Loading...
Note: File does not exist in v3.1.
1/*
2 * kernel/lockdep.c
3 *
4 * Runtime locking correctness validator
5 *
6 * Started by Ingo Molnar:
7 *
8 * Copyright (C) 2006,2007 Red Hat, Inc., Ingo Molnar <mingo@redhat.com>
9 * Copyright (C) 2007 Red Hat, Inc., Peter Zijlstra <pzijlstr@redhat.com>
10 *
11 * this code maps all the lock dependencies as they occur in a live kernel
12 * and will warn about the following classes of locking bugs:
13 *
14 * - lock inversion scenarios
15 * - circular lock dependencies
16 * - hardirq/softirq safe/unsafe locking bugs
17 *
18 * Bugs are reported even if the current locking scenario does not cause
19 * any deadlock at this point.
20 *
21 * I.e. if anytime in the past two locks were taken in a different order,
22 * even if it happened for another task, even if those were different
23 * locks (but of the same class as this lock), this code will detect it.
24 *
25 * Thanks to Arjan van de Ven for coming up with the initial idea of
26 * mapping lock dependencies runtime.
27 */
28#define DISABLE_BRANCH_PROFILING
29#include <linux/mutex.h>
30#include <linux/sched.h>
31#include <linux/delay.h>
32#include <linux/module.h>
33#include <linux/proc_fs.h>
34#include <linux/seq_file.h>
35#include <linux/spinlock.h>
36#include <linux/kallsyms.h>
37#include <linux/interrupt.h>
38#include <linux/stacktrace.h>
39#include <linux/debug_locks.h>
40#include <linux/irqflags.h>
41#include <linux/utsname.h>
42#include <linux/hash.h>
43#include <linux/ftrace.h>
44#include <linux/stringify.h>
45#include <linux/bitops.h>
46#include <linux/gfp.h>
47#include <linux/kmemcheck.h>
48
49#include <asm/sections.h>
50
51#include "lockdep_internals.h"
52
53#define CREATE_TRACE_POINTS
54#include <trace/events/lock.h>
55
56#ifdef CONFIG_PROVE_LOCKING
57int prove_locking = 1;
58module_param(prove_locking, int, 0644);
59#else
60#define prove_locking 0
61#endif
62
63#ifdef CONFIG_LOCK_STAT
64int lock_stat = 1;
65module_param(lock_stat, int, 0644);
66#else
67#define lock_stat 0
68#endif
69
70/*
71 * lockdep_lock: protects the lockdep graph, the hashes and the
72 * class/list/hash allocators.
73 *
74 * This is one of the rare exceptions where it's justified
75 * to use a raw spinlock - we really dont want the spinlock
76 * code to recurse back into the lockdep code...
77 */
78static arch_spinlock_t lockdep_lock = (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
79
80static int graph_lock(void)
81{
82 arch_spin_lock(&lockdep_lock);
83 /*
84 * Make sure that if another CPU detected a bug while
85 * walking the graph we dont change it (while the other
86 * CPU is busy printing out stuff with the graph lock
87 * dropped already)
88 */
89 if (!debug_locks) {
90 arch_spin_unlock(&lockdep_lock);
91 return 0;
92 }
93 /* prevent any recursions within lockdep from causing deadlocks */
94 current->lockdep_recursion++;
95 return 1;
96}
97
98static inline int graph_unlock(void)
99{
100 if (debug_locks && !arch_spin_is_locked(&lockdep_lock)) {
101 /*
102 * The lockdep graph lock isn't locked while we expect it to
103 * be, we're confused now, bye!
104 */
105 return DEBUG_LOCKS_WARN_ON(1);
106 }
107
108 current->lockdep_recursion--;
109 arch_spin_unlock(&lockdep_lock);
110 return 0;
111}
112
113/*
114 * Turn lock debugging off and return with 0 if it was off already,
115 * and also release the graph lock:
116 */
117static inline int debug_locks_off_graph_unlock(void)
118{
119 int ret = debug_locks_off();
120
121 arch_spin_unlock(&lockdep_lock);
122
123 return ret;
124}
125
126static int lockdep_initialized;
127
128unsigned long nr_list_entries;
129static struct lock_list list_entries[MAX_LOCKDEP_ENTRIES];
130
131/*
132 * All data structures here are protected by the global debug_lock.
133 *
134 * Mutex key structs only get allocated, once during bootup, and never
135 * get freed - this significantly simplifies the debugging code.
136 */
137unsigned long nr_lock_classes;
138static struct lock_class lock_classes[MAX_LOCKDEP_KEYS];
139
140static inline struct lock_class *hlock_class(struct held_lock *hlock)
141{
142 if (!hlock->class_idx) {
143 /*
144 * Someone passed in garbage, we give up.
145 */
146 DEBUG_LOCKS_WARN_ON(1);
147 return NULL;
148 }
149 return lock_classes + hlock->class_idx - 1;
150}
151
152#ifdef CONFIG_LOCK_STAT
153static DEFINE_PER_CPU(struct lock_class_stats[MAX_LOCKDEP_KEYS],
154 cpu_lock_stats);
155
156static inline u64 lockstat_clock(void)
157{
158 return local_clock();
159}
160
161static int lock_point(unsigned long points[], unsigned long ip)
162{
163 int i;
164
165 for (i = 0; i < LOCKSTAT_POINTS; i++) {
166 if (points[i] == 0) {
167 points[i] = ip;
168 break;
169 }
170 if (points[i] == ip)
171 break;
172 }
173
174 return i;
175}
176
177static void lock_time_inc(struct lock_time *lt, u64 time)
178{
179 if (time > lt->max)
180 lt->max = time;
181
182 if (time < lt->min || !lt->nr)
183 lt->min = time;
184
185 lt->total += time;
186 lt->nr++;
187}
188
189static inline void lock_time_add(struct lock_time *src, struct lock_time *dst)
190{
191 if (!src->nr)
192 return;
193
194 if (src->max > dst->max)
195 dst->max = src->max;
196
197 if (src->min < dst->min || !dst->nr)
198 dst->min = src->min;
199
200 dst->total += src->total;
201 dst->nr += src->nr;
202}
203
204struct lock_class_stats lock_stats(struct lock_class *class)
205{
206 struct lock_class_stats stats;
207 int cpu, i;
208
209 memset(&stats, 0, sizeof(struct lock_class_stats));
210 for_each_possible_cpu(cpu) {
211 struct lock_class_stats *pcs =
212 &per_cpu(cpu_lock_stats, cpu)[class - lock_classes];
213
214 for (i = 0; i < ARRAY_SIZE(stats.contention_point); i++)
215 stats.contention_point[i] += pcs->contention_point[i];
216
217 for (i = 0; i < ARRAY_SIZE(stats.contending_point); i++)
218 stats.contending_point[i] += pcs->contending_point[i];
219
220 lock_time_add(&pcs->read_waittime, &stats.read_waittime);
221 lock_time_add(&pcs->write_waittime, &stats.write_waittime);
222
223 lock_time_add(&pcs->read_holdtime, &stats.read_holdtime);
224 lock_time_add(&pcs->write_holdtime, &stats.write_holdtime);
225
226 for (i = 0; i < ARRAY_SIZE(stats.bounces); i++)
227 stats.bounces[i] += pcs->bounces[i];
228 }
229
230 return stats;
231}
232
233void clear_lock_stats(struct lock_class *class)
234{
235 int cpu;
236
237 for_each_possible_cpu(cpu) {
238 struct lock_class_stats *cpu_stats =
239 &per_cpu(cpu_lock_stats, cpu)[class - lock_classes];
240
241 memset(cpu_stats, 0, sizeof(struct lock_class_stats));
242 }
243 memset(class->contention_point, 0, sizeof(class->contention_point));
244 memset(class->contending_point, 0, sizeof(class->contending_point));
245}
246
247static struct lock_class_stats *get_lock_stats(struct lock_class *class)
248{
249 return &get_cpu_var(cpu_lock_stats)[class - lock_classes];
250}
251
252static void put_lock_stats(struct lock_class_stats *stats)
253{
254 put_cpu_var(cpu_lock_stats);
255}
256
257static void lock_release_holdtime(struct held_lock *hlock)
258{
259 struct lock_class_stats *stats;
260 u64 holdtime;
261
262 if (!lock_stat)
263 return;
264
265 holdtime = lockstat_clock() - hlock->holdtime_stamp;
266
267 stats = get_lock_stats(hlock_class(hlock));
268 if (hlock->read)
269 lock_time_inc(&stats->read_holdtime, holdtime);
270 else
271 lock_time_inc(&stats->write_holdtime, holdtime);
272 put_lock_stats(stats);
273}
274#else
275static inline void lock_release_holdtime(struct held_lock *hlock)
276{
277}
278#endif
279
280/*
281 * We keep a global list of all lock classes. The list only grows,
282 * never shrinks. The list is only accessed with the lockdep
283 * spinlock lock held.
284 */
285LIST_HEAD(all_lock_classes);
286
287/*
288 * The lockdep classes are in a hash-table as well, for fast lookup:
289 */
290#define CLASSHASH_BITS (MAX_LOCKDEP_KEYS_BITS - 1)
291#define CLASSHASH_SIZE (1UL << CLASSHASH_BITS)
292#define __classhashfn(key) hash_long((unsigned long)key, CLASSHASH_BITS)
293#define classhashentry(key) (classhash_table + __classhashfn((key)))
294
295static struct list_head classhash_table[CLASSHASH_SIZE];
296
297/*
298 * We put the lock dependency chains into a hash-table as well, to cache
299 * their existence:
300 */
301#define CHAINHASH_BITS (MAX_LOCKDEP_CHAINS_BITS-1)
302#define CHAINHASH_SIZE (1UL << CHAINHASH_BITS)
303#define __chainhashfn(chain) hash_long(chain, CHAINHASH_BITS)
304#define chainhashentry(chain) (chainhash_table + __chainhashfn((chain)))
305
306static struct list_head chainhash_table[CHAINHASH_SIZE];
307
308/*
309 * The hash key of the lock dependency chains is a hash itself too:
310 * it's a hash of all locks taken up to that lock, including that lock.
311 * It's a 64-bit hash, because it's important for the keys to be
312 * unique.
313 */
314#define iterate_chain_key(key1, key2) \
315 (((key1) << MAX_LOCKDEP_KEYS_BITS) ^ \
316 ((key1) >> (64-MAX_LOCKDEP_KEYS_BITS)) ^ \
317 (key2))
318
319void lockdep_off(void)
320{
321 current->lockdep_recursion++;
322}
323EXPORT_SYMBOL(lockdep_off);
324
325void lockdep_on(void)
326{
327 current->lockdep_recursion--;
328}
329EXPORT_SYMBOL(lockdep_on);
330
331/*
332 * Debugging switches:
333 */
334
335#define VERBOSE 0
336#define VERY_VERBOSE 0
337
338#if VERBOSE
339# define HARDIRQ_VERBOSE 1
340# define SOFTIRQ_VERBOSE 1
341# define RECLAIM_VERBOSE 1
342#else
343# define HARDIRQ_VERBOSE 0
344# define SOFTIRQ_VERBOSE 0
345# define RECLAIM_VERBOSE 0
346#endif
347
348#if VERBOSE || HARDIRQ_VERBOSE || SOFTIRQ_VERBOSE || RECLAIM_VERBOSE
349/*
350 * Quick filtering for interesting events:
351 */
352static int class_filter(struct lock_class *class)
353{
354#if 0
355 /* Example */
356 if (class->name_version == 1 &&
357 !strcmp(class->name, "lockname"))
358 return 1;
359 if (class->name_version == 1 &&
360 !strcmp(class->name, "&struct->lockfield"))
361 return 1;
362#endif
363 /* Filter everything else. 1 would be to allow everything else */
364 return 0;
365}
366#endif
367
368static int verbose(struct lock_class *class)
369{
370#if VERBOSE
371 return class_filter(class);
372#endif
373 return 0;
374}
375
376/*
377 * Stack-trace: tightly packed array of stack backtrace
378 * addresses. Protected by the graph_lock.
379 */
380unsigned long nr_stack_trace_entries;
381static unsigned long stack_trace[MAX_STACK_TRACE_ENTRIES];
382
383static void print_lockdep_off(const char *bug_msg)
384{
385 printk(KERN_DEBUG "%s\n", bug_msg);
386 printk(KERN_DEBUG "turning off the locking correctness validator.\n");
387 printk(KERN_DEBUG "Please attach the output of /proc/lock_stat to the bug report\n");
388}
389
390static int save_trace(struct stack_trace *trace)
391{
392 trace->nr_entries = 0;
393 trace->max_entries = MAX_STACK_TRACE_ENTRIES - nr_stack_trace_entries;
394 trace->entries = stack_trace + nr_stack_trace_entries;
395
396 trace->skip = 3;
397
398 save_stack_trace(trace);
399
400 /*
401 * Some daft arches put -1 at the end to indicate its a full trace.
402 *
403 * <rant> this is buggy anyway, since it takes a whole extra entry so a
404 * complete trace that maxes out the entries provided will be reported
405 * as incomplete, friggin useless </rant>
406 */
407 if (trace->nr_entries != 0 &&
408 trace->entries[trace->nr_entries-1] == ULONG_MAX)
409 trace->nr_entries--;
410
411 trace->max_entries = trace->nr_entries;
412
413 nr_stack_trace_entries += trace->nr_entries;
414
415 if (nr_stack_trace_entries >= MAX_STACK_TRACE_ENTRIES-1) {
416 if (!debug_locks_off_graph_unlock())
417 return 0;
418
419 print_lockdep_off("BUG: MAX_STACK_TRACE_ENTRIES too low!");
420 dump_stack();
421
422 return 0;
423 }
424
425 return 1;
426}
427
428unsigned int nr_hardirq_chains;
429unsigned int nr_softirq_chains;
430unsigned int nr_process_chains;
431unsigned int max_lockdep_depth;
432
433#ifdef CONFIG_DEBUG_LOCKDEP
434/*
435 * We cannot printk in early bootup code. Not even early_printk()
436 * might work. So we mark any initialization errors and printk
437 * about it later on, in lockdep_info().
438 */
439static int lockdep_init_error;
440static const char *lock_init_error;
441static unsigned long lockdep_init_trace_data[20];
442static struct stack_trace lockdep_init_trace = {
443 .max_entries = ARRAY_SIZE(lockdep_init_trace_data),
444 .entries = lockdep_init_trace_data,
445};
446
447/*
448 * Various lockdep statistics:
449 */
450DEFINE_PER_CPU(struct lockdep_stats, lockdep_stats);
451#endif
452
453/*
454 * Locking printouts:
455 */
456
457#define __USAGE(__STATE) \
458 [LOCK_USED_IN_##__STATE] = "IN-"__stringify(__STATE)"-W", \
459 [LOCK_ENABLED_##__STATE] = __stringify(__STATE)"-ON-W", \
460 [LOCK_USED_IN_##__STATE##_READ] = "IN-"__stringify(__STATE)"-R",\
461 [LOCK_ENABLED_##__STATE##_READ] = __stringify(__STATE)"-ON-R",
462
463static const char *usage_str[] =
464{
465#define LOCKDEP_STATE(__STATE) __USAGE(__STATE)
466#include "lockdep_states.h"
467#undef LOCKDEP_STATE
468 [LOCK_USED] = "INITIAL USE",
469};
470
471const char * __get_key_name(struct lockdep_subclass_key *key, char *str)
472{
473 return kallsyms_lookup((unsigned long)key, NULL, NULL, NULL, str);
474}
475
476static inline unsigned long lock_flag(enum lock_usage_bit bit)
477{
478 return 1UL << bit;
479}
480
481static char get_usage_char(struct lock_class *class, enum lock_usage_bit bit)
482{
483 char c = '.';
484
485 if (class->usage_mask & lock_flag(bit + 2))
486 c = '+';
487 if (class->usage_mask & lock_flag(bit)) {
488 c = '-';
489 if (class->usage_mask & lock_flag(bit + 2))
490 c = '?';
491 }
492
493 return c;
494}
495
496void get_usage_chars(struct lock_class *class, char usage[LOCK_USAGE_CHARS])
497{
498 int i = 0;
499
500#define LOCKDEP_STATE(__STATE) \
501 usage[i++] = get_usage_char(class, LOCK_USED_IN_##__STATE); \
502 usage[i++] = get_usage_char(class, LOCK_USED_IN_##__STATE##_READ);
503#include "lockdep_states.h"
504#undef LOCKDEP_STATE
505
506 usage[i] = '\0';
507}
508
509static void __print_lock_name(struct lock_class *class)
510{
511 char str[KSYM_NAME_LEN];
512 const char *name;
513
514 name = class->name;
515 if (!name) {
516 name = __get_key_name(class->key, str);
517 printk("%s", name);
518 } else {
519 printk("%s", name);
520 if (class->name_version > 1)
521 printk("#%d", class->name_version);
522 if (class->subclass)
523 printk("/%d", class->subclass);
524 }
525}
526
527static void print_lock_name(struct lock_class *class)
528{
529 char usage[LOCK_USAGE_CHARS];
530
531 get_usage_chars(class, usage);
532
533 printk(" (");
534 __print_lock_name(class);
535 printk("){%s}", usage);
536}
537
538static void print_lockdep_cache(struct lockdep_map *lock)
539{
540 const char *name;
541 char str[KSYM_NAME_LEN];
542
543 name = lock->name;
544 if (!name)
545 name = __get_key_name(lock->key->subkeys, str);
546
547 printk("%s", name);
548}
549
550static void print_lock(struct held_lock *hlock)
551{
552 print_lock_name(hlock_class(hlock));
553 printk(", at: ");
554 print_ip_sym(hlock->acquire_ip);
555}
556
557static void lockdep_print_held_locks(struct task_struct *curr)
558{
559 int i, depth = curr->lockdep_depth;
560
561 if (!depth) {
562 printk("no locks held by %s/%d.\n", curr->comm, task_pid_nr(curr));
563 return;
564 }
565 printk("%d lock%s held by %s/%d:\n",
566 depth, depth > 1 ? "s" : "", curr->comm, task_pid_nr(curr));
567
568 for (i = 0; i < depth; i++) {
569 printk(" #%d: ", i);
570 print_lock(curr->held_locks + i);
571 }
572}
573
574static void print_kernel_ident(void)
575{
576 printk("%s %.*s %s\n", init_utsname()->release,
577 (int)strcspn(init_utsname()->version, " "),
578 init_utsname()->version,
579 print_tainted());
580}
581
582static int very_verbose(struct lock_class *class)
583{
584#if VERY_VERBOSE
585 return class_filter(class);
586#endif
587 return 0;
588}
589
590/*
591 * Is this the address of a static object:
592 */
593#ifdef __KERNEL__
594static int static_obj(void *obj)
595{
596 unsigned long start = (unsigned long) &_stext,
597 end = (unsigned long) &_end,
598 addr = (unsigned long) obj;
599
600 /*
601 * static variable?
602 */
603 if ((addr >= start) && (addr < end))
604 return 1;
605
606 if (arch_is_kernel_data(addr))
607 return 1;
608
609 /*
610 * in-kernel percpu var?
611 */
612 if (is_kernel_percpu_address(addr))
613 return 1;
614
615 /*
616 * module static or percpu var?
617 */
618 return is_module_address(addr) || is_module_percpu_address(addr);
619}
620#endif
621
622/*
623 * To make lock name printouts unique, we calculate a unique
624 * class->name_version generation counter:
625 */
626static int count_matching_names(struct lock_class *new_class)
627{
628 struct lock_class *class;
629 int count = 0;
630
631 if (!new_class->name)
632 return 0;
633
634 list_for_each_entry(class, &all_lock_classes, lock_entry) {
635 if (new_class->key - new_class->subclass == class->key)
636 return class->name_version;
637 if (class->name && !strcmp(class->name, new_class->name))
638 count = max(count, class->name_version);
639 }
640
641 return count + 1;
642}
643
644/*
645 * Register a lock's class in the hash-table, if the class is not present
646 * yet. Otherwise we look it up. We cache the result in the lock object
647 * itself, so actual lookup of the hash should be once per lock object.
648 */
649static inline struct lock_class *
650look_up_lock_class(struct lockdep_map *lock, unsigned int subclass)
651{
652 struct lockdep_subclass_key *key;
653 struct list_head *hash_head;
654 struct lock_class *class;
655
656#ifdef CONFIG_DEBUG_LOCKDEP
657 /*
658 * If the architecture calls into lockdep before initializing
659 * the hashes then we'll warn about it later. (we cannot printk
660 * right now)
661 */
662 if (unlikely(!lockdep_initialized)) {
663 lockdep_init();
664 lockdep_init_error = 1;
665 lock_init_error = lock->name;
666 save_stack_trace(&lockdep_init_trace);
667 }
668#endif
669
670 if (unlikely(subclass >= MAX_LOCKDEP_SUBCLASSES)) {
671 debug_locks_off();
672 printk(KERN_ERR
673 "BUG: looking up invalid subclass: %u\n", subclass);
674 printk(KERN_ERR
675 "turning off the locking correctness validator.\n");
676 dump_stack();
677 return NULL;
678 }
679
680 /*
681 * Static locks do not have their class-keys yet - for them the key
682 * is the lock object itself:
683 */
684 if (unlikely(!lock->key))
685 lock->key = (void *)lock;
686
687 /*
688 * NOTE: the class-key must be unique. For dynamic locks, a static
689 * lock_class_key variable is passed in through the mutex_init()
690 * (or spin_lock_init()) call - which acts as the key. For static
691 * locks we use the lock object itself as the key.
692 */
693 BUILD_BUG_ON(sizeof(struct lock_class_key) >
694 sizeof(struct lockdep_map));
695
696 key = lock->key->subkeys + subclass;
697
698 hash_head = classhashentry(key);
699
700 /*
701 * We can walk the hash lockfree, because the hash only
702 * grows, and we are careful when adding entries to the end:
703 */
704 list_for_each_entry(class, hash_head, hash_entry) {
705 if (class->key == key) {
706 /*
707 * Huh! same key, different name? Did someone trample
708 * on some memory? We're most confused.
709 */
710 WARN_ON_ONCE(class->name != lock->name);
711 return class;
712 }
713 }
714
715 return NULL;
716}
717
718/*
719 * Register a lock's class in the hash-table, if the class is not present
720 * yet. Otherwise we look it up. We cache the result in the lock object
721 * itself, so actual lookup of the hash should be once per lock object.
722 */
723static inline struct lock_class *
724register_lock_class(struct lockdep_map *lock, unsigned int subclass, int force)
725{
726 struct lockdep_subclass_key *key;
727 struct list_head *hash_head;
728 struct lock_class *class;
729 unsigned long flags;
730
731 class = look_up_lock_class(lock, subclass);
732 if (likely(class))
733 goto out_set_class_cache;
734
735 /*
736 * Debug-check: all keys must be persistent!
737 */
738 if (!static_obj(lock->key)) {
739 debug_locks_off();
740 printk("INFO: trying to register non-static key.\n");
741 printk("the code is fine but needs lockdep annotation.\n");
742 printk("turning off the locking correctness validator.\n");
743 dump_stack();
744
745 return NULL;
746 }
747
748 key = lock->key->subkeys + subclass;
749 hash_head = classhashentry(key);
750
751 raw_local_irq_save(flags);
752 if (!graph_lock()) {
753 raw_local_irq_restore(flags);
754 return NULL;
755 }
756 /*
757 * We have to do the hash-walk again, to avoid races
758 * with another CPU:
759 */
760 list_for_each_entry(class, hash_head, hash_entry)
761 if (class->key == key)
762 goto out_unlock_set;
763 /*
764 * Allocate a new key from the static array, and add it to
765 * the hash:
766 */
767 if (nr_lock_classes >= MAX_LOCKDEP_KEYS) {
768 if (!debug_locks_off_graph_unlock()) {
769 raw_local_irq_restore(flags);
770 return NULL;
771 }
772 raw_local_irq_restore(flags);
773
774 print_lockdep_off("BUG: MAX_LOCKDEP_KEYS too low!");
775 dump_stack();
776 return NULL;
777 }
778 class = lock_classes + nr_lock_classes++;
779 debug_atomic_inc(nr_unused_locks);
780 class->key = key;
781 class->name = lock->name;
782 class->subclass = subclass;
783 INIT_LIST_HEAD(&class->lock_entry);
784 INIT_LIST_HEAD(&class->locks_before);
785 INIT_LIST_HEAD(&class->locks_after);
786 class->name_version = count_matching_names(class);
787 /*
788 * We use RCU's safe list-add method to make
789 * parallel walking of the hash-list safe:
790 */
791 list_add_tail_rcu(&class->hash_entry, hash_head);
792 /*
793 * Add it to the global list of classes:
794 */
795 list_add_tail_rcu(&class->lock_entry, &all_lock_classes);
796
797 if (verbose(class)) {
798 graph_unlock();
799 raw_local_irq_restore(flags);
800
801 printk("\nnew class %p: %s", class->key, class->name);
802 if (class->name_version > 1)
803 printk("#%d", class->name_version);
804 printk("\n");
805 dump_stack();
806
807 raw_local_irq_save(flags);
808 if (!graph_lock()) {
809 raw_local_irq_restore(flags);
810 return NULL;
811 }
812 }
813out_unlock_set:
814 graph_unlock();
815 raw_local_irq_restore(flags);
816
817out_set_class_cache:
818 if (!subclass || force)
819 lock->class_cache[0] = class;
820 else if (subclass < NR_LOCKDEP_CACHING_CLASSES)
821 lock->class_cache[subclass] = class;
822
823 /*
824 * Hash collision, did we smoke some? We found a class with a matching
825 * hash but the subclass -- which is hashed in -- didn't match.
826 */
827 if (DEBUG_LOCKS_WARN_ON(class->subclass != subclass))
828 return NULL;
829
830 return class;
831}
832
833#ifdef CONFIG_PROVE_LOCKING
834/*
835 * Allocate a lockdep entry. (assumes the graph_lock held, returns
836 * with NULL on failure)
837 */
838static struct lock_list *alloc_list_entry(void)
839{
840 if (nr_list_entries >= MAX_LOCKDEP_ENTRIES) {
841 if (!debug_locks_off_graph_unlock())
842 return NULL;
843
844 print_lockdep_off("BUG: MAX_LOCKDEP_ENTRIES too low!");
845 dump_stack();
846 return NULL;
847 }
848 return list_entries + nr_list_entries++;
849}
850
851/*
852 * Add a new dependency to the head of the list:
853 */
854static int add_lock_to_list(struct lock_class *class, struct lock_class *this,
855 struct list_head *head, unsigned long ip,
856 int distance, struct stack_trace *trace)
857{
858 struct lock_list *entry;
859 /*
860 * Lock not present yet - get a new dependency struct and
861 * add it to the list:
862 */
863 entry = alloc_list_entry();
864 if (!entry)
865 return 0;
866
867 entry->class = this;
868 entry->distance = distance;
869 entry->trace = *trace;
870 /*
871 * Since we never remove from the dependency list, the list can
872 * be walked lockless by other CPUs, it's only allocation
873 * that must be protected by the spinlock. But this also means
874 * we must make new entries visible only once writes to the
875 * entry become visible - hence the RCU op:
876 */
877 list_add_tail_rcu(&entry->entry, head);
878
879 return 1;
880}
881
882/*
883 * For good efficiency of modular, we use power of 2
884 */
885#define MAX_CIRCULAR_QUEUE_SIZE 4096UL
886#define CQ_MASK (MAX_CIRCULAR_QUEUE_SIZE-1)
887
888/*
889 * The circular_queue and helpers is used to implement the
890 * breadth-first search(BFS)algorithem, by which we can build
891 * the shortest path from the next lock to be acquired to the
892 * previous held lock if there is a circular between them.
893 */
894struct circular_queue {
895 unsigned long element[MAX_CIRCULAR_QUEUE_SIZE];
896 unsigned int front, rear;
897};
898
899static struct circular_queue lock_cq;
900
901unsigned int max_bfs_queue_depth;
902
903static unsigned int lockdep_dependency_gen_id;
904
905static inline void __cq_init(struct circular_queue *cq)
906{
907 cq->front = cq->rear = 0;
908 lockdep_dependency_gen_id++;
909}
910
911static inline int __cq_empty(struct circular_queue *cq)
912{
913 return (cq->front == cq->rear);
914}
915
916static inline int __cq_full(struct circular_queue *cq)
917{
918 return ((cq->rear + 1) & CQ_MASK) == cq->front;
919}
920
921static inline int __cq_enqueue(struct circular_queue *cq, unsigned long elem)
922{
923 if (__cq_full(cq))
924 return -1;
925
926 cq->element[cq->rear] = elem;
927 cq->rear = (cq->rear + 1) & CQ_MASK;
928 return 0;
929}
930
931static inline int __cq_dequeue(struct circular_queue *cq, unsigned long *elem)
932{
933 if (__cq_empty(cq))
934 return -1;
935
936 *elem = cq->element[cq->front];
937 cq->front = (cq->front + 1) & CQ_MASK;
938 return 0;
939}
940
941static inline unsigned int __cq_get_elem_count(struct circular_queue *cq)
942{
943 return (cq->rear - cq->front) & CQ_MASK;
944}
945
946static inline void mark_lock_accessed(struct lock_list *lock,
947 struct lock_list *parent)
948{
949 unsigned long nr;
950
951 nr = lock - list_entries;
952 WARN_ON(nr >= nr_list_entries); /* Out-of-bounds, input fail */
953 lock->parent = parent;
954 lock->class->dep_gen_id = lockdep_dependency_gen_id;
955}
956
957static inline unsigned long lock_accessed(struct lock_list *lock)
958{
959 unsigned long nr;
960
961 nr = lock - list_entries;
962 WARN_ON(nr >= nr_list_entries); /* Out-of-bounds, input fail */
963 return lock->class->dep_gen_id == lockdep_dependency_gen_id;
964}
965
966static inline struct lock_list *get_lock_parent(struct lock_list *child)
967{
968 return child->parent;
969}
970
971static inline int get_lock_depth(struct lock_list *child)
972{
973 int depth = 0;
974 struct lock_list *parent;
975
976 while ((parent = get_lock_parent(child))) {
977 child = parent;
978 depth++;
979 }
980 return depth;
981}
982
983static int __bfs(struct lock_list *source_entry,
984 void *data,
985 int (*match)(struct lock_list *entry, void *data),
986 struct lock_list **target_entry,
987 int forward)
988{
989 struct lock_list *entry;
990 struct list_head *head;
991 struct circular_queue *cq = &lock_cq;
992 int ret = 1;
993
994 if (match(source_entry, data)) {
995 *target_entry = source_entry;
996 ret = 0;
997 goto exit;
998 }
999
1000 if (forward)
1001 head = &source_entry->class->locks_after;
1002 else
1003 head = &source_entry->class->locks_before;
1004
1005 if (list_empty(head))
1006 goto exit;
1007
1008 __cq_init(cq);
1009 __cq_enqueue(cq, (unsigned long)source_entry);
1010
1011 while (!__cq_empty(cq)) {
1012 struct lock_list *lock;
1013
1014 __cq_dequeue(cq, (unsigned long *)&lock);
1015
1016 if (!lock->class) {
1017 ret = -2;
1018 goto exit;
1019 }
1020
1021 if (forward)
1022 head = &lock->class->locks_after;
1023 else
1024 head = &lock->class->locks_before;
1025
1026 list_for_each_entry(entry, head, entry) {
1027 if (!lock_accessed(entry)) {
1028 unsigned int cq_depth;
1029 mark_lock_accessed(entry, lock);
1030 if (match(entry, data)) {
1031 *target_entry = entry;
1032 ret = 0;
1033 goto exit;
1034 }
1035
1036 if (__cq_enqueue(cq, (unsigned long)entry)) {
1037 ret = -1;
1038 goto exit;
1039 }
1040 cq_depth = __cq_get_elem_count(cq);
1041 if (max_bfs_queue_depth < cq_depth)
1042 max_bfs_queue_depth = cq_depth;
1043 }
1044 }
1045 }
1046exit:
1047 return ret;
1048}
1049
1050static inline int __bfs_forwards(struct lock_list *src_entry,
1051 void *data,
1052 int (*match)(struct lock_list *entry, void *data),
1053 struct lock_list **target_entry)
1054{
1055 return __bfs(src_entry, data, match, target_entry, 1);
1056
1057}
1058
1059static inline int __bfs_backwards(struct lock_list *src_entry,
1060 void *data,
1061 int (*match)(struct lock_list *entry, void *data),
1062 struct lock_list **target_entry)
1063{
1064 return __bfs(src_entry, data, match, target_entry, 0);
1065
1066}
1067
1068/*
1069 * Recursive, forwards-direction lock-dependency checking, used for
1070 * both noncyclic checking and for hardirq-unsafe/softirq-unsafe
1071 * checking.
1072 */
1073
1074/*
1075 * Print a dependency chain entry (this is only done when a deadlock
1076 * has been detected):
1077 */
1078static noinline int
1079print_circular_bug_entry(struct lock_list *target, int depth)
1080{
1081 if (debug_locks_silent)
1082 return 0;
1083 printk("\n-> #%u", depth);
1084 print_lock_name(target->class);
1085 printk(":\n");
1086 print_stack_trace(&target->trace, 6);
1087
1088 return 0;
1089}
1090
1091static void
1092print_circular_lock_scenario(struct held_lock *src,
1093 struct held_lock *tgt,
1094 struct lock_list *prt)
1095{
1096 struct lock_class *source = hlock_class(src);
1097 struct lock_class *target = hlock_class(tgt);
1098 struct lock_class *parent = prt->class;
1099
1100 /*
1101 * A direct locking problem where unsafe_class lock is taken
1102 * directly by safe_class lock, then all we need to show
1103 * is the deadlock scenario, as it is obvious that the
1104 * unsafe lock is taken under the safe lock.
1105 *
1106 * But if there is a chain instead, where the safe lock takes
1107 * an intermediate lock (middle_class) where this lock is
1108 * not the same as the safe lock, then the lock chain is
1109 * used to describe the problem. Otherwise we would need
1110 * to show a different CPU case for each link in the chain
1111 * from the safe_class lock to the unsafe_class lock.
1112 */
1113 if (parent != source) {
1114 printk("Chain exists of:\n ");
1115 __print_lock_name(source);
1116 printk(" --> ");
1117 __print_lock_name(parent);
1118 printk(" --> ");
1119 __print_lock_name(target);
1120 printk("\n\n");
1121 }
1122
1123 printk(" Possible unsafe locking scenario:\n\n");
1124 printk(" CPU0 CPU1\n");
1125 printk(" ---- ----\n");
1126 printk(" lock(");
1127 __print_lock_name(target);
1128 printk(");\n");
1129 printk(" lock(");
1130 __print_lock_name(parent);
1131 printk(");\n");
1132 printk(" lock(");
1133 __print_lock_name(target);
1134 printk(");\n");
1135 printk(" lock(");
1136 __print_lock_name(source);
1137 printk(");\n");
1138 printk("\n *** DEADLOCK ***\n\n");
1139}
1140
1141/*
1142 * When a circular dependency is detected, print the
1143 * header first:
1144 */
1145static noinline int
1146print_circular_bug_header(struct lock_list *entry, unsigned int depth,
1147 struct held_lock *check_src,
1148 struct held_lock *check_tgt)
1149{
1150 struct task_struct *curr = current;
1151
1152 if (debug_locks_silent)
1153 return 0;
1154
1155 printk("\n");
1156 printk("======================================================\n");
1157 printk("[ INFO: possible circular locking dependency detected ]\n");
1158 print_kernel_ident();
1159 printk("-------------------------------------------------------\n");
1160 printk("%s/%d is trying to acquire lock:\n",
1161 curr->comm, task_pid_nr(curr));
1162 print_lock(check_src);
1163 printk("\nbut task is already holding lock:\n");
1164 print_lock(check_tgt);
1165 printk("\nwhich lock already depends on the new lock.\n\n");
1166 printk("\nthe existing dependency chain (in reverse order) is:\n");
1167
1168 print_circular_bug_entry(entry, depth);
1169
1170 return 0;
1171}
1172
1173static inline int class_equal(struct lock_list *entry, void *data)
1174{
1175 return entry->class == data;
1176}
1177
1178static noinline int print_circular_bug(struct lock_list *this,
1179 struct lock_list *target,
1180 struct held_lock *check_src,
1181 struct held_lock *check_tgt)
1182{
1183 struct task_struct *curr = current;
1184 struct lock_list *parent;
1185 struct lock_list *first_parent;
1186 int depth;
1187
1188 if (!debug_locks_off_graph_unlock() || debug_locks_silent)
1189 return 0;
1190
1191 if (!save_trace(&this->trace))
1192 return 0;
1193
1194 depth = get_lock_depth(target);
1195
1196 print_circular_bug_header(target, depth, check_src, check_tgt);
1197
1198 parent = get_lock_parent(target);
1199 first_parent = parent;
1200
1201 while (parent) {
1202 print_circular_bug_entry(parent, --depth);
1203 parent = get_lock_parent(parent);
1204 }
1205
1206 printk("\nother info that might help us debug this:\n\n");
1207 print_circular_lock_scenario(check_src, check_tgt,
1208 first_parent);
1209
1210 lockdep_print_held_locks(curr);
1211
1212 printk("\nstack backtrace:\n");
1213 dump_stack();
1214
1215 return 0;
1216}
1217
1218static noinline int print_bfs_bug(int ret)
1219{
1220 if (!debug_locks_off_graph_unlock())
1221 return 0;
1222
1223 /*
1224 * Breadth-first-search failed, graph got corrupted?
1225 */
1226 WARN(1, "lockdep bfs error:%d\n", ret);
1227
1228 return 0;
1229}
1230
1231static int noop_count(struct lock_list *entry, void *data)
1232{
1233 (*(unsigned long *)data)++;
1234 return 0;
1235}
1236
1237static unsigned long __lockdep_count_forward_deps(struct lock_list *this)
1238{
1239 unsigned long count = 0;
1240 struct lock_list *uninitialized_var(target_entry);
1241
1242 __bfs_forwards(this, (void *)&count, noop_count, &target_entry);
1243
1244 return count;
1245}
1246unsigned long lockdep_count_forward_deps(struct lock_class *class)
1247{
1248 unsigned long ret, flags;
1249 struct lock_list this;
1250
1251 this.parent = NULL;
1252 this.class = class;
1253
1254 local_irq_save(flags);
1255 arch_spin_lock(&lockdep_lock);
1256 ret = __lockdep_count_forward_deps(&this);
1257 arch_spin_unlock(&lockdep_lock);
1258 local_irq_restore(flags);
1259
1260 return ret;
1261}
1262
1263static unsigned long __lockdep_count_backward_deps(struct lock_list *this)
1264{
1265 unsigned long count = 0;
1266 struct lock_list *uninitialized_var(target_entry);
1267
1268 __bfs_backwards(this, (void *)&count, noop_count, &target_entry);
1269
1270 return count;
1271}
1272
1273unsigned long lockdep_count_backward_deps(struct lock_class *class)
1274{
1275 unsigned long ret, flags;
1276 struct lock_list this;
1277
1278 this.parent = NULL;
1279 this.class = class;
1280
1281 local_irq_save(flags);
1282 arch_spin_lock(&lockdep_lock);
1283 ret = __lockdep_count_backward_deps(&this);
1284 arch_spin_unlock(&lockdep_lock);
1285 local_irq_restore(flags);
1286
1287 return ret;
1288}
1289
1290/*
1291 * Prove that the dependency graph starting at <entry> can not
1292 * lead to <target>. Print an error and return 0 if it does.
1293 */
1294static noinline int
1295check_noncircular(struct lock_list *root, struct lock_class *target,
1296 struct lock_list **target_entry)
1297{
1298 int result;
1299
1300 debug_atomic_inc(nr_cyclic_checks);
1301
1302 result = __bfs_forwards(root, target, class_equal, target_entry);
1303
1304 return result;
1305}
1306
1307#if defined(CONFIG_TRACE_IRQFLAGS) && defined(CONFIG_PROVE_LOCKING)
1308/*
1309 * Forwards and backwards subgraph searching, for the purposes of
1310 * proving that two subgraphs can be connected by a new dependency
1311 * without creating any illegal irq-safe -> irq-unsafe lock dependency.
1312 */
1313
1314static inline int usage_match(struct lock_list *entry, void *bit)
1315{
1316 return entry->class->usage_mask & (1 << (enum lock_usage_bit)bit);
1317}
1318
1319
1320
1321/*
1322 * Find a node in the forwards-direction dependency sub-graph starting
1323 * at @root->class that matches @bit.
1324 *
1325 * Return 0 if such a node exists in the subgraph, and put that node
1326 * into *@target_entry.
1327 *
1328 * Return 1 otherwise and keep *@target_entry unchanged.
1329 * Return <0 on error.
1330 */
1331static int
1332find_usage_forwards(struct lock_list *root, enum lock_usage_bit bit,
1333 struct lock_list **target_entry)
1334{
1335 int result;
1336
1337 debug_atomic_inc(nr_find_usage_forwards_checks);
1338
1339 result = __bfs_forwards(root, (void *)bit, usage_match, target_entry);
1340
1341 return result;
1342}
1343
1344/*
1345 * Find a node in the backwards-direction dependency sub-graph starting
1346 * at @root->class that matches @bit.
1347 *
1348 * Return 0 if such a node exists in the subgraph, and put that node
1349 * into *@target_entry.
1350 *
1351 * Return 1 otherwise and keep *@target_entry unchanged.
1352 * Return <0 on error.
1353 */
1354static int
1355find_usage_backwards(struct lock_list *root, enum lock_usage_bit bit,
1356 struct lock_list **target_entry)
1357{
1358 int result;
1359
1360 debug_atomic_inc(nr_find_usage_backwards_checks);
1361
1362 result = __bfs_backwards(root, (void *)bit, usage_match, target_entry);
1363
1364 return result;
1365}
1366
1367static void print_lock_class_header(struct lock_class *class, int depth)
1368{
1369 int bit;
1370
1371 printk("%*s->", depth, "");
1372 print_lock_name(class);
1373 printk(" ops: %lu", class->ops);
1374 printk(" {\n");
1375
1376 for (bit = 0; bit < LOCK_USAGE_STATES; bit++) {
1377 if (class->usage_mask & (1 << bit)) {
1378 int len = depth;
1379
1380 len += printk("%*s %s", depth, "", usage_str[bit]);
1381 len += printk(" at:\n");
1382 print_stack_trace(class->usage_traces + bit, len);
1383 }
1384 }
1385 printk("%*s }\n", depth, "");
1386
1387 printk("%*s ... key at: ",depth,"");
1388 print_ip_sym((unsigned long)class->key);
1389}
1390
1391/*
1392 * printk the shortest lock dependencies from @start to @end in reverse order:
1393 */
1394static void __used
1395print_shortest_lock_dependencies(struct lock_list *leaf,
1396 struct lock_list *root)
1397{
1398 struct lock_list *entry = leaf;
1399 int depth;
1400
1401 /*compute depth from generated tree by BFS*/
1402 depth = get_lock_depth(leaf);
1403
1404 do {
1405 print_lock_class_header(entry->class, depth);
1406 printk("%*s ... acquired at:\n", depth, "");
1407 print_stack_trace(&entry->trace, 2);
1408 printk("\n");
1409
1410 if (depth == 0 && (entry != root)) {
1411 printk("lockdep:%s bad path found in chain graph\n", __func__);
1412 break;
1413 }
1414
1415 entry = get_lock_parent(entry);
1416 depth--;
1417 } while (entry && (depth >= 0));
1418
1419 return;
1420}
1421
1422static void
1423print_irq_lock_scenario(struct lock_list *safe_entry,
1424 struct lock_list *unsafe_entry,
1425 struct lock_class *prev_class,
1426 struct lock_class *next_class)
1427{
1428 struct lock_class *safe_class = safe_entry->class;
1429 struct lock_class *unsafe_class = unsafe_entry->class;
1430 struct lock_class *middle_class = prev_class;
1431
1432 if (middle_class == safe_class)
1433 middle_class = next_class;
1434
1435 /*
1436 * A direct locking problem where unsafe_class lock is taken
1437 * directly by safe_class lock, then all we need to show
1438 * is the deadlock scenario, as it is obvious that the
1439 * unsafe lock is taken under the safe lock.
1440 *
1441 * But if there is a chain instead, where the safe lock takes
1442 * an intermediate lock (middle_class) where this lock is
1443 * not the same as the safe lock, then the lock chain is
1444 * used to describe the problem. Otherwise we would need
1445 * to show a different CPU case for each link in the chain
1446 * from the safe_class lock to the unsafe_class lock.
1447 */
1448 if (middle_class != unsafe_class) {
1449 printk("Chain exists of:\n ");
1450 __print_lock_name(safe_class);
1451 printk(" --> ");
1452 __print_lock_name(middle_class);
1453 printk(" --> ");
1454 __print_lock_name(unsafe_class);
1455 printk("\n\n");
1456 }
1457
1458 printk(" Possible interrupt unsafe locking scenario:\n\n");
1459 printk(" CPU0 CPU1\n");
1460 printk(" ---- ----\n");
1461 printk(" lock(");
1462 __print_lock_name(unsafe_class);
1463 printk(");\n");
1464 printk(" local_irq_disable();\n");
1465 printk(" lock(");
1466 __print_lock_name(safe_class);
1467 printk(");\n");
1468 printk(" lock(");
1469 __print_lock_name(middle_class);
1470 printk(");\n");
1471 printk(" <Interrupt>\n");
1472 printk(" lock(");
1473 __print_lock_name(safe_class);
1474 printk(");\n");
1475 printk("\n *** DEADLOCK ***\n\n");
1476}
1477
1478static int
1479print_bad_irq_dependency(struct task_struct *curr,
1480 struct lock_list *prev_root,
1481 struct lock_list *next_root,
1482 struct lock_list *backwards_entry,
1483 struct lock_list *forwards_entry,
1484 struct held_lock *prev,
1485 struct held_lock *next,
1486 enum lock_usage_bit bit1,
1487 enum lock_usage_bit bit2,
1488 const char *irqclass)
1489{
1490 if (!debug_locks_off_graph_unlock() || debug_locks_silent)
1491 return 0;
1492
1493 printk("\n");
1494 printk("======================================================\n");
1495 printk("[ INFO: %s-safe -> %s-unsafe lock order detected ]\n",
1496 irqclass, irqclass);
1497 print_kernel_ident();
1498 printk("------------------------------------------------------\n");
1499 printk("%s/%d [HC%u[%lu]:SC%u[%lu]:HE%u:SE%u] is trying to acquire:\n",
1500 curr->comm, task_pid_nr(curr),
1501 curr->hardirq_context, hardirq_count() >> HARDIRQ_SHIFT,
1502 curr->softirq_context, softirq_count() >> SOFTIRQ_SHIFT,
1503 curr->hardirqs_enabled,
1504 curr->softirqs_enabled);
1505 print_lock(next);
1506
1507 printk("\nand this task is already holding:\n");
1508 print_lock(prev);
1509 printk("which would create a new lock dependency:\n");
1510 print_lock_name(hlock_class(prev));
1511 printk(" ->");
1512 print_lock_name(hlock_class(next));
1513 printk("\n");
1514
1515 printk("\nbut this new dependency connects a %s-irq-safe lock:\n",
1516 irqclass);
1517 print_lock_name(backwards_entry->class);
1518 printk("\n... which became %s-irq-safe at:\n", irqclass);
1519
1520 print_stack_trace(backwards_entry->class->usage_traces + bit1, 1);
1521
1522 printk("\nto a %s-irq-unsafe lock:\n", irqclass);
1523 print_lock_name(forwards_entry->class);
1524 printk("\n... which became %s-irq-unsafe at:\n", irqclass);
1525 printk("...");
1526
1527 print_stack_trace(forwards_entry->class->usage_traces + bit2, 1);
1528
1529 printk("\nother info that might help us debug this:\n\n");
1530 print_irq_lock_scenario(backwards_entry, forwards_entry,
1531 hlock_class(prev), hlock_class(next));
1532
1533 lockdep_print_held_locks(curr);
1534
1535 printk("\nthe dependencies between %s-irq-safe lock", irqclass);
1536 printk(" and the holding lock:\n");
1537 if (!save_trace(&prev_root->trace))
1538 return 0;
1539 print_shortest_lock_dependencies(backwards_entry, prev_root);
1540
1541 printk("\nthe dependencies between the lock to be acquired");
1542 printk(" and %s-irq-unsafe lock:\n", irqclass);
1543 if (!save_trace(&next_root->trace))
1544 return 0;
1545 print_shortest_lock_dependencies(forwards_entry, next_root);
1546
1547 printk("\nstack backtrace:\n");
1548 dump_stack();
1549
1550 return 0;
1551}
1552
1553static int
1554check_usage(struct task_struct *curr, struct held_lock *prev,
1555 struct held_lock *next, enum lock_usage_bit bit_backwards,
1556 enum lock_usage_bit bit_forwards, const char *irqclass)
1557{
1558 int ret;
1559 struct lock_list this, that;
1560 struct lock_list *uninitialized_var(target_entry);
1561 struct lock_list *uninitialized_var(target_entry1);
1562
1563 this.parent = NULL;
1564
1565 this.class = hlock_class(prev);
1566 ret = find_usage_backwards(&this, bit_backwards, &target_entry);
1567 if (ret < 0)
1568 return print_bfs_bug(ret);
1569 if (ret == 1)
1570 return ret;
1571
1572 that.parent = NULL;
1573 that.class = hlock_class(next);
1574 ret = find_usage_forwards(&that, bit_forwards, &target_entry1);
1575 if (ret < 0)
1576 return print_bfs_bug(ret);
1577 if (ret == 1)
1578 return ret;
1579
1580 return print_bad_irq_dependency(curr, &this, &that,
1581 target_entry, target_entry1,
1582 prev, next,
1583 bit_backwards, bit_forwards, irqclass);
1584}
1585
1586static const char *state_names[] = {
1587#define LOCKDEP_STATE(__STATE) \
1588 __stringify(__STATE),
1589#include "lockdep_states.h"
1590#undef LOCKDEP_STATE
1591};
1592
1593static const char *state_rnames[] = {
1594#define LOCKDEP_STATE(__STATE) \
1595 __stringify(__STATE)"-READ",
1596#include "lockdep_states.h"
1597#undef LOCKDEP_STATE
1598};
1599
1600static inline const char *state_name(enum lock_usage_bit bit)
1601{
1602 return (bit & 1) ? state_rnames[bit >> 2] : state_names[bit >> 2];
1603}
1604
1605static int exclusive_bit(int new_bit)
1606{
1607 /*
1608 * USED_IN
1609 * USED_IN_READ
1610 * ENABLED
1611 * ENABLED_READ
1612 *
1613 * bit 0 - write/read
1614 * bit 1 - used_in/enabled
1615 * bit 2+ state
1616 */
1617
1618 int state = new_bit & ~3;
1619 int dir = new_bit & 2;
1620
1621 /*
1622 * keep state, bit flip the direction and strip read.
1623 */
1624 return state | (dir ^ 2);
1625}
1626
1627static int check_irq_usage(struct task_struct *curr, struct held_lock *prev,
1628 struct held_lock *next, enum lock_usage_bit bit)
1629{
1630 /*
1631 * Prove that the new dependency does not connect a hardirq-safe
1632 * lock with a hardirq-unsafe lock - to achieve this we search
1633 * the backwards-subgraph starting at <prev>, and the
1634 * forwards-subgraph starting at <next>:
1635 */
1636 if (!check_usage(curr, prev, next, bit,
1637 exclusive_bit(bit), state_name(bit)))
1638 return 0;
1639
1640 bit++; /* _READ */
1641
1642 /*
1643 * Prove that the new dependency does not connect a hardirq-safe-read
1644 * lock with a hardirq-unsafe lock - to achieve this we search
1645 * the backwards-subgraph starting at <prev>, and the
1646 * forwards-subgraph starting at <next>:
1647 */
1648 if (!check_usage(curr, prev, next, bit,
1649 exclusive_bit(bit), state_name(bit)))
1650 return 0;
1651
1652 return 1;
1653}
1654
1655static int
1656check_prev_add_irq(struct task_struct *curr, struct held_lock *prev,
1657 struct held_lock *next)
1658{
1659#define LOCKDEP_STATE(__STATE) \
1660 if (!check_irq_usage(curr, prev, next, LOCK_USED_IN_##__STATE)) \
1661 return 0;
1662#include "lockdep_states.h"
1663#undef LOCKDEP_STATE
1664
1665 return 1;
1666}
1667
1668static void inc_chains(void)
1669{
1670 if (current->hardirq_context)
1671 nr_hardirq_chains++;
1672 else {
1673 if (current->softirq_context)
1674 nr_softirq_chains++;
1675 else
1676 nr_process_chains++;
1677 }
1678}
1679
1680#else
1681
1682static inline int
1683check_prev_add_irq(struct task_struct *curr, struct held_lock *prev,
1684 struct held_lock *next)
1685{
1686 return 1;
1687}
1688
1689static inline void inc_chains(void)
1690{
1691 nr_process_chains++;
1692}
1693
1694#endif
1695
1696static void
1697print_deadlock_scenario(struct held_lock *nxt,
1698 struct held_lock *prv)
1699{
1700 struct lock_class *next = hlock_class(nxt);
1701 struct lock_class *prev = hlock_class(prv);
1702
1703 printk(" Possible unsafe locking scenario:\n\n");
1704 printk(" CPU0\n");
1705 printk(" ----\n");
1706 printk(" lock(");
1707 __print_lock_name(prev);
1708 printk(");\n");
1709 printk(" lock(");
1710 __print_lock_name(next);
1711 printk(");\n");
1712 printk("\n *** DEADLOCK ***\n\n");
1713 printk(" May be due to missing lock nesting notation\n\n");
1714}
1715
1716static int
1717print_deadlock_bug(struct task_struct *curr, struct held_lock *prev,
1718 struct held_lock *next)
1719{
1720 if (!debug_locks_off_graph_unlock() || debug_locks_silent)
1721 return 0;
1722
1723 printk("\n");
1724 printk("=============================================\n");
1725 printk("[ INFO: possible recursive locking detected ]\n");
1726 print_kernel_ident();
1727 printk("---------------------------------------------\n");
1728 printk("%s/%d is trying to acquire lock:\n",
1729 curr->comm, task_pid_nr(curr));
1730 print_lock(next);
1731 printk("\nbut task is already holding lock:\n");
1732 print_lock(prev);
1733
1734 printk("\nother info that might help us debug this:\n");
1735 print_deadlock_scenario(next, prev);
1736 lockdep_print_held_locks(curr);
1737
1738 printk("\nstack backtrace:\n");
1739 dump_stack();
1740
1741 return 0;
1742}
1743
1744/*
1745 * Check whether we are holding such a class already.
1746 *
1747 * (Note that this has to be done separately, because the graph cannot
1748 * detect such classes of deadlocks.)
1749 *
1750 * Returns: 0 on deadlock detected, 1 on OK, 2 on recursive read
1751 */
1752static int
1753check_deadlock(struct task_struct *curr, struct held_lock *next,
1754 struct lockdep_map *next_instance, int read)
1755{
1756 struct held_lock *prev;
1757 struct held_lock *nest = NULL;
1758 int i;
1759
1760 for (i = 0; i < curr->lockdep_depth; i++) {
1761 prev = curr->held_locks + i;
1762
1763 if (prev->instance == next->nest_lock)
1764 nest = prev;
1765
1766 if (hlock_class(prev) != hlock_class(next))
1767 continue;
1768
1769 /*
1770 * Allow read-after-read recursion of the same
1771 * lock class (i.e. read_lock(lock)+read_lock(lock)):
1772 */
1773 if ((read == 2) && prev->read)
1774 return 2;
1775
1776 /*
1777 * We're holding the nest_lock, which serializes this lock's
1778 * nesting behaviour.
1779 */
1780 if (nest)
1781 return 2;
1782
1783 return print_deadlock_bug(curr, prev, next);
1784 }
1785 return 1;
1786}
1787
1788/*
1789 * There was a chain-cache miss, and we are about to add a new dependency
1790 * to a previous lock. We recursively validate the following rules:
1791 *
1792 * - would the adding of the <prev> -> <next> dependency create a
1793 * circular dependency in the graph? [== circular deadlock]
1794 *
1795 * - does the new prev->next dependency connect any hardirq-safe lock
1796 * (in the full backwards-subgraph starting at <prev>) with any
1797 * hardirq-unsafe lock (in the full forwards-subgraph starting at
1798 * <next>)? [== illegal lock inversion with hardirq contexts]
1799 *
1800 * - does the new prev->next dependency connect any softirq-safe lock
1801 * (in the full backwards-subgraph starting at <prev>) with any
1802 * softirq-unsafe lock (in the full forwards-subgraph starting at
1803 * <next>)? [== illegal lock inversion with softirq contexts]
1804 *
1805 * any of these scenarios could lead to a deadlock.
1806 *
1807 * Then if all the validations pass, we add the forwards and backwards
1808 * dependency.
1809 */
1810static int
1811check_prev_add(struct task_struct *curr, struct held_lock *prev,
1812 struct held_lock *next, int distance, int trylock_loop)
1813{
1814 struct lock_list *entry;
1815 int ret;
1816 struct lock_list this;
1817 struct lock_list *uninitialized_var(target_entry);
1818 /*
1819 * Static variable, serialized by the graph_lock().
1820 *
1821 * We use this static variable to save the stack trace in case
1822 * we call into this function multiple times due to encountering
1823 * trylocks in the held lock stack.
1824 */
1825 static struct stack_trace trace;
1826
1827 /*
1828 * Prove that the new <prev> -> <next> dependency would not
1829 * create a circular dependency in the graph. (We do this by
1830 * forward-recursing into the graph starting at <next>, and
1831 * checking whether we can reach <prev>.)
1832 *
1833 * We are using global variables to control the recursion, to
1834 * keep the stackframe size of the recursive functions low:
1835 */
1836 this.class = hlock_class(next);
1837 this.parent = NULL;
1838 ret = check_noncircular(&this, hlock_class(prev), &target_entry);
1839 if (unlikely(!ret))
1840 return print_circular_bug(&this, target_entry, next, prev);
1841 else if (unlikely(ret < 0))
1842 return print_bfs_bug(ret);
1843
1844 if (!check_prev_add_irq(curr, prev, next))
1845 return 0;
1846
1847 /*
1848 * For recursive read-locks we do all the dependency checks,
1849 * but we dont store read-triggered dependencies (only
1850 * write-triggered dependencies). This ensures that only the
1851 * write-side dependencies matter, and that if for example a
1852 * write-lock never takes any other locks, then the reads are
1853 * equivalent to a NOP.
1854 */
1855 if (next->read == 2 || prev->read == 2)
1856 return 1;
1857 /*
1858 * Is the <prev> -> <next> dependency already present?
1859 *
1860 * (this may occur even though this is a new chain: consider
1861 * e.g. the L1 -> L2 -> L3 -> L4 and the L5 -> L1 -> L2 -> L3
1862 * chains - the second one will be new, but L1 already has
1863 * L2 added to its dependency list, due to the first chain.)
1864 */
1865 list_for_each_entry(entry, &hlock_class(prev)->locks_after, entry) {
1866 if (entry->class == hlock_class(next)) {
1867 if (distance == 1)
1868 entry->distance = 1;
1869 return 2;
1870 }
1871 }
1872
1873 if (!trylock_loop && !save_trace(&trace))
1874 return 0;
1875
1876 /*
1877 * Ok, all validations passed, add the new lock
1878 * to the previous lock's dependency list:
1879 */
1880 ret = add_lock_to_list(hlock_class(prev), hlock_class(next),
1881 &hlock_class(prev)->locks_after,
1882 next->acquire_ip, distance, &trace);
1883
1884 if (!ret)
1885 return 0;
1886
1887 ret = add_lock_to_list(hlock_class(next), hlock_class(prev),
1888 &hlock_class(next)->locks_before,
1889 next->acquire_ip, distance, &trace);
1890 if (!ret)
1891 return 0;
1892
1893 /*
1894 * Debugging printouts:
1895 */
1896 if (verbose(hlock_class(prev)) || verbose(hlock_class(next))) {
1897 graph_unlock();
1898 printk("\n new dependency: ");
1899 print_lock_name(hlock_class(prev));
1900 printk(" => ");
1901 print_lock_name(hlock_class(next));
1902 printk("\n");
1903 dump_stack();
1904 return graph_lock();
1905 }
1906 return 1;
1907}
1908
1909/*
1910 * Add the dependency to all directly-previous locks that are 'relevant'.
1911 * The ones that are relevant are (in increasing distance from curr):
1912 * all consecutive trylock entries and the final non-trylock entry - or
1913 * the end of this context's lock-chain - whichever comes first.
1914 */
1915static int
1916check_prevs_add(struct task_struct *curr, struct held_lock *next)
1917{
1918 int depth = curr->lockdep_depth;
1919 int trylock_loop = 0;
1920 struct held_lock *hlock;
1921
1922 /*
1923 * Debugging checks.
1924 *
1925 * Depth must not be zero for a non-head lock:
1926 */
1927 if (!depth)
1928 goto out_bug;
1929 /*
1930 * At least two relevant locks must exist for this
1931 * to be a head:
1932 */
1933 if (curr->held_locks[depth].irq_context !=
1934 curr->held_locks[depth-1].irq_context)
1935 goto out_bug;
1936
1937 for (;;) {
1938 int distance = curr->lockdep_depth - depth + 1;
1939 hlock = curr->held_locks + depth - 1;
1940 /*
1941 * Only non-recursive-read entries get new dependencies
1942 * added:
1943 */
1944 if (hlock->read != 2 && hlock->check) {
1945 if (!check_prev_add(curr, hlock, next,
1946 distance, trylock_loop))
1947 return 0;
1948 /*
1949 * Stop after the first non-trylock entry,
1950 * as non-trylock entries have added their
1951 * own direct dependencies already, so this
1952 * lock is connected to them indirectly:
1953 */
1954 if (!hlock->trylock)
1955 break;
1956 }
1957 depth--;
1958 /*
1959 * End of lock-stack?
1960 */
1961 if (!depth)
1962 break;
1963 /*
1964 * Stop the search if we cross into another context:
1965 */
1966 if (curr->held_locks[depth].irq_context !=
1967 curr->held_locks[depth-1].irq_context)
1968 break;
1969 trylock_loop = 1;
1970 }
1971 return 1;
1972out_bug:
1973 if (!debug_locks_off_graph_unlock())
1974 return 0;
1975
1976 /*
1977 * Clearly we all shouldn't be here, but since we made it we
1978 * can reliable say we messed up our state. See the above two
1979 * gotos for reasons why we could possibly end up here.
1980 */
1981 WARN_ON(1);
1982
1983 return 0;
1984}
1985
1986unsigned long nr_lock_chains;
1987struct lock_chain lock_chains[MAX_LOCKDEP_CHAINS];
1988int nr_chain_hlocks;
1989static u16 chain_hlocks[MAX_LOCKDEP_CHAIN_HLOCKS];
1990
1991struct lock_class *lock_chain_get_class(struct lock_chain *chain, int i)
1992{
1993 return lock_classes + chain_hlocks[chain->base + i];
1994}
1995
1996/*
1997 * Look up a dependency chain. If the key is not present yet then
1998 * add it and return 1 - in this case the new dependency chain is
1999 * validated. If the key is already hashed, return 0.
2000 * (On return with 1 graph_lock is held.)
2001 */
2002static inline int lookup_chain_cache(struct task_struct *curr,
2003 struct held_lock *hlock,
2004 u64 chain_key)
2005{
2006 struct lock_class *class = hlock_class(hlock);
2007 struct list_head *hash_head = chainhashentry(chain_key);
2008 struct lock_chain *chain;
2009 struct held_lock *hlock_curr;
2010 int i, j;
2011
2012 /*
2013 * We might need to take the graph lock, ensure we've got IRQs
2014 * disabled to make this an IRQ-safe lock.. for recursion reasons
2015 * lockdep won't complain about its own locking errors.
2016 */
2017 if (DEBUG_LOCKS_WARN_ON(!irqs_disabled()))
2018 return 0;
2019 /*
2020 * We can walk it lock-free, because entries only get added
2021 * to the hash:
2022 */
2023 list_for_each_entry(chain, hash_head, entry) {
2024 if (chain->chain_key == chain_key) {
2025cache_hit:
2026 debug_atomic_inc(chain_lookup_hits);
2027 if (very_verbose(class))
2028 printk("\nhash chain already cached, key: "
2029 "%016Lx tail class: [%p] %s\n",
2030 (unsigned long long)chain_key,
2031 class->key, class->name);
2032 return 0;
2033 }
2034 }
2035 if (very_verbose(class))
2036 printk("\nnew hash chain, key: %016Lx tail class: [%p] %s\n",
2037 (unsigned long long)chain_key, class->key, class->name);
2038 /*
2039 * Allocate a new chain entry from the static array, and add
2040 * it to the hash:
2041 */
2042 if (!graph_lock())
2043 return 0;
2044 /*
2045 * We have to walk the chain again locked - to avoid duplicates:
2046 */
2047 list_for_each_entry(chain, hash_head, entry) {
2048 if (chain->chain_key == chain_key) {
2049 graph_unlock();
2050 goto cache_hit;
2051 }
2052 }
2053 if (unlikely(nr_lock_chains >= MAX_LOCKDEP_CHAINS)) {
2054 if (!debug_locks_off_graph_unlock())
2055 return 0;
2056
2057 print_lockdep_off("BUG: MAX_LOCKDEP_CHAINS too low!");
2058 dump_stack();
2059 return 0;
2060 }
2061 chain = lock_chains + nr_lock_chains++;
2062 chain->chain_key = chain_key;
2063 chain->irq_context = hlock->irq_context;
2064 /* Find the first held_lock of current chain */
2065 for (i = curr->lockdep_depth - 1; i >= 0; i--) {
2066 hlock_curr = curr->held_locks + i;
2067 if (hlock_curr->irq_context != hlock->irq_context)
2068 break;
2069 }
2070 i++;
2071 chain->depth = curr->lockdep_depth + 1 - i;
2072 if (likely(nr_chain_hlocks + chain->depth <= MAX_LOCKDEP_CHAIN_HLOCKS)) {
2073 chain->base = nr_chain_hlocks;
2074 nr_chain_hlocks += chain->depth;
2075 for (j = 0; j < chain->depth - 1; j++, i++) {
2076 int lock_id = curr->held_locks[i].class_idx - 1;
2077 chain_hlocks[chain->base + j] = lock_id;
2078 }
2079 chain_hlocks[chain->base + j] = class - lock_classes;
2080 }
2081 list_add_tail_rcu(&chain->entry, hash_head);
2082 debug_atomic_inc(chain_lookup_misses);
2083 inc_chains();
2084
2085 return 1;
2086}
2087
2088static int validate_chain(struct task_struct *curr, struct lockdep_map *lock,
2089 struct held_lock *hlock, int chain_head, u64 chain_key)
2090{
2091 /*
2092 * Trylock needs to maintain the stack of held locks, but it
2093 * does not add new dependencies, because trylock can be done
2094 * in any order.
2095 *
2096 * We look up the chain_key and do the O(N^2) check and update of
2097 * the dependencies only if this is a new dependency chain.
2098 * (If lookup_chain_cache() returns with 1 it acquires
2099 * graph_lock for us)
2100 */
2101 if (!hlock->trylock && hlock->check &&
2102 lookup_chain_cache(curr, hlock, chain_key)) {
2103 /*
2104 * Check whether last held lock:
2105 *
2106 * - is irq-safe, if this lock is irq-unsafe
2107 * - is softirq-safe, if this lock is hardirq-unsafe
2108 *
2109 * And check whether the new lock's dependency graph
2110 * could lead back to the previous lock.
2111 *
2112 * any of these scenarios could lead to a deadlock. If
2113 * All validations
2114 */
2115 int ret = check_deadlock(curr, hlock, lock, hlock->read);
2116
2117 if (!ret)
2118 return 0;
2119 /*
2120 * Mark recursive read, as we jump over it when
2121 * building dependencies (just like we jump over
2122 * trylock entries):
2123 */
2124 if (ret == 2)
2125 hlock->read = 2;
2126 /*
2127 * Add dependency only if this lock is not the head
2128 * of the chain, and if it's not a secondary read-lock:
2129 */
2130 if (!chain_head && ret != 2)
2131 if (!check_prevs_add(curr, hlock))
2132 return 0;
2133 graph_unlock();
2134 } else
2135 /* after lookup_chain_cache(): */
2136 if (unlikely(!debug_locks))
2137 return 0;
2138
2139 return 1;
2140}
2141#else
2142static inline int validate_chain(struct task_struct *curr,
2143 struct lockdep_map *lock, struct held_lock *hlock,
2144 int chain_head, u64 chain_key)
2145{
2146 return 1;
2147}
2148#endif
2149
2150/*
2151 * We are building curr_chain_key incrementally, so double-check
2152 * it from scratch, to make sure that it's done correctly:
2153 */
2154static void check_chain_key(struct task_struct *curr)
2155{
2156#ifdef CONFIG_DEBUG_LOCKDEP
2157 struct held_lock *hlock, *prev_hlock = NULL;
2158 unsigned int i, id;
2159 u64 chain_key = 0;
2160
2161 for (i = 0; i < curr->lockdep_depth; i++) {
2162 hlock = curr->held_locks + i;
2163 if (chain_key != hlock->prev_chain_key) {
2164 debug_locks_off();
2165 /*
2166 * We got mighty confused, our chain keys don't match
2167 * with what we expect, someone trample on our task state?
2168 */
2169 WARN(1, "hm#1, depth: %u [%u], %016Lx != %016Lx\n",
2170 curr->lockdep_depth, i,
2171 (unsigned long long)chain_key,
2172 (unsigned long long)hlock->prev_chain_key);
2173 return;
2174 }
2175 id = hlock->class_idx - 1;
2176 /*
2177 * Whoops ran out of static storage again?
2178 */
2179 if (DEBUG_LOCKS_WARN_ON(id >= MAX_LOCKDEP_KEYS))
2180 return;
2181
2182 if (prev_hlock && (prev_hlock->irq_context !=
2183 hlock->irq_context))
2184 chain_key = 0;
2185 chain_key = iterate_chain_key(chain_key, id);
2186 prev_hlock = hlock;
2187 }
2188 if (chain_key != curr->curr_chain_key) {
2189 debug_locks_off();
2190 /*
2191 * More smoking hash instead of calculating it, damn see these
2192 * numbers float.. I bet that a pink elephant stepped on my memory.
2193 */
2194 WARN(1, "hm#2, depth: %u [%u], %016Lx != %016Lx\n",
2195 curr->lockdep_depth, i,
2196 (unsigned long long)chain_key,
2197 (unsigned long long)curr->curr_chain_key);
2198 }
2199#endif
2200}
2201
2202static void
2203print_usage_bug_scenario(struct held_lock *lock)
2204{
2205 struct lock_class *class = hlock_class(lock);
2206
2207 printk(" Possible unsafe locking scenario:\n\n");
2208 printk(" CPU0\n");
2209 printk(" ----\n");
2210 printk(" lock(");
2211 __print_lock_name(class);
2212 printk(");\n");
2213 printk(" <Interrupt>\n");
2214 printk(" lock(");
2215 __print_lock_name(class);
2216 printk(");\n");
2217 printk("\n *** DEADLOCK ***\n\n");
2218}
2219
2220static int
2221print_usage_bug(struct task_struct *curr, struct held_lock *this,
2222 enum lock_usage_bit prev_bit, enum lock_usage_bit new_bit)
2223{
2224 if (!debug_locks_off_graph_unlock() || debug_locks_silent)
2225 return 0;
2226
2227 printk("\n");
2228 printk("=================================\n");
2229 printk("[ INFO: inconsistent lock state ]\n");
2230 print_kernel_ident();
2231 printk("---------------------------------\n");
2232
2233 printk("inconsistent {%s} -> {%s} usage.\n",
2234 usage_str[prev_bit], usage_str[new_bit]);
2235
2236 printk("%s/%d [HC%u[%lu]:SC%u[%lu]:HE%u:SE%u] takes:\n",
2237 curr->comm, task_pid_nr(curr),
2238 trace_hardirq_context(curr), hardirq_count() >> HARDIRQ_SHIFT,
2239 trace_softirq_context(curr), softirq_count() >> SOFTIRQ_SHIFT,
2240 trace_hardirqs_enabled(curr),
2241 trace_softirqs_enabled(curr));
2242 print_lock(this);
2243
2244 printk("{%s} state was registered at:\n", usage_str[prev_bit]);
2245 print_stack_trace(hlock_class(this)->usage_traces + prev_bit, 1);
2246
2247 print_irqtrace_events(curr);
2248 printk("\nother info that might help us debug this:\n");
2249 print_usage_bug_scenario(this);
2250
2251 lockdep_print_held_locks(curr);
2252
2253 printk("\nstack backtrace:\n");
2254 dump_stack();
2255
2256 return 0;
2257}
2258
2259/*
2260 * Print out an error if an invalid bit is set:
2261 */
2262static inline int
2263valid_state(struct task_struct *curr, struct held_lock *this,
2264 enum lock_usage_bit new_bit, enum lock_usage_bit bad_bit)
2265{
2266 if (unlikely(hlock_class(this)->usage_mask & (1 << bad_bit)))
2267 return print_usage_bug(curr, this, bad_bit, new_bit);
2268 return 1;
2269}
2270
2271static int mark_lock(struct task_struct *curr, struct held_lock *this,
2272 enum lock_usage_bit new_bit);
2273
2274#if defined(CONFIG_TRACE_IRQFLAGS) && defined(CONFIG_PROVE_LOCKING)
2275
2276/*
2277 * print irq inversion bug:
2278 */
2279static int
2280print_irq_inversion_bug(struct task_struct *curr,
2281 struct lock_list *root, struct lock_list *other,
2282 struct held_lock *this, int forwards,
2283 const char *irqclass)
2284{
2285 struct lock_list *entry = other;
2286 struct lock_list *middle = NULL;
2287 int depth;
2288
2289 if (!debug_locks_off_graph_unlock() || debug_locks_silent)
2290 return 0;
2291
2292 printk("\n");
2293 printk("=========================================================\n");
2294 printk("[ INFO: possible irq lock inversion dependency detected ]\n");
2295 print_kernel_ident();
2296 printk("---------------------------------------------------------\n");
2297 printk("%s/%d just changed the state of lock:\n",
2298 curr->comm, task_pid_nr(curr));
2299 print_lock(this);
2300 if (forwards)
2301 printk("but this lock took another, %s-unsafe lock in the past:\n", irqclass);
2302 else
2303 printk("but this lock was taken by another, %s-safe lock in the past:\n", irqclass);
2304 print_lock_name(other->class);
2305 printk("\n\nand interrupts could create inverse lock ordering between them.\n\n");
2306
2307 printk("\nother info that might help us debug this:\n");
2308
2309 /* Find a middle lock (if one exists) */
2310 depth = get_lock_depth(other);
2311 do {
2312 if (depth == 0 && (entry != root)) {
2313 printk("lockdep:%s bad path found in chain graph\n", __func__);
2314 break;
2315 }
2316 middle = entry;
2317 entry = get_lock_parent(entry);
2318 depth--;
2319 } while (entry && entry != root && (depth >= 0));
2320 if (forwards)
2321 print_irq_lock_scenario(root, other,
2322 middle ? middle->class : root->class, other->class);
2323 else
2324 print_irq_lock_scenario(other, root,
2325 middle ? middle->class : other->class, root->class);
2326
2327 lockdep_print_held_locks(curr);
2328
2329 printk("\nthe shortest dependencies between 2nd lock and 1st lock:\n");
2330 if (!save_trace(&root->trace))
2331 return 0;
2332 print_shortest_lock_dependencies(other, root);
2333
2334 printk("\nstack backtrace:\n");
2335 dump_stack();
2336
2337 return 0;
2338}
2339
2340/*
2341 * Prove that in the forwards-direction subgraph starting at <this>
2342 * there is no lock matching <mask>:
2343 */
2344static int
2345check_usage_forwards(struct task_struct *curr, struct held_lock *this,
2346 enum lock_usage_bit bit, const char *irqclass)
2347{
2348 int ret;
2349 struct lock_list root;
2350 struct lock_list *uninitialized_var(target_entry);
2351
2352 root.parent = NULL;
2353 root.class = hlock_class(this);
2354 ret = find_usage_forwards(&root, bit, &target_entry);
2355 if (ret < 0)
2356 return print_bfs_bug(ret);
2357 if (ret == 1)
2358 return ret;
2359
2360 return print_irq_inversion_bug(curr, &root, target_entry,
2361 this, 1, irqclass);
2362}
2363
2364/*
2365 * Prove that in the backwards-direction subgraph starting at <this>
2366 * there is no lock matching <mask>:
2367 */
2368static int
2369check_usage_backwards(struct task_struct *curr, struct held_lock *this,
2370 enum lock_usage_bit bit, const char *irqclass)
2371{
2372 int ret;
2373 struct lock_list root;
2374 struct lock_list *uninitialized_var(target_entry);
2375
2376 root.parent = NULL;
2377 root.class = hlock_class(this);
2378 ret = find_usage_backwards(&root, bit, &target_entry);
2379 if (ret < 0)
2380 return print_bfs_bug(ret);
2381 if (ret == 1)
2382 return ret;
2383
2384 return print_irq_inversion_bug(curr, &root, target_entry,
2385 this, 0, irqclass);
2386}
2387
2388void print_irqtrace_events(struct task_struct *curr)
2389{
2390 printk("irq event stamp: %u\n", curr->irq_events);
2391 printk("hardirqs last enabled at (%u): ", curr->hardirq_enable_event);
2392 print_ip_sym(curr->hardirq_enable_ip);
2393 printk("hardirqs last disabled at (%u): ", curr->hardirq_disable_event);
2394 print_ip_sym(curr->hardirq_disable_ip);
2395 printk("softirqs last enabled at (%u): ", curr->softirq_enable_event);
2396 print_ip_sym(curr->softirq_enable_ip);
2397 printk("softirqs last disabled at (%u): ", curr->softirq_disable_event);
2398 print_ip_sym(curr->softirq_disable_ip);
2399}
2400
2401static int HARDIRQ_verbose(struct lock_class *class)
2402{
2403#if HARDIRQ_VERBOSE
2404 return class_filter(class);
2405#endif
2406 return 0;
2407}
2408
2409static int SOFTIRQ_verbose(struct lock_class *class)
2410{
2411#if SOFTIRQ_VERBOSE
2412 return class_filter(class);
2413#endif
2414 return 0;
2415}
2416
2417static int RECLAIM_FS_verbose(struct lock_class *class)
2418{
2419#if RECLAIM_VERBOSE
2420 return class_filter(class);
2421#endif
2422 return 0;
2423}
2424
2425#define STRICT_READ_CHECKS 1
2426
2427static int (*state_verbose_f[])(struct lock_class *class) = {
2428#define LOCKDEP_STATE(__STATE) \
2429 __STATE##_verbose,
2430#include "lockdep_states.h"
2431#undef LOCKDEP_STATE
2432};
2433
2434static inline int state_verbose(enum lock_usage_bit bit,
2435 struct lock_class *class)
2436{
2437 return state_verbose_f[bit >> 2](class);
2438}
2439
2440typedef int (*check_usage_f)(struct task_struct *, struct held_lock *,
2441 enum lock_usage_bit bit, const char *name);
2442
2443static int
2444mark_lock_irq(struct task_struct *curr, struct held_lock *this,
2445 enum lock_usage_bit new_bit)
2446{
2447 int excl_bit = exclusive_bit(new_bit);
2448 int read = new_bit & 1;
2449 int dir = new_bit & 2;
2450
2451 /*
2452 * mark USED_IN has to look forwards -- to ensure no dependency
2453 * has ENABLED state, which would allow recursion deadlocks.
2454 *
2455 * mark ENABLED has to look backwards -- to ensure no dependee
2456 * has USED_IN state, which, again, would allow recursion deadlocks.
2457 */
2458 check_usage_f usage = dir ?
2459 check_usage_backwards : check_usage_forwards;
2460
2461 /*
2462 * Validate that this particular lock does not have conflicting
2463 * usage states.
2464 */
2465 if (!valid_state(curr, this, new_bit, excl_bit))
2466 return 0;
2467
2468 /*
2469 * Validate that the lock dependencies don't have conflicting usage
2470 * states.
2471 */
2472 if ((!read || !dir || STRICT_READ_CHECKS) &&
2473 !usage(curr, this, excl_bit, state_name(new_bit & ~1)))
2474 return 0;
2475
2476 /*
2477 * Check for read in write conflicts
2478 */
2479 if (!read) {
2480 if (!valid_state(curr, this, new_bit, excl_bit + 1))
2481 return 0;
2482
2483 if (STRICT_READ_CHECKS &&
2484 !usage(curr, this, excl_bit + 1,
2485 state_name(new_bit + 1)))
2486 return 0;
2487 }
2488
2489 if (state_verbose(new_bit, hlock_class(this)))
2490 return 2;
2491
2492 return 1;
2493}
2494
2495enum mark_type {
2496#define LOCKDEP_STATE(__STATE) __STATE,
2497#include "lockdep_states.h"
2498#undef LOCKDEP_STATE
2499};
2500
2501/*
2502 * Mark all held locks with a usage bit:
2503 */
2504static int
2505mark_held_locks(struct task_struct *curr, enum mark_type mark)
2506{
2507 enum lock_usage_bit usage_bit;
2508 struct held_lock *hlock;
2509 int i;
2510
2511 for (i = 0; i < curr->lockdep_depth; i++) {
2512 hlock = curr->held_locks + i;
2513
2514 usage_bit = 2 + (mark << 2); /* ENABLED */
2515 if (hlock->read)
2516 usage_bit += 1; /* READ */
2517
2518 BUG_ON(usage_bit >= LOCK_USAGE_STATES);
2519
2520 if (!hlock->check)
2521 continue;
2522
2523 if (!mark_lock(curr, hlock, usage_bit))
2524 return 0;
2525 }
2526
2527 return 1;
2528}
2529
2530/*
2531 * Hardirqs will be enabled:
2532 */
2533static void __trace_hardirqs_on_caller(unsigned long ip)
2534{
2535 struct task_struct *curr = current;
2536
2537 /* we'll do an OFF -> ON transition: */
2538 curr->hardirqs_enabled = 1;
2539
2540 /*
2541 * We are going to turn hardirqs on, so set the
2542 * usage bit for all held locks:
2543 */
2544 if (!mark_held_locks(curr, HARDIRQ))
2545 return;
2546 /*
2547 * If we have softirqs enabled, then set the usage
2548 * bit for all held locks. (disabled hardirqs prevented
2549 * this bit from being set before)
2550 */
2551 if (curr->softirqs_enabled)
2552 if (!mark_held_locks(curr, SOFTIRQ))
2553 return;
2554
2555 curr->hardirq_enable_ip = ip;
2556 curr->hardirq_enable_event = ++curr->irq_events;
2557 debug_atomic_inc(hardirqs_on_events);
2558}
2559
2560__visible void trace_hardirqs_on_caller(unsigned long ip)
2561{
2562 time_hardirqs_on(CALLER_ADDR0, ip);
2563
2564 if (unlikely(!debug_locks || current->lockdep_recursion))
2565 return;
2566
2567 if (unlikely(current->hardirqs_enabled)) {
2568 /*
2569 * Neither irq nor preemption are disabled here
2570 * so this is racy by nature but losing one hit
2571 * in a stat is not a big deal.
2572 */
2573 __debug_atomic_inc(redundant_hardirqs_on);
2574 return;
2575 }
2576
2577 /*
2578 * We're enabling irqs and according to our state above irqs weren't
2579 * already enabled, yet we find the hardware thinks they are in fact
2580 * enabled.. someone messed up their IRQ state tracing.
2581 */
2582 if (DEBUG_LOCKS_WARN_ON(!irqs_disabled()))
2583 return;
2584
2585 /*
2586 * See the fine text that goes along with this variable definition.
2587 */
2588 if (DEBUG_LOCKS_WARN_ON(unlikely(early_boot_irqs_disabled)))
2589 return;
2590
2591 /*
2592 * Can't allow enabling interrupts while in an interrupt handler,
2593 * that's general bad form and such. Recursion, limited stack etc..
2594 */
2595 if (DEBUG_LOCKS_WARN_ON(current->hardirq_context))
2596 return;
2597
2598 current->lockdep_recursion = 1;
2599 __trace_hardirqs_on_caller(ip);
2600 current->lockdep_recursion = 0;
2601}
2602EXPORT_SYMBOL(trace_hardirqs_on_caller);
2603
2604void trace_hardirqs_on(void)
2605{
2606 trace_hardirqs_on_caller(CALLER_ADDR0);
2607}
2608EXPORT_SYMBOL(trace_hardirqs_on);
2609
2610/*
2611 * Hardirqs were disabled:
2612 */
2613__visible void trace_hardirqs_off_caller(unsigned long ip)
2614{
2615 struct task_struct *curr = current;
2616
2617 time_hardirqs_off(CALLER_ADDR0, ip);
2618
2619 if (unlikely(!debug_locks || current->lockdep_recursion))
2620 return;
2621
2622 /*
2623 * So we're supposed to get called after you mask local IRQs, but for
2624 * some reason the hardware doesn't quite think you did a proper job.
2625 */
2626 if (DEBUG_LOCKS_WARN_ON(!irqs_disabled()))
2627 return;
2628
2629 if (curr->hardirqs_enabled) {
2630 /*
2631 * We have done an ON -> OFF transition:
2632 */
2633 curr->hardirqs_enabled = 0;
2634 curr->hardirq_disable_ip = ip;
2635 curr->hardirq_disable_event = ++curr->irq_events;
2636 debug_atomic_inc(hardirqs_off_events);
2637 } else
2638 debug_atomic_inc(redundant_hardirqs_off);
2639}
2640EXPORT_SYMBOL(trace_hardirqs_off_caller);
2641
2642void trace_hardirqs_off(void)
2643{
2644 trace_hardirqs_off_caller(CALLER_ADDR0);
2645}
2646EXPORT_SYMBOL(trace_hardirqs_off);
2647
2648/*
2649 * Softirqs will be enabled:
2650 */
2651void trace_softirqs_on(unsigned long ip)
2652{
2653 struct task_struct *curr = current;
2654
2655 if (unlikely(!debug_locks || current->lockdep_recursion))
2656 return;
2657
2658 /*
2659 * We fancy IRQs being disabled here, see softirq.c, avoids
2660 * funny state and nesting things.
2661 */
2662 if (DEBUG_LOCKS_WARN_ON(!irqs_disabled()))
2663 return;
2664
2665 if (curr->softirqs_enabled) {
2666 debug_atomic_inc(redundant_softirqs_on);
2667 return;
2668 }
2669
2670 current->lockdep_recursion = 1;
2671 /*
2672 * We'll do an OFF -> ON transition:
2673 */
2674 curr->softirqs_enabled = 1;
2675 curr->softirq_enable_ip = ip;
2676 curr->softirq_enable_event = ++curr->irq_events;
2677 debug_atomic_inc(softirqs_on_events);
2678 /*
2679 * We are going to turn softirqs on, so set the
2680 * usage bit for all held locks, if hardirqs are
2681 * enabled too:
2682 */
2683 if (curr->hardirqs_enabled)
2684 mark_held_locks(curr, SOFTIRQ);
2685 current->lockdep_recursion = 0;
2686}
2687
2688/*
2689 * Softirqs were disabled:
2690 */
2691void trace_softirqs_off(unsigned long ip)
2692{
2693 struct task_struct *curr = current;
2694
2695 if (unlikely(!debug_locks || current->lockdep_recursion))
2696 return;
2697
2698 /*
2699 * We fancy IRQs being disabled here, see softirq.c
2700 */
2701 if (DEBUG_LOCKS_WARN_ON(!irqs_disabled()))
2702 return;
2703
2704 if (curr->softirqs_enabled) {
2705 /*
2706 * We have done an ON -> OFF transition:
2707 */
2708 curr->softirqs_enabled = 0;
2709 curr->softirq_disable_ip = ip;
2710 curr->softirq_disable_event = ++curr->irq_events;
2711 debug_atomic_inc(softirqs_off_events);
2712 /*
2713 * Whoops, we wanted softirqs off, so why aren't they?
2714 */
2715 DEBUG_LOCKS_WARN_ON(!softirq_count());
2716 } else
2717 debug_atomic_inc(redundant_softirqs_off);
2718}
2719
2720static void __lockdep_trace_alloc(gfp_t gfp_mask, unsigned long flags)
2721{
2722 struct task_struct *curr = current;
2723
2724 if (unlikely(!debug_locks))
2725 return;
2726
2727 /* no reclaim without waiting on it */
2728 if (!(gfp_mask & __GFP_WAIT))
2729 return;
2730
2731 /* this guy won't enter reclaim */
2732 if ((curr->flags & PF_MEMALLOC) && !(gfp_mask & __GFP_NOMEMALLOC))
2733 return;
2734
2735 /* We're only interested __GFP_FS allocations for now */
2736 if (!(gfp_mask & __GFP_FS))
2737 return;
2738
2739 /*
2740 * Oi! Can't be having __GFP_FS allocations with IRQs disabled.
2741 */
2742 if (DEBUG_LOCKS_WARN_ON(irqs_disabled_flags(flags)))
2743 return;
2744
2745 mark_held_locks(curr, RECLAIM_FS);
2746}
2747
2748static void check_flags(unsigned long flags);
2749
2750void lockdep_trace_alloc(gfp_t gfp_mask)
2751{
2752 unsigned long flags;
2753
2754 if (unlikely(current->lockdep_recursion))
2755 return;
2756
2757 raw_local_irq_save(flags);
2758 check_flags(flags);
2759 current->lockdep_recursion = 1;
2760 __lockdep_trace_alloc(gfp_mask, flags);
2761 current->lockdep_recursion = 0;
2762 raw_local_irq_restore(flags);
2763}
2764
2765static int mark_irqflags(struct task_struct *curr, struct held_lock *hlock)
2766{
2767 /*
2768 * If non-trylock use in a hardirq or softirq context, then
2769 * mark the lock as used in these contexts:
2770 */
2771 if (!hlock->trylock) {
2772 if (hlock->read) {
2773 if (curr->hardirq_context)
2774 if (!mark_lock(curr, hlock,
2775 LOCK_USED_IN_HARDIRQ_READ))
2776 return 0;
2777 if (curr->softirq_context)
2778 if (!mark_lock(curr, hlock,
2779 LOCK_USED_IN_SOFTIRQ_READ))
2780 return 0;
2781 } else {
2782 if (curr->hardirq_context)
2783 if (!mark_lock(curr, hlock, LOCK_USED_IN_HARDIRQ))
2784 return 0;
2785 if (curr->softirq_context)
2786 if (!mark_lock(curr, hlock, LOCK_USED_IN_SOFTIRQ))
2787 return 0;
2788 }
2789 }
2790 if (!hlock->hardirqs_off) {
2791 if (hlock->read) {
2792 if (!mark_lock(curr, hlock,
2793 LOCK_ENABLED_HARDIRQ_READ))
2794 return 0;
2795 if (curr->softirqs_enabled)
2796 if (!mark_lock(curr, hlock,
2797 LOCK_ENABLED_SOFTIRQ_READ))
2798 return 0;
2799 } else {
2800 if (!mark_lock(curr, hlock,
2801 LOCK_ENABLED_HARDIRQ))
2802 return 0;
2803 if (curr->softirqs_enabled)
2804 if (!mark_lock(curr, hlock,
2805 LOCK_ENABLED_SOFTIRQ))
2806 return 0;
2807 }
2808 }
2809
2810 /*
2811 * We reuse the irq context infrastructure more broadly as a general
2812 * context checking code. This tests GFP_FS recursion (a lock taken
2813 * during reclaim for a GFP_FS allocation is held over a GFP_FS
2814 * allocation).
2815 */
2816 if (!hlock->trylock && (curr->lockdep_reclaim_gfp & __GFP_FS)) {
2817 if (hlock->read) {
2818 if (!mark_lock(curr, hlock, LOCK_USED_IN_RECLAIM_FS_READ))
2819 return 0;
2820 } else {
2821 if (!mark_lock(curr, hlock, LOCK_USED_IN_RECLAIM_FS))
2822 return 0;
2823 }
2824 }
2825
2826 return 1;
2827}
2828
2829static int separate_irq_context(struct task_struct *curr,
2830 struct held_lock *hlock)
2831{
2832 unsigned int depth = curr->lockdep_depth;
2833
2834 /*
2835 * Keep track of points where we cross into an interrupt context:
2836 */
2837 hlock->irq_context = 2*(curr->hardirq_context ? 1 : 0) +
2838 curr->softirq_context;
2839 if (depth) {
2840 struct held_lock *prev_hlock;
2841
2842 prev_hlock = curr->held_locks + depth-1;
2843 /*
2844 * If we cross into another context, reset the
2845 * hash key (this also prevents the checking and the
2846 * adding of the dependency to 'prev'):
2847 */
2848 if (prev_hlock->irq_context != hlock->irq_context)
2849 return 1;
2850 }
2851 return 0;
2852}
2853
2854#else /* defined(CONFIG_TRACE_IRQFLAGS) && defined(CONFIG_PROVE_LOCKING) */
2855
2856static inline
2857int mark_lock_irq(struct task_struct *curr, struct held_lock *this,
2858 enum lock_usage_bit new_bit)
2859{
2860 WARN_ON(1); /* Impossible innit? when we don't have TRACE_IRQFLAG */
2861 return 1;
2862}
2863
2864static inline int mark_irqflags(struct task_struct *curr,
2865 struct held_lock *hlock)
2866{
2867 return 1;
2868}
2869
2870static inline int separate_irq_context(struct task_struct *curr,
2871 struct held_lock *hlock)
2872{
2873 return 0;
2874}
2875
2876void lockdep_trace_alloc(gfp_t gfp_mask)
2877{
2878}
2879
2880#endif /* defined(CONFIG_TRACE_IRQFLAGS) && defined(CONFIG_PROVE_LOCKING) */
2881
2882/*
2883 * Mark a lock with a usage bit, and validate the state transition:
2884 */
2885static int mark_lock(struct task_struct *curr, struct held_lock *this,
2886 enum lock_usage_bit new_bit)
2887{
2888 unsigned int new_mask = 1 << new_bit, ret = 1;
2889
2890 /*
2891 * If already set then do not dirty the cacheline,
2892 * nor do any checks:
2893 */
2894 if (likely(hlock_class(this)->usage_mask & new_mask))
2895 return 1;
2896
2897 if (!graph_lock())
2898 return 0;
2899 /*
2900 * Make sure we didn't race:
2901 */
2902 if (unlikely(hlock_class(this)->usage_mask & new_mask)) {
2903 graph_unlock();
2904 return 1;
2905 }
2906
2907 hlock_class(this)->usage_mask |= new_mask;
2908
2909 if (!save_trace(hlock_class(this)->usage_traces + new_bit))
2910 return 0;
2911
2912 switch (new_bit) {
2913#define LOCKDEP_STATE(__STATE) \
2914 case LOCK_USED_IN_##__STATE: \
2915 case LOCK_USED_IN_##__STATE##_READ: \
2916 case LOCK_ENABLED_##__STATE: \
2917 case LOCK_ENABLED_##__STATE##_READ:
2918#include "lockdep_states.h"
2919#undef LOCKDEP_STATE
2920 ret = mark_lock_irq(curr, this, new_bit);
2921 if (!ret)
2922 return 0;
2923 break;
2924 case LOCK_USED:
2925 debug_atomic_dec(nr_unused_locks);
2926 break;
2927 default:
2928 if (!debug_locks_off_graph_unlock())
2929 return 0;
2930 WARN_ON(1);
2931 return 0;
2932 }
2933
2934 graph_unlock();
2935
2936 /*
2937 * We must printk outside of the graph_lock:
2938 */
2939 if (ret == 2) {
2940 printk("\nmarked lock as {%s}:\n", usage_str[new_bit]);
2941 print_lock(this);
2942 print_irqtrace_events(curr);
2943 dump_stack();
2944 }
2945
2946 return ret;
2947}
2948
2949/*
2950 * Initialize a lock instance's lock-class mapping info:
2951 */
2952void lockdep_init_map(struct lockdep_map *lock, const char *name,
2953 struct lock_class_key *key, int subclass)
2954{
2955 int i;
2956
2957 kmemcheck_mark_initialized(lock, sizeof(*lock));
2958
2959 for (i = 0; i < NR_LOCKDEP_CACHING_CLASSES; i++)
2960 lock->class_cache[i] = NULL;
2961
2962#ifdef CONFIG_LOCK_STAT
2963 lock->cpu = raw_smp_processor_id();
2964#endif
2965
2966 /*
2967 * Can't be having no nameless bastards around this place!
2968 */
2969 if (DEBUG_LOCKS_WARN_ON(!name)) {
2970 lock->name = "NULL";
2971 return;
2972 }
2973
2974 lock->name = name;
2975
2976 /*
2977 * No key, no joy, we need to hash something.
2978 */
2979 if (DEBUG_LOCKS_WARN_ON(!key))
2980 return;
2981 /*
2982 * Sanity check, the lock-class key must be persistent:
2983 */
2984 if (!static_obj(key)) {
2985 printk("BUG: key %p not in .data!\n", key);
2986 /*
2987 * What it says above ^^^^^, I suggest you read it.
2988 */
2989 DEBUG_LOCKS_WARN_ON(1);
2990 return;
2991 }
2992 lock->key = key;
2993
2994 if (unlikely(!debug_locks))
2995 return;
2996
2997 if (subclass)
2998 register_lock_class(lock, subclass, 1);
2999}
3000EXPORT_SYMBOL_GPL(lockdep_init_map);
3001
3002struct lock_class_key __lockdep_no_validate__;
3003EXPORT_SYMBOL_GPL(__lockdep_no_validate__);
3004
3005static int
3006print_lock_nested_lock_not_held(struct task_struct *curr,
3007 struct held_lock *hlock,
3008 unsigned long ip)
3009{
3010 if (!debug_locks_off())
3011 return 0;
3012 if (debug_locks_silent)
3013 return 0;
3014
3015 printk("\n");
3016 printk("==================================\n");
3017 printk("[ BUG: Nested lock was not taken ]\n");
3018 print_kernel_ident();
3019 printk("----------------------------------\n");
3020
3021 printk("%s/%d is trying to lock:\n", curr->comm, task_pid_nr(curr));
3022 print_lock(hlock);
3023
3024 printk("\nbut this task is not holding:\n");
3025 printk("%s\n", hlock->nest_lock->name);
3026
3027 printk("\nstack backtrace:\n");
3028 dump_stack();
3029
3030 printk("\nother info that might help us debug this:\n");
3031 lockdep_print_held_locks(curr);
3032
3033 printk("\nstack backtrace:\n");
3034 dump_stack();
3035
3036 return 0;
3037}
3038
3039static int __lock_is_held(struct lockdep_map *lock);
3040
3041/*
3042 * This gets called for every mutex_lock*()/spin_lock*() operation.
3043 * We maintain the dependency maps and validate the locking attempt:
3044 */
3045static int __lock_acquire(struct lockdep_map *lock, unsigned int subclass,
3046 int trylock, int read, int check, int hardirqs_off,
3047 struct lockdep_map *nest_lock, unsigned long ip,
3048 int references)
3049{
3050 struct task_struct *curr = current;
3051 struct lock_class *class = NULL;
3052 struct held_lock *hlock;
3053 unsigned int depth, id;
3054 int chain_head = 0;
3055 int class_idx;
3056 u64 chain_key;
3057
3058 if (unlikely(!debug_locks))
3059 return 0;
3060
3061 /*
3062 * Lockdep should run with IRQs disabled, otherwise we could
3063 * get an interrupt which would want to take locks, which would
3064 * end up in lockdep and have you got a head-ache already?
3065 */
3066 if (DEBUG_LOCKS_WARN_ON(!irqs_disabled()))
3067 return 0;
3068
3069 if (!prove_locking || lock->key == &__lockdep_no_validate__)
3070 check = 0;
3071
3072 if (subclass < NR_LOCKDEP_CACHING_CLASSES)
3073 class = lock->class_cache[subclass];
3074 /*
3075 * Not cached?
3076 */
3077 if (unlikely(!class)) {
3078 class = register_lock_class(lock, subclass, 0);
3079 if (!class)
3080 return 0;
3081 }
3082 atomic_inc((atomic_t *)&class->ops);
3083 if (very_verbose(class)) {
3084 printk("\nacquire class [%p] %s", class->key, class->name);
3085 if (class->name_version > 1)
3086 printk("#%d", class->name_version);
3087 printk("\n");
3088 dump_stack();
3089 }
3090
3091 /*
3092 * Add the lock to the list of currently held locks.
3093 * (we dont increase the depth just yet, up until the
3094 * dependency checks are done)
3095 */
3096 depth = curr->lockdep_depth;
3097 /*
3098 * Ran out of static storage for our per-task lock stack again have we?
3099 */
3100 if (DEBUG_LOCKS_WARN_ON(depth >= MAX_LOCK_DEPTH))
3101 return 0;
3102
3103 class_idx = class - lock_classes + 1;
3104
3105 if (depth) {
3106 hlock = curr->held_locks + depth - 1;
3107 if (hlock->class_idx == class_idx && nest_lock) {
3108 if (hlock->references)
3109 hlock->references++;
3110 else
3111 hlock->references = 2;
3112
3113 return 1;
3114 }
3115 }
3116
3117 hlock = curr->held_locks + depth;
3118 /*
3119 * Plain impossible, we just registered it and checked it weren't no
3120 * NULL like.. I bet this mushroom I ate was good!
3121 */
3122 if (DEBUG_LOCKS_WARN_ON(!class))
3123 return 0;
3124 hlock->class_idx = class_idx;
3125 hlock->acquire_ip = ip;
3126 hlock->instance = lock;
3127 hlock->nest_lock = nest_lock;
3128 hlock->trylock = trylock;
3129 hlock->read = read;
3130 hlock->check = check;
3131 hlock->hardirqs_off = !!hardirqs_off;
3132 hlock->references = references;
3133#ifdef CONFIG_LOCK_STAT
3134 hlock->waittime_stamp = 0;
3135 hlock->holdtime_stamp = lockstat_clock();
3136#endif
3137
3138 if (check && !mark_irqflags(curr, hlock))
3139 return 0;
3140
3141 /* mark it as used: */
3142 if (!mark_lock(curr, hlock, LOCK_USED))
3143 return 0;
3144
3145 /*
3146 * Calculate the chain hash: it's the combined hash of all the
3147 * lock keys along the dependency chain. We save the hash value
3148 * at every step so that we can get the current hash easily
3149 * after unlock. The chain hash is then used to cache dependency
3150 * results.
3151 *
3152 * The 'key ID' is what is the most compact key value to drive
3153 * the hash, not class->key.
3154 */
3155 id = class - lock_classes;
3156 /*
3157 * Whoops, we did it again.. ran straight out of our static allocation.
3158 */
3159 if (DEBUG_LOCKS_WARN_ON(id >= MAX_LOCKDEP_KEYS))
3160 return 0;
3161
3162 chain_key = curr->curr_chain_key;
3163 if (!depth) {
3164 /*
3165 * How can we have a chain hash when we ain't got no keys?!
3166 */
3167 if (DEBUG_LOCKS_WARN_ON(chain_key != 0))
3168 return 0;
3169 chain_head = 1;
3170 }
3171
3172 hlock->prev_chain_key = chain_key;
3173 if (separate_irq_context(curr, hlock)) {
3174 chain_key = 0;
3175 chain_head = 1;
3176 }
3177 chain_key = iterate_chain_key(chain_key, id);
3178
3179 if (nest_lock && !__lock_is_held(nest_lock))
3180 return print_lock_nested_lock_not_held(curr, hlock, ip);
3181
3182 if (!validate_chain(curr, lock, hlock, chain_head, chain_key))
3183 return 0;
3184
3185 curr->curr_chain_key = chain_key;
3186 curr->lockdep_depth++;
3187 check_chain_key(curr);
3188#ifdef CONFIG_DEBUG_LOCKDEP
3189 if (unlikely(!debug_locks))
3190 return 0;
3191#endif
3192 if (unlikely(curr->lockdep_depth >= MAX_LOCK_DEPTH)) {
3193 debug_locks_off();
3194 print_lockdep_off("BUG: MAX_LOCK_DEPTH too low!");
3195 printk(KERN_DEBUG "depth: %i max: %lu!\n",
3196 curr->lockdep_depth, MAX_LOCK_DEPTH);
3197
3198 lockdep_print_held_locks(current);
3199 debug_show_all_locks();
3200 dump_stack();
3201
3202 return 0;
3203 }
3204
3205 if (unlikely(curr->lockdep_depth > max_lockdep_depth))
3206 max_lockdep_depth = curr->lockdep_depth;
3207
3208 return 1;
3209}
3210
3211static int
3212print_unlock_imbalance_bug(struct task_struct *curr, struct lockdep_map *lock,
3213 unsigned long ip)
3214{
3215 if (!debug_locks_off())
3216 return 0;
3217 if (debug_locks_silent)
3218 return 0;
3219
3220 printk("\n");
3221 printk("=====================================\n");
3222 printk("[ BUG: bad unlock balance detected! ]\n");
3223 print_kernel_ident();
3224 printk("-------------------------------------\n");
3225 printk("%s/%d is trying to release lock (",
3226 curr->comm, task_pid_nr(curr));
3227 print_lockdep_cache(lock);
3228 printk(") at:\n");
3229 print_ip_sym(ip);
3230 printk("but there are no more locks to release!\n");
3231 printk("\nother info that might help us debug this:\n");
3232 lockdep_print_held_locks(curr);
3233
3234 printk("\nstack backtrace:\n");
3235 dump_stack();
3236
3237 return 0;
3238}
3239
3240/*
3241 * Common debugging checks for both nested and non-nested unlock:
3242 */
3243static int check_unlock(struct task_struct *curr, struct lockdep_map *lock,
3244 unsigned long ip)
3245{
3246 if (unlikely(!debug_locks))
3247 return 0;
3248 /*
3249 * Lockdep should run with IRQs disabled, recursion, head-ache, etc..
3250 */
3251 if (DEBUG_LOCKS_WARN_ON(!irqs_disabled()))
3252 return 0;
3253
3254 if (curr->lockdep_depth <= 0)
3255 return print_unlock_imbalance_bug(curr, lock, ip);
3256
3257 return 1;
3258}
3259
3260static int match_held_lock(struct held_lock *hlock, struct lockdep_map *lock)
3261{
3262 if (hlock->instance == lock)
3263 return 1;
3264
3265 if (hlock->references) {
3266 struct lock_class *class = lock->class_cache[0];
3267
3268 if (!class)
3269 class = look_up_lock_class(lock, 0);
3270
3271 /*
3272 * If look_up_lock_class() failed to find a class, we're trying
3273 * to test if we hold a lock that has never yet been acquired.
3274 * Clearly if the lock hasn't been acquired _ever_, we're not
3275 * holding it either, so report failure.
3276 */
3277 if (!class)
3278 return 0;
3279
3280 /*
3281 * References, but not a lock we're actually ref-counting?
3282 * State got messed up, follow the sites that change ->references
3283 * and try to make sense of it.
3284 */
3285 if (DEBUG_LOCKS_WARN_ON(!hlock->nest_lock))
3286 return 0;
3287
3288 if (hlock->class_idx == class - lock_classes + 1)
3289 return 1;
3290 }
3291
3292 return 0;
3293}
3294
3295static int
3296__lock_set_class(struct lockdep_map *lock, const char *name,
3297 struct lock_class_key *key, unsigned int subclass,
3298 unsigned long ip)
3299{
3300 struct task_struct *curr = current;
3301 struct held_lock *hlock, *prev_hlock;
3302 struct lock_class *class;
3303 unsigned int depth;
3304 int i;
3305
3306 depth = curr->lockdep_depth;
3307 /*
3308 * This function is about (re)setting the class of a held lock,
3309 * yet we're not actually holding any locks. Naughty user!
3310 */
3311 if (DEBUG_LOCKS_WARN_ON(!depth))
3312 return 0;
3313
3314 prev_hlock = NULL;
3315 for (i = depth-1; i >= 0; i--) {
3316 hlock = curr->held_locks + i;
3317 /*
3318 * We must not cross into another context:
3319 */
3320 if (prev_hlock && prev_hlock->irq_context != hlock->irq_context)
3321 break;
3322 if (match_held_lock(hlock, lock))
3323 goto found_it;
3324 prev_hlock = hlock;
3325 }
3326 return print_unlock_imbalance_bug(curr, lock, ip);
3327
3328found_it:
3329 lockdep_init_map(lock, name, key, 0);
3330 class = register_lock_class(lock, subclass, 0);
3331 hlock->class_idx = class - lock_classes + 1;
3332
3333 curr->lockdep_depth = i;
3334 curr->curr_chain_key = hlock->prev_chain_key;
3335
3336 for (; i < depth; i++) {
3337 hlock = curr->held_locks + i;
3338 if (!__lock_acquire(hlock->instance,
3339 hlock_class(hlock)->subclass, hlock->trylock,
3340 hlock->read, hlock->check, hlock->hardirqs_off,
3341 hlock->nest_lock, hlock->acquire_ip,
3342 hlock->references))
3343 return 0;
3344 }
3345
3346 /*
3347 * I took it apart and put it back together again, except now I have
3348 * these 'spare' parts.. where shall I put them.
3349 */
3350 if (DEBUG_LOCKS_WARN_ON(curr->lockdep_depth != depth))
3351 return 0;
3352 return 1;
3353}
3354
3355/*
3356 * Remove the lock to the list of currently held locks in a
3357 * potentially non-nested (out of order) manner. This is a
3358 * relatively rare operation, as all the unlock APIs default
3359 * to nested mode (which uses lock_release()):
3360 */
3361static int
3362lock_release_non_nested(struct task_struct *curr,
3363 struct lockdep_map *lock, unsigned long ip)
3364{
3365 struct held_lock *hlock, *prev_hlock;
3366 unsigned int depth;
3367 int i;
3368
3369 /*
3370 * Check whether the lock exists in the current stack
3371 * of held locks:
3372 */
3373 depth = curr->lockdep_depth;
3374 /*
3375 * So we're all set to release this lock.. wait what lock? We don't
3376 * own any locks, you've been drinking again?
3377 */
3378 if (DEBUG_LOCKS_WARN_ON(!depth))
3379 return 0;
3380
3381 prev_hlock = NULL;
3382 for (i = depth-1; i >= 0; i--) {
3383 hlock = curr->held_locks + i;
3384 /*
3385 * We must not cross into another context:
3386 */
3387 if (prev_hlock && prev_hlock->irq_context != hlock->irq_context)
3388 break;
3389 if (match_held_lock(hlock, lock))
3390 goto found_it;
3391 prev_hlock = hlock;
3392 }
3393 return print_unlock_imbalance_bug(curr, lock, ip);
3394
3395found_it:
3396 if (hlock->instance == lock)
3397 lock_release_holdtime(hlock);
3398
3399 if (hlock->references) {
3400 hlock->references--;
3401 if (hlock->references) {
3402 /*
3403 * We had, and after removing one, still have
3404 * references, the current lock stack is still
3405 * valid. We're done!
3406 */
3407 return 1;
3408 }
3409 }
3410
3411 /*
3412 * We have the right lock to unlock, 'hlock' points to it.
3413 * Now we remove it from the stack, and add back the other
3414 * entries (if any), recalculating the hash along the way:
3415 */
3416
3417 curr->lockdep_depth = i;
3418 curr->curr_chain_key = hlock->prev_chain_key;
3419
3420 for (i++; i < depth; i++) {
3421 hlock = curr->held_locks + i;
3422 if (!__lock_acquire(hlock->instance,
3423 hlock_class(hlock)->subclass, hlock->trylock,
3424 hlock->read, hlock->check, hlock->hardirqs_off,
3425 hlock->nest_lock, hlock->acquire_ip,
3426 hlock->references))
3427 return 0;
3428 }
3429
3430 /*
3431 * We had N bottles of beer on the wall, we drank one, but now
3432 * there's not N-1 bottles of beer left on the wall...
3433 */
3434 if (DEBUG_LOCKS_WARN_ON(curr->lockdep_depth != depth - 1))
3435 return 0;
3436 return 1;
3437}
3438
3439/*
3440 * Remove the lock to the list of currently held locks - this gets
3441 * called on mutex_unlock()/spin_unlock*() (or on a failed
3442 * mutex_lock_interruptible()). This is done for unlocks that nest
3443 * perfectly. (i.e. the current top of the lock-stack is unlocked)
3444 */
3445static int lock_release_nested(struct task_struct *curr,
3446 struct lockdep_map *lock, unsigned long ip)
3447{
3448 struct held_lock *hlock;
3449 unsigned int depth;
3450
3451 /*
3452 * Pop off the top of the lock stack:
3453 */
3454 depth = curr->lockdep_depth - 1;
3455 hlock = curr->held_locks + depth;
3456
3457 /*
3458 * Is the unlock non-nested:
3459 */
3460 if (hlock->instance != lock || hlock->references)
3461 return lock_release_non_nested(curr, lock, ip);
3462 curr->lockdep_depth--;
3463
3464 /*
3465 * No more locks, but somehow we've got hash left over, who left it?
3466 */
3467 if (DEBUG_LOCKS_WARN_ON(!depth && (hlock->prev_chain_key != 0)))
3468 return 0;
3469
3470 curr->curr_chain_key = hlock->prev_chain_key;
3471
3472 lock_release_holdtime(hlock);
3473
3474#ifdef CONFIG_DEBUG_LOCKDEP
3475 hlock->prev_chain_key = 0;
3476 hlock->class_idx = 0;
3477 hlock->acquire_ip = 0;
3478 hlock->irq_context = 0;
3479#endif
3480 return 1;
3481}
3482
3483/*
3484 * Remove the lock to the list of currently held locks - this gets
3485 * called on mutex_unlock()/spin_unlock*() (or on a failed
3486 * mutex_lock_interruptible()). This is done for unlocks that nest
3487 * perfectly. (i.e. the current top of the lock-stack is unlocked)
3488 */
3489static void
3490__lock_release(struct lockdep_map *lock, int nested, unsigned long ip)
3491{
3492 struct task_struct *curr = current;
3493
3494 if (!check_unlock(curr, lock, ip))
3495 return;
3496
3497 if (nested) {
3498 if (!lock_release_nested(curr, lock, ip))
3499 return;
3500 } else {
3501 if (!lock_release_non_nested(curr, lock, ip))
3502 return;
3503 }
3504
3505 check_chain_key(curr);
3506}
3507
3508static int __lock_is_held(struct lockdep_map *lock)
3509{
3510 struct task_struct *curr = current;
3511 int i;
3512
3513 for (i = 0; i < curr->lockdep_depth; i++) {
3514 struct held_lock *hlock = curr->held_locks + i;
3515
3516 if (match_held_lock(hlock, lock))
3517 return 1;
3518 }
3519
3520 return 0;
3521}
3522
3523/*
3524 * Check whether we follow the irq-flags state precisely:
3525 */
3526static void check_flags(unsigned long flags)
3527{
3528#if defined(CONFIG_PROVE_LOCKING) && defined(CONFIG_DEBUG_LOCKDEP) && \
3529 defined(CONFIG_TRACE_IRQFLAGS)
3530 if (!debug_locks)
3531 return;
3532
3533 if (irqs_disabled_flags(flags)) {
3534 if (DEBUG_LOCKS_WARN_ON(current->hardirqs_enabled)) {
3535 printk("possible reason: unannotated irqs-off.\n");
3536 }
3537 } else {
3538 if (DEBUG_LOCKS_WARN_ON(!current->hardirqs_enabled)) {
3539 printk("possible reason: unannotated irqs-on.\n");
3540 }
3541 }
3542
3543 /*
3544 * We dont accurately track softirq state in e.g.
3545 * hardirq contexts (such as on 4KSTACKS), so only
3546 * check if not in hardirq contexts:
3547 */
3548 if (!hardirq_count()) {
3549 if (softirq_count()) {
3550 /* like the above, but with softirqs */
3551 DEBUG_LOCKS_WARN_ON(current->softirqs_enabled);
3552 } else {
3553 /* lick the above, does it taste good? */
3554 DEBUG_LOCKS_WARN_ON(!current->softirqs_enabled);
3555 }
3556 }
3557
3558 if (!debug_locks)
3559 print_irqtrace_events(current);
3560#endif
3561}
3562
3563void lock_set_class(struct lockdep_map *lock, const char *name,
3564 struct lock_class_key *key, unsigned int subclass,
3565 unsigned long ip)
3566{
3567 unsigned long flags;
3568
3569 if (unlikely(current->lockdep_recursion))
3570 return;
3571
3572 raw_local_irq_save(flags);
3573 current->lockdep_recursion = 1;
3574 check_flags(flags);
3575 if (__lock_set_class(lock, name, key, subclass, ip))
3576 check_chain_key(current);
3577 current->lockdep_recursion = 0;
3578 raw_local_irq_restore(flags);
3579}
3580EXPORT_SYMBOL_GPL(lock_set_class);
3581
3582/*
3583 * We are not always called with irqs disabled - do that here,
3584 * and also avoid lockdep recursion:
3585 */
3586void lock_acquire(struct lockdep_map *lock, unsigned int subclass,
3587 int trylock, int read, int check,
3588 struct lockdep_map *nest_lock, unsigned long ip)
3589{
3590 unsigned long flags;
3591
3592 if (unlikely(current->lockdep_recursion))
3593 return;
3594
3595 raw_local_irq_save(flags);
3596 check_flags(flags);
3597
3598 current->lockdep_recursion = 1;
3599 trace_lock_acquire(lock, subclass, trylock, read, check, nest_lock, ip);
3600 __lock_acquire(lock, subclass, trylock, read, check,
3601 irqs_disabled_flags(flags), nest_lock, ip, 0);
3602 current->lockdep_recursion = 0;
3603 raw_local_irq_restore(flags);
3604}
3605EXPORT_SYMBOL_GPL(lock_acquire);
3606
3607void lock_release(struct lockdep_map *lock, int nested,
3608 unsigned long ip)
3609{
3610 unsigned long flags;
3611
3612 if (unlikely(current->lockdep_recursion))
3613 return;
3614
3615 raw_local_irq_save(flags);
3616 check_flags(flags);
3617 current->lockdep_recursion = 1;
3618 trace_lock_release(lock, ip);
3619 __lock_release(lock, nested, ip);
3620 current->lockdep_recursion = 0;
3621 raw_local_irq_restore(flags);
3622}
3623EXPORT_SYMBOL_GPL(lock_release);
3624
3625int lock_is_held(struct lockdep_map *lock)
3626{
3627 unsigned long flags;
3628 int ret = 0;
3629
3630 if (unlikely(current->lockdep_recursion))
3631 return 1; /* avoid false negative lockdep_assert_held() */
3632
3633 raw_local_irq_save(flags);
3634 check_flags(flags);
3635
3636 current->lockdep_recursion = 1;
3637 ret = __lock_is_held(lock);
3638 current->lockdep_recursion = 0;
3639 raw_local_irq_restore(flags);
3640
3641 return ret;
3642}
3643EXPORT_SYMBOL_GPL(lock_is_held);
3644
3645void lockdep_set_current_reclaim_state(gfp_t gfp_mask)
3646{
3647 current->lockdep_reclaim_gfp = gfp_mask;
3648}
3649
3650void lockdep_clear_current_reclaim_state(void)
3651{
3652 current->lockdep_reclaim_gfp = 0;
3653}
3654
3655#ifdef CONFIG_LOCK_STAT
3656static int
3657print_lock_contention_bug(struct task_struct *curr, struct lockdep_map *lock,
3658 unsigned long ip)
3659{
3660 if (!debug_locks_off())
3661 return 0;
3662 if (debug_locks_silent)
3663 return 0;
3664
3665 printk("\n");
3666 printk("=================================\n");
3667 printk("[ BUG: bad contention detected! ]\n");
3668 print_kernel_ident();
3669 printk("---------------------------------\n");
3670 printk("%s/%d is trying to contend lock (",
3671 curr->comm, task_pid_nr(curr));
3672 print_lockdep_cache(lock);
3673 printk(") at:\n");
3674 print_ip_sym(ip);
3675 printk("but there are no locks held!\n");
3676 printk("\nother info that might help us debug this:\n");
3677 lockdep_print_held_locks(curr);
3678
3679 printk("\nstack backtrace:\n");
3680 dump_stack();
3681
3682 return 0;
3683}
3684
3685static void
3686__lock_contended(struct lockdep_map *lock, unsigned long ip)
3687{
3688 struct task_struct *curr = current;
3689 struct held_lock *hlock, *prev_hlock;
3690 struct lock_class_stats *stats;
3691 unsigned int depth;
3692 int i, contention_point, contending_point;
3693
3694 depth = curr->lockdep_depth;
3695 /*
3696 * Whee, we contended on this lock, except it seems we're not
3697 * actually trying to acquire anything much at all..
3698 */
3699 if (DEBUG_LOCKS_WARN_ON(!depth))
3700 return;
3701
3702 prev_hlock = NULL;
3703 for (i = depth-1; i >= 0; i--) {
3704 hlock = curr->held_locks + i;
3705 /*
3706 * We must not cross into another context:
3707 */
3708 if (prev_hlock && prev_hlock->irq_context != hlock->irq_context)
3709 break;
3710 if (match_held_lock(hlock, lock))
3711 goto found_it;
3712 prev_hlock = hlock;
3713 }
3714 print_lock_contention_bug(curr, lock, ip);
3715 return;
3716
3717found_it:
3718 if (hlock->instance != lock)
3719 return;
3720
3721 hlock->waittime_stamp = lockstat_clock();
3722
3723 contention_point = lock_point(hlock_class(hlock)->contention_point, ip);
3724 contending_point = lock_point(hlock_class(hlock)->contending_point,
3725 lock->ip);
3726
3727 stats = get_lock_stats(hlock_class(hlock));
3728 if (contention_point < LOCKSTAT_POINTS)
3729 stats->contention_point[contention_point]++;
3730 if (contending_point < LOCKSTAT_POINTS)
3731 stats->contending_point[contending_point]++;
3732 if (lock->cpu != smp_processor_id())
3733 stats->bounces[bounce_contended + !!hlock->read]++;
3734 put_lock_stats(stats);
3735}
3736
3737static void
3738__lock_acquired(struct lockdep_map *lock, unsigned long ip)
3739{
3740 struct task_struct *curr = current;
3741 struct held_lock *hlock, *prev_hlock;
3742 struct lock_class_stats *stats;
3743 unsigned int depth;
3744 u64 now, waittime = 0;
3745 int i, cpu;
3746
3747 depth = curr->lockdep_depth;
3748 /*
3749 * Yay, we acquired ownership of this lock we didn't try to
3750 * acquire, how the heck did that happen?
3751 */
3752 if (DEBUG_LOCKS_WARN_ON(!depth))
3753 return;
3754
3755 prev_hlock = NULL;
3756 for (i = depth-1; i >= 0; i--) {
3757 hlock = curr->held_locks + i;
3758 /*
3759 * We must not cross into another context:
3760 */
3761 if (prev_hlock && prev_hlock->irq_context != hlock->irq_context)
3762 break;
3763 if (match_held_lock(hlock, lock))
3764 goto found_it;
3765 prev_hlock = hlock;
3766 }
3767 print_lock_contention_bug(curr, lock, _RET_IP_);
3768 return;
3769
3770found_it:
3771 if (hlock->instance != lock)
3772 return;
3773
3774 cpu = smp_processor_id();
3775 if (hlock->waittime_stamp) {
3776 now = lockstat_clock();
3777 waittime = now - hlock->waittime_stamp;
3778 hlock->holdtime_stamp = now;
3779 }
3780
3781 trace_lock_acquired(lock, ip);
3782
3783 stats = get_lock_stats(hlock_class(hlock));
3784 if (waittime) {
3785 if (hlock->read)
3786 lock_time_inc(&stats->read_waittime, waittime);
3787 else
3788 lock_time_inc(&stats->write_waittime, waittime);
3789 }
3790 if (lock->cpu != cpu)
3791 stats->bounces[bounce_acquired + !!hlock->read]++;
3792 put_lock_stats(stats);
3793
3794 lock->cpu = cpu;
3795 lock->ip = ip;
3796}
3797
3798void lock_contended(struct lockdep_map *lock, unsigned long ip)
3799{
3800 unsigned long flags;
3801
3802 if (unlikely(!lock_stat))
3803 return;
3804
3805 if (unlikely(current->lockdep_recursion))
3806 return;
3807
3808 raw_local_irq_save(flags);
3809 check_flags(flags);
3810 current->lockdep_recursion = 1;
3811 trace_lock_contended(lock, ip);
3812 __lock_contended(lock, ip);
3813 current->lockdep_recursion = 0;
3814 raw_local_irq_restore(flags);
3815}
3816EXPORT_SYMBOL_GPL(lock_contended);
3817
3818void lock_acquired(struct lockdep_map *lock, unsigned long ip)
3819{
3820 unsigned long flags;
3821
3822 if (unlikely(!lock_stat))
3823 return;
3824
3825 if (unlikely(current->lockdep_recursion))
3826 return;
3827
3828 raw_local_irq_save(flags);
3829 check_flags(flags);
3830 current->lockdep_recursion = 1;
3831 __lock_acquired(lock, ip);
3832 current->lockdep_recursion = 0;
3833 raw_local_irq_restore(flags);
3834}
3835EXPORT_SYMBOL_GPL(lock_acquired);
3836#endif
3837
3838/*
3839 * Used by the testsuite, sanitize the validator state
3840 * after a simulated failure:
3841 */
3842
3843void lockdep_reset(void)
3844{
3845 unsigned long flags;
3846 int i;
3847
3848 raw_local_irq_save(flags);
3849 current->curr_chain_key = 0;
3850 current->lockdep_depth = 0;
3851 current->lockdep_recursion = 0;
3852 memset(current->held_locks, 0, MAX_LOCK_DEPTH*sizeof(struct held_lock));
3853 nr_hardirq_chains = 0;
3854 nr_softirq_chains = 0;
3855 nr_process_chains = 0;
3856 debug_locks = 1;
3857 for (i = 0; i < CHAINHASH_SIZE; i++)
3858 INIT_LIST_HEAD(chainhash_table + i);
3859 raw_local_irq_restore(flags);
3860}
3861
3862static void zap_class(struct lock_class *class)
3863{
3864 int i;
3865
3866 /*
3867 * Remove all dependencies this lock is
3868 * involved in:
3869 */
3870 for (i = 0; i < nr_list_entries; i++) {
3871 if (list_entries[i].class == class)
3872 list_del_rcu(&list_entries[i].entry);
3873 }
3874 /*
3875 * Unhash the class and remove it from the all_lock_classes list:
3876 */
3877 list_del_rcu(&class->hash_entry);
3878 list_del_rcu(&class->lock_entry);
3879
3880 class->key = NULL;
3881}
3882
3883static inline int within(const void *addr, void *start, unsigned long size)
3884{
3885 return addr >= start && addr < start + size;
3886}
3887
3888void lockdep_free_key_range(void *start, unsigned long size)
3889{
3890 struct lock_class *class, *next;
3891 struct list_head *head;
3892 unsigned long flags;
3893 int i;
3894 int locked;
3895
3896 raw_local_irq_save(flags);
3897 locked = graph_lock();
3898
3899 /*
3900 * Unhash all classes that were created by this module:
3901 */
3902 for (i = 0; i < CLASSHASH_SIZE; i++) {
3903 head = classhash_table + i;
3904 if (list_empty(head))
3905 continue;
3906 list_for_each_entry_safe(class, next, head, hash_entry) {
3907 if (within(class->key, start, size))
3908 zap_class(class);
3909 else if (within(class->name, start, size))
3910 zap_class(class);
3911 }
3912 }
3913
3914 if (locked)
3915 graph_unlock();
3916 raw_local_irq_restore(flags);
3917}
3918
3919void lockdep_reset_lock(struct lockdep_map *lock)
3920{
3921 struct lock_class *class, *next;
3922 struct list_head *head;
3923 unsigned long flags;
3924 int i, j;
3925 int locked;
3926
3927 raw_local_irq_save(flags);
3928
3929 /*
3930 * Remove all classes this lock might have:
3931 */
3932 for (j = 0; j < MAX_LOCKDEP_SUBCLASSES; j++) {
3933 /*
3934 * If the class exists we look it up and zap it:
3935 */
3936 class = look_up_lock_class(lock, j);
3937 if (class)
3938 zap_class(class);
3939 }
3940 /*
3941 * Debug check: in the end all mapped classes should
3942 * be gone.
3943 */
3944 locked = graph_lock();
3945 for (i = 0; i < CLASSHASH_SIZE; i++) {
3946 head = classhash_table + i;
3947 if (list_empty(head))
3948 continue;
3949 list_for_each_entry_safe(class, next, head, hash_entry) {
3950 int match = 0;
3951
3952 for (j = 0; j < NR_LOCKDEP_CACHING_CLASSES; j++)
3953 match |= class == lock->class_cache[j];
3954
3955 if (unlikely(match)) {
3956 if (debug_locks_off_graph_unlock()) {
3957 /*
3958 * We all just reset everything, how did it match?
3959 */
3960 WARN_ON(1);
3961 }
3962 goto out_restore;
3963 }
3964 }
3965 }
3966 if (locked)
3967 graph_unlock();
3968
3969out_restore:
3970 raw_local_irq_restore(flags);
3971}
3972
3973void lockdep_init(void)
3974{
3975 int i;
3976
3977 /*
3978 * Some architectures have their own start_kernel()
3979 * code which calls lockdep_init(), while we also
3980 * call lockdep_init() from the start_kernel() itself,
3981 * and we want to initialize the hashes only once:
3982 */
3983 if (lockdep_initialized)
3984 return;
3985
3986 for (i = 0; i < CLASSHASH_SIZE; i++)
3987 INIT_LIST_HEAD(classhash_table + i);
3988
3989 for (i = 0; i < CHAINHASH_SIZE; i++)
3990 INIT_LIST_HEAD(chainhash_table + i);
3991
3992 lockdep_initialized = 1;
3993}
3994
3995void __init lockdep_info(void)
3996{
3997 printk("Lock dependency validator: Copyright (c) 2006 Red Hat, Inc., Ingo Molnar\n");
3998
3999 printk("... MAX_LOCKDEP_SUBCLASSES: %lu\n", MAX_LOCKDEP_SUBCLASSES);
4000 printk("... MAX_LOCK_DEPTH: %lu\n", MAX_LOCK_DEPTH);
4001 printk("... MAX_LOCKDEP_KEYS: %lu\n", MAX_LOCKDEP_KEYS);
4002 printk("... CLASSHASH_SIZE: %lu\n", CLASSHASH_SIZE);
4003 printk("... MAX_LOCKDEP_ENTRIES: %lu\n", MAX_LOCKDEP_ENTRIES);
4004 printk("... MAX_LOCKDEP_CHAINS: %lu\n", MAX_LOCKDEP_CHAINS);
4005 printk("... CHAINHASH_SIZE: %lu\n", CHAINHASH_SIZE);
4006
4007 printk(" memory used by lock dependency info: %lu kB\n",
4008 (sizeof(struct lock_class) * MAX_LOCKDEP_KEYS +
4009 sizeof(struct list_head) * CLASSHASH_SIZE +
4010 sizeof(struct lock_list) * MAX_LOCKDEP_ENTRIES +
4011 sizeof(struct lock_chain) * MAX_LOCKDEP_CHAINS +
4012 sizeof(struct list_head) * CHAINHASH_SIZE
4013#ifdef CONFIG_PROVE_LOCKING
4014 + sizeof(struct circular_queue)
4015#endif
4016 ) / 1024
4017 );
4018
4019 printk(" per task-struct memory footprint: %lu bytes\n",
4020 sizeof(struct held_lock) * MAX_LOCK_DEPTH);
4021
4022#ifdef CONFIG_DEBUG_LOCKDEP
4023 if (lockdep_init_error) {
4024 printk("WARNING: lockdep init error! lock-%s was acquired"
4025 "before lockdep_init\n", lock_init_error);
4026 printk("Call stack leading to lockdep invocation was:\n");
4027 print_stack_trace(&lockdep_init_trace, 0);
4028 }
4029#endif
4030}
4031
4032static void
4033print_freed_lock_bug(struct task_struct *curr, const void *mem_from,
4034 const void *mem_to, struct held_lock *hlock)
4035{
4036 if (!debug_locks_off())
4037 return;
4038 if (debug_locks_silent)
4039 return;
4040
4041 printk("\n");
4042 printk("=========================\n");
4043 printk("[ BUG: held lock freed! ]\n");
4044 print_kernel_ident();
4045 printk("-------------------------\n");
4046 printk("%s/%d is freeing memory %p-%p, with a lock still held there!\n",
4047 curr->comm, task_pid_nr(curr), mem_from, mem_to-1);
4048 print_lock(hlock);
4049 lockdep_print_held_locks(curr);
4050
4051 printk("\nstack backtrace:\n");
4052 dump_stack();
4053}
4054
4055static inline int not_in_range(const void* mem_from, unsigned long mem_len,
4056 const void* lock_from, unsigned long lock_len)
4057{
4058 return lock_from + lock_len <= mem_from ||
4059 mem_from + mem_len <= lock_from;
4060}
4061
4062/*
4063 * Called when kernel memory is freed (or unmapped), or if a lock
4064 * is destroyed or reinitialized - this code checks whether there is
4065 * any held lock in the memory range of <from> to <to>:
4066 */
4067void debug_check_no_locks_freed(const void *mem_from, unsigned long mem_len)
4068{
4069 struct task_struct *curr = current;
4070 struct held_lock *hlock;
4071 unsigned long flags;
4072 int i;
4073
4074 if (unlikely(!debug_locks))
4075 return;
4076
4077 local_irq_save(flags);
4078 for (i = 0; i < curr->lockdep_depth; i++) {
4079 hlock = curr->held_locks + i;
4080
4081 if (not_in_range(mem_from, mem_len, hlock->instance,
4082 sizeof(*hlock->instance)))
4083 continue;
4084
4085 print_freed_lock_bug(curr, mem_from, mem_from + mem_len, hlock);
4086 break;
4087 }
4088 local_irq_restore(flags);
4089}
4090EXPORT_SYMBOL_GPL(debug_check_no_locks_freed);
4091
4092static void print_held_locks_bug(void)
4093{
4094 if (!debug_locks_off())
4095 return;
4096 if (debug_locks_silent)
4097 return;
4098
4099 printk("\n");
4100 printk("=====================================\n");
4101 printk("[ BUG: %s/%d still has locks held! ]\n",
4102 current->comm, task_pid_nr(current));
4103 print_kernel_ident();
4104 printk("-------------------------------------\n");
4105 lockdep_print_held_locks(current);
4106 printk("\nstack backtrace:\n");
4107 dump_stack();
4108}
4109
4110void debug_check_no_locks_held(void)
4111{
4112 if (unlikely(current->lockdep_depth > 0))
4113 print_held_locks_bug();
4114}
4115EXPORT_SYMBOL_GPL(debug_check_no_locks_held);
4116
4117#ifdef __KERNEL__
4118void debug_show_all_locks(void)
4119{
4120 struct task_struct *g, *p;
4121 int count = 10;
4122 int unlock = 1;
4123
4124 if (unlikely(!debug_locks)) {
4125 printk("INFO: lockdep is turned off.\n");
4126 return;
4127 }
4128 printk("\nShowing all locks held in the system:\n");
4129
4130 /*
4131 * Here we try to get the tasklist_lock as hard as possible,
4132 * if not successful after 2 seconds we ignore it (but keep
4133 * trying). This is to enable a debug printout even if a
4134 * tasklist_lock-holding task deadlocks or crashes.
4135 */
4136retry:
4137 if (!read_trylock(&tasklist_lock)) {
4138 if (count == 10)
4139 printk("hm, tasklist_lock locked, retrying... ");
4140 if (count) {
4141 count--;
4142 printk(" #%d", 10-count);
4143 mdelay(200);
4144 goto retry;
4145 }
4146 printk(" ignoring it.\n");
4147 unlock = 0;
4148 } else {
4149 if (count != 10)
4150 printk(KERN_CONT " locked it.\n");
4151 }
4152
4153 do_each_thread(g, p) {
4154 /*
4155 * It's not reliable to print a task's held locks
4156 * if it's not sleeping (or if it's not the current
4157 * task):
4158 */
4159 if (p->state == TASK_RUNNING && p != current)
4160 continue;
4161 if (p->lockdep_depth)
4162 lockdep_print_held_locks(p);
4163 if (!unlock)
4164 if (read_trylock(&tasklist_lock))
4165 unlock = 1;
4166 } while_each_thread(g, p);
4167
4168 printk("\n");
4169 printk("=============================================\n\n");
4170
4171 if (unlock)
4172 read_unlock(&tasklist_lock);
4173}
4174EXPORT_SYMBOL_GPL(debug_show_all_locks);
4175#endif
4176
4177/*
4178 * Careful: only use this function if you are sure that
4179 * the task cannot run in parallel!
4180 */
4181void debug_show_held_locks(struct task_struct *task)
4182{
4183 if (unlikely(!debug_locks)) {
4184 printk("INFO: lockdep is turned off.\n");
4185 return;
4186 }
4187 lockdep_print_held_locks(task);
4188}
4189EXPORT_SYMBOL_GPL(debug_show_held_locks);
4190
4191asmlinkage __visible void lockdep_sys_exit(void)
4192{
4193 struct task_struct *curr = current;
4194
4195 if (unlikely(curr->lockdep_depth)) {
4196 if (!debug_locks_off())
4197 return;
4198 printk("\n");
4199 printk("================================================\n");
4200 printk("[ BUG: lock held when returning to user space! ]\n");
4201 print_kernel_ident();
4202 printk("------------------------------------------------\n");
4203 printk("%s/%d is leaving the kernel with locks still held!\n",
4204 curr->comm, curr->pid);
4205 lockdep_print_held_locks(curr);
4206 }
4207}
4208
4209void lockdep_rcu_suspicious(const char *file, const int line, const char *s)
4210{
4211 struct task_struct *curr = current;
4212
4213#ifndef CONFIG_PROVE_RCU_REPEATEDLY
4214 if (!debug_locks_off())
4215 return;
4216#endif /* #ifdef CONFIG_PROVE_RCU_REPEATEDLY */
4217 /* Note: the following can be executed concurrently, so be careful. */
4218 printk("\n");
4219 printk("===============================\n");
4220 printk("[ INFO: suspicious RCU usage. ]\n");
4221 print_kernel_ident();
4222 printk("-------------------------------\n");
4223 printk("%s:%d %s!\n", file, line, s);
4224 printk("\nother info that might help us debug this:\n\n");
4225 printk("\n%srcu_scheduler_active = %d, debug_locks = %d\n",
4226 !rcu_lockdep_current_cpu_online()
4227 ? "RCU used illegally from offline CPU!\n"
4228 : !rcu_is_watching()
4229 ? "RCU used illegally from idle CPU!\n"
4230 : "",
4231 rcu_scheduler_active, debug_locks);
4232
4233 /*
4234 * If a CPU is in the RCU-free window in idle (ie: in the section
4235 * between rcu_idle_enter() and rcu_idle_exit(), then RCU
4236 * considers that CPU to be in an "extended quiescent state",
4237 * which means that RCU will be completely ignoring that CPU.
4238 * Therefore, rcu_read_lock() and friends have absolutely no
4239 * effect on a CPU running in that state. In other words, even if
4240 * such an RCU-idle CPU has called rcu_read_lock(), RCU might well
4241 * delete data structures out from under it. RCU really has no
4242 * choice here: we need to keep an RCU-free window in idle where
4243 * the CPU may possibly enter into low power mode. This way we can
4244 * notice an extended quiescent state to other CPUs that started a grace
4245 * period. Otherwise we would delay any grace period as long as we run
4246 * in the idle task.
4247 *
4248 * So complain bitterly if someone does call rcu_read_lock(),
4249 * rcu_read_lock_bh() and so on from extended quiescent states.
4250 */
4251 if (!rcu_is_watching())
4252 printk("RCU used illegally from extended quiescent state!\n");
4253
4254 lockdep_print_held_locks(curr);
4255 printk("\nstack backtrace:\n");
4256 dump_stack();
4257}
4258EXPORT_SYMBOL_GPL(lockdep_rcu_suspicious);