4 * Runtime locking correctness validator
6 * Started by Ingo Molnar:
8 * Copyright (C) 2006,2007 Red Hat, Inc., Ingo Molnar <mingo@redhat.com>
9 * Copyright (C) 2007 Red Hat, Inc., Peter Zijlstra
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:
14 * - lock inversion scenarios
15 * - circular lock dependencies
16 * - hardirq/softirq safe/unsafe locking bugs
18 * Bugs are reported even if the current locking scenario does not cause
19 * any deadlock at this point.
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.
25 * Thanks to Arjan van de Ven for coming up with the initial idea of
26 * mapping lock dependencies runtime.
28 #define DISABLE_BRANCH_PROFILING
29 #include <linux/mutex.h>
30 #include <linux/sched.h>
31 #include <linux/sched/clock.h>
32 #include <linux/sched/task.h>
33 #include <linux/delay.h>
34 #include <linux/module.h>
35 #include <linux/proc_fs.h>
36 #include <linux/seq_file.h>
37 #include <linux/spinlock.h>
38 #include <linux/kallsyms.h>
39 #include <linux/interrupt.h>
40 #include <linux/stacktrace.h>
41 #include <linux/debug_locks.h>
42 #include <linux/irqflags.h>
43 #include <linux/utsname.h>
44 #include <linux/hash.h>
45 #include <linux/ftrace.h>
46 #include <linux/stringify.h>
47 #include <linux/bitops.h>
48 #include <linux/gfp.h>
49 #include <linux/kmemcheck.h>
50 #include <linux/random.h>
51 #include <linux/jhash.h>
53 #include <asm/sections.h>
55 #include "lockdep_internals.h"
57 #define CREATE_TRACE_POINTS
58 #include <trace/events/lock.h>
60 #ifdef CONFIG_PROVE_LOCKING
61 int prove_locking = 1;
62 module_param(prove_locking, int, 0644);
64 #define prove_locking 0
67 #ifdef CONFIG_LOCK_STAT
69 module_param(lock_stat, int, 0644);
75 * lockdep_lock: protects the lockdep graph, the hashes and the
76 * class/list/hash allocators.
78 * This is one of the rare exceptions where it's justified
79 * to use a raw spinlock - we really dont want the spinlock
80 * code to recurse back into the lockdep code...
82 static arch_spinlock_t lockdep_lock = (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
84 static int graph_lock(void)
86 arch_spin_lock(&lockdep_lock);
88 * Make sure that if another CPU detected a bug while
89 * walking the graph we dont change it (while the other
90 * CPU is busy printing out stuff with the graph lock
94 arch_spin_unlock(&lockdep_lock);
97 /* prevent any recursions within lockdep from causing deadlocks */
98 current->lockdep_recursion++;
102 static inline int graph_unlock(void)
104 if (debug_locks && !arch_spin_is_locked(&lockdep_lock)) {
106 * The lockdep graph lock isn't locked while we expect it to
107 * be, we're confused now, bye!
109 return DEBUG_LOCKS_WARN_ON(1);
112 current->lockdep_recursion--;
113 arch_spin_unlock(&lockdep_lock);
118 * Turn lock debugging off and return with 0 if it was off already,
119 * and also release the graph lock:
121 static inline int debug_locks_off_graph_unlock(void)
123 int ret = debug_locks_off();
125 arch_spin_unlock(&lockdep_lock);
130 unsigned long nr_list_entries;
131 static struct lock_list list_entries[MAX_LOCKDEP_ENTRIES];
134 * All data structures here are protected by the global debug_lock.
136 * Mutex key structs only get allocated, once during bootup, and never
137 * get freed - this significantly simplifies the debugging code.
139 unsigned long nr_lock_classes;
140 static struct lock_class lock_classes[MAX_LOCKDEP_KEYS];
142 static inline struct lock_class *hlock_class(struct held_lock *hlock)
144 if (!hlock->class_idx) {
146 * Someone passed in garbage, we give up.
148 DEBUG_LOCKS_WARN_ON(1);
151 return lock_classes + hlock->class_idx - 1;
154 #ifdef CONFIG_LOCK_STAT
155 static DEFINE_PER_CPU(struct lock_class_stats[MAX_LOCKDEP_KEYS], cpu_lock_stats);
157 static inline u64 lockstat_clock(void)
159 return local_clock();
162 static int lock_point(unsigned long points[], unsigned long ip)
166 for (i = 0; i < LOCKSTAT_POINTS; i++) {
167 if (points[i] == 0) {
178 static void lock_time_inc(struct lock_time *lt, u64 time)
183 if (time < lt->min || !lt->nr)
190 static inline void lock_time_add(struct lock_time *src, struct lock_time *dst)
195 if (src->max > dst->max)
198 if (src->min < dst->min || !dst->nr)
201 dst->total += src->total;
205 struct lock_class_stats lock_stats(struct lock_class *class)
207 struct lock_class_stats stats;
210 memset(&stats, 0, sizeof(struct lock_class_stats));
211 for_each_possible_cpu(cpu) {
212 struct lock_class_stats *pcs =
213 &per_cpu(cpu_lock_stats, cpu)[class - lock_classes];
215 for (i = 0; i < ARRAY_SIZE(stats.contention_point); i++)
216 stats.contention_point[i] += pcs->contention_point[i];
218 for (i = 0; i < ARRAY_SIZE(stats.contending_point); i++)
219 stats.contending_point[i] += pcs->contending_point[i];
221 lock_time_add(&pcs->read_waittime, &stats.read_waittime);
222 lock_time_add(&pcs->write_waittime, &stats.write_waittime);
224 lock_time_add(&pcs->read_holdtime, &stats.read_holdtime);
225 lock_time_add(&pcs->write_holdtime, &stats.write_holdtime);
227 for (i = 0; i < ARRAY_SIZE(stats.bounces); i++)
228 stats.bounces[i] += pcs->bounces[i];
234 void clear_lock_stats(struct lock_class *class)
238 for_each_possible_cpu(cpu) {
239 struct lock_class_stats *cpu_stats =
240 &per_cpu(cpu_lock_stats, cpu)[class - lock_classes];
242 memset(cpu_stats, 0, sizeof(struct lock_class_stats));
244 memset(class->contention_point, 0, sizeof(class->contention_point));
245 memset(class->contending_point, 0, sizeof(class->contending_point));
248 static struct lock_class_stats *get_lock_stats(struct lock_class *class)
250 return &get_cpu_var(cpu_lock_stats)[class - lock_classes];
253 static void put_lock_stats(struct lock_class_stats *stats)
255 put_cpu_var(cpu_lock_stats);
258 static void lock_release_holdtime(struct held_lock *hlock)
260 struct lock_class_stats *stats;
266 holdtime = lockstat_clock() - hlock->holdtime_stamp;
268 stats = get_lock_stats(hlock_class(hlock));
270 lock_time_inc(&stats->read_holdtime, holdtime);
272 lock_time_inc(&stats->write_holdtime, holdtime);
273 put_lock_stats(stats);
276 static inline void lock_release_holdtime(struct held_lock *hlock)
282 * We keep a global list of all lock classes. The list only grows,
283 * never shrinks. The list is only accessed with the lockdep
284 * spinlock lock held.
286 LIST_HEAD(all_lock_classes);
289 * The lockdep classes are in a hash-table as well, for fast lookup:
291 #define CLASSHASH_BITS (MAX_LOCKDEP_KEYS_BITS - 1)
292 #define CLASSHASH_SIZE (1UL << CLASSHASH_BITS)
293 #define __classhashfn(key) hash_long((unsigned long)key, CLASSHASH_BITS)
294 #define classhashentry(key) (classhash_table + __classhashfn((key)))
296 static struct hlist_head classhash_table[CLASSHASH_SIZE];
299 * We put the lock dependency chains into a hash-table as well, to cache
302 #define CHAINHASH_BITS (MAX_LOCKDEP_CHAINS_BITS-1)
303 #define CHAINHASH_SIZE (1UL << CHAINHASH_BITS)
304 #define __chainhashfn(chain) hash_long(chain, CHAINHASH_BITS)
305 #define chainhashentry(chain) (chainhash_table + __chainhashfn((chain)))
307 static struct hlist_head chainhash_table[CHAINHASH_SIZE];
310 * The hash key of the lock dependency chains is a hash itself too:
311 * it's a hash of all locks taken up to that lock, including that lock.
312 * It's a 64-bit hash, because it's important for the keys to be
315 static inline u64 iterate_chain_key(u64 key, u32 idx)
317 u32 k0 = key, k1 = key >> 32;
319 __jhash_mix(idx, k0, k1); /* Macro that modifies arguments! */
321 return k0 | (u64)k1 << 32;
324 void lockdep_off(void)
326 current->lockdep_recursion++;
328 EXPORT_SYMBOL(lockdep_off);
330 void lockdep_on(void)
332 current->lockdep_recursion--;
334 EXPORT_SYMBOL(lockdep_on);
337 * Debugging switches:
341 #define VERY_VERBOSE 0
344 # define HARDIRQ_VERBOSE 1
345 # define SOFTIRQ_VERBOSE 1
346 # define RECLAIM_VERBOSE 1
348 # define HARDIRQ_VERBOSE 0
349 # define SOFTIRQ_VERBOSE 0
350 # define RECLAIM_VERBOSE 0
353 #if VERBOSE || HARDIRQ_VERBOSE || SOFTIRQ_VERBOSE || RECLAIM_VERBOSE
355 * Quick filtering for interesting events:
357 static int class_filter(struct lock_class *class)
361 if (class->name_version == 1 &&
362 !strcmp(class->name, "lockname"))
364 if (class->name_version == 1 &&
365 !strcmp(class->name, "&struct->lockfield"))
368 /* Filter everything else. 1 would be to allow everything else */
373 static int verbose(struct lock_class *class)
376 return class_filter(class);
382 * Stack-trace: tightly packed array of stack backtrace
383 * addresses. Protected by the graph_lock.
385 unsigned long nr_stack_trace_entries;
386 static unsigned long stack_trace[MAX_STACK_TRACE_ENTRIES];
388 static void print_lockdep_off(const char *bug_msg)
390 printk(KERN_DEBUG "%s\n", bug_msg);
391 printk(KERN_DEBUG "turning off the locking correctness validator.\n");
392 #ifdef CONFIG_LOCK_STAT
393 printk(KERN_DEBUG "Please attach the output of /proc/lock_stat to the bug report\n");
397 static int save_trace(struct stack_trace *trace)
399 trace->nr_entries = 0;
400 trace->max_entries = MAX_STACK_TRACE_ENTRIES - nr_stack_trace_entries;
401 trace->entries = stack_trace + nr_stack_trace_entries;
405 save_stack_trace(trace);
408 * Some daft arches put -1 at the end to indicate its a full trace.
410 * <rant> this is buggy anyway, since it takes a whole extra entry so a
411 * complete trace that maxes out the entries provided will be reported
412 * as incomplete, friggin useless </rant>
414 if (trace->nr_entries != 0 &&
415 trace->entries[trace->nr_entries-1] == ULONG_MAX)
418 trace->max_entries = trace->nr_entries;
420 nr_stack_trace_entries += trace->nr_entries;
422 if (nr_stack_trace_entries >= MAX_STACK_TRACE_ENTRIES-1) {
423 if (!debug_locks_off_graph_unlock())
426 print_lockdep_off("BUG: MAX_STACK_TRACE_ENTRIES too low!");
435 unsigned int nr_hardirq_chains;
436 unsigned int nr_softirq_chains;
437 unsigned int nr_process_chains;
438 unsigned int max_lockdep_depth;
440 #ifdef CONFIG_DEBUG_LOCKDEP
442 * Various lockdep statistics:
444 DEFINE_PER_CPU(struct lockdep_stats, lockdep_stats);
451 #define __USAGE(__STATE) \
452 [LOCK_USED_IN_##__STATE] = "IN-"__stringify(__STATE)"-W", \
453 [LOCK_ENABLED_##__STATE] = __stringify(__STATE)"-ON-W", \
454 [LOCK_USED_IN_##__STATE##_READ] = "IN-"__stringify(__STATE)"-R",\
455 [LOCK_ENABLED_##__STATE##_READ] = __stringify(__STATE)"-ON-R",
457 static const char *usage_str[] =
459 #define LOCKDEP_STATE(__STATE) __USAGE(__STATE)
460 #include "lockdep_states.h"
462 [LOCK_USED] = "INITIAL USE",
465 const char * __get_key_name(struct lockdep_subclass_key *key, char *str)
467 return kallsyms_lookup((unsigned long)key, NULL, NULL, NULL, str);
470 static inline unsigned long lock_flag(enum lock_usage_bit bit)
475 static char get_usage_char(struct lock_class *class, enum lock_usage_bit bit)
479 if (class->usage_mask & lock_flag(bit + 2))
481 if (class->usage_mask & lock_flag(bit)) {
483 if (class->usage_mask & lock_flag(bit + 2))
490 void get_usage_chars(struct lock_class *class, char usage[LOCK_USAGE_CHARS])
494 #define LOCKDEP_STATE(__STATE) \
495 usage[i++] = get_usage_char(class, LOCK_USED_IN_##__STATE); \
496 usage[i++] = get_usage_char(class, LOCK_USED_IN_##__STATE##_READ);
497 #include "lockdep_states.h"
503 static void __print_lock_name(struct lock_class *class)
505 char str[KSYM_NAME_LEN];
510 name = __get_key_name(class->key, str);
511 printk(KERN_CONT "%s", name);
513 printk(KERN_CONT "%s", name);
514 if (class->name_version > 1)
515 printk(KERN_CONT "#%d", class->name_version);
517 printk(KERN_CONT "/%d", class->subclass);
521 static void print_lock_name(struct lock_class *class)
523 char usage[LOCK_USAGE_CHARS];
525 get_usage_chars(class, usage);
527 printk(KERN_CONT " (");
528 __print_lock_name(class);
529 printk(KERN_CONT "){%s}", usage);
532 static void print_lockdep_cache(struct lockdep_map *lock)
535 char str[KSYM_NAME_LEN];
539 name = __get_key_name(lock->key->subkeys, str);
541 printk(KERN_CONT "%s", name);
544 static void print_lock(struct held_lock *hlock)
547 * We can be called locklessly through debug_show_all_locks() so be
548 * extra careful, the hlock might have been released and cleared.
550 unsigned int class_idx = hlock->class_idx;
552 /* Don't re-read hlock->class_idx, can't use READ_ONCE() on bitfields: */
555 if (!class_idx || (class_idx - 1) >= MAX_LOCKDEP_KEYS) {
556 printk(KERN_CONT "<RELEASED>\n");
560 print_lock_name(lock_classes + class_idx - 1);
561 printk(KERN_CONT ", at: [<%p>] %pS\n",
562 (void *)hlock->acquire_ip, (void *)hlock->acquire_ip);
565 static void lockdep_print_held_locks(struct task_struct *curr)
567 int i, depth = curr->lockdep_depth;
570 printk("no locks held by %s/%d.\n", curr->comm, task_pid_nr(curr));
573 printk("%d lock%s held by %s/%d:\n",
574 depth, depth > 1 ? "s" : "", curr->comm, task_pid_nr(curr));
576 for (i = 0; i < depth; i++) {
578 print_lock(curr->held_locks + i);
582 static void print_kernel_ident(void)
584 printk("%s %.*s %s\n", init_utsname()->release,
585 (int)strcspn(init_utsname()->version, " "),
586 init_utsname()->version,
590 static int very_verbose(struct lock_class *class)
593 return class_filter(class);
599 * Is this the address of a static object:
602 static int static_obj(void *obj)
604 unsigned long start = (unsigned long) &_stext,
605 end = (unsigned long) &_end,
606 addr = (unsigned long) obj;
611 if ((addr >= start) && (addr < end))
614 if (arch_is_kernel_data(addr))
618 * in-kernel percpu var?
620 if (is_kernel_percpu_address(addr))
624 * module static or percpu var?
626 return is_module_address(addr) || is_module_percpu_address(addr);
631 * To make lock name printouts unique, we calculate a unique
632 * class->name_version generation counter:
634 static int count_matching_names(struct lock_class *new_class)
636 struct lock_class *class;
639 if (!new_class->name)
642 list_for_each_entry_rcu(class, &all_lock_classes, lock_entry) {
643 if (new_class->key - new_class->subclass == class->key)
644 return class->name_version;
645 if (class->name && !strcmp(class->name, new_class->name))
646 count = max(count, class->name_version);
653 * Register a lock's class in the hash-table, if the class is not present
654 * yet. Otherwise we look it up. We cache the result in the lock object
655 * itself, so actual lookup of the hash should be once per lock object.
657 static inline struct lock_class *
658 look_up_lock_class(struct lockdep_map *lock, unsigned int subclass)
660 struct lockdep_subclass_key *key;
661 struct hlist_head *hash_head;
662 struct lock_class *class;
663 bool is_static = false;
665 if (unlikely(subclass >= MAX_LOCKDEP_SUBCLASSES)) {
668 "BUG: looking up invalid subclass: %u\n", subclass);
670 "turning off the locking correctness validator.\n");
676 * Static locks do not have their class-keys yet - for them the key
677 * is the lock object itself. If the lock is in the per cpu area,
678 * the canonical address of the lock (per cpu offset removed) is
681 if (unlikely(!lock->key)) {
682 unsigned long can_addr, addr = (unsigned long)lock;
684 if (__is_kernel_percpu_address(addr, &can_addr))
685 lock->key = (void *)can_addr;
686 else if (__is_module_percpu_address(addr, &can_addr))
687 lock->key = (void *)can_addr;
688 else if (static_obj(lock))
689 lock->key = (void *)lock;
691 return ERR_PTR(-EINVAL);
696 * NOTE: the class-key must be unique. For dynamic locks, a static
697 * lock_class_key variable is passed in through the mutex_init()
698 * (or spin_lock_init()) call - which acts as the key. For static
699 * locks we use the lock object itself as the key.
701 BUILD_BUG_ON(sizeof(struct lock_class_key) >
702 sizeof(struct lockdep_map));
704 key = lock->key->subkeys + subclass;
706 hash_head = classhashentry(key);
709 * We do an RCU walk of the hash, see lockdep_free_key_range().
711 if (DEBUG_LOCKS_WARN_ON(!irqs_disabled()))
714 hlist_for_each_entry_rcu(class, hash_head, hash_entry) {
715 if (class->key == key) {
717 * Huh! same key, different name? Did someone trample
718 * on some memory? We're most confused.
720 WARN_ON_ONCE(class->name != lock->name);
725 return is_static || static_obj(lock->key) ? NULL : ERR_PTR(-EINVAL);
729 * Register a lock's class in the hash-table, if the class is not present
730 * yet. Otherwise we look it up. We cache the result in the lock object
731 * itself, so actual lookup of the hash should be once per lock object.
733 static struct lock_class *
734 register_lock_class(struct lockdep_map *lock, unsigned int subclass, int force)
736 struct lockdep_subclass_key *key;
737 struct hlist_head *hash_head;
738 struct lock_class *class;
740 DEBUG_LOCKS_WARN_ON(!irqs_disabled());
742 class = look_up_lock_class(lock, subclass);
743 if (likely(!IS_ERR_OR_NULL(class)))
744 goto out_set_class_cache;
747 * Debug-check: all keys must be persistent!
751 printk("INFO: trying to register non-static key.\n");
752 printk("the code is fine but needs lockdep annotation.\n");
753 printk("turning off the locking correctness validator.\n");
758 key = lock->key->subkeys + subclass;
759 hash_head = classhashentry(key);
765 * We have to do the hash-walk again, to avoid races
768 hlist_for_each_entry_rcu(class, hash_head, hash_entry) {
769 if (class->key == key)
774 * Allocate a new key from the static array, and add it to
777 if (nr_lock_classes >= MAX_LOCKDEP_KEYS) {
778 if (!debug_locks_off_graph_unlock()) {
782 print_lockdep_off("BUG: MAX_LOCKDEP_KEYS too low!");
786 class = lock_classes + nr_lock_classes++;
787 debug_atomic_inc(nr_unused_locks);
789 class->name = lock->name;
790 class->subclass = subclass;
791 INIT_LIST_HEAD(&class->lock_entry);
792 INIT_LIST_HEAD(&class->locks_before);
793 INIT_LIST_HEAD(&class->locks_after);
794 class->name_version = count_matching_names(class);
796 * We use RCU's safe list-add method to make
797 * parallel walking of the hash-list safe:
799 hlist_add_head_rcu(&class->hash_entry, hash_head);
801 * Add it to the global list of classes:
803 list_add_tail_rcu(&class->lock_entry, &all_lock_classes);
805 if (verbose(class)) {
808 printk("\nnew class %p: %s", class->key, class->name);
809 if (class->name_version > 1)
810 printk(KERN_CONT "#%d", class->name_version);
811 printk(KERN_CONT "\n");
822 if (!subclass || force)
823 lock->class_cache[0] = class;
824 else if (subclass < NR_LOCKDEP_CACHING_CLASSES)
825 lock->class_cache[subclass] = class;
828 * Hash collision, did we smoke some? We found a class with a matching
829 * hash but the subclass -- which is hashed in -- didn't match.
831 if (DEBUG_LOCKS_WARN_ON(class->subclass != subclass))
837 #ifdef CONFIG_PROVE_LOCKING
839 * Allocate a lockdep entry. (assumes the graph_lock held, returns
840 * with NULL on failure)
842 static struct lock_list *alloc_list_entry(void)
844 if (nr_list_entries >= MAX_LOCKDEP_ENTRIES) {
845 if (!debug_locks_off_graph_unlock())
848 print_lockdep_off("BUG: MAX_LOCKDEP_ENTRIES too low!");
852 return list_entries + nr_list_entries++;
856 * Add a new dependency to the head of the list:
858 static int add_lock_to_list(struct lock_class *this, struct list_head *head,
859 unsigned long ip, int distance,
860 struct stack_trace *trace)
862 struct lock_list *entry;
864 * Lock not present yet - get a new dependency struct and
865 * add it to the list:
867 entry = alloc_list_entry();
872 entry->distance = distance;
873 entry->trace = *trace;
875 * Both allocation and removal are done under the graph lock; but
876 * iteration is under RCU-sched; see look_up_lock_class() and
877 * lockdep_free_key_range().
879 list_add_tail_rcu(&entry->entry, head);
885 * For good efficiency of modular, we use power of 2
887 #define MAX_CIRCULAR_QUEUE_SIZE 4096UL
888 #define CQ_MASK (MAX_CIRCULAR_QUEUE_SIZE-1)
891 * The circular_queue and helpers is used to implement the
892 * breadth-first search(BFS)algorithem, by which we can build
893 * the shortest path from the next lock to be acquired to the
894 * previous held lock if there is a circular between them.
896 struct circular_queue {
897 unsigned long element[MAX_CIRCULAR_QUEUE_SIZE];
898 unsigned int front, rear;
901 static struct circular_queue lock_cq;
903 unsigned int max_bfs_queue_depth;
905 static unsigned int lockdep_dependency_gen_id;
907 static inline void __cq_init(struct circular_queue *cq)
909 cq->front = cq->rear = 0;
910 lockdep_dependency_gen_id++;
913 static inline int __cq_empty(struct circular_queue *cq)
915 return (cq->front == cq->rear);
918 static inline int __cq_full(struct circular_queue *cq)
920 return ((cq->rear + 1) & CQ_MASK) == cq->front;
923 static inline int __cq_enqueue(struct circular_queue *cq, unsigned long elem)
928 cq->element[cq->rear] = elem;
929 cq->rear = (cq->rear + 1) & CQ_MASK;
933 static inline int __cq_dequeue(struct circular_queue *cq, unsigned long *elem)
938 *elem = cq->element[cq->front];
939 cq->front = (cq->front + 1) & CQ_MASK;
943 static inline unsigned int __cq_get_elem_count(struct circular_queue *cq)
945 return (cq->rear - cq->front) & CQ_MASK;
948 static inline void mark_lock_accessed(struct lock_list *lock,
949 struct lock_list *parent)
953 nr = lock - list_entries;
954 WARN_ON(nr >= nr_list_entries); /* Out-of-bounds, input fail */
955 lock->parent = parent;
956 lock->class->dep_gen_id = lockdep_dependency_gen_id;
959 static inline unsigned long lock_accessed(struct lock_list *lock)
963 nr = lock - list_entries;
964 WARN_ON(nr >= nr_list_entries); /* Out-of-bounds, input fail */
965 return lock->class->dep_gen_id == lockdep_dependency_gen_id;
968 static inline struct lock_list *get_lock_parent(struct lock_list *child)
970 return child->parent;
973 static inline int get_lock_depth(struct lock_list *child)
976 struct lock_list *parent;
978 while ((parent = get_lock_parent(child))) {
985 static int __bfs(struct lock_list *source_entry,
987 int (*match)(struct lock_list *entry, void *data),
988 struct lock_list **target_entry,
991 struct lock_list *entry;
992 struct list_head *head;
993 struct circular_queue *cq = &lock_cq;
996 if (match(source_entry, data)) {
997 *target_entry = source_entry;
1003 head = &source_entry->class->locks_after;
1005 head = &source_entry->class->locks_before;
1007 if (list_empty(head))
1011 __cq_enqueue(cq, (unsigned long)source_entry);
1013 while (!__cq_empty(cq)) {
1014 struct lock_list *lock;
1016 __cq_dequeue(cq, (unsigned long *)&lock);
1024 head = &lock->class->locks_after;
1026 head = &lock->class->locks_before;
1028 DEBUG_LOCKS_WARN_ON(!irqs_disabled());
1030 list_for_each_entry_rcu(entry, head, entry) {
1031 if (!lock_accessed(entry)) {
1032 unsigned int cq_depth;
1033 mark_lock_accessed(entry, lock);
1034 if (match(entry, data)) {
1035 *target_entry = entry;
1040 if (__cq_enqueue(cq, (unsigned long)entry)) {
1044 cq_depth = __cq_get_elem_count(cq);
1045 if (max_bfs_queue_depth < cq_depth)
1046 max_bfs_queue_depth = cq_depth;
1054 static inline int __bfs_forwards(struct lock_list *src_entry,
1056 int (*match)(struct lock_list *entry, void *data),
1057 struct lock_list **target_entry)
1059 return __bfs(src_entry, data, match, target_entry, 1);
1063 static inline int __bfs_backwards(struct lock_list *src_entry,
1065 int (*match)(struct lock_list *entry, void *data),
1066 struct lock_list **target_entry)
1068 return __bfs(src_entry, data, match, target_entry, 0);
1073 * Recursive, forwards-direction lock-dependency checking, used for
1074 * both noncyclic checking and for hardirq-unsafe/softirq-unsafe
1079 * Print a dependency chain entry (this is only done when a deadlock
1080 * has been detected):
1083 print_circular_bug_entry(struct lock_list *target, int depth)
1085 if (debug_locks_silent)
1087 printk("\n-> #%u", depth);
1088 print_lock_name(target->class);
1089 printk(KERN_CONT ":\n");
1090 print_stack_trace(&target->trace, 6);
1096 print_circular_lock_scenario(struct held_lock *src,
1097 struct held_lock *tgt,
1098 struct lock_list *prt)
1100 struct lock_class *source = hlock_class(src);
1101 struct lock_class *target = hlock_class(tgt);
1102 struct lock_class *parent = prt->class;
1105 * A direct locking problem where unsafe_class lock is taken
1106 * directly by safe_class lock, then all we need to show
1107 * is the deadlock scenario, as it is obvious that the
1108 * unsafe lock is taken under the safe lock.
1110 * But if there is a chain instead, where the safe lock takes
1111 * an intermediate lock (middle_class) where this lock is
1112 * not the same as the safe lock, then the lock chain is
1113 * used to describe the problem. Otherwise we would need
1114 * to show a different CPU case for each link in the chain
1115 * from the safe_class lock to the unsafe_class lock.
1117 if (parent != source) {
1118 printk("Chain exists of:\n ");
1119 __print_lock_name(source);
1120 printk(KERN_CONT " --> ");
1121 __print_lock_name(parent);
1122 printk(KERN_CONT " --> ");
1123 __print_lock_name(target);
1124 printk(KERN_CONT "\n\n");
1127 printk(" Possible unsafe locking scenario:\n\n");
1128 printk(" CPU0 CPU1\n");
1129 printk(" ---- ----\n");
1131 __print_lock_name(target);
1132 printk(KERN_CONT ");\n");
1134 __print_lock_name(parent);
1135 printk(KERN_CONT ");\n");
1137 __print_lock_name(target);
1138 printk(KERN_CONT ");\n");
1140 __print_lock_name(source);
1141 printk(KERN_CONT ");\n");
1142 printk("\n *** DEADLOCK ***\n\n");
1146 * When a circular dependency is detected, print the
1150 print_circular_bug_header(struct lock_list *entry, unsigned int depth,
1151 struct held_lock *check_src,
1152 struct held_lock *check_tgt)
1154 struct task_struct *curr = current;
1156 if (debug_locks_silent)
1160 printk("======================================================\n");
1161 printk("[ INFO: possible circular locking dependency detected ]\n");
1162 print_kernel_ident();
1163 printk("-------------------------------------------------------\n");
1164 printk("%s/%d is trying to acquire lock:\n",
1165 curr->comm, task_pid_nr(curr));
1166 print_lock(check_src);
1167 printk("\nbut task is already holding lock:\n");
1168 print_lock(check_tgt);
1169 printk("\nwhich lock already depends on the new lock.\n\n");
1170 printk("\nthe existing dependency chain (in reverse order) is:\n");
1172 print_circular_bug_entry(entry, depth);
1177 static inline int class_equal(struct lock_list *entry, void *data)
1179 return entry->class == data;
1182 static noinline int print_circular_bug(struct lock_list *this,
1183 struct lock_list *target,
1184 struct held_lock *check_src,
1185 struct held_lock *check_tgt)
1187 struct task_struct *curr = current;
1188 struct lock_list *parent;
1189 struct lock_list *first_parent;
1192 if (!debug_locks_off_graph_unlock() || debug_locks_silent)
1195 if (!save_trace(&this->trace))
1198 depth = get_lock_depth(target);
1200 print_circular_bug_header(target, depth, check_src, check_tgt);
1202 parent = get_lock_parent(target);
1203 first_parent = parent;
1206 print_circular_bug_entry(parent, --depth);
1207 parent = get_lock_parent(parent);
1210 printk("\nother info that might help us debug this:\n\n");
1211 print_circular_lock_scenario(check_src, check_tgt,
1214 lockdep_print_held_locks(curr);
1216 printk("\nstack backtrace:\n");
1222 static noinline int print_bfs_bug(int ret)
1224 if (!debug_locks_off_graph_unlock())
1228 * Breadth-first-search failed, graph got corrupted?
1230 WARN(1, "lockdep bfs error:%d\n", ret);
1235 static int noop_count(struct lock_list *entry, void *data)
1237 (*(unsigned long *)data)++;
1241 static unsigned long __lockdep_count_forward_deps(struct lock_list *this)
1243 unsigned long count = 0;
1244 struct lock_list *uninitialized_var(target_entry);
1246 __bfs_forwards(this, (void *)&count, noop_count, &target_entry);
1250 unsigned long lockdep_count_forward_deps(struct lock_class *class)
1252 unsigned long ret, flags;
1253 struct lock_list this;
1258 local_irq_save(flags);
1259 arch_spin_lock(&lockdep_lock);
1260 ret = __lockdep_count_forward_deps(&this);
1261 arch_spin_unlock(&lockdep_lock);
1262 local_irq_restore(flags);
1267 static unsigned long __lockdep_count_backward_deps(struct lock_list *this)
1269 unsigned long count = 0;
1270 struct lock_list *uninitialized_var(target_entry);
1272 __bfs_backwards(this, (void *)&count, noop_count, &target_entry);
1277 unsigned long lockdep_count_backward_deps(struct lock_class *class)
1279 unsigned long ret, flags;
1280 struct lock_list this;
1285 local_irq_save(flags);
1286 arch_spin_lock(&lockdep_lock);
1287 ret = __lockdep_count_backward_deps(&this);
1288 arch_spin_unlock(&lockdep_lock);
1289 local_irq_restore(flags);
1295 * Prove that the dependency graph starting at <entry> can not
1296 * lead to <target>. Print an error and return 0 if it does.
1299 check_noncircular(struct lock_list *root, struct lock_class *target,
1300 struct lock_list **target_entry)
1304 debug_atomic_inc(nr_cyclic_checks);
1306 result = __bfs_forwards(root, target, class_equal, target_entry);
1311 #if defined(CONFIG_TRACE_IRQFLAGS) && defined(CONFIG_PROVE_LOCKING)
1313 * Forwards and backwards subgraph searching, for the purposes of
1314 * proving that two subgraphs can be connected by a new dependency
1315 * without creating any illegal irq-safe -> irq-unsafe lock dependency.
1318 static inline int usage_match(struct lock_list *entry, void *bit)
1320 return entry->class->usage_mask & (1 << (enum lock_usage_bit)bit);
1326 * Find a node in the forwards-direction dependency sub-graph starting
1327 * at @root->class that matches @bit.
1329 * Return 0 if such a node exists in the subgraph, and put that node
1330 * into *@target_entry.
1332 * Return 1 otherwise and keep *@target_entry unchanged.
1333 * Return <0 on error.
1336 find_usage_forwards(struct lock_list *root, enum lock_usage_bit bit,
1337 struct lock_list **target_entry)
1341 debug_atomic_inc(nr_find_usage_forwards_checks);
1343 result = __bfs_forwards(root, (void *)bit, usage_match, target_entry);
1349 * Find a node in the backwards-direction dependency sub-graph starting
1350 * at @root->class that matches @bit.
1352 * Return 0 if such a node exists in the subgraph, and put that node
1353 * into *@target_entry.
1355 * Return 1 otherwise and keep *@target_entry unchanged.
1356 * Return <0 on error.
1359 find_usage_backwards(struct lock_list *root, enum lock_usage_bit bit,
1360 struct lock_list **target_entry)
1364 debug_atomic_inc(nr_find_usage_backwards_checks);
1366 result = __bfs_backwards(root, (void *)bit, usage_match, target_entry);
1371 static void print_lock_class_header(struct lock_class *class, int depth)
1375 printk("%*s->", depth, "");
1376 print_lock_name(class);
1377 printk(KERN_CONT " ops: %lu", class->ops);
1378 printk(KERN_CONT " {\n");
1380 for (bit = 0; bit < LOCK_USAGE_STATES; bit++) {
1381 if (class->usage_mask & (1 << bit)) {
1384 len += printk("%*s %s", depth, "", usage_str[bit]);
1385 len += printk(KERN_CONT " at:\n");
1386 print_stack_trace(class->usage_traces + bit, len);
1389 printk("%*s }\n", depth, "");
1391 printk("%*s ... key at: [<%p>] %pS\n",
1392 depth, "", class->key, class->key);
1396 * printk the shortest lock dependencies from @start to @end in reverse order:
1399 print_shortest_lock_dependencies(struct lock_list *leaf,
1400 struct lock_list *root)
1402 struct lock_list *entry = leaf;
1405 /*compute depth from generated tree by BFS*/
1406 depth = get_lock_depth(leaf);
1409 print_lock_class_header(entry->class, depth);
1410 printk("%*s ... acquired at:\n", depth, "");
1411 print_stack_trace(&entry->trace, 2);
1414 if (depth == 0 && (entry != root)) {
1415 printk("lockdep:%s bad path found in chain graph\n", __func__);
1419 entry = get_lock_parent(entry);
1421 } while (entry && (depth >= 0));
1427 print_irq_lock_scenario(struct lock_list *safe_entry,
1428 struct lock_list *unsafe_entry,
1429 struct lock_class *prev_class,
1430 struct lock_class *next_class)
1432 struct lock_class *safe_class = safe_entry->class;
1433 struct lock_class *unsafe_class = unsafe_entry->class;
1434 struct lock_class *middle_class = prev_class;
1436 if (middle_class == safe_class)
1437 middle_class = next_class;
1440 * A direct locking problem where unsafe_class lock is taken
1441 * directly by safe_class lock, then all we need to show
1442 * is the deadlock scenario, as it is obvious that the
1443 * unsafe lock is taken under the safe lock.
1445 * But if there is a chain instead, where the safe lock takes
1446 * an intermediate lock (middle_class) where this lock is
1447 * not the same as the safe lock, then the lock chain is
1448 * used to describe the problem. Otherwise we would need
1449 * to show a different CPU case for each link in the chain
1450 * from the safe_class lock to the unsafe_class lock.
1452 if (middle_class != unsafe_class) {
1453 printk("Chain exists of:\n ");
1454 __print_lock_name(safe_class);
1455 printk(KERN_CONT " --> ");
1456 __print_lock_name(middle_class);
1457 printk(KERN_CONT " --> ");
1458 __print_lock_name(unsafe_class);
1459 printk(KERN_CONT "\n\n");
1462 printk(" Possible interrupt unsafe locking scenario:\n\n");
1463 printk(" CPU0 CPU1\n");
1464 printk(" ---- ----\n");
1466 __print_lock_name(unsafe_class);
1467 printk(KERN_CONT ");\n");
1468 printk(" local_irq_disable();\n");
1470 __print_lock_name(safe_class);
1471 printk(KERN_CONT ");\n");
1473 __print_lock_name(middle_class);
1474 printk(KERN_CONT ");\n");
1475 printk(" <Interrupt>\n");
1477 __print_lock_name(safe_class);
1478 printk(KERN_CONT ");\n");
1479 printk("\n *** DEADLOCK ***\n\n");
1483 print_bad_irq_dependency(struct task_struct *curr,
1484 struct lock_list *prev_root,
1485 struct lock_list *next_root,
1486 struct lock_list *backwards_entry,
1487 struct lock_list *forwards_entry,
1488 struct held_lock *prev,
1489 struct held_lock *next,
1490 enum lock_usage_bit bit1,
1491 enum lock_usage_bit bit2,
1492 const char *irqclass)
1494 if (!debug_locks_off_graph_unlock() || debug_locks_silent)
1498 printk("======================================================\n");
1499 printk("[ INFO: %s-safe -> %s-unsafe lock order detected ]\n",
1500 irqclass, irqclass);
1501 print_kernel_ident();
1502 printk("------------------------------------------------------\n");
1503 printk("%s/%d [HC%u[%lu]:SC%u[%lu]:HE%u:SE%u] is trying to acquire:\n",
1504 curr->comm, task_pid_nr(curr),
1505 curr->hardirq_context, hardirq_count() >> HARDIRQ_SHIFT,
1506 curr->softirq_context, softirq_count() >> SOFTIRQ_SHIFT,
1507 curr->hardirqs_enabled,
1508 curr->softirqs_enabled);
1511 printk("\nand this task is already holding:\n");
1513 printk("which would create a new lock dependency:\n");
1514 print_lock_name(hlock_class(prev));
1515 printk(KERN_CONT " ->");
1516 print_lock_name(hlock_class(next));
1517 printk(KERN_CONT "\n");
1519 printk("\nbut this new dependency connects a %s-irq-safe lock:\n",
1521 print_lock_name(backwards_entry->class);
1522 printk("\n... which became %s-irq-safe at:\n", irqclass);
1524 print_stack_trace(backwards_entry->class->usage_traces + bit1, 1);
1526 printk("\nto a %s-irq-unsafe lock:\n", irqclass);
1527 print_lock_name(forwards_entry->class);
1528 printk("\n... which became %s-irq-unsafe at:\n", irqclass);
1531 print_stack_trace(forwards_entry->class->usage_traces + bit2, 1);
1533 printk("\nother info that might help us debug this:\n\n");
1534 print_irq_lock_scenario(backwards_entry, forwards_entry,
1535 hlock_class(prev), hlock_class(next));
1537 lockdep_print_held_locks(curr);
1539 printk("\nthe dependencies between %s-irq-safe lock and the holding lock:\n", irqclass);
1540 if (!save_trace(&prev_root->trace))
1542 print_shortest_lock_dependencies(backwards_entry, prev_root);
1544 printk("\nthe dependencies between the lock to be acquired");
1545 printk(" and %s-irq-unsafe lock:\n", irqclass);
1546 if (!save_trace(&next_root->trace))
1548 print_shortest_lock_dependencies(forwards_entry, next_root);
1550 printk("\nstack backtrace:\n");
1557 check_usage(struct task_struct *curr, struct held_lock *prev,
1558 struct held_lock *next, enum lock_usage_bit bit_backwards,
1559 enum lock_usage_bit bit_forwards, const char *irqclass)
1562 struct lock_list this, that;
1563 struct lock_list *uninitialized_var(target_entry);
1564 struct lock_list *uninitialized_var(target_entry1);
1568 this.class = hlock_class(prev);
1569 ret = find_usage_backwards(&this, bit_backwards, &target_entry);
1571 return print_bfs_bug(ret);
1576 that.class = hlock_class(next);
1577 ret = find_usage_forwards(&that, bit_forwards, &target_entry1);
1579 return print_bfs_bug(ret);
1583 return print_bad_irq_dependency(curr, &this, &that,
1584 target_entry, target_entry1,
1586 bit_backwards, bit_forwards, irqclass);
1589 static const char *state_names[] = {
1590 #define LOCKDEP_STATE(__STATE) \
1591 __stringify(__STATE),
1592 #include "lockdep_states.h"
1593 #undef LOCKDEP_STATE
1596 static const char *state_rnames[] = {
1597 #define LOCKDEP_STATE(__STATE) \
1598 __stringify(__STATE)"-READ",
1599 #include "lockdep_states.h"
1600 #undef LOCKDEP_STATE
1603 static inline const char *state_name(enum lock_usage_bit bit)
1605 return (bit & 1) ? state_rnames[bit >> 2] : state_names[bit >> 2];
1608 static int exclusive_bit(int new_bit)
1616 * bit 0 - write/read
1617 * bit 1 - used_in/enabled
1621 int state = new_bit & ~3;
1622 int dir = new_bit & 2;
1625 * keep state, bit flip the direction and strip read.
1627 return state | (dir ^ 2);
1630 static int check_irq_usage(struct task_struct *curr, struct held_lock *prev,
1631 struct held_lock *next, enum lock_usage_bit bit)
1634 * Prove that the new dependency does not connect a hardirq-safe
1635 * lock with a hardirq-unsafe lock - to achieve this we search
1636 * the backwards-subgraph starting at <prev>, and the
1637 * forwards-subgraph starting at <next>:
1639 if (!check_usage(curr, prev, next, bit,
1640 exclusive_bit(bit), state_name(bit)))
1646 * Prove that the new dependency does not connect a hardirq-safe-read
1647 * lock with a hardirq-unsafe lock - to achieve this we search
1648 * the backwards-subgraph starting at <prev>, and the
1649 * forwards-subgraph starting at <next>:
1651 if (!check_usage(curr, prev, next, bit,
1652 exclusive_bit(bit), state_name(bit)))
1659 check_prev_add_irq(struct task_struct *curr, struct held_lock *prev,
1660 struct held_lock *next)
1662 #define LOCKDEP_STATE(__STATE) \
1663 if (!check_irq_usage(curr, prev, next, LOCK_USED_IN_##__STATE)) \
1665 #include "lockdep_states.h"
1666 #undef LOCKDEP_STATE
1671 static void inc_chains(void)
1673 if (current->hardirq_context)
1674 nr_hardirq_chains++;
1676 if (current->softirq_context)
1677 nr_softirq_chains++;
1679 nr_process_chains++;
1686 check_prev_add_irq(struct task_struct *curr, struct held_lock *prev,
1687 struct held_lock *next)
1692 static inline void inc_chains(void)
1694 nr_process_chains++;
1700 print_deadlock_scenario(struct held_lock *nxt,
1701 struct held_lock *prv)
1703 struct lock_class *next = hlock_class(nxt);
1704 struct lock_class *prev = hlock_class(prv);
1706 printk(" Possible unsafe locking scenario:\n\n");
1710 __print_lock_name(prev);
1711 printk(KERN_CONT ");\n");
1713 __print_lock_name(next);
1714 printk(KERN_CONT ");\n");
1715 printk("\n *** DEADLOCK ***\n\n");
1716 printk(" May be due to missing lock nesting notation\n\n");
1720 print_deadlock_bug(struct task_struct *curr, struct held_lock *prev,
1721 struct held_lock *next)
1723 if (!debug_locks_off_graph_unlock() || debug_locks_silent)
1727 printk("=============================================\n");
1728 printk("[ INFO: possible recursive locking detected ]\n");
1729 print_kernel_ident();
1730 printk("---------------------------------------------\n");
1731 printk("%s/%d is trying to acquire lock:\n",
1732 curr->comm, task_pid_nr(curr));
1734 printk("\nbut task is already holding lock:\n");
1737 printk("\nother info that might help us debug this:\n");
1738 print_deadlock_scenario(next, prev);
1739 lockdep_print_held_locks(curr);
1741 printk("\nstack backtrace:\n");
1748 * Check whether we are holding such a class already.
1750 * (Note that this has to be done separately, because the graph cannot
1751 * detect such classes of deadlocks.)
1753 * Returns: 0 on deadlock detected, 1 on OK, 2 on recursive read
1756 check_deadlock(struct task_struct *curr, struct held_lock *next,
1757 struct lockdep_map *next_instance, int read)
1759 struct held_lock *prev;
1760 struct held_lock *nest = NULL;
1763 for (i = 0; i < curr->lockdep_depth; i++) {
1764 prev = curr->held_locks + i;
1766 if (prev->instance == next->nest_lock)
1769 if (hlock_class(prev) != hlock_class(next))
1773 * Allow read-after-read recursion of the same
1774 * lock class (i.e. read_lock(lock)+read_lock(lock)):
1776 if ((read == 2) && prev->read)
1780 * We're holding the nest_lock, which serializes this lock's
1781 * nesting behaviour.
1786 return print_deadlock_bug(curr, prev, next);
1792 * There was a chain-cache miss, and we are about to add a new dependency
1793 * to a previous lock. We recursively validate the following rules:
1795 * - would the adding of the <prev> -> <next> dependency create a
1796 * circular dependency in the graph? [== circular deadlock]
1798 * - does the new prev->next dependency connect any hardirq-safe lock
1799 * (in the full backwards-subgraph starting at <prev>) with any
1800 * hardirq-unsafe lock (in the full forwards-subgraph starting at
1801 * <next>)? [== illegal lock inversion with hardirq contexts]
1803 * - does the new prev->next dependency connect any softirq-safe lock
1804 * (in the full backwards-subgraph starting at <prev>) with any
1805 * softirq-unsafe lock (in the full forwards-subgraph starting at
1806 * <next>)? [== illegal lock inversion with softirq contexts]
1808 * any of these scenarios could lead to a deadlock.
1810 * Then if all the validations pass, we add the forwards and backwards
1814 check_prev_add(struct task_struct *curr, struct held_lock *prev,
1815 struct held_lock *next, int distance, int *stack_saved)
1817 struct lock_list *entry;
1819 struct lock_list this;
1820 struct lock_list *uninitialized_var(target_entry);
1822 * Static variable, serialized by the graph_lock().
1824 * We use this static variable to save the stack trace in case
1825 * we call into this function multiple times due to encountering
1826 * trylocks in the held lock stack.
1828 static struct stack_trace trace;
1831 * Prove that the new <prev> -> <next> dependency would not
1832 * create a circular dependency in the graph. (We do this by
1833 * forward-recursing into the graph starting at <next>, and
1834 * checking whether we can reach <prev>.)
1836 * We are using global variables to control the recursion, to
1837 * keep the stackframe size of the recursive functions low:
1839 this.class = hlock_class(next);
1841 ret = check_noncircular(&this, hlock_class(prev), &target_entry);
1843 return print_circular_bug(&this, target_entry, next, prev);
1844 else if (unlikely(ret < 0))
1845 return print_bfs_bug(ret);
1847 if (!check_prev_add_irq(curr, prev, next))
1851 * For recursive read-locks we do all the dependency checks,
1852 * but we dont store read-triggered dependencies (only
1853 * write-triggered dependencies). This ensures that only the
1854 * write-side dependencies matter, and that if for example a
1855 * write-lock never takes any other locks, then the reads are
1856 * equivalent to a NOP.
1858 if (next->read == 2 || prev->read == 2)
1861 * Is the <prev> -> <next> dependency already present?
1863 * (this may occur even though this is a new chain: consider
1864 * e.g. the L1 -> L2 -> L3 -> L4 and the L5 -> L1 -> L2 -> L3
1865 * chains - the second one will be new, but L1 already has
1866 * L2 added to its dependency list, due to the first chain.)
1868 list_for_each_entry(entry, &hlock_class(prev)->locks_after, entry) {
1869 if (entry->class == hlock_class(next)) {
1871 entry->distance = 1;
1876 if (!*stack_saved) {
1877 if (!save_trace(&trace))
1883 * Ok, all validations passed, add the new lock
1884 * to the previous lock's dependency list:
1886 ret = add_lock_to_list(hlock_class(next),
1887 &hlock_class(prev)->locks_after,
1888 next->acquire_ip, distance, &trace);
1893 ret = add_lock_to_list(hlock_class(prev),
1894 &hlock_class(next)->locks_before,
1895 next->acquire_ip, distance, &trace);
1900 * Debugging printouts:
1902 if (verbose(hlock_class(prev)) || verbose(hlock_class(next))) {
1903 /* We drop graph lock, so another thread can overwrite trace. */
1906 printk("\n new dependency: ");
1907 print_lock_name(hlock_class(prev));
1908 printk(KERN_CONT " => ");
1909 print_lock_name(hlock_class(next));
1910 printk(KERN_CONT "\n");
1912 return graph_lock();
1918 * Add the dependency to all directly-previous locks that are 'relevant'.
1919 * The ones that are relevant are (in increasing distance from curr):
1920 * all consecutive trylock entries and the final non-trylock entry - or
1921 * the end of this context's lock-chain - whichever comes first.
1924 check_prevs_add(struct task_struct *curr, struct held_lock *next)
1926 int depth = curr->lockdep_depth;
1927 int stack_saved = 0;
1928 struct held_lock *hlock;
1933 * Depth must not be zero for a non-head lock:
1938 * At least two relevant locks must exist for this
1941 if (curr->held_locks[depth].irq_context !=
1942 curr->held_locks[depth-1].irq_context)
1946 int distance = curr->lockdep_depth - depth + 1;
1947 hlock = curr->held_locks + depth - 1;
1949 * Only non-recursive-read entries get new dependencies
1952 if (hlock->read != 2 && hlock->check) {
1953 if (!check_prev_add(curr, hlock, next,
1954 distance, &stack_saved))
1957 * Stop after the first non-trylock entry,
1958 * as non-trylock entries have added their
1959 * own direct dependencies already, so this
1960 * lock is connected to them indirectly:
1962 if (!hlock->trylock)
1967 * End of lock-stack?
1972 * Stop the search if we cross into another context:
1974 if (curr->held_locks[depth].irq_context !=
1975 curr->held_locks[depth-1].irq_context)
1980 if (!debug_locks_off_graph_unlock())
1984 * Clearly we all shouldn't be here, but since we made it we
1985 * can reliable say we messed up our state. See the above two
1986 * gotos for reasons why we could possibly end up here.
1993 unsigned long nr_lock_chains;
1994 struct lock_chain lock_chains[MAX_LOCKDEP_CHAINS];
1995 int nr_chain_hlocks;
1996 static u16 chain_hlocks[MAX_LOCKDEP_CHAIN_HLOCKS];
1998 struct lock_class *lock_chain_get_class(struct lock_chain *chain, int i)
2000 return lock_classes + chain_hlocks[chain->base + i];
2004 * Returns the index of the first held_lock of the current chain
2006 static inline int get_first_held_lock(struct task_struct *curr,
2007 struct held_lock *hlock)
2010 struct held_lock *hlock_curr;
2012 for (i = curr->lockdep_depth - 1; i >= 0; i--) {
2013 hlock_curr = curr->held_locks + i;
2014 if (hlock_curr->irq_context != hlock->irq_context)
2022 #ifdef CONFIG_DEBUG_LOCKDEP
2024 * Returns the next chain_key iteration
2026 static u64 print_chain_key_iteration(int class_idx, u64 chain_key)
2028 u64 new_chain_key = iterate_chain_key(chain_key, class_idx);
2030 printk(" class_idx:%d -> chain_key:%016Lx",
2032 (unsigned long long)new_chain_key);
2033 return new_chain_key;
2037 print_chain_keys_held_locks(struct task_struct *curr, struct held_lock *hlock_next)
2039 struct held_lock *hlock;
2041 int depth = curr->lockdep_depth;
2044 printk("depth: %u\n", depth + 1);
2045 for (i = get_first_held_lock(curr, hlock_next); i < depth; i++) {
2046 hlock = curr->held_locks + i;
2047 chain_key = print_chain_key_iteration(hlock->class_idx, chain_key);
2052 print_chain_key_iteration(hlock_next->class_idx, chain_key);
2053 print_lock(hlock_next);
2056 static void print_chain_keys_chain(struct lock_chain *chain)
2062 printk("depth: %u\n", chain->depth);
2063 for (i = 0; i < chain->depth; i++) {
2064 class_id = chain_hlocks[chain->base + i];
2065 chain_key = print_chain_key_iteration(class_id + 1, chain_key);
2067 print_lock_name(lock_classes + class_id);
2072 static void print_collision(struct task_struct *curr,
2073 struct held_lock *hlock_next,
2074 struct lock_chain *chain)
2077 printk("======================\n");
2078 printk("[chain_key collision ]\n");
2079 print_kernel_ident();
2080 printk("----------------------\n");
2081 printk("%s/%d: ", current->comm, task_pid_nr(current));
2082 printk("Hash chain already cached but the contents don't match!\n");
2084 printk("Held locks:");
2085 print_chain_keys_held_locks(curr, hlock_next);
2087 printk("Locks in cached chain:");
2088 print_chain_keys_chain(chain);
2090 printk("\nstack backtrace:\n");
2096 * Checks whether the chain and the current held locks are consistent
2097 * in depth and also in content. If they are not it most likely means
2098 * that there was a collision during the calculation of the chain_key.
2099 * Returns: 0 not passed, 1 passed
2101 static int check_no_collision(struct task_struct *curr,
2102 struct held_lock *hlock,
2103 struct lock_chain *chain)
2105 #ifdef CONFIG_DEBUG_LOCKDEP
2108 i = get_first_held_lock(curr, hlock);
2110 if (DEBUG_LOCKS_WARN_ON(chain->depth != curr->lockdep_depth - (i - 1))) {
2111 print_collision(curr, hlock, chain);
2115 for (j = 0; j < chain->depth - 1; j++, i++) {
2116 id = curr->held_locks[i].class_idx - 1;
2118 if (DEBUG_LOCKS_WARN_ON(chain_hlocks[chain->base + j] != id)) {
2119 print_collision(curr, hlock, chain);
2128 * Look up a dependency chain. If the key is not present yet then
2129 * add it and return 1 - in this case the new dependency chain is
2130 * validated. If the key is already hashed, return 0.
2131 * (On return with 1 graph_lock is held.)
2133 static inline int lookup_chain_cache(struct task_struct *curr,
2134 struct held_lock *hlock,
2137 struct lock_class *class = hlock_class(hlock);
2138 struct hlist_head *hash_head = chainhashentry(chain_key);
2139 struct lock_chain *chain;
2143 * We might need to take the graph lock, ensure we've got IRQs
2144 * disabled to make this an IRQ-safe lock.. for recursion reasons
2145 * lockdep won't complain about its own locking errors.
2147 if (DEBUG_LOCKS_WARN_ON(!irqs_disabled()))
2150 * We can walk it lock-free, because entries only get added
2153 hlist_for_each_entry_rcu(chain, hash_head, entry) {
2154 if (chain->chain_key == chain_key) {
2156 debug_atomic_inc(chain_lookup_hits);
2157 if (!check_no_collision(curr, hlock, chain))
2160 if (very_verbose(class))
2161 printk("\nhash chain already cached, key: "
2162 "%016Lx tail class: [%p] %s\n",
2163 (unsigned long long)chain_key,
2164 class->key, class->name);
2168 if (very_verbose(class))
2169 printk("\nnew hash chain, key: %016Lx tail class: [%p] %s\n",
2170 (unsigned long long)chain_key, class->key, class->name);
2172 * Allocate a new chain entry from the static array, and add
2178 * We have to walk the chain again locked - to avoid duplicates:
2180 hlist_for_each_entry(chain, hash_head, entry) {
2181 if (chain->chain_key == chain_key) {
2186 if (unlikely(nr_lock_chains >= MAX_LOCKDEP_CHAINS)) {
2187 if (!debug_locks_off_graph_unlock())
2190 print_lockdep_off("BUG: MAX_LOCKDEP_CHAINS too low!");
2194 chain = lock_chains + nr_lock_chains++;
2195 chain->chain_key = chain_key;
2196 chain->irq_context = hlock->irq_context;
2197 i = get_first_held_lock(curr, hlock);
2198 chain->depth = curr->lockdep_depth + 1 - i;
2200 BUILD_BUG_ON((1UL << 24) <= ARRAY_SIZE(chain_hlocks));
2201 BUILD_BUG_ON((1UL << 6) <= ARRAY_SIZE(curr->held_locks));
2202 BUILD_BUG_ON((1UL << 8*sizeof(chain_hlocks[0])) <= ARRAY_SIZE(lock_classes));
2204 if (likely(nr_chain_hlocks + chain->depth <= MAX_LOCKDEP_CHAIN_HLOCKS)) {
2205 chain->base = nr_chain_hlocks;
2206 for (j = 0; j < chain->depth - 1; j++, i++) {
2207 int lock_id = curr->held_locks[i].class_idx - 1;
2208 chain_hlocks[chain->base + j] = lock_id;
2210 chain_hlocks[chain->base + j] = class - lock_classes;
2213 if (nr_chain_hlocks < MAX_LOCKDEP_CHAIN_HLOCKS)
2214 nr_chain_hlocks += chain->depth;
2216 #ifdef CONFIG_DEBUG_LOCKDEP
2218 * Important for check_no_collision().
2220 if (unlikely(nr_chain_hlocks > MAX_LOCKDEP_CHAIN_HLOCKS)) {
2221 if (!debug_locks_off_graph_unlock())
2224 print_lockdep_off("BUG: MAX_LOCKDEP_CHAIN_HLOCKS too low!");
2230 hlist_add_head_rcu(&chain->entry, hash_head);
2231 debug_atomic_inc(chain_lookup_misses);
2237 static int validate_chain(struct task_struct *curr, struct lockdep_map *lock,
2238 struct held_lock *hlock, int chain_head, u64 chain_key)
2241 * Trylock needs to maintain the stack of held locks, but it
2242 * does not add new dependencies, because trylock can be done
2245 * We look up the chain_key and do the O(N^2) check and update of
2246 * the dependencies only if this is a new dependency chain.
2247 * (If lookup_chain_cache() returns with 1 it acquires
2248 * graph_lock for us)
2250 if (!hlock->trylock && hlock->check &&
2251 lookup_chain_cache(curr, hlock, chain_key)) {
2253 * Check whether last held lock:
2255 * - is irq-safe, if this lock is irq-unsafe
2256 * - is softirq-safe, if this lock is hardirq-unsafe
2258 * And check whether the new lock's dependency graph
2259 * could lead back to the previous lock.
2261 * any of these scenarios could lead to a deadlock. If
2264 int ret = check_deadlock(curr, hlock, lock, hlock->read);
2269 * Mark recursive read, as we jump over it when
2270 * building dependencies (just like we jump over
2276 * Add dependency only if this lock is not the head
2277 * of the chain, and if it's not a secondary read-lock:
2279 if (!chain_head && ret != 2)
2280 if (!check_prevs_add(curr, hlock))
2284 /* after lookup_chain_cache(): */
2285 if (unlikely(!debug_locks))
2291 static inline int validate_chain(struct task_struct *curr,
2292 struct lockdep_map *lock, struct held_lock *hlock,
2293 int chain_head, u64 chain_key)
2300 * We are building curr_chain_key incrementally, so double-check
2301 * it from scratch, to make sure that it's done correctly:
2303 static void check_chain_key(struct task_struct *curr)
2305 #ifdef CONFIG_DEBUG_LOCKDEP
2306 struct held_lock *hlock, *prev_hlock = NULL;
2310 for (i = 0; i < curr->lockdep_depth; i++) {
2311 hlock = curr->held_locks + i;
2312 if (chain_key != hlock->prev_chain_key) {
2315 * We got mighty confused, our chain keys don't match
2316 * with what we expect, someone trample on our task state?
2318 WARN(1, "hm#1, depth: %u [%u], %016Lx != %016Lx\n",
2319 curr->lockdep_depth, i,
2320 (unsigned long long)chain_key,
2321 (unsigned long long)hlock->prev_chain_key);
2325 * Whoops ran out of static storage again?
2327 if (DEBUG_LOCKS_WARN_ON(hlock->class_idx > MAX_LOCKDEP_KEYS))
2330 if (prev_hlock && (prev_hlock->irq_context !=
2331 hlock->irq_context))
2333 chain_key = iterate_chain_key(chain_key, hlock->class_idx);
2336 if (chain_key != curr->curr_chain_key) {
2339 * More smoking hash instead of calculating it, damn see these
2340 * numbers float.. I bet that a pink elephant stepped on my memory.
2342 WARN(1, "hm#2, depth: %u [%u], %016Lx != %016Lx\n",
2343 curr->lockdep_depth, i,
2344 (unsigned long long)chain_key,
2345 (unsigned long long)curr->curr_chain_key);
2351 print_usage_bug_scenario(struct held_lock *lock)
2353 struct lock_class *class = hlock_class(lock);
2355 printk(" Possible unsafe locking scenario:\n\n");
2359 __print_lock_name(class);
2360 printk(KERN_CONT ");\n");
2361 printk(" <Interrupt>\n");
2363 __print_lock_name(class);
2364 printk(KERN_CONT ");\n");
2365 printk("\n *** DEADLOCK ***\n\n");
2369 print_usage_bug(struct task_struct *curr, struct held_lock *this,
2370 enum lock_usage_bit prev_bit, enum lock_usage_bit new_bit)
2372 if (!debug_locks_off_graph_unlock() || debug_locks_silent)
2376 printk("=================================\n");
2377 printk("[ INFO: inconsistent lock state ]\n");
2378 print_kernel_ident();
2379 printk("---------------------------------\n");
2381 printk("inconsistent {%s} -> {%s} usage.\n",
2382 usage_str[prev_bit], usage_str[new_bit]);
2384 printk("%s/%d [HC%u[%lu]:SC%u[%lu]:HE%u:SE%u] takes:\n",
2385 curr->comm, task_pid_nr(curr),
2386 trace_hardirq_context(curr), hardirq_count() >> HARDIRQ_SHIFT,
2387 trace_softirq_context(curr), softirq_count() >> SOFTIRQ_SHIFT,
2388 trace_hardirqs_enabled(curr),
2389 trace_softirqs_enabled(curr));
2392 printk("{%s} state was registered at:\n", usage_str[prev_bit]);
2393 print_stack_trace(hlock_class(this)->usage_traces + prev_bit, 1);
2395 print_irqtrace_events(curr);
2396 printk("\nother info that might help us debug this:\n");
2397 print_usage_bug_scenario(this);
2399 lockdep_print_held_locks(curr);
2401 printk("\nstack backtrace:\n");
2408 * Print out an error if an invalid bit is set:
2411 valid_state(struct task_struct *curr, struct held_lock *this,
2412 enum lock_usage_bit new_bit, enum lock_usage_bit bad_bit)
2414 if (unlikely(hlock_class(this)->usage_mask & (1 << bad_bit)))
2415 return print_usage_bug(curr, this, bad_bit, new_bit);
2419 static int mark_lock(struct task_struct *curr, struct held_lock *this,
2420 enum lock_usage_bit new_bit);
2422 #if defined(CONFIG_TRACE_IRQFLAGS) && defined(CONFIG_PROVE_LOCKING)
2425 * print irq inversion bug:
2428 print_irq_inversion_bug(struct task_struct *curr,
2429 struct lock_list *root, struct lock_list *other,
2430 struct held_lock *this, int forwards,
2431 const char *irqclass)
2433 struct lock_list *entry = other;
2434 struct lock_list *middle = NULL;
2437 if (!debug_locks_off_graph_unlock() || debug_locks_silent)
2441 printk("=========================================================\n");
2442 printk("[ INFO: possible irq lock inversion dependency detected ]\n");
2443 print_kernel_ident();
2444 printk("---------------------------------------------------------\n");
2445 printk("%s/%d just changed the state of lock:\n",
2446 curr->comm, task_pid_nr(curr));
2449 printk("but this lock took another, %s-unsafe lock in the past:\n", irqclass);
2451 printk("but this lock was taken by another, %s-safe lock in the past:\n", irqclass);
2452 print_lock_name(other->class);
2453 printk("\n\nand interrupts could create inverse lock ordering between them.\n\n");
2455 printk("\nother info that might help us debug this:\n");
2457 /* Find a middle lock (if one exists) */
2458 depth = get_lock_depth(other);
2460 if (depth == 0 && (entry != root)) {
2461 printk("lockdep:%s bad path found in chain graph\n", __func__);
2465 entry = get_lock_parent(entry);
2467 } while (entry && entry != root && (depth >= 0));
2469 print_irq_lock_scenario(root, other,
2470 middle ? middle->class : root->class, other->class);
2472 print_irq_lock_scenario(other, root,
2473 middle ? middle->class : other->class, root->class);
2475 lockdep_print_held_locks(curr);
2477 printk("\nthe shortest dependencies between 2nd lock and 1st lock:\n");
2478 if (!save_trace(&root->trace))
2480 print_shortest_lock_dependencies(other, root);
2482 printk("\nstack backtrace:\n");
2489 * Prove that in the forwards-direction subgraph starting at <this>
2490 * there is no lock matching <mask>:
2493 check_usage_forwards(struct task_struct *curr, struct held_lock *this,
2494 enum lock_usage_bit bit, const char *irqclass)
2497 struct lock_list root;
2498 struct lock_list *uninitialized_var(target_entry);
2501 root.class = hlock_class(this);
2502 ret = find_usage_forwards(&root, bit, &target_entry);
2504 return print_bfs_bug(ret);
2508 return print_irq_inversion_bug(curr, &root, target_entry,
2513 * Prove that in the backwards-direction subgraph starting at <this>
2514 * there is no lock matching <mask>:
2517 check_usage_backwards(struct task_struct *curr, struct held_lock *this,
2518 enum lock_usage_bit bit, const char *irqclass)
2521 struct lock_list root;
2522 struct lock_list *uninitialized_var(target_entry);
2525 root.class = hlock_class(this);
2526 ret = find_usage_backwards(&root, bit, &target_entry);
2528 return print_bfs_bug(ret);
2532 return print_irq_inversion_bug(curr, &root, target_entry,
2536 void print_irqtrace_events(struct task_struct *curr)
2538 printk("irq event stamp: %u\n", curr->irq_events);
2539 printk("hardirqs last enabled at (%u): [<%p>] %pS\n",
2540 curr->hardirq_enable_event, (void *)curr->hardirq_enable_ip,
2541 (void *)curr->hardirq_enable_ip);
2542 printk("hardirqs last disabled at (%u): [<%p>] %pS\n",
2543 curr->hardirq_disable_event, (void *)curr->hardirq_disable_ip,
2544 (void *)curr->hardirq_disable_ip);
2545 printk("softirqs last enabled at (%u): [<%p>] %pS\n",
2546 curr->softirq_enable_event, (void *)curr->softirq_enable_ip,
2547 (void *)curr->softirq_enable_ip);
2548 printk("softirqs last disabled at (%u): [<%p>] %pS\n",
2549 curr->softirq_disable_event, (void *)curr->softirq_disable_ip,
2550 (void *)curr->softirq_disable_ip);
2553 static int HARDIRQ_verbose(struct lock_class *class)
2556 return class_filter(class);
2561 static int SOFTIRQ_verbose(struct lock_class *class)
2564 return class_filter(class);
2569 static int RECLAIM_FS_verbose(struct lock_class *class)
2572 return class_filter(class);
2577 #define STRICT_READ_CHECKS 1
2579 static int (*state_verbose_f[])(struct lock_class *class) = {
2580 #define LOCKDEP_STATE(__STATE) \
2582 #include "lockdep_states.h"
2583 #undef LOCKDEP_STATE
2586 static inline int state_verbose(enum lock_usage_bit bit,
2587 struct lock_class *class)
2589 return state_verbose_f[bit >> 2](class);
2592 typedef int (*check_usage_f)(struct task_struct *, struct held_lock *,
2593 enum lock_usage_bit bit, const char *name);
2596 mark_lock_irq(struct task_struct *curr, struct held_lock *this,
2597 enum lock_usage_bit new_bit)
2599 int excl_bit = exclusive_bit(new_bit);
2600 int read = new_bit & 1;
2601 int dir = new_bit & 2;
2604 * mark USED_IN has to look forwards -- to ensure no dependency
2605 * has ENABLED state, which would allow recursion deadlocks.
2607 * mark ENABLED has to look backwards -- to ensure no dependee
2608 * has USED_IN state, which, again, would allow recursion deadlocks.
2610 check_usage_f usage = dir ?
2611 check_usage_backwards : check_usage_forwards;
2614 * Validate that this particular lock does not have conflicting
2617 if (!valid_state(curr, this, new_bit, excl_bit))
2621 * Validate that the lock dependencies don't have conflicting usage
2624 if ((!read || !dir || STRICT_READ_CHECKS) &&
2625 !usage(curr, this, excl_bit, state_name(new_bit & ~1)))
2629 * Check for read in write conflicts
2632 if (!valid_state(curr, this, new_bit, excl_bit + 1))
2635 if (STRICT_READ_CHECKS &&
2636 !usage(curr, this, excl_bit + 1,
2637 state_name(new_bit + 1)))
2641 if (state_verbose(new_bit, hlock_class(this)))
2648 #define LOCKDEP_STATE(__STATE) __STATE,
2649 #include "lockdep_states.h"
2650 #undef LOCKDEP_STATE
2654 * Mark all held locks with a usage bit:
2657 mark_held_locks(struct task_struct *curr, enum mark_type mark)
2659 enum lock_usage_bit usage_bit;
2660 struct held_lock *hlock;
2663 for (i = 0; i < curr->lockdep_depth; i++) {
2664 hlock = curr->held_locks + i;
2666 usage_bit = 2 + (mark << 2); /* ENABLED */
2668 usage_bit += 1; /* READ */
2670 BUG_ON(usage_bit >= LOCK_USAGE_STATES);
2675 if (!mark_lock(curr, hlock, usage_bit))
2683 * Hardirqs will be enabled:
2685 static void __trace_hardirqs_on_caller(unsigned long ip)
2687 struct task_struct *curr = current;
2689 /* we'll do an OFF -> ON transition: */
2690 curr->hardirqs_enabled = 1;
2693 * We are going to turn hardirqs on, so set the
2694 * usage bit for all held locks:
2696 if (!mark_held_locks(curr, HARDIRQ))
2699 * If we have softirqs enabled, then set the usage
2700 * bit for all held locks. (disabled hardirqs prevented
2701 * this bit from being set before)
2703 if (curr->softirqs_enabled)
2704 if (!mark_held_locks(curr, SOFTIRQ))
2707 curr->hardirq_enable_ip = ip;
2708 curr->hardirq_enable_event = ++curr->irq_events;
2709 debug_atomic_inc(hardirqs_on_events);
2712 __visible void trace_hardirqs_on_caller(unsigned long ip)
2714 time_hardirqs_on(CALLER_ADDR0, ip);
2716 if (unlikely(!debug_locks || current->lockdep_recursion))
2719 if (unlikely(current->hardirqs_enabled)) {
2721 * Neither irq nor preemption are disabled here
2722 * so this is racy by nature but losing one hit
2723 * in a stat is not a big deal.
2725 __debug_atomic_inc(redundant_hardirqs_on);
2730 * We're enabling irqs and according to our state above irqs weren't
2731 * already enabled, yet we find the hardware thinks they are in fact
2732 * enabled.. someone messed up their IRQ state tracing.
2734 if (DEBUG_LOCKS_WARN_ON(!irqs_disabled()))
2738 * See the fine text that goes along with this variable definition.
2740 if (DEBUG_LOCKS_WARN_ON(unlikely(early_boot_irqs_disabled)))
2744 * Can't allow enabling interrupts while in an interrupt handler,
2745 * that's general bad form and such. Recursion, limited stack etc..
2747 if (DEBUG_LOCKS_WARN_ON(current->hardirq_context))
2750 current->lockdep_recursion = 1;
2751 __trace_hardirqs_on_caller(ip);
2752 current->lockdep_recursion = 0;
2754 EXPORT_SYMBOL(trace_hardirqs_on_caller);
2756 void trace_hardirqs_on(void)
2758 trace_hardirqs_on_caller(CALLER_ADDR0);
2760 EXPORT_SYMBOL(trace_hardirqs_on);
2763 * Hardirqs were disabled:
2765 __visible void trace_hardirqs_off_caller(unsigned long ip)
2767 struct task_struct *curr = current;
2769 time_hardirqs_off(CALLER_ADDR0, ip);
2771 if (unlikely(!debug_locks || current->lockdep_recursion))
2775 * So we're supposed to get called after you mask local IRQs, but for
2776 * some reason the hardware doesn't quite think you did a proper job.
2778 if (DEBUG_LOCKS_WARN_ON(!irqs_disabled()))
2781 if (curr->hardirqs_enabled) {
2783 * We have done an ON -> OFF transition:
2785 curr->hardirqs_enabled = 0;
2786 curr->hardirq_disable_ip = ip;
2787 curr->hardirq_disable_event = ++curr->irq_events;
2788 debug_atomic_inc(hardirqs_off_events);
2790 debug_atomic_inc(redundant_hardirqs_off);
2792 EXPORT_SYMBOL(trace_hardirqs_off_caller);
2794 void trace_hardirqs_off(void)
2796 trace_hardirqs_off_caller(CALLER_ADDR0);
2798 EXPORT_SYMBOL(trace_hardirqs_off);
2801 * Softirqs will be enabled:
2803 void trace_softirqs_on(unsigned long ip)
2805 struct task_struct *curr = current;
2807 if (unlikely(!debug_locks || current->lockdep_recursion))
2811 * We fancy IRQs being disabled here, see softirq.c, avoids
2812 * funny state and nesting things.
2814 if (DEBUG_LOCKS_WARN_ON(!irqs_disabled()))
2817 if (curr->softirqs_enabled) {
2818 debug_atomic_inc(redundant_softirqs_on);
2822 current->lockdep_recursion = 1;
2824 * We'll do an OFF -> ON transition:
2826 curr->softirqs_enabled = 1;
2827 curr->softirq_enable_ip = ip;
2828 curr->softirq_enable_event = ++curr->irq_events;
2829 debug_atomic_inc(softirqs_on_events);
2831 * We are going to turn softirqs on, so set the
2832 * usage bit for all held locks, if hardirqs are
2835 if (curr->hardirqs_enabled)
2836 mark_held_locks(curr, SOFTIRQ);
2837 current->lockdep_recursion = 0;
2841 * Softirqs were disabled:
2843 void trace_softirqs_off(unsigned long ip)
2845 struct task_struct *curr = current;
2847 if (unlikely(!debug_locks || current->lockdep_recursion))
2851 * We fancy IRQs being disabled here, see softirq.c
2853 if (DEBUG_LOCKS_WARN_ON(!irqs_disabled()))
2856 if (curr->softirqs_enabled) {
2858 * We have done an ON -> OFF transition:
2860 curr->softirqs_enabled = 0;
2861 curr->softirq_disable_ip = ip;
2862 curr->softirq_disable_event = ++curr->irq_events;
2863 debug_atomic_inc(softirqs_off_events);
2865 * Whoops, we wanted softirqs off, so why aren't they?
2867 DEBUG_LOCKS_WARN_ON(!softirq_count());
2869 debug_atomic_inc(redundant_softirqs_off);
2872 static void __lockdep_trace_alloc(gfp_t gfp_mask, unsigned long flags)
2874 struct task_struct *curr = current;
2876 if (unlikely(!debug_locks))
2879 /* no reclaim without waiting on it */
2880 if (!(gfp_mask & __GFP_DIRECT_RECLAIM))
2883 /* this guy won't enter reclaim */
2884 if ((curr->flags & PF_MEMALLOC) && !(gfp_mask & __GFP_NOMEMALLOC))
2887 /* We're only interested __GFP_FS allocations for now */
2888 if (!(gfp_mask & __GFP_FS))
2892 * Oi! Can't be having __GFP_FS allocations with IRQs disabled.
2894 if (DEBUG_LOCKS_WARN_ON(irqs_disabled_flags(flags)))
2897 mark_held_locks(curr, RECLAIM_FS);
2900 static void check_flags(unsigned long flags);
2902 void lockdep_trace_alloc(gfp_t gfp_mask)
2904 unsigned long flags;
2906 if (unlikely(current->lockdep_recursion))
2909 raw_local_irq_save(flags);
2911 current->lockdep_recursion = 1;
2912 __lockdep_trace_alloc(gfp_mask, flags);
2913 current->lockdep_recursion = 0;
2914 raw_local_irq_restore(flags);
2917 static int mark_irqflags(struct task_struct *curr, struct held_lock *hlock)
2920 * If non-trylock use in a hardirq or softirq context, then
2921 * mark the lock as used in these contexts:
2923 if (!hlock->trylock) {
2925 if (curr->hardirq_context)
2926 if (!mark_lock(curr, hlock,
2927 LOCK_USED_IN_HARDIRQ_READ))
2929 if (curr->softirq_context)
2930 if (!mark_lock(curr, hlock,
2931 LOCK_USED_IN_SOFTIRQ_READ))
2934 if (curr->hardirq_context)
2935 if (!mark_lock(curr, hlock, LOCK_USED_IN_HARDIRQ))
2937 if (curr->softirq_context)
2938 if (!mark_lock(curr, hlock, LOCK_USED_IN_SOFTIRQ))
2942 if (!hlock->hardirqs_off) {
2944 if (!mark_lock(curr, hlock,
2945 LOCK_ENABLED_HARDIRQ_READ))
2947 if (curr->softirqs_enabled)
2948 if (!mark_lock(curr, hlock,
2949 LOCK_ENABLED_SOFTIRQ_READ))
2952 if (!mark_lock(curr, hlock,
2953 LOCK_ENABLED_HARDIRQ))
2955 if (curr->softirqs_enabled)
2956 if (!mark_lock(curr, hlock,
2957 LOCK_ENABLED_SOFTIRQ))
2963 * We reuse the irq context infrastructure more broadly as a general
2964 * context checking code. This tests GFP_FS recursion (a lock taken
2965 * during reclaim for a GFP_FS allocation is held over a GFP_FS
2968 if (!hlock->trylock && (curr->lockdep_reclaim_gfp & __GFP_FS)) {
2970 if (!mark_lock(curr, hlock, LOCK_USED_IN_RECLAIM_FS_READ))
2973 if (!mark_lock(curr, hlock, LOCK_USED_IN_RECLAIM_FS))
2981 static inline unsigned int task_irq_context(struct task_struct *task)
2983 return 2 * !!task->hardirq_context + !!task->softirq_context;
2986 static int separate_irq_context(struct task_struct *curr,
2987 struct held_lock *hlock)
2989 unsigned int depth = curr->lockdep_depth;
2992 * Keep track of points where we cross into an interrupt context:
2995 struct held_lock *prev_hlock;
2997 prev_hlock = curr->held_locks + depth-1;
2999 * If we cross into another context, reset the
3000 * hash key (this also prevents the checking and the
3001 * adding of the dependency to 'prev'):
3003 if (prev_hlock->irq_context != hlock->irq_context)
3009 #else /* defined(CONFIG_TRACE_IRQFLAGS) && defined(CONFIG_PROVE_LOCKING) */
3012 int mark_lock_irq(struct task_struct *curr, struct held_lock *this,
3013 enum lock_usage_bit new_bit)
3015 WARN_ON(1); /* Impossible innit? when we don't have TRACE_IRQFLAG */
3019 static inline int mark_irqflags(struct task_struct *curr,
3020 struct held_lock *hlock)
3025 static inline unsigned int task_irq_context(struct task_struct *task)
3030 static inline int separate_irq_context(struct task_struct *curr,
3031 struct held_lock *hlock)
3036 void lockdep_trace_alloc(gfp_t gfp_mask)
3040 #endif /* defined(CONFIG_TRACE_IRQFLAGS) && defined(CONFIG_PROVE_LOCKING) */
3043 * Mark a lock with a usage bit, and validate the state transition:
3045 static int mark_lock(struct task_struct *curr, struct held_lock *this,
3046 enum lock_usage_bit new_bit)
3048 unsigned int new_mask = 1 << new_bit, ret = 1;
3051 * If already set then do not dirty the cacheline,
3052 * nor do any checks:
3054 if (likely(hlock_class(this)->usage_mask & new_mask))
3060 * Make sure we didn't race:
3062 if (unlikely(hlock_class(this)->usage_mask & new_mask)) {
3067 hlock_class(this)->usage_mask |= new_mask;
3069 if (!save_trace(hlock_class(this)->usage_traces + new_bit))
3073 #define LOCKDEP_STATE(__STATE) \
3074 case LOCK_USED_IN_##__STATE: \
3075 case LOCK_USED_IN_##__STATE##_READ: \
3076 case LOCK_ENABLED_##__STATE: \
3077 case LOCK_ENABLED_##__STATE##_READ:
3078 #include "lockdep_states.h"
3079 #undef LOCKDEP_STATE
3080 ret = mark_lock_irq(curr, this, new_bit);
3085 debug_atomic_dec(nr_unused_locks);
3088 if (!debug_locks_off_graph_unlock())
3097 * We must printk outside of the graph_lock:
3100 printk("\nmarked lock as {%s}:\n", usage_str[new_bit]);
3102 print_irqtrace_events(curr);
3110 * Initialize a lock instance's lock-class mapping info:
3112 void lockdep_init_map(struct lockdep_map *lock, const char *name,
3113 struct lock_class_key *key, int subclass)
3117 kmemcheck_mark_initialized(lock, sizeof(*lock));
3119 for (i = 0; i < NR_LOCKDEP_CACHING_CLASSES; i++)
3120 lock->class_cache[i] = NULL;
3122 #ifdef CONFIG_LOCK_STAT
3123 lock->cpu = raw_smp_processor_id();
3127 * Can't be having no nameless bastards around this place!
3129 if (DEBUG_LOCKS_WARN_ON(!name)) {
3130 lock->name = "NULL";
3137 * No key, no joy, we need to hash something.
3139 if (DEBUG_LOCKS_WARN_ON(!key))
3142 * Sanity check, the lock-class key must be persistent:
3144 if (!static_obj(key)) {
3145 printk("BUG: key %p not in .data!\n", key);
3147 * What it says above ^^^^^, I suggest you read it.
3149 DEBUG_LOCKS_WARN_ON(1);
3154 if (unlikely(!debug_locks))
3158 unsigned long flags;
3160 if (DEBUG_LOCKS_WARN_ON(current->lockdep_recursion))
3163 raw_local_irq_save(flags);
3164 current->lockdep_recursion = 1;
3165 register_lock_class(lock, subclass, 1);
3166 current->lockdep_recursion = 0;
3167 raw_local_irq_restore(flags);
3170 EXPORT_SYMBOL_GPL(lockdep_init_map);
3172 struct lock_class_key __lockdep_no_validate__;
3173 EXPORT_SYMBOL_GPL(__lockdep_no_validate__);
3176 print_lock_nested_lock_not_held(struct task_struct *curr,
3177 struct held_lock *hlock,
3180 if (!debug_locks_off())
3182 if (debug_locks_silent)
3186 printk("==================================\n");
3187 printk("[ BUG: Nested lock was not taken ]\n");
3188 print_kernel_ident();
3189 printk("----------------------------------\n");
3191 printk("%s/%d is trying to lock:\n", curr->comm, task_pid_nr(curr));
3194 printk("\nbut this task is not holding:\n");
3195 printk("%s\n", hlock->nest_lock->name);
3197 printk("\nstack backtrace:\n");
3200 printk("\nother info that might help us debug this:\n");
3201 lockdep_print_held_locks(curr);
3203 printk("\nstack backtrace:\n");
3209 static int __lock_is_held(struct lockdep_map *lock, int read);
3212 * This gets called for every mutex_lock*()/spin_lock*() operation.
3213 * We maintain the dependency maps and validate the locking attempt:
3215 static int __lock_acquire(struct lockdep_map *lock, unsigned int subclass,
3216 int trylock, int read, int check, int hardirqs_off,
3217 struct lockdep_map *nest_lock, unsigned long ip,
3218 int references, int pin_count)
3220 struct task_struct *curr = current;
3221 struct lock_class *class = NULL;
3222 struct held_lock *hlock;
3228 if (unlikely(!debug_locks))
3232 * Lockdep should run with IRQs disabled, otherwise we could
3233 * get an interrupt which would want to take locks, which would
3234 * end up in lockdep and have you got a head-ache already?
3236 if (DEBUG_LOCKS_WARN_ON(!irqs_disabled()))
3239 if (!prove_locking || lock->key == &__lockdep_no_validate__)
3242 if (subclass < NR_LOCKDEP_CACHING_CLASSES)
3243 class = lock->class_cache[subclass];
3247 if (unlikely(!class)) {
3248 class = register_lock_class(lock, subclass, 0);
3252 atomic_inc((atomic_t *)&class->ops);
3253 if (very_verbose(class)) {
3254 printk("\nacquire class [%p] %s", class->key, class->name);
3255 if (class->name_version > 1)
3256 printk(KERN_CONT "#%d", class->name_version);
3257 printk(KERN_CONT "\n");
3262 * Add the lock to the list of currently held locks.
3263 * (we dont increase the depth just yet, up until the
3264 * dependency checks are done)
3266 depth = curr->lockdep_depth;
3268 * Ran out of static storage for our per-task lock stack again have we?
3270 if (DEBUG_LOCKS_WARN_ON(depth >= MAX_LOCK_DEPTH))
3273 class_idx = class - lock_classes + 1;
3276 hlock = curr->held_locks + depth - 1;
3277 if (hlock->class_idx == class_idx && nest_lock) {
3278 if (hlock->references) {
3280 * Check: unsigned int references:12, overflow.
3282 if (DEBUG_LOCKS_WARN_ON(hlock->references == (1 << 12)-1))
3285 hlock->references++;
3287 hlock->references = 2;
3294 hlock = curr->held_locks + depth;
3296 * Plain impossible, we just registered it and checked it weren't no
3297 * NULL like.. I bet this mushroom I ate was good!
3299 if (DEBUG_LOCKS_WARN_ON(!class))
3301 hlock->class_idx = class_idx;
3302 hlock->acquire_ip = ip;
3303 hlock->instance = lock;
3304 hlock->nest_lock = nest_lock;
3305 hlock->irq_context = task_irq_context(curr);
3306 hlock->trylock = trylock;
3308 hlock->check = check;
3309 hlock->hardirqs_off = !!hardirqs_off;
3310 hlock->references = references;
3311 #ifdef CONFIG_LOCK_STAT
3312 hlock->waittime_stamp = 0;
3313 hlock->holdtime_stamp = lockstat_clock();
3315 hlock->pin_count = pin_count;
3317 if (check && !mark_irqflags(curr, hlock))
3320 /* mark it as used: */
3321 if (!mark_lock(curr, hlock, LOCK_USED))
3325 * Calculate the chain hash: it's the combined hash of all the
3326 * lock keys along the dependency chain. We save the hash value
3327 * at every step so that we can get the current hash easily
3328 * after unlock. The chain hash is then used to cache dependency
3331 * The 'key ID' is what is the most compact key value to drive
3332 * the hash, not class->key.
3335 * Whoops, we did it again.. ran straight out of our static allocation.
3337 if (DEBUG_LOCKS_WARN_ON(class_idx > MAX_LOCKDEP_KEYS))
3340 chain_key = curr->curr_chain_key;
3343 * How can we have a chain hash when we ain't got no keys?!
3345 if (DEBUG_LOCKS_WARN_ON(chain_key != 0))
3350 hlock->prev_chain_key = chain_key;
3351 if (separate_irq_context(curr, hlock)) {
3355 chain_key = iterate_chain_key(chain_key, class_idx);
3357 if (nest_lock && !__lock_is_held(nest_lock, -1))
3358 return print_lock_nested_lock_not_held(curr, hlock, ip);
3360 if (!validate_chain(curr, lock, hlock, chain_head, chain_key))
3363 curr->curr_chain_key = chain_key;
3364 curr->lockdep_depth++;
3365 check_chain_key(curr);
3366 #ifdef CONFIG_DEBUG_LOCKDEP
3367 if (unlikely(!debug_locks))
3370 if (unlikely(curr->lockdep_depth >= MAX_LOCK_DEPTH)) {
3372 print_lockdep_off("BUG: MAX_LOCK_DEPTH too low!");
3373 printk(KERN_DEBUG "depth: %i max: %lu!\n",
3374 curr->lockdep_depth, MAX_LOCK_DEPTH);
3376 lockdep_print_held_locks(current);
3377 debug_show_all_locks();
3383 if (unlikely(curr->lockdep_depth > max_lockdep_depth))
3384 max_lockdep_depth = curr->lockdep_depth;
3390 print_unlock_imbalance_bug(struct task_struct *curr, struct lockdep_map *lock,
3393 if (!debug_locks_off())
3395 if (debug_locks_silent)
3399 printk("=====================================\n");
3400 printk("[ BUG: bad unlock balance detected! ]\n");
3401 print_kernel_ident();
3402 printk("-------------------------------------\n");
3403 printk("%s/%d is trying to release lock (",
3404 curr->comm, task_pid_nr(curr));
3405 print_lockdep_cache(lock);
3406 printk(KERN_CONT ") at:\n");
3408 printk("but there are no more locks to release!\n");
3409 printk("\nother info that might help us debug this:\n");
3410 lockdep_print_held_locks(curr);
3412 printk("\nstack backtrace:\n");
3418 static int match_held_lock(struct held_lock *hlock, struct lockdep_map *lock)
3420 if (hlock->instance == lock)
3423 if (hlock->references) {
3424 struct lock_class *class = lock->class_cache[0];
3427 class = look_up_lock_class(lock, 0);
3430 * If look_up_lock_class() failed to find a class, we're trying
3431 * to test if we hold a lock that has never yet been acquired.
3432 * Clearly if the lock hasn't been acquired _ever_, we're not
3433 * holding it either, so report failure.
3435 if (IS_ERR_OR_NULL(class))
3439 * References, but not a lock we're actually ref-counting?
3440 * State got messed up, follow the sites that change ->references
3441 * and try to make sense of it.
3443 if (DEBUG_LOCKS_WARN_ON(!hlock->nest_lock))
3446 if (hlock->class_idx == class - lock_classes + 1)
3453 /* @depth must not be zero */
3454 static struct held_lock *find_held_lock(struct task_struct *curr,
3455 struct lockdep_map *lock,
3456 unsigned int depth, int *idx)
3458 struct held_lock *ret, *hlock, *prev_hlock;
3462 hlock = curr->held_locks + i;
3464 if (match_held_lock(hlock, lock))
3468 for (i--, prev_hlock = hlock--;
3470 i--, prev_hlock = hlock--) {
3472 * We must not cross into another context:
3474 if (prev_hlock->irq_context != hlock->irq_context) {
3478 if (match_held_lock(hlock, lock)) {
3489 static int reacquire_held_locks(struct task_struct *curr, unsigned int depth,
3492 struct held_lock *hlock;
3494 for (hlock = curr->held_locks + idx; idx < depth; idx++, hlock++) {
3495 if (!__lock_acquire(hlock->instance,
3496 hlock_class(hlock)->subclass,
3498 hlock->read, hlock->check,
3499 hlock->hardirqs_off,
3500 hlock->nest_lock, hlock->acquire_ip,
3501 hlock->references, hlock->pin_count))
3508 __lock_set_class(struct lockdep_map *lock, const char *name,
3509 struct lock_class_key *key, unsigned int subclass,
3512 struct task_struct *curr = current;
3513 struct held_lock *hlock;
3514 struct lock_class *class;
3518 depth = curr->lockdep_depth;
3520 * This function is about (re)setting the class of a held lock,
3521 * yet we're not actually holding any locks. Naughty user!
3523 if (DEBUG_LOCKS_WARN_ON(!depth))
3526 hlock = find_held_lock(curr, lock, depth, &i);
3528 return print_unlock_imbalance_bug(curr, lock, ip);
3530 lockdep_init_map(lock, name, key, 0);
3531 class = register_lock_class(lock, subclass, 0);
3532 hlock->class_idx = class - lock_classes + 1;
3534 curr->lockdep_depth = i;
3535 curr->curr_chain_key = hlock->prev_chain_key;
3537 if (reacquire_held_locks(curr, depth, i))
3541 * I took it apart and put it back together again, except now I have
3542 * these 'spare' parts.. where shall I put them.
3544 if (DEBUG_LOCKS_WARN_ON(curr->lockdep_depth != depth))
3549 static int __lock_downgrade(struct lockdep_map *lock, unsigned long ip)
3551 struct task_struct *curr = current;
3552 struct held_lock *hlock;
3556 depth = curr->lockdep_depth;
3558 * This function is about (re)setting the class of a held lock,
3559 * yet we're not actually holding any locks. Naughty user!
3561 if (DEBUG_LOCKS_WARN_ON(!depth))
3564 hlock = find_held_lock(curr, lock, depth, &i);
3566 return print_unlock_imbalance_bug(curr, lock, ip);
3568 curr->lockdep_depth = i;
3569 curr->curr_chain_key = hlock->prev_chain_key;
3571 WARN(hlock->read, "downgrading a read lock");
3573 hlock->acquire_ip = ip;
3575 if (reacquire_held_locks(curr, depth, i))
3579 * I took it apart and put it back together again, except now I have
3580 * these 'spare' parts.. where shall I put them.
3582 if (DEBUG_LOCKS_WARN_ON(curr->lockdep_depth != depth))
3588 * Remove the lock to the list of currently held locks - this gets
3589 * called on mutex_unlock()/spin_unlock*() (or on a failed
3590 * mutex_lock_interruptible()).
3592 * @nested is an hysterical artifact, needs a tree wide cleanup.
3595 __lock_release(struct lockdep_map *lock, int nested, unsigned long ip)
3597 struct task_struct *curr = current;
3598 struct held_lock *hlock;
3602 if (unlikely(!debug_locks))
3605 depth = curr->lockdep_depth;
3607 * So we're all set to release this lock.. wait what lock? We don't
3608 * own any locks, you've been drinking again?
3610 if (DEBUG_LOCKS_WARN_ON(depth <= 0))
3611 return print_unlock_imbalance_bug(curr, lock, ip);
3614 * Check whether the lock exists in the current stack
3617 hlock = find_held_lock(curr, lock, depth, &i);
3619 return print_unlock_imbalance_bug(curr, lock, ip);
3621 if (hlock->instance == lock)
3622 lock_release_holdtime(hlock);
3624 WARN(hlock->pin_count, "releasing a pinned lock\n");
3626 if (hlock->references) {
3627 hlock->references--;
3628 if (hlock->references) {
3630 * We had, and after removing one, still have
3631 * references, the current lock stack is still
3632 * valid. We're done!
3639 * We have the right lock to unlock, 'hlock' points to it.
3640 * Now we remove it from the stack, and add back the other
3641 * entries (if any), recalculating the hash along the way:
3644 curr->lockdep_depth = i;
3645 curr->curr_chain_key = hlock->prev_chain_key;
3647 if (reacquire_held_locks(curr, depth, i + 1))
3651 * We had N bottles of beer on the wall, we drank one, but now
3652 * there's not N-1 bottles of beer left on the wall...
3654 if (DEBUG_LOCKS_WARN_ON(curr->lockdep_depth != depth - 1))
3660 static int __lock_is_held(struct lockdep_map *lock, int read)
3662 struct task_struct *curr = current;
3665 for (i = 0; i < curr->lockdep_depth; i++) {
3666 struct held_lock *hlock = curr->held_locks + i;
3668 if (match_held_lock(hlock, lock)) {
3669 if (read == -1 || hlock->read == read)
3679 static struct pin_cookie __lock_pin_lock(struct lockdep_map *lock)
3681 struct pin_cookie cookie = NIL_COOKIE;
3682 struct task_struct *curr = current;
3685 if (unlikely(!debug_locks))
3688 for (i = 0; i < curr->lockdep_depth; i++) {
3689 struct held_lock *hlock = curr->held_locks + i;
3691 if (match_held_lock(hlock, lock)) {
3693 * Grab 16bits of randomness; this is sufficient to not
3694 * be guessable and still allows some pin nesting in
3695 * our u32 pin_count.
3697 cookie.val = 1 + (prandom_u32() >> 16);
3698 hlock->pin_count += cookie.val;
3703 WARN(1, "pinning an unheld lock\n");
3707 static void __lock_repin_lock(struct lockdep_map *lock, struct pin_cookie cookie)
3709 struct task_struct *curr = current;
3712 if (unlikely(!debug_locks))
3715 for (i = 0; i < curr->lockdep_depth; i++) {
3716 struct held_lock *hlock = curr->held_locks + i;
3718 if (match_held_lock(hlock, lock)) {
3719 hlock->pin_count += cookie.val;
3724 WARN(1, "pinning an unheld lock\n");
3727 static void __lock_unpin_lock(struct lockdep_map *lock, struct pin_cookie cookie)
3729 struct task_struct *curr = current;
3732 if (unlikely(!debug_locks))
3735 for (i = 0; i < curr->lockdep_depth; i++) {
3736 struct held_lock *hlock = curr->held_locks + i;
3738 if (match_held_lock(hlock, lock)) {
3739 if (WARN(!hlock->pin_count, "unpinning an unpinned lock\n"))
3742 hlock->pin_count -= cookie.val;
3744 if (WARN((int)hlock->pin_count < 0, "pin count corrupted\n"))
3745 hlock->pin_count = 0;
3751 WARN(1, "unpinning an unheld lock\n");
3755 * Check whether we follow the irq-flags state precisely:
3757 static void check_flags(unsigned long flags)
3759 #if defined(CONFIG_PROVE_LOCKING) && defined(CONFIG_DEBUG_LOCKDEP) && \
3760 defined(CONFIG_TRACE_IRQFLAGS)
3764 if (irqs_disabled_flags(flags)) {
3765 if (DEBUG_LOCKS_WARN_ON(current->hardirqs_enabled)) {
3766 printk("possible reason: unannotated irqs-off.\n");
3769 if (DEBUG_LOCKS_WARN_ON(!current->hardirqs_enabled)) {
3770 printk("possible reason: unannotated irqs-on.\n");
3775 * We dont accurately track softirq state in e.g.
3776 * hardirq contexts (such as on 4KSTACKS), so only
3777 * check if not in hardirq contexts:
3779 if (!hardirq_count()) {
3780 if (softirq_count()) {
3781 /* like the above, but with softirqs */
3782 DEBUG_LOCKS_WARN_ON(current->softirqs_enabled);
3784 /* lick the above, does it taste good? */
3785 DEBUG_LOCKS_WARN_ON(!current->softirqs_enabled);
3790 print_irqtrace_events(current);
3794 void lock_set_class(struct lockdep_map *lock, const char *name,
3795 struct lock_class_key *key, unsigned int subclass,
3798 unsigned long flags;
3800 if (unlikely(current->lockdep_recursion))
3803 raw_local_irq_save(flags);
3804 current->lockdep_recursion = 1;
3806 if (__lock_set_class(lock, name, key, subclass, ip))
3807 check_chain_key(current);
3808 current->lockdep_recursion = 0;
3809 raw_local_irq_restore(flags);
3811 EXPORT_SYMBOL_GPL(lock_set_class);
3813 void lock_downgrade(struct lockdep_map *lock, unsigned long ip)
3815 unsigned long flags;
3817 if (unlikely(current->lockdep_recursion))
3820 raw_local_irq_save(flags);
3821 current->lockdep_recursion = 1;
3823 if (__lock_downgrade(lock, ip))
3824 check_chain_key(current);
3825 current->lockdep_recursion = 0;
3826 raw_local_irq_restore(flags);
3828 EXPORT_SYMBOL_GPL(lock_downgrade);
3831 * We are not always called with irqs disabled - do that here,
3832 * and also avoid lockdep recursion:
3834 void lock_acquire(struct lockdep_map *lock, unsigned int subclass,
3835 int trylock, int read, int check,
3836 struct lockdep_map *nest_lock, unsigned long ip)
3838 unsigned long flags;
3840 if (unlikely(current->lockdep_recursion))
3843 raw_local_irq_save(flags);
3846 current->lockdep_recursion = 1;
3847 trace_lock_acquire(lock, subclass, trylock, read, check, nest_lock, ip);
3848 __lock_acquire(lock, subclass, trylock, read, check,
3849 irqs_disabled_flags(flags), nest_lock, ip, 0, 0);
3850 current->lockdep_recursion = 0;
3851 raw_local_irq_restore(flags);
3853 EXPORT_SYMBOL_GPL(lock_acquire);
3855 void lock_release(struct lockdep_map *lock, int nested,
3858 unsigned long flags;
3860 if (unlikely(current->lockdep_recursion))
3863 raw_local_irq_save(flags);
3865 current->lockdep_recursion = 1;
3866 trace_lock_release(lock, ip);
3867 if (__lock_release(lock, nested, ip))
3868 check_chain_key(current);
3869 current->lockdep_recursion = 0;
3870 raw_local_irq_restore(flags);
3872 EXPORT_SYMBOL_GPL(lock_release);
3874 int lock_is_held_type(struct lockdep_map *lock, int read)
3876 unsigned long flags;
3879 if (unlikely(current->lockdep_recursion))
3880 return 1; /* avoid false negative lockdep_assert_held() */
3882 raw_local_irq_save(flags);
3885 current->lockdep_recursion = 1;
3886 ret = __lock_is_held(lock, read);
3887 current->lockdep_recursion = 0;
3888 raw_local_irq_restore(flags);
3892 EXPORT_SYMBOL_GPL(lock_is_held_type);
3894 struct pin_cookie lock_pin_lock(struct lockdep_map *lock)
3896 struct pin_cookie cookie = NIL_COOKIE;
3897 unsigned long flags;
3899 if (unlikely(current->lockdep_recursion))
3902 raw_local_irq_save(flags);
3905 current->lockdep_recursion = 1;
3906 cookie = __lock_pin_lock(lock);
3907 current->lockdep_recursion = 0;
3908 raw_local_irq_restore(flags);
3912 EXPORT_SYMBOL_GPL(lock_pin_lock);
3914 void lock_repin_lock(struct lockdep_map *lock, struct pin_cookie cookie)
3916 unsigned long flags;
3918 if (unlikely(current->lockdep_recursion))
3921 raw_local_irq_save(flags);
3924 current->lockdep_recursion = 1;
3925 __lock_repin_lock(lock, cookie);
3926 current->lockdep_recursion = 0;
3927 raw_local_irq_restore(flags);
3929 EXPORT_SYMBOL_GPL(lock_repin_lock);
3931 void lock_unpin_lock(struct lockdep_map *lock, struct pin_cookie cookie)
3933 unsigned long flags;
3935 if (unlikely(current->lockdep_recursion))
3938 raw_local_irq_save(flags);
3941 current->lockdep_recursion = 1;
3942 __lock_unpin_lock(lock, cookie);
3943 current->lockdep_recursion = 0;
3944 raw_local_irq_restore(flags);
3946 EXPORT_SYMBOL_GPL(lock_unpin_lock);
3948 void lockdep_set_current_reclaim_state(gfp_t gfp_mask)
3950 current->lockdep_reclaim_gfp = gfp_mask;
3952 EXPORT_SYMBOL_GPL(lockdep_set_current_reclaim_state);
3954 void lockdep_clear_current_reclaim_state(void)
3956 current->lockdep_reclaim_gfp = 0;
3958 EXPORT_SYMBOL_GPL(lockdep_clear_current_reclaim_state);
3960 #ifdef CONFIG_LOCK_STAT
3962 print_lock_contention_bug(struct task_struct *curr, struct lockdep_map *lock,
3965 if (!debug_locks_off())
3967 if (debug_locks_silent)
3971 printk("=================================\n");
3972 printk("[ BUG: bad contention detected! ]\n");
3973 print_kernel_ident();
3974 printk("---------------------------------\n");
3975 printk("%s/%d is trying to contend lock (",
3976 curr->comm, task_pid_nr(curr));
3977 print_lockdep_cache(lock);
3978 printk(KERN_CONT ") at:\n");
3980 printk("but there are no locks held!\n");
3981 printk("\nother info that might help us debug this:\n");
3982 lockdep_print_held_locks(curr);
3984 printk("\nstack backtrace:\n");
3991 __lock_contended(struct lockdep_map *lock, unsigned long ip)
3993 struct task_struct *curr = current;
3994 struct held_lock *hlock;
3995 struct lock_class_stats *stats;
3997 int i, contention_point, contending_point;
3999 depth = curr->lockdep_depth;
4001 * Whee, we contended on this lock, except it seems we're not
4002 * actually trying to acquire anything much at all..
4004 if (DEBUG_LOCKS_WARN_ON(!depth))
4007 hlock = find_held_lock(curr, lock, depth, &i);
4009 print_lock_contention_bug(curr, lock, ip);
4013 if (hlock->instance != lock)
4016 hlock->waittime_stamp = lockstat_clock();
4018 contention_point = lock_point(hlock_class(hlock)->contention_point, ip);
4019 contending_point = lock_point(hlock_class(hlock)->contending_point,
4022 stats = get_lock_stats(hlock_class(hlock));
4023 if (contention_point < LOCKSTAT_POINTS)
4024 stats->contention_point[contention_point]++;
4025 if (contending_point < LOCKSTAT_POINTS)
4026 stats->contending_point[contending_point]++;
4027 if (lock->cpu != smp_processor_id())
4028 stats->bounces[bounce_contended + !!hlock->read]++;
4029 put_lock_stats(stats);
4033 __lock_acquired(struct lockdep_map *lock, unsigned long ip)
4035 struct task_struct *curr = current;
4036 struct held_lock *hlock;
4037 struct lock_class_stats *stats;
4039 u64 now, waittime = 0;
4042 depth = curr->lockdep_depth;
4044 * Yay, we acquired ownership of this lock we didn't try to
4045 * acquire, how the heck did that happen?
4047 if (DEBUG_LOCKS_WARN_ON(!depth))
4050 hlock = find_held_lock(curr, lock, depth, &i);
4052 print_lock_contention_bug(curr, lock, _RET_IP_);
4056 if (hlock->instance != lock)
4059 cpu = smp_processor_id();
4060 if (hlock->waittime_stamp) {
4061 now = lockstat_clock();
4062 waittime = now - hlock->waittime_stamp;
4063 hlock->holdtime_stamp = now;
4066 trace_lock_acquired(lock, ip);
4068 stats = get_lock_stats(hlock_class(hlock));
4071 lock_time_inc(&stats->read_waittime, waittime);
4073 lock_time_inc(&stats->write_waittime, waittime);
4075 if (lock->cpu != cpu)
4076 stats->bounces[bounce_acquired + !!hlock->read]++;
4077 put_lock_stats(stats);
4083 void lock_contended(struct lockdep_map *lock, unsigned long ip)
4085 unsigned long flags;
4087 if (unlikely(!lock_stat))
4090 if (unlikely(current->lockdep_recursion))
4093 raw_local_irq_save(flags);
4095 current->lockdep_recursion = 1;
4096 trace_lock_contended(lock, ip);
4097 __lock_contended(lock, ip);
4098 current->lockdep_recursion = 0;
4099 raw_local_irq_restore(flags);
4101 EXPORT_SYMBOL_GPL(lock_contended);
4103 void lock_acquired(struct lockdep_map *lock, unsigned long ip)
4105 unsigned long flags;
4107 if (unlikely(!lock_stat))
4110 if (unlikely(current->lockdep_recursion))
4113 raw_local_irq_save(flags);
4115 current->lockdep_recursion = 1;
4116 __lock_acquired(lock, ip);
4117 current->lockdep_recursion = 0;
4118 raw_local_irq_restore(flags);
4120 EXPORT_SYMBOL_GPL(lock_acquired);
4124 * Used by the testsuite, sanitize the validator state
4125 * after a simulated failure:
4128 void lockdep_reset(void)
4130 unsigned long flags;
4133 raw_local_irq_save(flags);
4134 current->curr_chain_key = 0;
4135 current->lockdep_depth = 0;
4136 current->lockdep_recursion = 0;
4137 memset(current->held_locks, 0, MAX_LOCK_DEPTH*sizeof(struct held_lock));
4138 nr_hardirq_chains = 0;
4139 nr_softirq_chains = 0;
4140 nr_process_chains = 0;
4142 for (i = 0; i < CHAINHASH_SIZE; i++)
4143 INIT_HLIST_HEAD(chainhash_table + i);
4144 raw_local_irq_restore(flags);
4147 static void zap_class(struct lock_class *class)
4152 * Remove all dependencies this lock is
4155 for (i = 0; i < nr_list_entries; i++) {
4156 if (list_entries[i].class == class)
4157 list_del_rcu(&list_entries[i].entry);
4160 * Unhash the class and remove it from the all_lock_classes list:
4162 hlist_del_rcu(&class->hash_entry);
4163 list_del_rcu(&class->lock_entry);
4165 RCU_INIT_POINTER(class->key, NULL);
4166 RCU_INIT_POINTER(class->name, NULL);
4169 static inline int within(const void *addr, void *start, unsigned long size)
4171 return addr >= start && addr < start + size;
4175 * Used in module.c to remove lock classes from memory that is going to be
4176 * freed; and possibly re-used by other modules.
4178 * We will have had one sync_sched() before getting here, so we're guaranteed
4179 * nobody will look up these exact classes -- they're properly dead but still
4182 void lockdep_free_key_range(void *start, unsigned long size)
4184 struct lock_class *class;
4185 struct hlist_head *head;
4186 unsigned long flags;
4190 raw_local_irq_save(flags);
4191 locked = graph_lock();
4194 * Unhash all classes that were created by this module:
4196 for (i = 0; i < CLASSHASH_SIZE; i++) {
4197 head = classhash_table + i;
4198 hlist_for_each_entry_rcu(class, head, hash_entry) {
4199 if (within(class->key, start, size))
4201 else if (within(class->name, start, size))
4208 raw_local_irq_restore(flags);
4211 * Wait for any possible iterators from look_up_lock_class() to pass
4212 * before continuing to free the memory they refer to.
4214 * sync_sched() is sufficient because the read-side is IRQ disable.
4216 synchronize_sched();
4219 * XXX at this point we could return the resources to the pool;
4220 * instead we leak them. We would need to change to bitmap allocators
4221 * instead of the linear allocators we have now.
4225 void lockdep_reset_lock(struct lockdep_map *lock)
4227 struct lock_class *class;
4228 struct hlist_head *head;
4229 unsigned long flags;
4233 raw_local_irq_save(flags);
4236 * Remove all classes this lock might have:
4238 for (j = 0; j < MAX_LOCKDEP_SUBCLASSES; j++) {
4240 * If the class exists we look it up and zap it:
4242 class = look_up_lock_class(lock, j);
4243 if (!IS_ERR_OR_NULL(class))
4247 * Debug check: in the end all mapped classes should
4250 locked = graph_lock();
4251 for (i = 0; i < CLASSHASH_SIZE; i++) {
4252 head = classhash_table + i;
4253 hlist_for_each_entry_rcu(class, head, hash_entry) {
4256 for (j = 0; j < NR_LOCKDEP_CACHING_CLASSES; j++)
4257 match |= class == lock->class_cache[j];
4259 if (unlikely(match)) {
4260 if (debug_locks_off_graph_unlock()) {
4262 * We all just reset everything, how did it match?
4274 raw_local_irq_restore(flags);
4277 void __init lockdep_info(void)
4279 printk("Lock dependency validator: Copyright (c) 2006 Red Hat, Inc., Ingo Molnar\n");
4281 printk("... MAX_LOCKDEP_SUBCLASSES: %lu\n", MAX_LOCKDEP_SUBCLASSES);
4282 printk("... MAX_LOCK_DEPTH: %lu\n", MAX_LOCK_DEPTH);
4283 printk("... MAX_LOCKDEP_KEYS: %lu\n", MAX_LOCKDEP_KEYS);
4284 printk("... CLASSHASH_SIZE: %lu\n", CLASSHASH_SIZE);
4285 printk("... MAX_LOCKDEP_ENTRIES: %lu\n", MAX_LOCKDEP_ENTRIES);
4286 printk("... MAX_LOCKDEP_CHAINS: %lu\n", MAX_LOCKDEP_CHAINS);
4287 printk("... CHAINHASH_SIZE: %lu\n", CHAINHASH_SIZE);
4289 printk(" memory used by lock dependency info: %lu kB\n",
4290 (sizeof(struct lock_class) * MAX_LOCKDEP_KEYS +
4291 sizeof(struct list_head) * CLASSHASH_SIZE +
4292 sizeof(struct lock_list) * MAX_LOCKDEP_ENTRIES +
4293 sizeof(struct lock_chain) * MAX_LOCKDEP_CHAINS +
4294 sizeof(struct list_head) * CHAINHASH_SIZE
4295 #ifdef CONFIG_PROVE_LOCKING
4296 + sizeof(struct circular_queue)
4301 printk(" per task-struct memory footprint: %lu bytes\n",
4302 sizeof(struct held_lock) * MAX_LOCK_DEPTH);
4306 print_freed_lock_bug(struct task_struct *curr, const void *mem_from,
4307 const void *mem_to, struct held_lock *hlock)
4309 if (!debug_locks_off())
4311 if (debug_locks_silent)
4315 printk("=========================\n");
4316 printk("[ BUG: held lock freed! ]\n");
4317 print_kernel_ident();
4318 printk("-------------------------\n");
4319 printk("%s/%d is freeing memory %p-%p, with a lock still held there!\n",
4320 curr->comm, task_pid_nr(curr), mem_from, mem_to-1);
4322 lockdep_print_held_locks(curr);
4324 printk("\nstack backtrace:\n");
4328 static inline int not_in_range(const void* mem_from, unsigned long mem_len,
4329 const void* lock_from, unsigned long lock_len)
4331 return lock_from + lock_len <= mem_from ||
4332 mem_from + mem_len <= lock_from;
4336 * Called when kernel memory is freed (or unmapped), or if a lock
4337 * is destroyed or reinitialized - this code checks whether there is
4338 * any held lock in the memory range of <from> to <to>:
4340 void debug_check_no_locks_freed(const void *mem_from, unsigned long mem_len)
4342 struct task_struct *curr = current;
4343 struct held_lock *hlock;
4344 unsigned long flags;
4347 if (unlikely(!debug_locks))
4350 local_irq_save(flags);
4351 for (i = 0; i < curr->lockdep_depth; i++) {
4352 hlock = curr->held_locks + i;
4354 if (not_in_range(mem_from, mem_len, hlock->instance,
4355 sizeof(*hlock->instance)))
4358 print_freed_lock_bug(curr, mem_from, mem_from + mem_len, hlock);
4361 local_irq_restore(flags);
4363 EXPORT_SYMBOL_GPL(debug_check_no_locks_freed);
4365 static void print_held_locks_bug(void)
4367 if (!debug_locks_off())
4369 if (debug_locks_silent)
4373 printk("=====================================\n");
4374 printk("[ BUG: %s/%d still has locks held! ]\n",
4375 current->comm, task_pid_nr(current));
4376 print_kernel_ident();
4377 printk("-------------------------------------\n");
4378 lockdep_print_held_locks(current);
4379 printk("\nstack backtrace:\n");
4383 void debug_check_no_locks_held(void)
4385 if (unlikely(current->lockdep_depth > 0))
4386 print_held_locks_bug();
4388 EXPORT_SYMBOL_GPL(debug_check_no_locks_held);
4391 void debug_show_all_locks(void)
4393 struct task_struct *g, *p;
4397 if (unlikely(!debug_locks)) {
4398 printk("INFO: lockdep is turned off.\n");
4401 printk("\nShowing all locks held in the system:\n");
4404 * Here we try to get the tasklist_lock as hard as possible,
4405 * if not successful after 2 seconds we ignore it (but keep
4406 * trying). This is to enable a debug printout even if a
4407 * tasklist_lock-holding task deadlocks or crashes.
4410 if (!read_trylock(&tasklist_lock)) {
4412 printk("hm, tasklist_lock locked, retrying... ");
4415 printk(" #%d", 10-count);
4419 printk(" ignoring it.\n");
4423 printk(KERN_CONT " locked it.\n");
4426 do_each_thread(g, p) {
4428 * It's not reliable to print a task's held locks
4429 * if it's not sleeping (or if it's not the current
4432 if (p->state == TASK_RUNNING && p != current)
4434 if (p->lockdep_depth)
4435 lockdep_print_held_locks(p);
4437 if (read_trylock(&tasklist_lock))
4439 } while_each_thread(g, p);
4442 printk("=============================================\n\n");
4445 read_unlock(&tasklist_lock);
4447 EXPORT_SYMBOL_GPL(debug_show_all_locks);
4451 * Careful: only use this function if you are sure that
4452 * the task cannot run in parallel!
4454 void debug_show_held_locks(struct task_struct *task)
4456 if (unlikely(!debug_locks)) {
4457 printk("INFO: lockdep is turned off.\n");
4460 lockdep_print_held_locks(task);
4462 EXPORT_SYMBOL_GPL(debug_show_held_locks);
4464 asmlinkage __visible void lockdep_sys_exit(void)
4466 struct task_struct *curr = current;
4468 if (unlikely(curr->lockdep_depth)) {
4469 if (!debug_locks_off())
4472 printk("================================================\n");
4473 printk("[ BUG: lock held when returning to user space! ]\n");
4474 print_kernel_ident();
4475 printk("------------------------------------------------\n");
4476 printk("%s/%d is leaving the kernel with locks still held!\n",
4477 curr->comm, curr->pid);
4478 lockdep_print_held_locks(curr);
4482 void lockdep_rcu_suspicious(const char *file, const int line, const char *s)
4484 struct task_struct *curr = current;
4486 #ifndef CONFIG_PROVE_RCU_REPEATEDLY
4487 if (!debug_locks_off())
4489 #endif /* #ifdef CONFIG_PROVE_RCU_REPEATEDLY */
4490 /* Note: the following can be executed concurrently, so be careful. */
4492 pr_err("===============================\n");
4493 pr_err("[ ERR: suspicious RCU usage. ]\n");
4494 print_kernel_ident();
4495 pr_err("-------------------------------\n");
4496 pr_err("%s:%d %s!\n", file, line, s);
4497 pr_err("\nother info that might help us debug this:\n\n");
4498 pr_err("\n%srcu_scheduler_active = %d, debug_locks = %d\n",
4499 !rcu_lockdep_current_cpu_online()
4500 ? "RCU used illegally from offline CPU!\n"
4501 : !rcu_is_watching()
4502 ? "RCU used illegally from idle CPU!\n"
4504 rcu_scheduler_active, debug_locks);
4507 * If a CPU is in the RCU-free window in idle (ie: in the section
4508 * between rcu_idle_enter() and rcu_idle_exit(), then RCU
4509 * considers that CPU to be in an "extended quiescent state",
4510 * which means that RCU will be completely ignoring that CPU.
4511 * Therefore, rcu_read_lock() and friends have absolutely no
4512 * effect on a CPU running in that state. In other words, even if
4513 * such an RCU-idle CPU has called rcu_read_lock(), RCU might well
4514 * delete data structures out from under it. RCU really has no
4515 * choice here: we need to keep an RCU-free window in idle where
4516 * the CPU may possibly enter into low power mode. This way we can
4517 * notice an extended quiescent state to other CPUs that started a grace
4518 * period. Otherwise we would delay any grace period as long as we run
4521 * So complain bitterly if someone does call rcu_read_lock(),
4522 * rcu_read_lock_bh() and so on from extended quiescent states.
4524 if (!rcu_is_watching())
4525 printk("RCU used illegally from extended quiescent state!\n");
4527 lockdep_print_held_locks(curr);
4528 printk("\nstack backtrace:\n");
4531 EXPORT_SYMBOL_GPL(lockdep_rcu_suspicious);