3417 	/*
3418 	 * We'll do an OFF -> ON transition:
3419 	 */
3420 	curr->softirqs_enabled = 1;
3421 	curr->softirq_enable_ip = ip;
3422 	curr->softirq_enable_event = ++curr->irq_events;
3423 	debug_atomic_inc(softirqs_on_events);
3424 	/*
3425 	 * We are going to turn softirqs on, so set the
3426 	 * usage bit for all held locks, if hardirqs are
3427 	 * enabled too:
3428 	 */
3429 	if (curr->hardirqs_enabled)
3430 		mark_held_locks(curr, LOCK_ENABLED_SOFTIRQ);
3431 	current->lockdep_recursion = 0;
3432 }
3433 
3434 /*
3435  * Softirqs were disabled:
3436  */
3437 void trace_softirqs_off(unsigned long ip)
3438 {
3439 	struct task_struct *curr = current;
3440 
3441 	if (unlikely(!debug_locks || current->lockdep_recursion))
3442 		return;
3443 
3444 	/*
3445 	 * We fancy IRQs being disabled here, see softirq.c
3446 	 */
3447 	if (DEBUG_LOCKS_WARN_ON(!irqs_disabled()))
3448 		return;
3449 
3450 	if (curr->softirqs_enabled) {
3451 		/*
3452 		 * We have done an ON -> OFF transition:
3453 		 */
3454 		curr->softirqs_enabled = 0;
3455 		curr->softirq_disable_ip = ip;
3456 		curr->softirq_disable_event = ++curr->irq_events;
3457 		debug_atomic_inc(softirqs_off_events);
3458 		/*
3459 		 * Whoops, we wanted softirqs off, so why aren't they?
3460 		 */
3461 		DEBUG_LOCKS_WARN_ON(!softirq_count());
3462 	} else
3463 		debug_atomic_inc(redundant_softirqs_off);
3464 }
3465 
3466 static int
3467 mark_usage(struct task_struct *curr, struct held_lock *hlock, int check)
3468 {
3469 	if (!check)
3470 		goto lock_used;
3471 
3472 	/*
3473 	 * If non-trylock use in a hardirq or softirq context, then
3474 	 * mark the lock as used in these contexts:
3475 	 */
3476 	if (!hlock->trylock) {
3477 		if (hlock->read) {
3478 			if (curr->hardirq_context)
3479 				if (!mark_lock(curr, hlock,
3480 						LOCK_USED_IN_HARDIRQ_READ))
3481 					return 0;
3482 			if (curr->softirq_context)
3483 				if (!mark_lock(curr, hlock,
3484 						LOCK_USED_IN_SOFTIRQ_READ))
3485 					return 0;
3486 		} else {
3487 			if (curr->hardirq_context)
3488 				if (!mark_lock(curr, hlock, LOCK_USED_IN_HARDIRQ))
3489 					return 0;
3490 			if (curr->softirq_context)
3491 				if (!mark_lock(curr, hlock, LOCK_USED_IN_SOFTIRQ))
3492 					return 0;
3493 		}
3494 	}
3495 	if (!hlock->hardirqs_off) {
3496 		if (hlock->read) {
3497 			if (!mark_lock(curr, hlock,
3498 					LOCK_ENABLED_HARDIRQ_READ))
3499 				return 0;
3500 			if (curr->softirqs_enabled)
3501 				if (!mark_lock(curr, hlock,
3502 						LOCK_ENABLED_SOFTIRQ_READ))
3503 					return 0;
3504 		} else {
3505 			if (!mark_lock(curr, hlock,
3506 					LOCK_ENABLED_HARDIRQ))
3507 				return 0;
3508 			if (curr->softirqs_enabled)
3509 				if (!mark_lock(curr, hlock,
3510 						LOCK_ENABLED_SOFTIRQ))
3511 					return 0;
3512 		}
3513 	}
3514 
3515 lock_used:
3516 	/* mark it as used: */
3517 	if (!mark_lock(curr, hlock, LOCK_USED))
3518 		return 0;
3519 
3520 	return 1;
3521 }
3522 
3523 static inline unsigned int task_irq_context(struct task_struct *task)
3524 {
3525 	return 2 * !!task->hardirq_context + !!task->softirq_context;
3526 }
3527 
3528 static int separate_irq_context(struct task_struct *curr,
3529 		struct held_lock *hlock)
3530 {
3531 	unsigned int depth = curr->lockdep_depth;
3532 
3533 	/*
3534 	 * Keep track of points where we cross into an interrupt context:
3535 	 */
3536 	if (depth) {
3537 		struct held_lock *prev_hlock;
3538 
3539 		prev_hlock = curr->held_locks + depth-1;
3540 		/*
3541 		 * If we cross into another context, reset the
3542 		 * hash key (this also prevents the checking and the
3543 		 * adding of the dependency to 'prev'):
3544 		 */
3545 		if (prev_hlock->irq_context != hlock->irq_context)
3546 			return 1;
3547 	}
3548 	return 0;
3549 }
3550 
3551 /*
3552  * Mark a lock with a usage bit, and validate the state transition:
3553  */
3554 static int mark_lock(struct task_struct *curr, struct held_lock *this,
3555 			     enum lock_usage_bit new_bit)
3556 {
3557 	unsigned int new_mask = 1 << new_bit, ret = 1;
3558 
3559 	if (new_bit >= LOCK_USAGE_STATES) {
3560 		DEBUG_LOCKS_WARN_ON(1);
3561 		return 0;
3562 	}
3563 
3564 	/*
3565 	 * If already set then do not dirty the cacheline,
3566 	 * nor do any checks:
3567 	 */
3568 	if (likely(hlock_class(this)->usage_mask & new_mask))
3569 		return 1;
3570 
3571 	if (!graph_lock())
3572 		return 0;
3573 	/*
3574 	 * Make sure we didn't race:
3575 	 */
3576 	if (unlikely(hlock_class(this)->usage_mask & new_mask)) {
3577 		graph_unlock();
3578 		return 1;
3579 	}
3580 
3581 	hlock_class(this)->usage_mask |= new_mask;
3582 
3583 	if (!save_trace(hlock_class(this)->usage_traces + new_bit))
3584 		return 0;
3585 
3586 	switch (new_bit) {
3587 	case LOCK_USED:
3588 		debug_atomic_dec(nr_unused_locks);
3589 		break;
3590 	default:
3591 		ret = mark_lock_irq(curr, this, new_bit);
3592 		if (!ret)
3593 			return 0;
3594 	}
3595 
3596 	graph_unlock();
3597 
3598 	/*
3599 	 * We must printk outside of the graph_lock:
3600 	 */
3601 	if (ret == 2) {
3602 		printk("\nmarked lock as {%s}:\n", usage_str[new_bit]);
3603 		print_lock(this);
3604 		print_irqtrace_events(curr);
3605 		dump_stack();
3606 	}
3607 
3608 	return ret;
3609 }
3610 
3611 #else /* CONFIG_PROVE_LOCKING */
3612 
3613 static inline int
3614 mark_usage(struct task_struct *curr, struct held_lock *hlock, int check)
3615 {
3616 	return 1;
3617 }
3618 
3619 static inline unsigned int task_irq_context(struct task_struct *task)
3620 {
3621 	return 0;
3622 }
3623 
3624 static inline int separate_irq_context(struct task_struct *curr,
3625 		struct held_lock *hlock)
3626 {
3627 	return 0;
3628 }
3629 
3630 #endif /* CONFIG_PROVE_LOCKING */
3631 
3632 /*
3633  * Initialize a lock instance's lock-class mapping info:
3634  */
3635 void lockdep_init_map(struct lockdep_map *lock, const char *name,
3636 		      struct lock_class_key *key, int subclass)
3637 {
3638 	int i;
3639 
3640 	for (i = 0; i < NR_LOCKDEP_CACHING_CLASSES; i++)
3641 		lock->class_cache[i] = NULL;
3642 
3643 #ifdef CONFIG_LOCK_STAT
3644 	lock->cpu = raw_smp_processor_id();
3645 #endif
3646 
3647 	/*
3648 	 * Can't be having no nameless bastards around this place!
3649 	 */
3650 	if (DEBUG_LOCKS_WARN_ON(!name)) {
3651 		lock->name = "NULL";
3652 		return;
3653 	}
3654 
3655 	lock->name = name;
3656 
3657 	/*
3658 	 * No key, no joy, we need to hash something.
3659 	 */
3660 	if (DEBUG_LOCKS_WARN_ON(!key))
3661 		return;
3662 	/*
3663 	 * Sanity check, the lock-class key must either have been allocated
3664 	 * statically or must have been registered as a dynamic key.
3665 	 */
3666 	if (!static_obj(key) && !is_dynamic_key(key)) {
3667 		if (debug_locks)
3668 			printk(KERN_ERR "BUG: key %px has not been registered!\n", key);
3669 		DEBUG_LOCKS_WARN_ON(1);
3670 		return;
3671 	}
3672 	lock->key = key;
3673 
3674 	if (unlikely(!debug_locks))
3675 		return;
3676 
3677 	if (subclass) {
3678 		unsigned long flags;
3679 
3680 		if (DEBUG_LOCKS_WARN_ON(current->lockdep_recursion))
3681 			return;
3682 
3683 		raw_local_irq_save(flags);
3684 		current->lockdep_recursion = 1;
3685 		register_lock_class(lock, subclass, 1);
3686 		current->lockdep_recursion = 0;
3687 		raw_local_irq_restore(flags);
3688 	}
3689 }
3690 EXPORT_SYMBOL_GPL(lockdep_init_map);
3691 
3692 struct lock_class_key __lockdep_no_validate__;
3693 EXPORT_SYMBOL_GPL(__lockdep_no_validate__);
3694 
3695 static void
3696 print_lock_nested_lock_not_held(struct task_struct *curr,
3697 				struct held_lock *hlock,
3698 				unsigned long ip)
3699 {
3700 	if (!debug_locks_off())
3701 		return;
3702 	if (debug_locks_silent)
3703 		return;
3704 
3705 	pr_warn("\n");
3706 	pr_warn("==================================\n");
3707 	pr_warn("WARNING: Nested lock was not taken\n");
3708 	print_kernel_ident();
3709 	pr_warn("----------------------------------\n");
3710 
3711 	pr_warn("%s/%d is trying to lock:\n", curr->comm, task_pid_nr(curr));
3712 	print_lock(hlock);
3713 
3714 	pr_warn("\nbut this task is not holding:\n");
3715 	pr_warn("%s\n", hlock->nest_lock->name);
3716 
3717 	pr_warn("\nstack backtrace:\n");
3718 	dump_stack();
3719 
3720 	pr_warn("\nother info that might help us debug this:\n");
3721 	lockdep_print_held_locks(curr);
3722 
3723 	pr_warn("\nstack backtrace:\n");
3724 	dump_stack();
3725 }
3726 
3727 static int __lock_is_held(const struct lockdep_map *lock, int read);
3728 
3729 /*
3730  * This gets called for every mutex_lock*()/spin_lock*() operation.
3731  * We maintain the dependency maps and validate the locking attempt:
3732  *
3733  * The callers must make sure that IRQs are disabled before calling it,
3734  * otherwise we could get an interrupt which would want to take locks,
3735  * which would end up in lockdep again.
3736  */
3737 static int __lock_acquire(struct lockdep_map *lock, unsigned int subclass,
3738 			  int trylock, int read, int check, int hardirqs_off,
3739 			  struct lockdep_map *nest_lock, unsigned long ip,
3740 			  int references, int pin_count)
3741 {
3742 	struct task_struct *curr = current;
3743 	struct lock_class *class = NULL;
3744 	struct held_lock *hlock;
3745 	unsigned int depth;
3746 	int chain_head = 0;
3747 	int class_idx;
3748 	u64 chain_key;
3749 
3750 	if (unlikely(!debug_locks))
3751 		return 0;
3752 
3753 	if (!prove_locking || lock->key == &__lockdep_no_validate__)
3754 		check = 0;
3755 
3756 	if (subclass < NR_LOCKDEP_CACHING_CLASSES)
3757 		class = lock->class_cache[subclass];
3758 	/*
3759 	 * Not cached?
3760 	 */
3761 	if (unlikely(!class)) {
3762 		class = register_lock_class(lock, subclass, 0);
3763 		if (!class)
3764 			return 0;
3765 	}
3766 
3767 	debug_class_ops_inc(class);
3768 
3769 	if (very_verbose(class)) {
3770 		printk("\nacquire class [%px] %s", class->key, class->name);
3771 		if (class->name_version > 1)
3772 			printk(KERN_CONT "#%d", class->name_version);
3773 		printk(KERN_CONT "\n");
3774 		dump_stack();
3775 	}
3776 
3777 	/*
3778 	 * Add the lock to the list of currently held locks.
3779 	 * (we dont increase the depth just yet, up until the
3780 	 * dependency checks are done)
3781 	 */
3782 	depth = curr->lockdep_depth;
3783 	/*
3784 	 * Ran out of static storage for our per-task lock stack again have we?
3785 	 */
3786 	if (DEBUG_LOCKS_WARN_ON(depth >= MAX_LOCK_DEPTH))
3787 		return 0;
3788 
3789 	class_idx = class - lock_classes;
3790 
3791 	if (depth) {
3792 		hlock = curr->held_locks + depth - 1;
3793 		if (hlock->class_idx == class_idx && nest_lock) {
3794 			if (!references)
3795 				references++;
3796 
3797 			if (!hlock->references)
3798 				hlock->references++;
3799 
3800 			hlock->references += references;
3801 
3802 			/* Overflow */
3803 			if (DEBUG_LOCKS_WARN_ON(hlock->references < references))
3804 				return 0;
3805 
3806 			return 2;
3807 		}
3808 	}
3809 
3810 	hlock = curr->held_locks + depth;
3811 	/*
3812 	 * Plain impossible, we just registered it and checked it weren't no
3813 	 * NULL like.. I bet this mushroom I ate was good!
3814 	 */
3815 	if (DEBUG_LOCKS_WARN_ON(!class))
3816 		return 0;
3817 	hlock->class_idx = class_idx;
3818 	hlock->acquire_ip = ip;
3819 	hlock->instance = lock;
3820 	hlock->nest_lock = nest_lock;
3821 	hlock->irq_context = task_irq_context(curr);
3822 	hlock->trylock = trylock;
3823 	hlock->read = read;
3824 	hlock->check = check;
3825 	hlock->hardirqs_off = !!hardirqs_off;
3826 	hlock->references = references;
3827 #ifdef CONFIG_LOCK_STAT
3828 	hlock->waittime_stamp = 0;
3829 	hlock->holdtime_stamp = lockstat_clock();
3830 #endif
3831 	hlock->pin_count = pin_count;
3832 
3833 	/* Initialize the lock usage bit */
3834 	if (!mark_usage(curr, hlock, check))
3835 		return 0;
3836 
3837 	/*
3838 	 * Calculate the chain hash: it's the combined hash of all the
3839 	 * lock keys along the dependency chain. We save the hash value
3840 	 * at every step so that we can get the current hash easily
3841 	 * after unlock. The chain hash is then used to cache dependency
3842 	 * results.
3843 	 *
3844 	 * The 'key ID' is what is the most compact key value to drive
3845 	 * the hash, not class->key.
3846 	 */
3847 	/*
3848 	 * Whoops, we did it again.. class_idx is invalid.
3849 	 */
3850 	if (DEBUG_LOCKS_WARN_ON(!test_bit(class_idx, lock_classes_in_use)))
3851 		return 0;
3852 
3853 	chain_key = curr->curr_chain_key;
3854 	if (!depth) {
3855 		/*
3856 		 * How can we have a chain hash when we ain't got no keys?!
3857 		 */
3858 		if (DEBUG_LOCKS_WARN_ON(chain_key != INITIAL_CHAIN_KEY))
3859 			return 0;
3860 		chain_head = 1;
3861 	}
3862 
3863 	hlock->prev_chain_key = chain_key;
3864 	if (separate_irq_context(curr, hlock)) {
3865 		chain_key = INITIAL_CHAIN_KEY;
3866 		chain_head = 1;
3867 	}
3868 	chain_key = iterate_chain_key(chain_key, class_idx);
3869 
3870 	if (nest_lock && !__lock_is_held(nest_lock, -1)) {
3871 		print_lock_nested_lock_not_held(curr, hlock, ip);
3872 		return 0;
3873 	}
3874 
3875 	if (!debug_locks_silent) {
3876 		WARN_ON_ONCE(depth && !hlock_class(hlock - 1)->key);
3877 		WARN_ON_ONCE(!hlock_class(hlock)->key);
3878 	}
3879 
3880 	if (!validate_chain(curr, hlock, chain_head, chain_key))
3881 		return 0;
3882 
3883 	curr->curr_chain_key = chain_key;
3884 	curr->lockdep_depth++;
3885 	check_chain_key(curr);
3886 #ifdef CONFIG_DEBUG_LOCKDEP
3887 	if (unlikely(!debug_locks))
3888 		return 0;
3889 #endif
3890 	if (unlikely(curr->lockdep_depth >= MAX_LOCK_DEPTH)) {
3891 		debug_locks_off();
3892 		print_lockdep_off("BUG: MAX_LOCK_DEPTH too low!");
3893 		printk(KERN_DEBUG "depth: %i  max: %lu!\n",
3894 		       curr->lockdep_depth, MAX_LOCK_DEPTH);
3895 
3896 		lockdep_print_held_locks(current);
3897 		debug_show_all_locks();
3898 		dump_stack();
3899 
3900 		return 0;
3901 	}
3902 
3903 	if (unlikely(curr->lockdep_depth > max_lockdep_depth))
3904 		max_lockdep_depth = curr->lockdep_depth;
3905 
3906 	return 1;
3907 }
3908 
3909 static void print_unlock_imbalance_bug(struct task_struct *curr,
3910 				       struct lockdep_map *lock,
3911 				       unsigned long ip)
3912 {
3913 	if (!debug_locks_off())
3914 		return;
3915 	if (debug_locks_silent)
3916 		return;
3917 
3918 	pr_warn("\n");
3919 	pr_warn("=====================================\n");
3920 	pr_warn("WARNING: bad unlock balance detected!\n");
3921 	print_kernel_ident();
3922 	pr_warn("-------------------------------------\n");
3923 	pr_warn("%s/%d is trying to release lock (",
3924 		curr->comm, task_pid_nr(curr));
3925 	print_lockdep_cache(lock);
3926 	pr_cont(") at:\n");
3927 	print_ip_sym(ip);
3928 	pr_warn("but there are no more locks to release!\n");
3929 	pr_warn("\nother info that might help us debug this:\n");
3930 	lockdep_print_held_locks(curr);
3931 
3932 	pr_warn("\nstack backtrace:\n");
3933 	dump_stack();
3934 }
3935 
3936 static int match_held_lock(const struct held_lock *hlock,
3937 					const struct lockdep_map *lock)
3938 {
3939 	if (hlock->instance == lock)
3940 		return 1;
3941 
3942 	if (hlock->references) {
3943 		const struct lock_class *class = lock->class_cache[0];
3944 
3945 		if (!class)
3946 			class = look_up_lock_class(lock, 0);
3947 
3948 		/*
3949 		 * If look_up_lock_class() failed to find a class, we're trying
3950 		 * to test if we hold a lock that has never yet been acquired.
3951 		 * Clearly if the lock hasn't been acquired _ever_, we're not
3952 		 * holding it either, so report failure.
3953 		 */
3954 		if (!class)
3955 			return 0;
3956 
3957 		/*
3958 		 * References, but not a lock we're actually ref-counting?
3959 		 * State got messed up, follow the sites that change ->references
3960 		 * and try to make sense of it.
3961 		 */
3962 		if (DEBUG_LOCKS_WARN_ON(!hlock->nest_lock))
3963 			return 0;
3964 
3965 		if (hlock->class_idx == class - lock_classes)
3966 			return 1;
3967 	}
3968 
3969 	return 0;
3970 }
3971 
3972 /* @depth must not be zero */
3973 static struct held_lock *find_held_lock(struct task_struct *curr,
3974 					struct lockdep_map *lock,
3975 					unsigned int depth, int *idx)
3976 {
3977 	struct held_lock *ret, *hlock, *prev_hlock;
3978 	int i;
3979 
3980 	i = depth - 1;
3981 	hlock = curr->held_locks + i;
3982 	ret = hlock;
3983 	if (match_held_lock(hlock, lock))
3984 		goto out;
3985 
3986 	ret = NULL;
3987 	for (i--, prev_hlock = hlock--;
3988 	     i >= 0;
3989 	     i--, prev_hlock = hlock--) {
3990 		/*
3991 		 * We must not cross into another context:
3992 		 */
3993 		if (prev_hlock->irq_context != hlock->irq_context) {
3994 			ret = NULL;
3995 			break;
3996 		}
3997 		if (match_held_lock(hlock, lock)) {
3998 			ret = hlock;
3999 			break;
4000 		}
4001 	}
4002 
4003 out:
4004 	*idx = i;
4005 	return ret;
4006 }
4007 
4008 static int reacquire_held_locks(struct task_struct *curr, unsigned int depth,
4009 				int idx, unsigned int *merged)
4010 {
4011 	struct held_lock *hlock;
4012 	int first_idx = idx;
4013 
4014 	if (DEBUG_LOCKS_WARN_ON(!irqs_disabled()))
4015 		return 0;
4016 
4017 	for (hlock = curr->held_locks + idx; idx < depth; idx++, hlock++) {
4018 		switch (__lock_acquire(hlock->instance,
4019 				    hlock_class(hlock)->subclass,
4020 				    hlock->trylock,
4021 				    hlock->read, hlock->check,
4022 				    hlock->hardirqs_off,
4023 				    hlock->nest_lock, hlock->acquire_ip,
4024 				    hlock->references, hlock->pin_count)) {
4025 		case 0:
4026 			return 1;
4027 		case 1:
4028 			break;
4029 		case 2:
4030 			*merged += (idx == first_idx);
4031 			break;
4032 		default:
4033 			WARN_ON(1);
4034 			return 0;
4035 		}
4036 	}
4037 	return 0;
4038 }
4039 
4040 static int
4041 __lock_set_class(struct lockdep_map *lock, const char *name,
4042 		 struct lock_class_key *key, unsigned int subclass,
4043 		 unsigned long ip)
4044 {
4045 	struct task_struct *curr = current;
4046 	unsigned int depth, merged = 0;
4047 	struct held_lock *hlock;
4048 	struct lock_class *class;
4049 	int i;
4050 
4051 	if (unlikely(!debug_locks))
4052 		return 0;
4053 
4054 	depth = curr->lockdep_depth;
4055 	/*
4056 	 * This function is about (re)setting the class of a held lock,
4057 	 * yet we're not actually holding any locks. Naughty user!
4058 	 */
4059 	if (DEBUG_LOCKS_WARN_ON(!depth))
4060 		return 0;
4061 
4062 	hlock = find_held_lock(curr, lock, depth, &i);
4063 	if (!hlock) {
4064 		print_unlock_imbalance_bug(curr, lock, ip);
4065 		return 0;
4066 	}
4067 
4068 	lockdep_init_map(lock, name, key, 0);
4069 	class = register_lock_class(lock, subclass, 0);
4070 	hlock->class_idx = class - lock_classes;
4071 
4072 	curr->lockdep_depth = i;
4073 	curr->curr_chain_key = hlock->prev_chain_key;
4074 
4075 	if (reacquire_held_locks(curr, depth, i, &merged))
4076 		return 0;
4077 
4078 	/*
4079 	 * I took it apart and put it back together again, except now I have
4080 	 * these 'spare' parts.. where shall I put them.
4081 	 */
4082 	if (DEBUG_LOCKS_WARN_ON(curr->lockdep_depth != depth - merged))
4083 		return 0;
4084 	return 1;
4085 }
4086 
4087 static int __lock_downgrade(struct lockdep_map *lock, unsigned long ip)
4088 {
4089 	struct task_struct *curr = current;
4090 	unsigned int depth, merged = 0;
4091 	struct held_lock *hlock;
4092 	int i;
4093 
4094 	if (unlikely(!debug_locks))
4095 		return 0;
4096 
4097 	depth = curr->lockdep_depth;
4098 	/*
4099 	 * This function is about (re)setting the class of a held lock,
4100 	 * yet we're not actually holding any locks. Naughty user!
4101 	 */
4102 	if (DEBUG_LOCKS_WARN_ON(!depth))
4103 		return 0;
4104 
4105 	hlock = find_held_lock(curr, lock, depth, &i);
4106 	if (!hlock) {
4107 		print_unlock_imbalance_bug(curr, lock, ip);
4108 		return 0;
4109 	}
4110 
4111 	curr->lockdep_depth = i;
4112 	curr->curr_chain_key = hlock->prev_chain_key;
4113 
4114 	WARN(hlock->read, "downgrading a read lock");
4115 	hlock->read = 1;
4116 	hlock->acquire_ip = ip;
4117 
4118 	if (reacquire_held_locks(curr, depth, i, &merged))
4119 		return 0;
4120 
4121 	/* Merging can't happen with unchanged classes.. */
4122 	if (DEBUG_LOCKS_WARN_ON(merged))
4123 		return 0;
4124 
4125 	/*
4126 	 * I took it apart and put it back together again, except now I have
4127 	 * these 'spare' parts.. where shall I put them.
4128 	 */
4129 	if (DEBUG_LOCKS_WARN_ON(curr->lockdep_depth != depth))
4130 		return 0;
4131 
4132 	return 1;
4133 }
4134 
4135 /*
4136  * Remove the lock to the list of currently held locks - this gets
4137  * called on mutex_unlock()/spin_unlock*() (or on a failed
4138  * mutex_lock_interruptible()).
4139  *
4140  * @nested is an hysterical artifact, needs a tree wide cleanup.
4141  */
4142 static int
4143 __lock_release(struct lockdep_map *lock, unsigned long ip)
4144 {
4145 	struct task_struct *curr = current;
4146 	unsigned int depth, merged = 1;
4147 	struct held_lock *hlock;
4148 	int i;
4149 
4150 	if (unlikely(!debug_locks))
4151 		return 0;
4152 
4153 	depth = curr->lockdep_depth;
4154 	/*
4155 	 * So we're all set to release this lock.. wait what lock? We don't
4156 	 * own any locks, you've been drinking again?
4157 	 */
4158 	if (depth <= 0) {
4159 		print_unlock_imbalance_bug(curr, lock, ip);
4160 		return 0;
4161 	}
4162 
4163 	/*
4164 	 * Check whether the lock exists in the current stack
4165 	 * of held locks:
4166 	 */
4167 	hlock = find_held_lock(curr, lock, depth, &i);
4168 	if (!hlock) {
4169 		print_unlock_imbalance_bug(curr, lock, ip);
4170 		return 0;
4171 	}
4172 
4173 	if (hlock->instance == lock)
4174 		lock_release_holdtime(hlock);
4175 
4176 	WARN(hlock->pin_count, "releasing a pinned lock\n");
4177 
4178 	if (hlock->references) {
4179 		hlock->references--;
4180 		if (hlock->references) {
4181 			/*
4182 			 * We had, and after removing one, still have
4183 			 * references, the current lock stack is still
4184 			 * valid. We're done!
4185 			 */
4186 			return 1;
4187 		}
4188 	}
4189 
4190 	/*
4191 	 * We have the right lock to unlock, 'hlock' points to it.
4192 	 * Now we remove it from the stack, and add back the other
4193 	 * entries (if any), recalculating the hash along the way:
4194 	 */
4195 
4196 	curr->lockdep_depth = i;
4197 	curr->curr_chain_key = hlock->prev_chain_key;
4198 
4199 	/*
4200 	 * The most likely case is when the unlock is on the innermost
4201 	 * lock. In this case, we are done!
4202 	 */
4203 	if (i == depth-1)
4204 		return 1;
4205 
4206 	if (reacquire_held_locks(curr, depth, i + 1, &merged))
4207 		return 0;
4208 
4209 	/*
4210 	 * We had N bottles of beer on the wall, we drank one, but now
4211 	 * there's not N-1 bottles of beer left on the wall...
4212 	 * Pouring two of the bottles together is acceptable.
4213 	 */
4214 	DEBUG_LOCKS_WARN_ON(curr->lockdep_depth != depth - merged);
4215 
4216 	/*
4217 	 * Since reacquire_held_locks() would have called check_chain_key()
4218 	 * indirectly via __lock_acquire(), we don't need to do it again
4219 	 * on return.
4220 	 */
4221 	return 0;
4222 }
4223 
4224 static nokprobe_inline
4225 int __lock_is_held(const struct lockdep_map *lock, int read)
4226 {
4227 	struct task_struct *curr = current;
4228 	int i;
4229 
4230 	for (i = 0; i < curr->lockdep_depth; i++) {
4231 		struct held_lock *hlock = curr->held_locks + i;
4232 
4233 		if (match_held_lock(hlock, lock)) {
4234 			if (read == -1 || hlock->read == read)
4235 				return 1;
4236 
4237 			return 0;
4238 		}
4239 	}
4240 
4241 	return 0;
4242 }
4243 
4244 static struct pin_cookie __lock_pin_lock(struct lockdep_map *lock)
4245 {
4246 	struct pin_cookie cookie = NIL_COOKIE;
4247 	struct task_struct *curr = current;
4248 	int i;
4249 
4250 	if (unlikely(!debug_locks))
4251 		return cookie;
4252 
4253 	for (i = 0; i < curr->lockdep_depth; i++) {
4254 		struct held_lock *hlock = curr->held_locks + i;
4255 
4256 		if (match_held_lock(hlock, lock)) {
4257 			/*
4258 			 * Grab 16bits of randomness; this is sufficient to not
4259 			 * be guessable and still allows some pin nesting in
4260 			 * our u32 pin_count.
4261 			 */
4262 			cookie.val = 1 + (prandom_u32() >> 16);
4263 			hlock->pin_count += cookie.val;
4264 			return cookie;
4265 		}
4266 	}
4267 
4268 	WARN(1, "pinning an unheld lock\n");
4269 	return cookie;
4270 }
4271 
4272 static void __lock_repin_lock(struct lockdep_map *lock, struct pin_cookie cookie)
4273 {
4274 	struct task_struct *curr = current;
4275 	int i;
4276 
4277 	if (unlikely(!debug_locks))
4278 		return;
4279 
4280 	for (i = 0; i < curr->lockdep_depth; i++) {
4281 		struct held_lock *hlock = curr->held_locks + i;
4282 
4283 		if (match_held_lock(hlock, lock)) {
4284 			hlock->pin_count += cookie.val;
4285 			return;
4286 		}
4287 	}
4288 
4289 	WARN(1, "pinning an unheld lock\n");
4290 }
4291 
4292 static void __lock_unpin_lock(struct lockdep_map *lock, struct pin_cookie cookie)
4293 {
4294 	struct task_struct *curr = current;
4295 	int i;
4296 
4297 	if (unlikely(!debug_locks))
4298 		return;
4299 
4300 	for (i = 0; i < curr->lockdep_depth; i++) {
4301 		struct held_lock *hlock = curr->held_locks + i;
4302 
4303 		if (match_held_lock(hlock, lock)) {
4304 			if (WARN(!hlock->pin_count, "unpinning an unpinned lock\n"))
4305 				return;
4306 
4307 			hlock->pin_count -= cookie.val;
4308 
4309 			if (WARN((int)hlock->pin_count < 0, "pin count corrupted\n"))
4310 				hlock->pin_count = 0;
4311 
4312 			return;
4313 		}
4314 	}
4315 
4316 	WARN(1, "unpinning an unheld lock\n");
4317 }
4318 
4319 /*
4320  * Check whether we follow the irq-flags state precisely:
4321  */
4322 static void check_flags(unsigned long flags)
4323 {
4324 #if defined(CONFIG_PROVE_LOCKING) && defined(CONFIG_DEBUG_LOCKDEP)
4325 	if (!debug_locks)
4326 		return;
4327 
4328 	if (irqs_disabled_flags(flags)) {
4329 		if (DEBUG_LOCKS_WARN_ON(current->hardirqs_enabled)) {
4330 			printk("possible reason: unannotated irqs-off.\n");
4331 		}
4332 	} else {
4333 		if (DEBUG_LOCKS_WARN_ON(!current->hardirqs_enabled)) {
4334 			printk("possible reason: unannotated irqs-on.\n");
4335 		}
4336 	}
4337 
4338 	/*
4339 	 * We dont accurately track softirq state in e.g.
4340 	 * hardirq contexts (such as on 4KSTACKS), so only
4341 	 * check if not in hardirq contexts:
4342 	 */
4343 	if (!hardirq_count()) {
4344 		if (softirq_count()) {
4345 			/* like the above, but with softirqs */
4346 			DEBUG_LOCKS_WARN_ON(current->softirqs_enabled);
4347 		} else {
4348 			/* lick the above, does it taste good? */
4349 			DEBUG_LOCKS_WARN_ON(!current->softirqs_enabled);
4350 		}
4351 	}
4352 
4353 	if (!debug_locks)
4354 		print_irqtrace_events(current);
4355 #endif
4356 }
4357 
4358 void lock_set_class(struct lockdep_map *lock, const char *name,
4359 		    struct lock_class_key *key, unsigned int subclass,
4360 		    unsigned long ip)
4361 {
4362 	unsigned long flags;
4363 
4364 	if (unlikely(current->lockdep_recursion))
4365 		return;
4366 
4367 	raw_local_irq_save(flags);
4368 	current->lockdep_recursion = 1;
4369 	check_flags(flags);
4370 	if (__lock_set_class(lock, name, key, subclass, ip))
4371 		check_chain_key(current);
4372 	current->lockdep_recursion = 0;
4373 	raw_local_irq_restore(flags);
4374 }
4375 EXPORT_SYMBOL_GPL(lock_set_class);
4376 
4377 void lock_downgrade(struct lockdep_map *lock, unsigned long ip)
4378 {
4379 	unsigned long flags;
4380 
4381 	if (unlikely(current->lockdep_recursion))
4382 		return;
4383 
4384 	raw_local_irq_save(flags);
4385 	current->lockdep_recursion = 1;
4386 	check_flags(flags);
4387 	if (__lock_downgrade(lock, ip))
4388 		check_chain_key(current);
4389 	current->lockdep_recursion = 0;
4390 	raw_local_irq_restore(flags);
4391 }
4392 EXPORT_SYMBOL_GPL(lock_downgrade);
4393 
4394 /*
4395  * We are not always called with irqs disabled - do that here,
4396  * and also avoid lockdep recursion:
4397  */
4398 void lock_acquire(struct lockdep_map *lock, unsigned int subclass,
4399 			  int trylock, int read, int check,
4400 			  struct lockdep_map *nest_lock, unsigned long ip)
4401 {
4402 	unsigned long flags;
4403 
4404 	if (unlikely(current->lockdep_recursion))
4405 		return;
4406 
4407 	raw_local_irq_save(flags);
4408 	check_flags(flags);
4409 
4410 	current->lockdep_recursion = 1;
4411 	trace_lock_acquire(lock, subclass, trylock, read, check, nest_lock, ip);
4412 	__lock_acquire(lock, subclass, trylock, read, check,
4413 		       irqs_disabled_flags(flags), nest_lock, ip, 0, 0);
4414 	current->lockdep_recursion = 0;
4415 	raw_local_irq_restore(flags);
4416 }
4417 EXPORT_SYMBOL_GPL(lock_acquire);
4418 
4419 void lock_release(struct lockdep_map *lock, int nested,
4420 			  unsigned long ip)
4421 {
4422 	unsigned long flags;
4423 
4424 	if (unlikely(current->lockdep_recursion))
4425 		return;
4426 
4427 	raw_local_irq_save(flags);
4428 	check_flags(flags);
4429 	current->lockdep_recursion = 1;
4430 	trace_lock_release(lock, ip);
4431 	if (__lock_release(lock, ip))
4432 		check_chain_key(current);
4433 	current->lockdep_recursion = 0;
4434 	raw_local_irq_restore(flags);
4435 }
4436 EXPORT_SYMBOL_GPL(lock_release);
4437 
4438 int lock_is_held_type(const struct lockdep_map *lock, int read)
4439 {
4440 	unsigned long flags;
4441 	int ret = 0;
4442 
4443 	if (unlikely(current->lockdep_recursion))
4444 		return 1; /* avoid false negative lockdep_assert_held() */
4445 
4446 	raw_local_irq_save(flags);
4447 	check_flags(flags);
4448 
4449 	current->lockdep_recursion = 1;
4450 	ret = __lock_is_held(lock, read);
4451 	current->lockdep_recursion = 0;
4452 	raw_local_irq_restore(flags);
4453 
4454 	return ret;
4455 }
4456 EXPORT_SYMBOL_GPL(lock_is_held_type);
4457 NOKPROBE_SYMBOL(lock_is_held_type);
4458 
4459 struct pin_cookie lock_pin_lock(struct lockdep_map *lock)
4460 {
4461 	struct pin_cookie cookie = NIL_COOKIE;
4462 	unsigned long flags;
4463 
4464 	if (unlikely(current->lockdep_recursion))
4465 		return cookie;
4466 
4467 	raw_local_irq_save(flags);
4468 	check_flags(flags);
4469 
4470 	current->lockdep_recursion = 1;
4471 	cookie = __lock_pin_lock(lock);
4472 	current->lockdep_recursion = 0;
4473 	raw_local_irq_restore(flags);
4474 
4475 	return cookie;
4476 }
4477 EXPORT_SYMBOL_GPL(lock_pin_lock);
4478 
4479 void lock_repin_lock(struct lockdep_map *lock, struct pin_cookie cookie)
4480 {
4481 	unsigned long flags;
4482 
4483 	if (unlikely(current->lockdep_recursion))
4484 		return;
4485 
4486 	raw_local_irq_save(flags);
4487 	check_flags(flags);
4488 
4489 	current->lockdep_recursion = 1;
4490 	__lock_repin_lock(lock, cookie);
4491 	current->lockdep_recursion = 0;
4492 	raw_local_irq_restore(flags);
4493 }
4494 EXPORT_SYMBOL_GPL(lock_repin_lock);
4495 
4496 void lock_unpin_lock(struct lockdep_map *lock, struct pin_cookie cookie)
4497 {
4498 	unsigned long flags;
4499 
4500 	if (unlikely(current->lockdep_recursion))
4501 		return;
4502 
4503 	raw_local_irq_save(flags);
4504 	check_flags(flags);
4505 
4506 	current->lockdep_recursion = 1;
4507 	__lock_unpin_lock(lock, cookie);
4508 	current->lockdep_recursion = 0;
4509 	raw_local_irq_restore(flags);
4510 }
4511 EXPORT_SYMBOL_GPL(lock_unpin_lock);
4512 
4513 #ifdef CONFIG_LOCK_STAT
4514 static void print_lock_contention_bug(struct task_struct *curr,
4515 				      struct lockdep_map *lock,
4516 				      unsigned long ip)
4517 {
4518 	if (!debug_locks_off())
4519 		return;
4520 	if (debug_locks_silent)
4521 		return;
4522 
4523 	pr_warn("\n");
4524 	pr_warn("=================================\n");
4525 	pr_warn("WARNING: bad contention detected!\n");
4526 	print_kernel_ident();
4527 	pr_warn("---------------------------------\n");
4528 	pr_warn("%s/%d is trying to contend lock (",
4529 		curr->comm, task_pid_nr(curr));
4530 	print_lockdep_cache(lock);
4531 	pr_cont(") at:\n");
4532 	print_ip_sym(ip);
4533 	pr_warn("but there are no locks held!\n");
4534 	pr_warn("\nother info that might help us debug this:\n");
4535 	lockdep_print_held_locks(curr);
4536 
4537 	pr_warn("\nstack backtrace:\n");
4538 	dump_stack();
4539 }
4540 
4541 static void
4542 __lock_contended(struct lockdep_map *lock, unsigned long ip)
4543 {
4544 	struct task_struct *curr = current;
4545 	struct held_lock *hlock;
4546 	struct lock_class_stats *stats;
4547 	unsigned int depth;
4548 	int i, contention_point, contending_point;
4549 
4550 	depth = curr->lockdep_depth;
4551 	/*
4552 	 * Whee, we contended on this lock, except it seems we're not
4553 	 * actually trying to acquire anything much at all..
4554 	 */
4555 	if (DEBUG_LOCKS_WARN_ON(!depth))
4556 		return;
4557 
4558 	hlock = find_held_lock(curr, lock, depth, &i);
4559 	if (!hlock) {
4560 		print_lock_contention_bug(curr, lock, ip);
4561 		return;
4562 	}
4563 
4564 	if (hlock->instance != lock)
4565 		return;
4566 
4567 	hlock->waittime_stamp = lockstat_clock();
4568 
4569 	contention_point = lock_point(hlock_class(hlock)->contention_point, ip);
4570 	contending_point = lock_point(hlock_class(hlock)->contending_point,
4571 				      lock->ip);
4572 
4573 	stats = get_lock_stats(hlock_class(hlock));
4574 	if (contention_point < LOCKSTAT_POINTS)
4575 		stats->contention_point[contention_point]++;
4576 	if (contending_point < LOCKSTAT_POINTS)
4577 		stats->contending_point[contending_point]++;
4578 	if (lock->cpu != smp_processor_id())
4579 		stats->bounces[bounce_contended + !!hlock->read]++;
4580 }
4581 
4582 static void
4583 __lock_acquired(struct lockdep_map *lock, unsigned long ip)
4584 {
4585 	struct task_struct *curr = current;
4586 	struct held_lock *hlock;
4587 	struct lock_class_stats *stats;
4588 	unsigned int depth;
4589 	u64 now, waittime = 0;
4590 	int i, cpu;
4591 
4592 	depth = curr->lockdep_depth;
4593 	/*
4594 	 * Yay, we acquired ownership of this lock we didn't try to
4595 	 * acquire, how the heck did that happen?
4596 	 */
4597 	if (DEBUG_LOCKS_WARN_ON(!depth))
4598 		return;
4599 
4600 	hlock = find_held_lock(curr, lock, depth, &i);
4601 	if (!hlock) {
4602 		print_lock_contention_bug(curr, lock, _RET_IP_);
4603 		return;
4604 	}
4605 
4606 	if (hlock->instance != lock)
4607 		return;
4608 
4609 	cpu = smp_processor_id();
4610 	if (hlock->waittime_stamp) {
4611 		now = lockstat_clock();
4612 		waittime = now - hlock->waittime_stamp;
4613 		hlock->holdtime_stamp = now;
4614 	}
4615 
4616 	trace_lock_acquired(lock, ip);
4617 
4618 	stats = get_lock_stats(hlock_class(hlock));
4619 	if (waittime) {
4620 		if (hlock->read)
4621 			lock_time_inc(&stats->read_waittime, waittime);
4622 		else
4623 			lock_time_inc(&stats->write_waittime, waittime);
4624 	}
4625 	if (lock->cpu != cpu)
4626 		stats->bounces[bounce_acquired + !!hlock->read]++;
4627 
4628 	lock->cpu = cpu;
4629 	lock->ip = ip;
4630 }
4631 
4632 void lock_contended(struct lockdep_map *lock, unsigned long ip)
4633 {
4634 	unsigned long flags;
4635 
4636 	if (unlikely(!lock_stat || !debug_locks))
4637 		return;
4638 
4639 	if (unlikely(current->lockdep_recursion))
4640 		return;
4641 
4642 	raw_local_irq_save(flags);
4643 	check_flags(flags);
4644 	current->lockdep_recursion = 1;
4645 	trace_lock_contended(lock, ip);
4646 	__lock_contended(lock, ip);
4647 	current->lockdep_recursion = 0;
4648 	raw_local_irq_restore(flags);
4649 }
4650 EXPORT_SYMBOL_GPL(lock_contended);
4651 
4652 void lock_acquired(struct lockdep_map *lock, unsigned long ip)
4653 {
4654 	unsigned long flags;
4655 
4656 	if (unlikely(!lock_stat || !debug_locks))
4657 		return;
4658 
4659 	if (unlikely(current->lockdep_recursion))
4660 		return;
4661 
4662 	raw_local_irq_save(flags);
4663 	check_flags(flags);
4664 	current->lockdep_recursion = 1;
4665 	__lock_acquired(lock, ip);
4666 	current->lockdep_recursion = 0;
4667 	raw_local_irq_restore(flags);
4668 }
4669 EXPORT_SYMBOL_GPL(lock_acquired);
4670 #endif
4671 
4672 /*
4673  * Used by the testsuite, sanitize the validator state
4674  * after a simulated failure:
4675  */
4676 
4677 void lockdep_reset(void)
4678 {
4679 	unsigned long flags;
4680 	int i;
4681 
4682 	raw_local_irq_save(flags);
4683 	lockdep_init_task(current);
4684 	memset(current->held_locks, 0, MAX_LOCK_DEPTH*sizeof(struct held_lock));
4685 	nr_hardirq_chains = 0;
4686 	nr_softirq_chains = 0;
4687 	nr_process_chains = 0;
4688 	debug_locks = 1;
4689 	for (i = 0; i < CHAINHASH_SIZE; i++)
4690 		INIT_HLIST_HEAD(chainhash_table + i);
4691 	raw_local_irq_restore(flags);
4692 }
4693 
4694 /* Remove a class from a lock chain. Must be called with the graph lock held. */
4695 static void remove_class_from_lock_chain(struct pending_free *pf,
4696 					 struct lock_chain *chain,
4697 					 struct lock_class *class)
4698 {
4699 #ifdef CONFIG_PROVE_LOCKING
4700 	struct lock_chain *new_chain;
4701 	u64 chain_key;
4702 	int i;
4703 
4704 	for (i = chain->base; i < chain->base + chain->depth; i++) {
4705 		if (chain_hlocks[i] != class - lock_classes)
4706 			continue;
4707 		/* The code below leaks one chain_hlock[] entry. */
4708 		if (--chain->depth > 0) {
4709 			memmove(&chain_hlocks[i], &chain_hlocks[i + 1],
4710 				(chain->base + chain->depth - i) *
4711 				sizeof(chain_hlocks[0]));
4712 		}
4713 		/*
4714 		 * Each lock class occurs at most once in a lock chain so once
4715 		 * we found a match we can break out of this loop.
4716 		 */
4717 		goto recalc;
4718 	}
4719 	/* Since the chain has not been modified, return. */
4720 	return;
4721 
4722 recalc:
4723 	chain_key = INITIAL_CHAIN_KEY;
4724 	for (i = chain->base; i < chain->base + chain->depth; i++)
4725 		chain_key = iterate_chain_key(chain_key, chain_hlocks[i]);
4726 	if (chain->depth && chain->chain_key == chain_key)
4727 		return;
4728 	/* Overwrite the chain key for concurrent RCU readers. */
4729 	WRITE_ONCE(chain->chain_key, chain_key);
4730 	/*
4731 	 * Note: calling hlist_del_rcu() from inside a
4732 	 * hlist_for_each_entry_rcu() loop is safe.
4733 	 */
4734 	hlist_del_rcu(&chain->entry);
4735 	__set_bit(chain - lock_chains, pf->lock_chains_being_freed);
4736 	if (chain->depth == 0)
4737 		return;
4738 	/*
4739 	 * If the modified lock chain matches an existing lock chain, drop
4740 	 * the modified lock chain.
4741 	 */
4742 	if (lookup_chain_cache(chain_key))
4743 		return;
4744 	new_chain = alloc_lock_chain();
4745 	if (WARN_ON_ONCE(!new_chain)) {
4746 		debug_locks_off();
4747 		return;
4748 	}
4749 	*new_chain = *chain;
4750 	hlist_add_head_rcu(&new_chain->entry, chainhashentry(chain_key));
4751 #endif
4752 }
4753 
4754 /* Must be called with the graph lock held. */
4755 static void remove_class_from_lock_chains(struct pending_free *pf,
4756 					  struct lock_class *class)
4757 {
4758 	struct lock_chain *chain;
4759 	struct hlist_head *head;
4760 	int i;
4761 
4762 	for (i = 0; i < ARRAY_SIZE(chainhash_table); i++) {
4763 		head = chainhash_table + i;
4764 		hlist_for_each_entry_rcu(chain, head, entry) {
4765 			remove_class_from_lock_chain(pf, chain, class);
4766 		}
4767 	}
4768 }
4769 
4770 /*
4771  * Remove all references to a lock class. The caller must hold the graph lock.
4772  */
4773 static void zap_class(struct pending_free *pf, struct lock_class *class)
4774 {
4775 	struct lock_list *entry;
4776 	int i;
4777 
4778 	WARN_ON_ONCE(!class->key);
4779 
4780 	/*
4781 	 * Remove all dependencies this lock is
4782 	 * involved in:
4783 	 */
4784 	for_each_set_bit(i, list_entries_in_use, ARRAY_SIZE(list_entries)) {
4785 		entry = list_entries + i;
4786 		if (entry->class != class && entry->links_to != class)
4787 			continue;
4788 		__clear_bit(i, list_entries_in_use);
4789 		nr_list_entries--;
4790 		list_del_rcu(&entry->entry);
4791 	}
4792 	if (list_empty(&class->locks_after) &&
4793 	    list_empty(&class->locks_before)) {
4794 		list_move_tail(&class->lock_entry, &pf->zapped);
4795 		hlist_del_rcu(&class->hash_entry);
4796 		WRITE_ONCE(class->key, NULL);
4797 		WRITE_ONCE(class->name, NULL);
4798 		nr_lock_classes--;
4799 		__clear_bit(class - lock_classes, lock_classes_in_use);
4800 	} else {
4801 		WARN_ONCE(true, "%s() failed for class %s\n", __func__,
4802 			  class->name);
4803 	}
4804 
4805 	remove_class_from_lock_chains(pf, class);
4806 }
4807 
4808 static void reinit_class(struct lock_class *class)
4809 {
4810 	void *const p = class;
4811 	const unsigned int offset = offsetof(struct lock_class, key);
4812 
4813 	WARN_ON_ONCE(!class->lock_entry.next);
4814 	WARN_ON_ONCE(!list_empty(&class->locks_after));
4815 	WARN_ON_ONCE(!list_empty(&class->locks_before));
4816 	memset(p + offset, 0, sizeof(*class) - offset);
4817 	WARN_ON_ONCE(!class->lock_entry.next);
4818 	WARN_ON_ONCE(!list_empty(&class->locks_after));
4819 	WARN_ON_ONCE(!list_empty(&class->locks_before));
4820 }
4821 
4822 static inline int within(const void *addr, void *start, unsigned long size)
4823 {
4824 	return addr >= start && addr < start + size;
4825 }
4826 
4827 static bool inside_selftest(void)
4828 {
4829 	return current == lockdep_selftest_task_struct;
4830 }
4831 
4832 /* The caller must hold the graph lock. */
4833 static struct pending_free *get_pending_free(void)
4834 {
4835 	return delayed_free.pf + delayed_free.index;
4836 }
4837 
4838 static void free_zapped_rcu(struct rcu_head *cb);
4839 
4840 /*
4841  * Schedule an RCU callback if no RCU callback is pending. Must be called with
4842  * the graph lock held.
4843  */
4844 static void call_rcu_zapped(struct pending_free *pf)
4845 {
4846 	WARN_ON_ONCE(inside_selftest());
4847 
4848 	if (list_empty(&pf->zapped))
4849 		return;
4850 
4851 	if (delayed_free.scheduled)
4852 		return;
4853 
4854 	delayed_free.scheduled = true;
4855 
4856 	WARN_ON_ONCE(delayed_free.pf + delayed_free.index != pf);
4857 	delayed_free.index ^= 1;
4858 
4859 	call_rcu(&delayed_free.rcu_head, free_zapped_rcu);
4860 }
4861 
4862 /* The caller must hold the graph lock. May be called from RCU context. */
4863 static void __free_zapped_classes(struct pending_free *pf)
4864 {
4865 	struct lock_class *class;
4866 
4867 	check_data_structures();
4868 
4869 	list_for_each_entry(class, &pf->zapped, lock_entry)
4870 		reinit_class(class);
4871 
4872 	list_splice_init(&pf->zapped, &free_lock_classes);
4873 
4874 #ifdef CONFIG_PROVE_LOCKING
4875 	bitmap_andnot(lock_chains_in_use, lock_chains_in_use,
4876 		      pf->lock_chains_being_freed, ARRAY_SIZE(lock_chains));
4877 	bitmap_clear(pf->lock_chains_being_freed, 0, ARRAY_SIZE(lock_chains));
4878 #endif
4879 }
4880 
4881 static void free_zapped_rcu(struct rcu_head *ch)
4882 {
4883 	struct pending_free *pf;
4884 	unsigned long flags;
4885 
4886 	if (WARN_ON_ONCE(ch != &delayed_free.rcu_head))
4887 		return;
4888 
4889 	raw_local_irq_save(flags);
4890 	arch_spin_lock(&lockdep_lock);
4891 	current->lockdep_recursion = 1;
4892 
4893 	/* closed head */
4894 	pf = delayed_free.pf + (delayed_free.index ^ 1);
4895 	__free_zapped_classes(pf);
4896 	delayed_free.scheduled = false;
4897 
4898 	/*
4899 	 * If there's anything on the open list, close and start a new callback.
4900 	 */
4901 	call_rcu_zapped(delayed_free.pf + delayed_free.index);
4902 
4903 	current->lockdep_recursion = 0;
4904 	arch_spin_unlock(&lockdep_lock);
4905 	raw_local_irq_restore(flags);
4906 }
4907 
4908 /*
4909  * Remove all lock classes from the class hash table and from the
4910  * all_lock_classes list whose key or name is in the address range [start,
4911  * start + size). Move these lock classes to the zapped_classes list. Must
4912  * be called with the graph lock held.
4913  */
4914 static void __lockdep_free_key_range(struct pending_free *pf, void *start,
4915 				     unsigned long size)
4916 {
4917 	struct lock_class *class;
4918 	struct hlist_head *head;
4919 	int i;
4920 
4921 	/* Unhash all classes that were created by a module. */
4922 	for (i = 0; i < CLASSHASH_SIZE; i++) {
4923 		head = classhash_table + i;
4924 		hlist_for_each_entry_rcu(class, head, hash_entry) {
4925 			if (!within(class->key, start, size) &&
4926 			    !within(class->name, start, size))
4927 				continue;
4928 			zap_class(pf, class);
4929 		}
4930 	}
4931 }
4932 
4933 /*
4934  * Used in module.c to remove lock classes from memory that is going to be
4935  * freed; and possibly re-used by other modules.
4936  *
4937  * We will have had one synchronize_rcu() before getting here, so we're
4938  * guaranteed nobody will look up these exact classes -- they're properly dead
4939  * but still allocated.
4940  */
4941 static void lockdep_free_key_range_reg(void *start, unsigned long size)
4942 {
4943 	struct pending_free *pf;
4944 	unsigned long flags;
4945 
4946 	init_data_structures_once();
4947 
4948 	raw_local_irq_save(flags);
4949 	arch_spin_lock(&lockdep_lock);
4950 	current->lockdep_recursion = 1;
4951 	pf = get_pending_free();
4952 	__lockdep_free_key_range(pf, start, size);
4953 	call_rcu_zapped(pf);
4954 	current->lockdep_recursion = 0;
4955 	arch_spin_unlock(&lockdep_lock);
4956 	raw_local_irq_restore(flags);
4957 
4958 	/*
4959 	 * Wait for any possible iterators from look_up_lock_class() to pass
4960 	 * before continuing to free the memory they refer to.
4961 	 */
4962 	synchronize_rcu();
4963 }
4964 
4965 /*
4966  * Free all lockdep keys in the range [start, start+size). Does not sleep.
4967  * Ignores debug_locks. Must only be used by the lockdep selftests.
4968  */
4969 static void lockdep_free_key_range_imm(void *start, unsigned long size)
4970 {
4971 	struct pending_free *pf = delayed_free.pf;
4972 	unsigned long flags;
4973 
4974 	init_data_structures_once();
4975 
4976 	raw_local_irq_save(flags);
4977 	arch_spin_lock(&lockdep_lock);
4978 	__lockdep_free_key_range(pf, start, size);
4979 	__free_zapped_classes(pf);
4980 	arch_spin_unlock(&lockdep_lock);
4981 	raw_local_irq_restore(flags);
4982 }
4983 
4984 void lockdep_free_key_range(void *start, unsigned long size)
4985 {
4986 	init_data_structures_once();
4987 
4988 	if (inside_selftest())
4989 		lockdep_free_key_range_imm(start, size);
4990 	else
4991 		lockdep_free_key_range_reg(start, size);
4992 }
4993 
4994 /*
4995  * Check whether any element of the @lock->class_cache[] array refers to a
4996  * registered lock class. The caller must hold either the graph lock or the
4997  * RCU read lock.
4998  */
4999 static bool lock_class_cache_is_registered(struct lockdep_map *lock)
5000 {
5001 	struct lock_class *class;
5002 	struct hlist_head *head;
5003 	int i, j;
5004 
5005 	for (i = 0; i < CLASSHASH_SIZE; i++) {
5006 		head = classhash_table + i;
5007 		hlist_for_each_entry_rcu(class, head, hash_entry) {
5008 			for (j = 0; j < NR_LOCKDEP_CACHING_CLASSES; j++)
5009 				if (lock->class_cache[j] == class)
5010 					return true;
5011 		}
5012 	}
5013 	return false;
5014 }
5015 
5016 /* The caller must hold the graph lock. Does not sleep. */
5017 static void __lockdep_reset_lock(struct pending_free *pf,
5018 				 struct lockdep_map *lock)
5019 {
5020 	struct lock_class *class;
5021 	int j;
5022 
5023 	/*
5024 	 * Remove all classes this lock might have:
5025 	 */
5026 	for (j = 0; j < MAX_LOCKDEP_SUBCLASSES; j++) {
5027 		/*
5028 		 * If the class exists we look it up and zap it:
5029 		 */
5030 		class = look_up_lock_class(lock, j);
5031 		if (class)
5032 			zap_class(pf, class);
5033 	}
5034 	/*
5035 	 * Debug check: in the end all mapped classes should
5036 	 * be gone.
5037 	 */
5038 	if (WARN_ON_ONCE(lock_class_cache_is_registered(lock)))
5039 		debug_locks_off();
5040 }
5041 
5042 /*
5043  * Remove all information lockdep has about a lock if debug_locks == 1. Free
5044  * released data structures from RCU context.
5045  */
5046 static void lockdep_reset_lock_reg(struct lockdep_map *lock)
5047 {
5048 	struct pending_free *pf;
5049 	unsigned long flags;
5050 	int locked;
5051 
5052 	raw_local_irq_save(flags);
5053 	locked = graph_lock();
5054 	if (!locked)
5055 		goto out_irq;
5056 
5057 	pf = get_pending_free();
5058 	__lockdep_reset_lock(pf, lock);
5059 	call_rcu_zapped(pf);
5060 
5061 	graph_unlock();
5062 out_irq:
5063 	raw_local_irq_restore(flags);
5064 }
5065 
5066 /*
5067  * Reset a lock. Does not sleep. Ignores debug_locks. Must only be used by the
5068  * lockdep selftests.
5069  */
5070 static void lockdep_reset_lock_imm(struct lockdep_map *lock)
5071 {
5072 	struct pending_free *pf = delayed_free.pf;
5073 	unsigned long flags;
5074 
5075 	raw_local_irq_save(flags);
5076 	arch_spin_lock(&lockdep_lock);
5077 	__lockdep_reset_lock(pf, lock);
5078 	__free_zapped_classes(pf);
5079 	arch_spin_unlock(&lockdep_lock);
5080 	raw_local_irq_restore(flags);
5081 }
5082 
5083 void lockdep_reset_lock(struct lockdep_map *lock)
5084 {
5085 	init_data_structures_once();
5086 
5087 	if (inside_selftest())
5088 		lockdep_reset_lock_imm(lock);
5089 	else
5090 		lockdep_reset_lock_reg(lock);
5091 }
5092 
5093 /* Unregister a dynamically allocated key. */
5094 void lockdep_unregister_key(struct lock_class_key *key)
5095 {
5096 	struct hlist_head *hash_head = keyhashentry(key);
5097 	struct lock_class_key *k;
5098 	struct pending_free *pf;
5099 	unsigned long flags;
5100 	bool found = false;
5101 
5102 	might_sleep();
5103 
5104 	if (WARN_ON_ONCE(static_obj(key)))
5105 		return;
5106 
5107 	raw_local_irq_save(flags);
5108 	if (!graph_lock())
5109 		goto out_irq;
5110 
5111 	pf = get_pending_free();
5112 	hlist_for_each_entry_rcu(k, hash_head, hash_entry) {
5113 		if (k == key) {
5114 			hlist_del_rcu(&k->hash_entry);
5115 			found = true;
5116 			break;
5117 		}
5118 	}
5119 	WARN_ON_ONCE(!found);
5120 	__lockdep_free_key_range(pf, key, 1);
5121 	call_rcu_zapped(pf);
5122 	graph_unlock();
5123 out_irq:
5124 	raw_local_irq_restore(flags);
5125 
5126 	/* Wait until is_dynamic_key() has finished accessing k->hash_entry. */
5127 	synchronize_rcu();
5128 }
5129 EXPORT_SYMBOL_GPL(lockdep_unregister_key);
5130 
5131 void __init lockdep_init(void)
5132 {
5133 	printk("Lock dependency validator: Copyright (c) 2006 Red Hat, Inc., Ingo Molnar\n");
5134 
5135 	printk("... MAX_LOCKDEP_SUBCLASSES:  %lu\n", MAX_LOCKDEP_SUBCLASSES);
5136 	printk("... MAX_LOCK_DEPTH:          %lu\n", MAX_LOCK_DEPTH);
5137 	printk("... MAX_LOCKDEP_KEYS:        %lu\n", MAX_LOCKDEP_KEYS);
5138 	printk("... CLASSHASH_SIZE:          %lu\n", CLASSHASH_SIZE);
5139 	printk("... MAX_LOCKDEP_ENTRIES:     %lu\n", MAX_LOCKDEP_ENTRIES);
5140 	printk("... MAX_LOCKDEP_CHAINS:      %lu\n", MAX_LOCKDEP_CHAINS);
5141 	printk("... CHAINHASH_SIZE:          %lu\n", CHAINHASH_SIZE);
5142 
5143 	printk(" memory used by lock dependency info: %zu kB\n",
5144 	       (sizeof(lock_classes) +
5145 		sizeof(lock_classes_in_use) +
5146 		sizeof(classhash_table) +
5147 		sizeof(list_entries) +
5148 		sizeof(list_entries_in_use) +
5149 		sizeof(chainhash_table) +
5150 		sizeof(delayed_free)
5151 #ifdef CONFIG_PROVE_LOCKING
5152 		+ sizeof(lock_cq)
5153 		+ sizeof(lock_chains)
5154 		+ sizeof(lock_chains_in_use)
5155 		+ sizeof(chain_hlocks)
5156 #endif
5157 		) / 1024
5158 		);
5159 
5160 	printk(" per task-struct memory footprint: %zu bytes\n",
5161 	       sizeof(((struct task_struct *)NULL)->held_locks));
5162 }
5163 
5164 static void
5165 print_freed_lock_bug(struct task_struct *curr, const void *mem_from,
5166 		     const void *mem_to, struct held_lock *hlock)
5167 {
5168 	if (!debug_locks_off())
5169 		return;
5170 	if (debug_locks_silent)
5171 		return;
5172 
5173 	pr_warn("\n");
5174 	pr_warn("=========================\n");
5175 	pr_warn("WARNING: held lock freed!\n");
5176 	print_kernel_ident();
5177 	pr_warn("-------------------------\n");
5178 	pr_warn("%s/%d is freeing memory %px-%px, with a lock still held there!\n",
5179 		curr->comm, task_pid_nr(curr), mem_from, mem_to-1);
5180 	print_lock(hlock);
5181 	lockdep_print_held_locks(curr);
5182 
5183 	pr_warn("\nstack backtrace:\n");
5184 	dump_stack();
5185 }
5186 
5187 static inline int not_in_range(const void* mem_from, unsigned long mem_len,
5188 				const void* lock_from, unsigned long lock_len)
5189 {
5190 	return lock_from + lock_len <= mem_from ||
5191 		mem_from + mem_len <= lock_from;
5192 }
5193 
5194 /*
5195  * Called when kernel memory is freed (or unmapped), or if a lock
5196  * is destroyed or reinitialized - this code checks whether there is
5197  * any held lock in the memory range of <from> to <to>:
5198  */
5199 void debug_check_no_locks_freed(const void *mem_from, unsigned long mem_len)
5200 {
5201 	struct task_struct *curr = current;
5202 	struct held_lock *hlock;
5203 	unsigned long flags;
5204 	int i;
5205 
5206 	if (unlikely(!debug_locks))
5207 		return;
5208 
5209 	raw_local_irq_save(flags);
5210 	for (i = 0; i < curr->lockdep_depth; i++) {
5211 		hlock = curr->held_locks + i;
5212 
5213 		if (not_in_range(mem_from, mem_len, hlock->instance,
5214 					sizeof(*hlock->instance)))
5215 			continue;
5216 
5217 		print_freed_lock_bug(curr, mem_from, mem_from + mem_len, hlock);
5218 		break;
5219 	}
5220 	raw_local_irq_restore(flags);
5221 }
5222 EXPORT_SYMBOL_GPL(debug_check_no_locks_freed);
5223 
5224 static void print_held_locks_bug(void)
5225 {
5226 	if (!debug_locks_off())
5227 		return;
5228 	if (debug_locks_silent)
5229 		return;
5230 
5231 	pr_warn("\n");
5232 	pr_warn("====================================\n");
5233 	pr_warn("WARNING: %s/%d still has locks held!\n",
5234 	       current->comm, task_pid_nr(current));
5235 	print_kernel_ident();
5236 	pr_warn("------------------------------------\n");
5237 	lockdep_print_held_locks(current);
5238 	pr_warn("\nstack backtrace:\n");
5239 	dump_stack();
5240 }
5241 
5242 void debug_check_no_locks_held(void)
5243 {
5244 	if (unlikely(current->lockdep_depth > 0))
5245 		print_held_locks_bug();
5246 }
5247 EXPORT_SYMBOL_GPL(debug_check_no_locks_held);
5248 
5249 #ifdef __KERNEL__
5250 void debug_show_all_locks(void)
5251 {
5252 	struct task_struct *g, *p;
5253 
5254 	if (unlikely(!debug_locks)) {
5255 		pr_warn("INFO: lockdep is turned off.\n");
5256 		return;
5257 	}
5258 	pr_warn("\nShowing all locks held in the system:\n");
5259 
5260 	rcu_read_lock();
5261 	for_each_process_thread(g, p) {
5262 		if (!p->lockdep_depth)
5263 			continue;
5264 		lockdep_print_held_locks(p);
5265 		touch_nmi_watchdog();
5266 		touch_all_softlockup_watchdogs();
5267 	}
5268 	rcu_read_unlock();
5269 
5270 	pr_warn("\n");
5271 	pr_warn("=============================================\n\n");
5272 }
5273 EXPORT_SYMBOL_GPL(debug_show_all_locks);
5274 #endif
5275 
5276 /*
5277  * Careful: only use this function if you are sure that
5278  * the task cannot run in parallel!
5279  */
5280 void debug_show_held_locks(struct task_struct *task)
5281 {
5282 	if (unlikely(!debug_locks)) {
5283 		printk("INFO: lockdep is turned off.\n");
5284 		return;
5285 	}
5286 	lockdep_print_held_locks(task);
5287 }
5288 EXPORT_SYMBOL_GPL(debug_show_held_locks);
5289 
5290 asmlinkage __visible void lockdep_sys_exit(void)
5291 {
5292 	struct task_struct *curr = current;
5293 
5294 	if (unlikely(curr->lockdep_depth)) {
5295 		if (!debug_locks_off())
5296 			return;
5297 		pr_warn("\n");
5298 		pr_warn("================================================\n");
5299 		pr_warn("WARNING: lock held when returning to user space!\n");
5300 		print_kernel_ident();
5301 		pr_warn("------------------------------------------------\n");
5302 		pr_warn("%s/%d is leaving the kernel with locks still held!\n",
5303 				curr->comm, curr->pid);
5304 		lockdep_print_held_locks(curr);
5305 	}
5306 
5307 	/*
5308 	 * The lock history for each syscall should be independent. So wipe the
5309 	 * slate clean on return to userspace.
5310 	 */
5311 	lockdep_invariant_state(false);
5312 }
5313 
5314 void lockdep_rcu_suspicious(const char *file, const int line, const char *s)
5315 {
5316 	struct task_struct *curr = current;
5317 
5318 	/* Note: the following can be executed concurrently, so be careful. */
5319 	pr_warn("\n");
5320 	pr_warn("=============================\n");
5321 	pr_warn("WARNING: suspicious RCU usage\n");
5322 	print_kernel_ident();
5323 	pr_warn("-----------------------------\n");
5324 	pr_warn("%s:%d %s!\n", file, line, s);
5325 	pr_warn("\nother info that might help us debug this:\n\n");
5326 	pr_warn("\n%srcu_scheduler_active = %d, debug_locks = %d\n",
5327 	       !rcu_lockdep_current_cpu_online()
5328 			? "RCU used illegally from offline CPU!\n"
5329 			: !rcu_is_watching()
5330 				? "RCU used illegally from idle CPU!\n"
5331 				: "",
5332 	       rcu_scheduler_active, debug_locks);
5333 
5334 	/*
5335 	 * If a CPU is in the RCU-free window in idle (ie: in the section
5336 	 * between rcu_idle_enter() and rcu_idle_exit(), then RCU
5337 	 * considers that CPU to be in an "extended quiescent state",
5338 	 * which means that RCU will be completely ignoring that CPU.
5339 	 * Therefore, rcu_read_lock() and friends have absolutely no
5340 	 * effect on a CPU running in that state. In other words, even if
5341 	 * such an RCU-idle CPU has called rcu_read_lock(), RCU might well
5342 	 * delete data structures out from under it.  RCU really has no
5343 	 * choice here: we need to keep an RCU-free window in idle where
5344 	 * the CPU may possibly enter into low power mode. This way we can
5345 	 * notice an extended quiescent state to other CPUs that started a grace
5346 	 * period. Otherwise we would delay any grace period as long as we run
5347 	 * in the idle task.
5348 	 *
5349 	 * So complain bitterly if someone does call rcu_read_lock(),
5350 	 * rcu_read_lock_bh() and so on from extended quiescent states.
5351 	 */
5352 	if (!rcu_is_watching())
5353 		pr_warn("RCU used illegally from extended quiescent state!\n");
5354 
5355 	lockdep_print_held_locks(curr);
5356 	pr_warn("\nstack backtrace:\n");
5357 	dump_stack();
5358 }
5359 EXPORT_SYMBOL_GPL(lockdep_rcu_suspicious);
5360