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