4 * Runtime locking correctness validator
6 * Started by Ingo Molnar:
8 * Copyright (C) 2006 Red Hat, Inc., Ingo Molnar <mingo@redhat.com>
10 * this code maps all the lock dependencies as they occur in a live kernel
11 * and will warn about the following classes of locking bugs:
13 * - lock inversion scenarios
14 * - circular lock dependencies
15 * - hardirq/softirq safe/unsafe locking bugs
17 * Bugs are reported even if the current locking scenario does not cause
18 * any deadlock at this point.
20 * I.e. if anytime in the past two locks were taken in a different order,
21 * even if it happened for another task, even if those were different
22 * locks (but of the same class as this lock), this code will detect it.
24 * Thanks to Arjan van de Ven for coming up with the initial idea of
25 * mapping lock dependencies runtime.
27 #include <linux/mutex.h>
28 #include <linux/sched.h>
29 #include <linux/delay.h>
30 #include <linux/module.h>
31 #include <linux/proc_fs.h>
32 #include <linux/seq_file.h>
33 #include <linux/spinlock.h>
34 #include <linux/kallsyms.h>
35 #include <linux/interrupt.h>
36 #include <linux/stacktrace.h>
37 #include <linux/debug_locks.h>
38 #include <linux/irqflags.h>
39 #include <linux/utsname.h>
41 #include <asm/sections.h>
43 #include "lockdep_internals.h"
46 * lockdep_lock: protects the lockdep graph, the hashes and the
47 * class/list/hash allocators.
49 * This is one of the rare exceptions where it's justified
50 * to use a raw spinlock - we really dont want the spinlock
51 * code to recurse back into the lockdep code...
53 static raw_spinlock_t lockdep_lock = (raw_spinlock_t)__RAW_SPIN_LOCK_UNLOCKED;
55 static int graph_lock(void)
57 __raw_spin_lock(&lockdep_lock);
59 * Make sure that if another CPU detected a bug while
60 * walking the graph we dont change it (while the other
61 * CPU is busy printing out stuff with the graph lock
65 __raw_spin_unlock(&lockdep_lock);
71 static inline int graph_unlock(void)
73 if (debug_locks && !__raw_spin_is_locked(&lockdep_lock))
74 return DEBUG_LOCKS_WARN_ON(1);
76 __raw_spin_unlock(&lockdep_lock);
81 * Turn lock debugging off and return with 0 if it was off already,
82 * and also release the graph lock:
84 static inline int debug_locks_off_graph_unlock(void)
86 int ret = debug_locks_off();
88 __raw_spin_unlock(&lockdep_lock);
93 static int lockdep_initialized;
95 unsigned long nr_list_entries;
96 static struct lock_list list_entries[MAX_LOCKDEP_ENTRIES];
99 * Allocate a lockdep entry. (assumes the graph_lock held, returns
100 * with NULL on failure)
102 static struct lock_list *alloc_list_entry(void)
104 if (nr_list_entries >= MAX_LOCKDEP_ENTRIES) {
105 if (!debug_locks_off_graph_unlock())
108 printk("BUG: MAX_LOCKDEP_ENTRIES too low!\n");
109 printk("turning off the locking correctness validator.\n");
112 return list_entries + nr_list_entries++;
116 * All data structures here are protected by the global debug_lock.
118 * Mutex key structs only get allocated, once during bootup, and never
119 * get freed - this significantly simplifies the debugging code.
121 unsigned long nr_lock_classes;
122 static struct lock_class lock_classes[MAX_LOCKDEP_KEYS];
125 * We keep a global list of all lock classes. The list only grows,
126 * never shrinks. The list is only accessed with the lockdep
127 * spinlock lock held.
129 LIST_HEAD(all_lock_classes);
132 * The lockdep classes are in a hash-table as well, for fast lookup:
134 #define CLASSHASH_BITS (MAX_LOCKDEP_KEYS_BITS - 1)
135 #define CLASSHASH_SIZE (1UL << CLASSHASH_BITS)
136 #define CLASSHASH_MASK (CLASSHASH_SIZE - 1)
137 #define __classhashfn(key) ((((unsigned long)key >> CLASSHASH_BITS) + (unsigned long)key) & CLASSHASH_MASK)
138 #define classhashentry(key) (classhash_table + __classhashfn((key)))
140 static struct list_head classhash_table[CLASSHASH_SIZE];
142 unsigned long nr_lock_chains;
143 static struct lock_chain lock_chains[MAX_LOCKDEP_CHAINS];
146 * We put the lock dependency chains into a hash-table as well, to cache
149 #define CHAINHASH_BITS (MAX_LOCKDEP_CHAINS_BITS-1)
150 #define CHAINHASH_SIZE (1UL << CHAINHASH_BITS)
151 #define CHAINHASH_MASK (CHAINHASH_SIZE - 1)
152 #define __chainhashfn(chain) \
153 (((chain >> CHAINHASH_BITS) + chain) & CHAINHASH_MASK)
154 #define chainhashentry(chain) (chainhash_table + __chainhashfn((chain)))
156 static struct list_head chainhash_table[CHAINHASH_SIZE];
159 * The hash key of the lock dependency chains is a hash itself too:
160 * it's a hash of all locks taken up to that lock, including that lock.
161 * It's a 64-bit hash, because it's important for the keys to be
164 #define iterate_chain_key(key1, key2) \
165 (((key1) << MAX_LOCKDEP_KEYS_BITS) ^ \
166 ((key1) >> (64-MAX_LOCKDEP_KEYS_BITS)) ^ \
169 void lockdep_off(void)
171 current->lockdep_recursion++;
174 EXPORT_SYMBOL(lockdep_off);
176 void lockdep_on(void)
178 current->lockdep_recursion--;
181 EXPORT_SYMBOL(lockdep_on);
184 * Debugging switches:
188 #define VERY_VERBOSE 0
191 # define HARDIRQ_VERBOSE 1
192 # define SOFTIRQ_VERBOSE 1
194 # define HARDIRQ_VERBOSE 0
195 # define SOFTIRQ_VERBOSE 0
198 #if VERBOSE || HARDIRQ_VERBOSE || SOFTIRQ_VERBOSE
200 * Quick filtering for interesting events:
202 static int class_filter(struct lock_class *class)
206 if (class->name_version == 1 &&
207 !strcmp(class->name, "lockname"))
209 if (class->name_version == 1 &&
210 !strcmp(class->name, "&struct->lockfield"))
213 /* Filter everything else. 1 would be to allow everything else */
218 static int verbose(struct lock_class *class)
221 return class_filter(class);
226 #ifdef CONFIG_TRACE_IRQFLAGS
228 static int hardirq_verbose(struct lock_class *class)
231 return class_filter(class);
236 static int softirq_verbose(struct lock_class *class)
239 return class_filter(class);
247 * Stack-trace: tightly packed array of stack backtrace
248 * addresses. Protected by the graph_lock.
250 unsigned long nr_stack_trace_entries;
251 static unsigned long stack_trace[MAX_STACK_TRACE_ENTRIES];
253 static int save_trace(struct stack_trace *trace)
255 trace->nr_entries = 0;
256 trace->max_entries = MAX_STACK_TRACE_ENTRIES - nr_stack_trace_entries;
257 trace->entries = stack_trace + nr_stack_trace_entries;
260 trace->all_contexts = 0;
262 save_stack_trace(trace, NULL);
264 trace->max_entries = trace->nr_entries;
266 nr_stack_trace_entries += trace->nr_entries;
268 if (nr_stack_trace_entries == MAX_STACK_TRACE_ENTRIES) {
269 if (!debug_locks_off_graph_unlock())
272 printk("BUG: MAX_STACK_TRACE_ENTRIES too low!\n");
273 printk("turning off the locking correctness validator.\n");
282 unsigned int nr_hardirq_chains;
283 unsigned int nr_softirq_chains;
284 unsigned int nr_process_chains;
285 unsigned int max_lockdep_depth;
286 unsigned int max_recursion_depth;
288 #ifdef CONFIG_DEBUG_LOCKDEP
290 * We cannot printk in early bootup code. Not even early_printk()
291 * might work. So we mark any initialization errors and printk
292 * about it later on, in lockdep_info().
294 static int lockdep_init_error;
297 * Various lockdep statistics:
299 atomic_t chain_lookup_hits;
300 atomic_t chain_lookup_misses;
301 atomic_t hardirqs_on_events;
302 atomic_t hardirqs_off_events;
303 atomic_t redundant_hardirqs_on;
304 atomic_t redundant_hardirqs_off;
305 atomic_t softirqs_on_events;
306 atomic_t softirqs_off_events;
307 atomic_t redundant_softirqs_on;
308 atomic_t redundant_softirqs_off;
309 atomic_t nr_unused_locks;
310 atomic_t nr_cyclic_checks;
311 atomic_t nr_cyclic_check_recursions;
312 atomic_t nr_find_usage_forwards_checks;
313 atomic_t nr_find_usage_forwards_recursions;
314 atomic_t nr_find_usage_backwards_checks;
315 atomic_t nr_find_usage_backwards_recursions;
316 # define debug_atomic_inc(ptr) atomic_inc(ptr)
317 # define debug_atomic_dec(ptr) atomic_dec(ptr)
318 # define debug_atomic_read(ptr) atomic_read(ptr)
320 # define debug_atomic_inc(ptr) do { } while (0)
321 # define debug_atomic_dec(ptr) do { } while (0)
322 # define debug_atomic_read(ptr) 0
329 static const char *usage_str[] =
331 [LOCK_USED] = "initial-use ",
332 [LOCK_USED_IN_HARDIRQ] = "in-hardirq-W",
333 [LOCK_USED_IN_SOFTIRQ] = "in-softirq-W",
334 [LOCK_ENABLED_SOFTIRQS] = "softirq-on-W",
335 [LOCK_ENABLED_HARDIRQS] = "hardirq-on-W",
336 [LOCK_USED_IN_HARDIRQ_READ] = "in-hardirq-R",
337 [LOCK_USED_IN_SOFTIRQ_READ] = "in-softirq-R",
338 [LOCK_ENABLED_SOFTIRQS_READ] = "softirq-on-R",
339 [LOCK_ENABLED_HARDIRQS_READ] = "hardirq-on-R",
342 const char * __get_key_name(struct lockdep_subclass_key *key, char *str)
344 unsigned long offs, size;
347 return kallsyms_lookup((unsigned long)key, &size, &offs, &modname, str);
351 get_usage_chars(struct lock_class *class, char *c1, char *c2, char *c3, char *c4)
353 *c1 = '.', *c2 = '.', *c3 = '.', *c4 = '.';
355 if (class->usage_mask & LOCKF_USED_IN_HARDIRQ)
358 if (class->usage_mask & LOCKF_ENABLED_HARDIRQS)
361 if (class->usage_mask & LOCKF_USED_IN_SOFTIRQ)
364 if (class->usage_mask & LOCKF_ENABLED_SOFTIRQS)
367 if (class->usage_mask & LOCKF_ENABLED_HARDIRQS_READ)
369 if (class->usage_mask & LOCKF_USED_IN_HARDIRQ_READ) {
371 if (class->usage_mask & LOCKF_ENABLED_HARDIRQS_READ)
375 if (class->usage_mask & LOCKF_ENABLED_SOFTIRQS_READ)
377 if (class->usage_mask & LOCKF_USED_IN_SOFTIRQ_READ) {
379 if (class->usage_mask & LOCKF_ENABLED_SOFTIRQS_READ)
384 static void print_lock_name(struct lock_class *class)
386 char str[KSYM_NAME_LEN + 1], c1, c2, c3, c4;
389 get_usage_chars(class, &c1, &c2, &c3, &c4);
393 name = __get_key_name(class->key, str);
394 printk(" (%s", name);
396 printk(" (%s", name);
397 if (class->name_version > 1)
398 printk("#%d", class->name_version);
400 printk("/%d", class->subclass);
402 printk("){%c%c%c%c}", c1, c2, c3, c4);
405 static void print_lockdep_cache(struct lockdep_map *lock)
408 char str[KSYM_NAME_LEN + 1];
412 name = __get_key_name(lock->key->subkeys, str);
417 static void print_lock(struct held_lock *hlock)
419 print_lock_name(hlock->class);
421 print_ip_sym(hlock->acquire_ip);
424 static void lockdep_print_held_locks(struct task_struct *curr)
426 int i, depth = curr->lockdep_depth;
429 printk("no locks held by %s/%d.\n", curr->comm, curr->pid);
432 printk("%d lock%s held by %s/%d:\n",
433 depth, depth > 1 ? "s" : "", curr->comm, curr->pid);
435 for (i = 0; i < depth; i++) {
437 print_lock(curr->held_locks + i);
441 static void print_lock_class_header(struct lock_class *class, int depth)
445 printk("%*s->", depth, "");
446 print_lock_name(class);
447 printk(" ops: %lu", class->ops);
450 for (bit = 0; bit < LOCK_USAGE_STATES; bit++) {
451 if (class->usage_mask & (1 << bit)) {
454 len += printk("%*s %s", depth, "", usage_str[bit]);
455 len += printk(" at:\n");
456 print_stack_trace(class->usage_traces + bit, len);
459 printk("%*s }\n", depth, "");
461 printk("%*s ... key at: ",depth,"");
462 print_ip_sym((unsigned long)class->key);
466 * printk all lock dependencies starting at <entry>:
468 static void print_lock_dependencies(struct lock_class *class, int depth)
470 struct lock_list *entry;
472 if (DEBUG_LOCKS_WARN_ON(depth >= 20))
475 print_lock_class_header(class, depth);
477 list_for_each_entry(entry, &class->locks_after, entry) {
478 if (DEBUG_LOCKS_WARN_ON(!entry->class))
481 print_lock_dependencies(entry->class, depth + 1);
483 printk("%*s ... acquired at:\n",depth,"");
484 print_stack_trace(&entry->trace, 2);
490 * Add a new dependency to the head of the list:
492 static int add_lock_to_list(struct lock_class *class, struct lock_class *this,
493 struct list_head *head, unsigned long ip)
495 struct lock_list *entry;
497 * Lock not present yet - get a new dependency struct and
498 * add it to the list:
500 entry = alloc_list_entry();
505 if (!save_trace(&entry->trace))
509 * Since we never remove from the dependency list, the list can
510 * be walked lockless by other CPUs, it's only allocation
511 * that must be protected by the spinlock. But this also means
512 * we must make new entries visible only once writes to the
513 * entry become visible - hence the RCU op:
515 list_add_tail_rcu(&entry->entry, head);
521 * Recursive, forwards-direction lock-dependency checking, used for
522 * both noncyclic checking and for hardirq-unsafe/softirq-unsafe
525 * (to keep the stackframe of the recursive functions small we
526 * use these global variables, and we also mark various helper
527 * functions as noinline.)
529 static struct held_lock *check_source, *check_target;
532 * Print a dependency chain entry (this is only done when a deadlock
533 * has been detected):
536 print_circular_bug_entry(struct lock_list *target, unsigned int depth)
538 if (debug_locks_silent)
540 printk("\n-> #%u", depth);
541 print_lock_name(target->class);
543 print_stack_trace(&target->trace, 6);
548 static void print_kernel_version(void)
550 printk("%s %.*s\n", init_utsname()->release,
551 (int)strcspn(init_utsname()->version, " "),
552 init_utsname()->version);
556 * When a circular dependency is detected, print the
560 print_circular_bug_header(struct lock_list *entry, unsigned int depth)
562 struct task_struct *curr = current;
564 if (!debug_locks_off_graph_unlock() || debug_locks_silent)
567 printk("\n=======================================================\n");
568 printk( "[ INFO: possible circular locking dependency detected ]\n");
569 print_kernel_version();
570 printk( "-------------------------------------------------------\n");
571 printk("%s/%d is trying to acquire lock:\n",
572 curr->comm, curr->pid);
573 print_lock(check_source);
574 printk("\nbut task is already holding lock:\n");
575 print_lock(check_target);
576 printk("\nwhich lock already depends on the new lock.\n\n");
577 printk("\nthe existing dependency chain (in reverse order) is:\n");
579 print_circular_bug_entry(entry, depth);
584 static noinline int print_circular_bug_tail(void)
586 struct task_struct *curr = current;
587 struct lock_list this;
589 if (debug_locks_silent)
592 this.class = check_source->class;
593 if (!save_trace(&this.trace))
596 print_circular_bug_entry(&this, 0);
598 printk("\nother info that might help us debug this:\n\n");
599 lockdep_print_held_locks(curr);
601 printk("\nstack backtrace:\n");
607 #define RECURSION_LIMIT 40
609 static int noinline print_infinite_recursion_bug(void)
611 if (!debug_locks_off_graph_unlock())
620 * Prove that the dependency graph starting at <entry> can not
621 * lead to <target>. Print an error and return 0 if it does.
624 check_noncircular(struct lock_class *source, unsigned int depth)
626 struct lock_list *entry;
628 debug_atomic_inc(&nr_cyclic_check_recursions);
629 if (depth > max_recursion_depth)
630 max_recursion_depth = depth;
631 if (depth >= RECURSION_LIMIT)
632 return print_infinite_recursion_bug();
634 * Check this lock's dependency list:
636 list_for_each_entry(entry, &source->locks_after, entry) {
637 if (entry->class == check_target->class)
638 return print_circular_bug_header(entry, depth+1);
639 debug_atomic_inc(&nr_cyclic_checks);
640 if (!check_noncircular(entry->class, depth+1))
641 return print_circular_bug_entry(entry, depth+1);
646 static int very_verbose(struct lock_class *class)
649 return class_filter(class);
653 #ifdef CONFIG_TRACE_IRQFLAGS
656 * Forwards and backwards subgraph searching, for the purposes of
657 * proving that two subgraphs can be connected by a new dependency
658 * without creating any illegal irq-safe -> irq-unsafe lock dependency.
660 static enum lock_usage_bit find_usage_bit;
661 static struct lock_class *forwards_match, *backwards_match;
664 * Find a node in the forwards-direction dependency sub-graph starting
665 * at <source> that matches <find_usage_bit>.
667 * Return 2 if such a node exists in the subgraph, and put that node
668 * into <forwards_match>.
670 * Return 1 otherwise and keep <forwards_match> unchanged.
674 find_usage_forwards(struct lock_class *source, unsigned int depth)
676 struct lock_list *entry;
679 if (depth > max_recursion_depth)
680 max_recursion_depth = depth;
681 if (depth >= RECURSION_LIMIT)
682 return print_infinite_recursion_bug();
684 debug_atomic_inc(&nr_find_usage_forwards_checks);
685 if (source->usage_mask & (1 << find_usage_bit)) {
686 forwards_match = source;
691 * Check this lock's dependency list:
693 list_for_each_entry(entry, &source->locks_after, entry) {
694 debug_atomic_inc(&nr_find_usage_forwards_recursions);
695 ret = find_usage_forwards(entry->class, depth+1);
696 if (ret == 2 || ret == 0)
703 * Find a node in the backwards-direction dependency sub-graph starting
704 * at <source> that matches <find_usage_bit>.
706 * Return 2 if such a node exists in the subgraph, and put that node
707 * into <backwards_match>.
709 * Return 1 otherwise and keep <backwards_match> unchanged.
713 find_usage_backwards(struct lock_class *source, unsigned int depth)
715 struct lock_list *entry;
718 if (!__raw_spin_is_locked(&lockdep_lock))
719 return DEBUG_LOCKS_WARN_ON(1);
721 if (depth > max_recursion_depth)
722 max_recursion_depth = depth;
723 if (depth >= RECURSION_LIMIT)
724 return print_infinite_recursion_bug();
726 debug_atomic_inc(&nr_find_usage_backwards_checks);
727 if (source->usage_mask & (1 << find_usage_bit)) {
728 backwards_match = source;
733 * Check this lock's dependency list:
735 list_for_each_entry(entry, &source->locks_before, entry) {
736 debug_atomic_inc(&nr_find_usage_backwards_recursions);
737 ret = find_usage_backwards(entry->class, depth+1);
738 if (ret == 2 || ret == 0)
745 print_bad_irq_dependency(struct task_struct *curr,
746 struct held_lock *prev,
747 struct held_lock *next,
748 enum lock_usage_bit bit1,
749 enum lock_usage_bit bit2,
750 const char *irqclass)
752 if (!debug_locks_off_graph_unlock() || debug_locks_silent)
755 printk("\n======================================================\n");
756 printk( "[ INFO: %s-safe -> %s-unsafe lock order detected ]\n",
758 print_kernel_version();
759 printk( "------------------------------------------------------\n");
760 printk("%s/%d [HC%u[%lu]:SC%u[%lu]:HE%u:SE%u] is trying to acquire:\n",
761 curr->comm, curr->pid,
762 curr->hardirq_context, hardirq_count() >> HARDIRQ_SHIFT,
763 curr->softirq_context, softirq_count() >> SOFTIRQ_SHIFT,
764 curr->hardirqs_enabled,
765 curr->softirqs_enabled);
768 printk("\nand this task is already holding:\n");
770 printk("which would create a new lock dependency:\n");
771 print_lock_name(prev->class);
773 print_lock_name(next->class);
776 printk("\nbut this new dependency connects a %s-irq-safe lock:\n",
778 print_lock_name(backwards_match);
779 printk("\n... which became %s-irq-safe at:\n", irqclass);
781 print_stack_trace(backwards_match->usage_traces + bit1, 1);
783 printk("\nto a %s-irq-unsafe lock:\n", irqclass);
784 print_lock_name(forwards_match);
785 printk("\n... which became %s-irq-unsafe at:\n", irqclass);
788 print_stack_trace(forwards_match->usage_traces + bit2, 1);
790 printk("\nother info that might help us debug this:\n\n");
791 lockdep_print_held_locks(curr);
793 printk("\nthe %s-irq-safe lock's dependencies:\n", irqclass);
794 print_lock_dependencies(backwards_match, 0);
796 printk("\nthe %s-irq-unsafe lock's dependencies:\n", irqclass);
797 print_lock_dependencies(forwards_match, 0);
799 printk("\nstack backtrace:\n");
806 check_usage(struct task_struct *curr, struct held_lock *prev,
807 struct held_lock *next, enum lock_usage_bit bit_backwards,
808 enum lock_usage_bit bit_forwards, const char *irqclass)
812 find_usage_bit = bit_backwards;
813 /* fills in <backwards_match> */
814 ret = find_usage_backwards(prev->class, 0);
815 if (!ret || ret == 1)
818 find_usage_bit = bit_forwards;
819 ret = find_usage_forwards(next->class, 0);
820 if (!ret || ret == 1)
823 return print_bad_irq_dependency(curr, prev, next,
824 bit_backwards, bit_forwards, irqclass);
830 print_deadlock_bug(struct task_struct *curr, struct held_lock *prev,
831 struct held_lock *next)
833 if (!debug_locks_off_graph_unlock() || debug_locks_silent)
836 printk("\n=============================================\n");
837 printk( "[ INFO: possible recursive locking detected ]\n");
838 print_kernel_version();
839 printk( "---------------------------------------------\n");
840 printk("%s/%d is trying to acquire lock:\n",
841 curr->comm, curr->pid);
843 printk("\nbut task is already holding lock:\n");
846 printk("\nother info that might help us debug this:\n");
847 lockdep_print_held_locks(curr);
849 printk("\nstack backtrace:\n");
856 * Check whether we are holding such a class already.
858 * (Note that this has to be done separately, because the graph cannot
859 * detect such classes of deadlocks.)
861 * Returns: 0 on deadlock detected, 1 on OK, 2 on recursive read
864 check_deadlock(struct task_struct *curr, struct held_lock *next,
865 struct lockdep_map *next_instance, int read)
867 struct held_lock *prev;
870 for (i = 0; i < curr->lockdep_depth; i++) {
871 prev = curr->held_locks + i;
872 if (prev->class != next->class)
875 * Allow read-after-read recursion of the same
876 * lock class (i.e. read_lock(lock)+read_lock(lock)):
878 if ((read == 2) && prev->read)
880 return print_deadlock_bug(curr, prev, next);
886 * There was a chain-cache miss, and we are about to add a new dependency
887 * to a previous lock. We recursively validate the following rules:
889 * - would the adding of the <prev> -> <next> dependency create a
890 * circular dependency in the graph? [== circular deadlock]
892 * - does the new prev->next dependency connect any hardirq-safe lock
893 * (in the full backwards-subgraph starting at <prev>) with any
894 * hardirq-unsafe lock (in the full forwards-subgraph starting at
895 * <next>)? [== illegal lock inversion with hardirq contexts]
897 * - does the new prev->next dependency connect any softirq-safe lock
898 * (in the full backwards-subgraph starting at <prev>) with any
899 * softirq-unsafe lock (in the full forwards-subgraph starting at
900 * <next>)? [== illegal lock inversion with softirq contexts]
902 * any of these scenarios could lead to a deadlock.
904 * Then if all the validations pass, we add the forwards and backwards
908 check_prev_add(struct task_struct *curr, struct held_lock *prev,
909 struct held_lock *next)
911 struct lock_list *entry;
915 * Prove that the new <prev> -> <next> dependency would not
916 * create a circular dependency in the graph. (We do this by
917 * forward-recursing into the graph starting at <next>, and
918 * checking whether we can reach <prev>.)
920 * We are using global variables to control the recursion, to
921 * keep the stackframe size of the recursive functions low:
925 if (!(check_noncircular(next->class, 0)))
926 return print_circular_bug_tail();
928 #ifdef CONFIG_TRACE_IRQFLAGS
930 * Prove that the new dependency does not connect a hardirq-safe
931 * lock with a hardirq-unsafe lock - to achieve this we search
932 * the backwards-subgraph starting at <prev>, and the
933 * forwards-subgraph starting at <next>:
935 if (!check_usage(curr, prev, next, LOCK_USED_IN_HARDIRQ,
936 LOCK_ENABLED_HARDIRQS, "hard"))
940 * Prove that the new dependency does not connect a hardirq-safe-read
941 * lock with a hardirq-unsafe lock - to achieve this we search
942 * the backwards-subgraph starting at <prev>, and the
943 * forwards-subgraph starting at <next>:
945 if (!check_usage(curr, prev, next, LOCK_USED_IN_HARDIRQ_READ,
946 LOCK_ENABLED_HARDIRQS, "hard-read"))
950 * Prove that the new dependency does not connect a softirq-safe
951 * lock with a softirq-unsafe lock - to achieve this we search
952 * the backwards-subgraph starting at <prev>, and the
953 * forwards-subgraph starting at <next>:
955 if (!check_usage(curr, prev, next, LOCK_USED_IN_SOFTIRQ,
956 LOCK_ENABLED_SOFTIRQS, "soft"))
959 * Prove that the new dependency does not connect a softirq-safe-read
960 * lock with a softirq-unsafe lock - to achieve this we search
961 * the backwards-subgraph starting at <prev>, and the
962 * forwards-subgraph starting at <next>:
964 if (!check_usage(curr, prev, next, LOCK_USED_IN_SOFTIRQ_READ,
965 LOCK_ENABLED_SOFTIRQS, "soft"))
969 * For recursive read-locks we do all the dependency checks,
970 * but we dont store read-triggered dependencies (only
971 * write-triggered dependencies). This ensures that only the
972 * write-side dependencies matter, and that if for example a
973 * write-lock never takes any other locks, then the reads are
974 * equivalent to a NOP.
976 if (next->read == 2 || prev->read == 2)
979 * Is the <prev> -> <next> dependency already present?
981 * (this may occur even though this is a new chain: consider
982 * e.g. the L1 -> L2 -> L3 -> L4 and the L5 -> L1 -> L2 -> L3
983 * chains - the second one will be new, but L1 already has
984 * L2 added to its dependency list, due to the first chain.)
986 list_for_each_entry(entry, &prev->class->locks_after, entry) {
987 if (entry->class == next->class)
992 * Ok, all validations passed, add the new lock
993 * to the previous lock's dependency list:
995 ret = add_lock_to_list(prev->class, next->class,
996 &prev->class->locks_after, next->acquire_ip);
1000 ret = add_lock_to_list(next->class, prev->class,
1001 &next->class->locks_before, next->acquire_ip);
1006 * Debugging printouts:
1008 if (verbose(prev->class) || verbose(next->class)) {
1010 printk("\n new dependency: ");
1011 print_lock_name(prev->class);
1013 print_lock_name(next->class);
1016 return graph_lock();
1022 * Add the dependency to all directly-previous locks that are 'relevant'.
1023 * The ones that are relevant are (in increasing distance from curr):
1024 * all consecutive trylock entries and the final non-trylock entry - or
1025 * the end of this context's lock-chain - whichever comes first.
1028 check_prevs_add(struct task_struct *curr, struct held_lock *next)
1030 int depth = curr->lockdep_depth;
1031 struct held_lock *hlock;
1036 * Depth must not be zero for a non-head lock:
1041 * At least two relevant locks must exist for this
1044 if (curr->held_locks[depth].irq_context !=
1045 curr->held_locks[depth-1].irq_context)
1049 hlock = curr->held_locks + depth-1;
1051 * Only non-recursive-read entries get new dependencies
1054 if (hlock->read != 2) {
1055 if (!check_prev_add(curr, hlock, next))
1058 * Stop after the first non-trylock entry,
1059 * as non-trylock entries have added their
1060 * own direct dependencies already, so this
1061 * lock is connected to them indirectly:
1063 if (!hlock->trylock)
1068 * End of lock-stack?
1073 * Stop the search if we cross into another context:
1075 if (curr->held_locks[depth].irq_context !=
1076 curr->held_locks[depth-1].irq_context)
1081 if (!debug_locks_off_graph_unlock())
1091 * Is this the address of a static object:
1093 static int static_obj(void *obj)
1095 unsigned long start = (unsigned long) &_stext,
1096 end = (unsigned long) &_end,
1097 addr = (unsigned long) obj;
1105 if ((addr >= start) && (addr < end))
1112 for_each_possible_cpu(i) {
1113 start = (unsigned long) &__per_cpu_start + per_cpu_offset(i);
1114 end = (unsigned long) &__per_cpu_start + PERCPU_ENOUGH_ROOM
1115 + per_cpu_offset(i);
1117 if ((addr >= start) && (addr < end))
1125 return is_module_address(addr);
1129 * To make lock name printouts unique, we calculate a unique
1130 * class->name_version generation counter:
1132 static int count_matching_names(struct lock_class *new_class)
1134 struct lock_class *class;
1137 if (!new_class->name)
1140 list_for_each_entry(class, &all_lock_classes, lock_entry) {
1141 if (new_class->key - new_class->subclass == class->key)
1142 return class->name_version;
1143 if (class->name && !strcmp(class->name, new_class->name))
1144 count = max(count, class->name_version);
1151 * Register a lock's class in the hash-table, if the class is not present
1152 * yet. Otherwise we look it up. We cache the result in the lock object
1153 * itself, so actual lookup of the hash should be once per lock object.
1155 static inline struct lock_class *
1156 look_up_lock_class(struct lockdep_map *lock, unsigned int subclass)
1158 struct lockdep_subclass_key *key;
1159 struct list_head *hash_head;
1160 struct lock_class *class;
1162 #ifdef CONFIG_DEBUG_LOCKDEP
1164 * If the architecture calls into lockdep before initializing
1165 * the hashes then we'll warn about it later. (we cannot printk
1168 if (unlikely(!lockdep_initialized)) {
1170 lockdep_init_error = 1;
1175 * Static locks do not have their class-keys yet - for them the key
1176 * is the lock object itself:
1178 if (unlikely(!lock->key))
1179 lock->key = (void *)lock;
1182 * NOTE: the class-key must be unique. For dynamic locks, a static
1183 * lock_class_key variable is passed in through the mutex_init()
1184 * (or spin_lock_init()) call - which acts as the key. For static
1185 * locks we use the lock object itself as the key.
1187 BUILD_BUG_ON(sizeof(struct lock_class_key) > sizeof(struct lock_class));
1189 key = lock->key->subkeys + subclass;
1191 hash_head = classhashentry(key);
1194 * We can walk the hash lockfree, because the hash only
1195 * grows, and we are careful when adding entries to the end:
1197 list_for_each_entry(class, hash_head, hash_entry)
1198 if (class->key == key)
1205 * Register a lock's class in the hash-table, if the class is not present
1206 * yet. Otherwise we look it up. We cache the result in the lock object
1207 * itself, so actual lookup of the hash should be once per lock object.
1209 static inline struct lock_class *
1210 register_lock_class(struct lockdep_map *lock, unsigned int subclass, int force)
1212 struct lockdep_subclass_key *key;
1213 struct list_head *hash_head;
1214 struct lock_class *class;
1215 unsigned long flags;
1217 class = look_up_lock_class(lock, subclass);
1222 * Debug-check: all keys must be persistent!
1224 if (!static_obj(lock->key)) {
1226 printk("INFO: trying to register non-static key.\n");
1227 printk("the code is fine but needs lockdep annotation.\n");
1228 printk("turning off the locking correctness validator.\n");
1234 key = lock->key->subkeys + subclass;
1235 hash_head = classhashentry(key);
1237 raw_local_irq_save(flags);
1238 if (!graph_lock()) {
1239 raw_local_irq_restore(flags);
1243 * We have to do the hash-walk again, to avoid races
1246 list_for_each_entry(class, hash_head, hash_entry)
1247 if (class->key == key)
1248 goto out_unlock_set;
1250 * Allocate a new key from the static array, and add it to
1253 if (nr_lock_classes >= MAX_LOCKDEP_KEYS) {
1254 if (!debug_locks_off_graph_unlock()) {
1255 raw_local_irq_restore(flags);
1258 raw_local_irq_restore(flags);
1260 printk("BUG: MAX_LOCKDEP_KEYS too low!\n");
1261 printk("turning off the locking correctness validator.\n");
1264 class = lock_classes + nr_lock_classes++;
1265 debug_atomic_inc(&nr_unused_locks);
1267 class->name = lock->name;
1268 class->subclass = subclass;
1269 INIT_LIST_HEAD(&class->lock_entry);
1270 INIT_LIST_HEAD(&class->locks_before);
1271 INIT_LIST_HEAD(&class->locks_after);
1272 class->name_version = count_matching_names(class);
1274 * We use RCU's safe list-add method to make
1275 * parallel walking of the hash-list safe:
1277 list_add_tail_rcu(&class->hash_entry, hash_head);
1279 if (verbose(class)) {
1281 raw_local_irq_restore(flags);
1283 printk("\nnew class %p: %s", class->key, class->name);
1284 if (class->name_version > 1)
1285 printk("#%d", class->name_version);
1289 raw_local_irq_save(flags);
1290 if (!graph_lock()) {
1291 raw_local_irq_restore(flags);
1297 raw_local_irq_restore(flags);
1299 if (!subclass || force)
1300 lock->class_cache = class;
1302 if (DEBUG_LOCKS_WARN_ON(class->subclass != subclass))
1309 * Look up a dependency chain. If the key is not present yet then
1310 * add it and return 0 - in this case the new dependency chain is
1311 * validated. If the key is already hashed, return 1.
1313 static inline int lookup_chain_cache(u64 chain_key, struct lock_class *class)
1315 struct list_head *hash_head = chainhashentry(chain_key);
1316 struct lock_chain *chain;
1318 if (DEBUG_LOCKS_WARN_ON(!irqs_disabled()))
1321 * We can walk it lock-free, because entries only get added
1324 list_for_each_entry(chain, hash_head, entry) {
1325 if (chain->chain_key == chain_key) {
1327 debug_atomic_inc(&chain_lookup_hits);
1328 if (very_verbose(class))
1329 printk("\nhash chain already cached, key: "
1330 "%016Lx tail class: [%p] %s\n",
1331 (unsigned long long)chain_key,
1332 class->key, class->name);
1336 if (very_verbose(class))
1337 printk("\nnew hash chain, key: %016Lx tail class: [%p] %s\n",
1338 (unsigned long long)chain_key, class->key, class->name);
1340 * Allocate a new chain entry from the static array, and add
1346 * We have to walk the chain again locked - to avoid duplicates:
1348 list_for_each_entry(chain, hash_head, entry) {
1349 if (chain->chain_key == chain_key) {
1354 if (unlikely(nr_lock_chains >= MAX_LOCKDEP_CHAINS)) {
1355 if (!debug_locks_off_graph_unlock())
1358 printk("BUG: MAX_LOCKDEP_CHAINS too low!\n");
1359 printk("turning off the locking correctness validator.\n");
1362 chain = lock_chains + nr_lock_chains++;
1363 chain->chain_key = chain_key;
1364 list_add_tail_rcu(&chain->entry, hash_head);
1365 debug_atomic_inc(&chain_lookup_misses);
1366 #ifdef CONFIG_TRACE_IRQFLAGS
1367 if (current->hardirq_context)
1368 nr_hardirq_chains++;
1370 if (current->softirq_context)
1371 nr_softirq_chains++;
1373 nr_process_chains++;
1376 nr_process_chains++;
1383 * We are building curr_chain_key incrementally, so double-check
1384 * it from scratch, to make sure that it's done correctly:
1386 static void check_chain_key(struct task_struct *curr)
1388 #ifdef CONFIG_DEBUG_LOCKDEP
1389 struct held_lock *hlock, *prev_hlock = NULL;
1393 for (i = 0; i < curr->lockdep_depth; i++) {
1394 hlock = curr->held_locks + i;
1395 if (chain_key != hlock->prev_chain_key) {
1397 printk("hm#1, depth: %u [%u], %016Lx != %016Lx\n",
1398 curr->lockdep_depth, i,
1399 (unsigned long long)chain_key,
1400 (unsigned long long)hlock->prev_chain_key);
1404 id = hlock->class - lock_classes;
1405 if (DEBUG_LOCKS_WARN_ON(id >= MAX_LOCKDEP_KEYS))
1408 if (prev_hlock && (prev_hlock->irq_context !=
1409 hlock->irq_context))
1411 chain_key = iterate_chain_key(chain_key, id);
1414 if (chain_key != curr->curr_chain_key) {
1416 printk("hm#2, depth: %u [%u], %016Lx != %016Lx\n",
1417 curr->lockdep_depth, i,
1418 (unsigned long long)chain_key,
1419 (unsigned long long)curr->curr_chain_key);
1425 #ifdef CONFIG_TRACE_IRQFLAGS
1428 * print irq inversion bug:
1431 print_irq_inversion_bug(struct task_struct *curr, struct lock_class *other,
1432 struct held_lock *this, int forwards,
1433 const char *irqclass)
1435 if (!debug_locks_off_graph_unlock() || debug_locks_silent)
1438 printk("\n=========================================================\n");
1439 printk( "[ INFO: possible irq lock inversion dependency detected ]\n");
1440 print_kernel_version();
1441 printk( "---------------------------------------------------------\n");
1442 printk("%s/%d just changed the state of lock:\n",
1443 curr->comm, curr->pid);
1446 printk("but this lock took another, %s-irq-unsafe lock in the past:\n", irqclass);
1448 printk("but this lock was taken by another, %s-irq-safe lock in the past:\n", irqclass);
1449 print_lock_name(other);
1450 printk("\n\nand interrupts could create inverse lock ordering between them.\n\n");
1452 printk("\nother info that might help us debug this:\n");
1453 lockdep_print_held_locks(curr);
1455 printk("\nthe first lock's dependencies:\n");
1456 print_lock_dependencies(this->class, 0);
1458 printk("\nthe second lock's dependencies:\n");
1459 print_lock_dependencies(other, 0);
1461 printk("\nstack backtrace:\n");
1468 * Prove that in the forwards-direction subgraph starting at <this>
1469 * there is no lock matching <mask>:
1472 check_usage_forwards(struct task_struct *curr, struct held_lock *this,
1473 enum lock_usage_bit bit, const char *irqclass)
1477 find_usage_bit = bit;
1478 /* fills in <forwards_match> */
1479 ret = find_usage_forwards(this->class, 0);
1480 if (!ret || ret == 1)
1483 return print_irq_inversion_bug(curr, forwards_match, this, 1, irqclass);
1487 * Prove that in the backwards-direction subgraph starting at <this>
1488 * there is no lock matching <mask>:
1491 check_usage_backwards(struct task_struct *curr, struct held_lock *this,
1492 enum lock_usage_bit bit, const char *irqclass)
1496 find_usage_bit = bit;
1497 /* fills in <backwards_match> */
1498 ret = find_usage_backwards(this->class, 0);
1499 if (!ret || ret == 1)
1502 return print_irq_inversion_bug(curr, backwards_match, this, 0, irqclass);
1505 void print_irqtrace_events(struct task_struct *curr)
1507 printk("irq event stamp: %u\n", curr->irq_events);
1508 printk("hardirqs last enabled at (%u): ", curr->hardirq_enable_event);
1509 print_ip_sym(curr->hardirq_enable_ip);
1510 printk("hardirqs last disabled at (%u): ", curr->hardirq_disable_event);
1511 print_ip_sym(curr->hardirq_disable_ip);
1512 printk("softirqs last enabled at (%u): ", curr->softirq_enable_event);
1513 print_ip_sym(curr->softirq_enable_ip);
1514 printk("softirqs last disabled at (%u): ", curr->softirq_disable_event);
1515 print_ip_sym(curr->softirq_disable_ip);
1521 print_usage_bug(struct task_struct *curr, struct held_lock *this,
1522 enum lock_usage_bit prev_bit, enum lock_usage_bit new_bit)
1524 if (!debug_locks_off_graph_unlock() || debug_locks_silent)
1527 printk("\n=================================\n");
1528 printk( "[ INFO: inconsistent lock state ]\n");
1529 print_kernel_version();
1530 printk( "---------------------------------\n");
1532 printk("inconsistent {%s} -> {%s} usage.\n",
1533 usage_str[prev_bit], usage_str[new_bit]);
1535 printk("%s/%d [HC%u[%lu]:SC%u[%lu]:HE%u:SE%u] takes:\n",
1536 curr->comm, curr->pid,
1537 trace_hardirq_context(curr), hardirq_count() >> HARDIRQ_SHIFT,
1538 trace_softirq_context(curr), softirq_count() >> SOFTIRQ_SHIFT,
1539 trace_hardirqs_enabled(curr),
1540 trace_softirqs_enabled(curr));
1543 printk("{%s} state was registered at:\n", usage_str[prev_bit]);
1544 print_stack_trace(this->class->usage_traces + prev_bit, 1);
1546 print_irqtrace_events(curr);
1547 printk("\nother info that might help us debug this:\n");
1548 lockdep_print_held_locks(curr);
1550 printk("\nstack backtrace:\n");
1557 * Print out an error if an invalid bit is set:
1560 valid_state(struct task_struct *curr, struct held_lock *this,
1561 enum lock_usage_bit new_bit, enum lock_usage_bit bad_bit)
1563 if (unlikely(this->class->usage_mask & (1 << bad_bit)))
1564 return print_usage_bug(curr, this, bad_bit, new_bit);
1568 #define STRICT_READ_CHECKS 1
1571 * Mark a lock with a usage bit, and validate the state transition:
1573 static int mark_lock(struct task_struct *curr, struct held_lock *this,
1574 enum lock_usage_bit new_bit, unsigned long ip)
1576 unsigned int new_mask = 1 << new_bit, ret = 1;
1579 * If already set then do not dirty the cacheline,
1580 * nor do any checks:
1582 if (likely(this->class->usage_mask & new_mask))
1588 * Make sure we didnt race:
1590 if (unlikely(this->class->usage_mask & new_mask)) {
1595 this->class->usage_mask |= new_mask;
1597 #ifdef CONFIG_TRACE_IRQFLAGS
1598 if (new_bit == LOCK_ENABLED_HARDIRQS ||
1599 new_bit == LOCK_ENABLED_HARDIRQS_READ)
1600 ip = curr->hardirq_enable_ip;
1601 else if (new_bit == LOCK_ENABLED_SOFTIRQS ||
1602 new_bit == LOCK_ENABLED_SOFTIRQS_READ)
1603 ip = curr->softirq_enable_ip;
1605 if (!save_trace(this->class->usage_traces + new_bit))
1609 #ifdef CONFIG_TRACE_IRQFLAGS
1610 case LOCK_USED_IN_HARDIRQ:
1611 if (!valid_state(curr, this, new_bit, LOCK_ENABLED_HARDIRQS))
1613 if (!valid_state(curr, this, new_bit,
1614 LOCK_ENABLED_HARDIRQS_READ))
1617 * just marked it hardirq-safe, check that this lock
1618 * took no hardirq-unsafe lock in the past:
1620 if (!check_usage_forwards(curr, this,
1621 LOCK_ENABLED_HARDIRQS, "hard"))
1623 #if STRICT_READ_CHECKS
1625 * just marked it hardirq-safe, check that this lock
1626 * took no hardirq-unsafe-read lock in the past:
1628 if (!check_usage_forwards(curr, this,
1629 LOCK_ENABLED_HARDIRQS_READ, "hard-read"))
1632 if (hardirq_verbose(this->class))
1635 case LOCK_USED_IN_SOFTIRQ:
1636 if (!valid_state(curr, this, new_bit, LOCK_ENABLED_SOFTIRQS))
1638 if (!valid_state(curr, this, new_bit,
1639 LOCK_ENABLED_SOFTIRQS_READ))
1642 * just marked it softirq-safe, check that this lock
1643 * took no softirq-unsafe lock in the past:
1645 if (!check_usage_forwards(curr, this,
1646 LOCK_ENABLED_SOFTIRQS, "soft"))
1648 #if STRICT_READ_CHECKS
1650 * just marked it softirq-safe, check that this lock
1651 * took no softirq-unsafe-read lock in the past:
1653 if (!check_usage_forwards(curr, this,
1654 LOCK_ENABLED_SOFTIRQS_READ, "soft-read"))
1657 if (softirq_verbose(this->class))
1660 case LOCK_USED_IN_HARDIRQ_READ:
1661 if (!valid_state(curr, this, new_bit, LOCK_ENABLED_HARDIRQS))
1664 * just marked it hardirq-read-safe, check that this lock
1665 * took no hardirq-unsafe lock in the past:
1667 if (!check_usage_forwards(curr, this,
1668 LOCK_ENABLED_HARDIRQS, "hard"))
1670 if (hardirq_verbose(this->class))
1673 case LOCK_USED_IN_SOFTIRQ_READ:
1674 if (!valid_state(curr, this, new_bit, LOCK_ENABLED_SOFTIRQS))
1677 * just marked it softirq-read-safe, check that this lock
1678 * took no softirq-unsafe lock in the past:
1680 if (!check_usage_forwards(curr, this,
1681 LOCK_ENABLED_SOFTIRQS, "soft"))
1683 if (softirq_verbose(this->class))
1686 case LOCK_ENABLED_HARDIRQS:
1687 if (!valid_state(curr, this, new_bit, LOCK_USED_IN_HARDIRQ))
1689 if (!valid_state(curr, this, new_bit,
1690 LOCK_USED_IN_HARDIRQ_READ))
1693 * just marked it hardirq-unsafe, check that no hardirq-safe
1694 * lock in the system ever took it in the past:
1696 if (!check_usage_backwards(curr, this,
1697 LOCK_USED_IN_HARDIRQ, "hard"))
1699 #if STRICT_READ_CHECKS
1701 * just marked it hardirq-unsafe, check that no
1702 * hardirq-safe-read lock in the system ever took
1705 if (!check_usage_backwards(curr, this,
1706 LOCK_USED_IN_HARDIRQ_READ, "hard-read"))
1709 if (hardirq_verbose(this->class))
1712 case LOCK_ENABLED_SOFTIRQS:
1713 if (!valid_state(curr, this, new_bit, LOCK_USED_IN_SOFTIRQ))
1715 if (!valid_state(curr, this, new_bit,
1716 LOCK_USED_IN_SOFTIRQ_READ))
1719 * just marked it softirq-unsafe, check that no softirq-safe
1720 * lock in the system ever took it in the past:
1722 if (!check_usage_backwards(curr, this,
1723 LOCK_USED_IN_SOFTIRQ, "soft"))
1725 #if STRICT_READ_CHECKS
1727 * just marked it softirq-unsafe, check that no
1728 * softirq-safe-read lock in the system ever took
1731 if (!check_usage_backwards(curr, this,
1732 LOCK_USED_IN_SOFTIRQ_READ, "soft-read"))
1735 if (softirq_verbose(this->class))
1738 case LOCK_ENABLED_HARDIRQS_READ:
1739 if (!valid_state(curr, this, new_bit, LOCK_USED_IN_HARDIRQ))
1741 #if STRICT_READ_CHECKS
1743 * just marked it hardirq-read-unsafe, check that no
1744 * hardirq-safe lock in the system ever took it in the past:
1746 if (!check_usage_backwards(curr, this,
1747 LOCK_USED_IN_HARDIRQ, "hard"))
1750 if (hardirq_verbose(this->class))
1753 case LOCK_ENABLED_SOFTIRQS_READ:
1754 if (!valid_state(curr, this, new_bit, LOCK_USED_IN_SOFTIRQ))
1756 #if STRICT_READ_CHECKS
1758 * just marked it softirq-read-unsafe, check that no
1759 * softirq-safe lock in the system ever took it in the past:
1761 if (!check_usage_backwards(curr, this,
1762 LOCK_USED_IN_SOFTIRQ, "soft"))
1765 if (softirq_verbose(this->class))
1771 * Add it to the global list of classes:
1773 list_add_tail_rcu(&this->class->lock_entry, &all_lock_classes);
1774 debug_atomic_dec(&nr_unused_locks);
1777 if (!debug_locks_off_graph_unlock())
1786 * We must printk outside of the graph_lock:
1789 printk("\nmarked lock as {%s}:\n", usage_str[new_bit]);
1791 print_irqtrace_events(curr);
1798 #ifdef CONFIG_TRACE_IRQFLAGS
1800 * Mark all held locks with a usage bit:
1803 mark_held_locks(struct task_struct *curr, int hardirq, unsigned long ip)
1805 enum lock_usage_bit usage_bit;
1806 struct held_lock *hlock;
1809 for (i = 0; i < curr->lockdep_depth; i++) {
1810 hlock = curr->held_locks + i;
1814 usage_bit = LOCK_ENABLED_HARDIRQS_READ;
1816 usage_bit = LOCK_ENABLED_HARDIRQS;
1819 usage_bit = LOCK_ENABLED_SOFTIRQS_READ;
1821 usage_bit = LOCK_ENABLED_SOFTIRQS;
1823 if (!mark_lock(curr, hlock, usage_bit, ip))
1831 * Debugging helper: via this flag we know that we are in
1832 * 'early bootup code', and will warn about any invalid irqs-on event:
1834 static int early_boot_irqs_enabled;
1836 void early_boot_irqs_off(void)
1838 early_boot_irqs_enabled = 0;
1841 void early_boot_irqs_on(void)
1843 early_boot_irqs_enabled = 1;
1847 * Hardirqs will be enabled:
1849 void trace_hardirqs_on(void)
1851 struct task_struct *curr = current;
1854 if (unlikely(!debug_locks || current->lockdep_recursion))
1857 if (DEBUG_LOCKS_WARN_ON(unlikely(!early_boot_irqs_enabled)))
1860 if (unlikely(curr->hardirqs_enabled)) {
1861 debug_atomic_inc(&redundant_hardirqs_on);
1864 /* we'll do an OFF -> ON transition: */
1865 curr->hardirqs_enabled = 1;
1866 ip = (unsigned long) __builtin_return_address(0);
1868 if (DEBUG_LOCKS_WARN_ON(!irqs_disabled()))
1870 if (DEBUG_LOCKS_WARN_ON(current->hardirq_context))
1873 * We are going to turn hardirqs on, so set the
1874 * usage bit for all held locks:
1876 if (!mark_held_locks(curr, 1, ip))
1879 * If we have softirqs enabled, then set the usage
1880 * bit for all held locks. (disabled hardirqs prevented
1881 * this bit from being set before)
1883 if (curr->softirqs_enabled)
1884 if (!mark_held_locks(curr, 0, ip))
1887 curr->hardirq_enable_ip = ip;
1888 curr->hardirq_enable_event = ++curr->irq_events;
1889 debug_atomic_inc(&hardirqs_on_events);
1892 EXPORT_SYMBOL(trace_hardirqs_on);
1895 * Hardirqs were disabled:
1897 void trace_hardirqs_off(void)
1899 struct task_struct *curr = current;
1901 if (unlikely(!debug_locks || current->lockdep_recursion))
1904 if (DEBUG_LOCKS_WARN_ON(!irqs_disabled()))
1907 if (curr->hardirqs_enabled) {
1909 * We have done an ON -> OFF transition:
1911 curr->hardirqs_enabled = 0;
1912 curr->hardirq_disable_ip = _RET_IP_;
1913 curr->hardirq_disable_event = ++curr->irq_events;
1914 debug_atomic_inc(&hardirqs_off_events);
1916 debug_atomic_inc(&redundant_hardirqs_off);
1919 EXPORT_SYMBOL(trace_hardirqs_off);
1922 * Softirqs will be enabled:
1924 void trace_softirqs_on(unsigned long ip)
1926 struct task_struct *curr = current;
1928 if (unlikely(!debug_locks))
1931 if (DEBUG_LOCKS_WARN_ON(!irqs_disabled()))
1934 if (curr->softirqs_enabled) {
1935 debug_atomic_inc(&redundant_softirqs_on);
1940 * We'll do an OFF -> ON transition:
1942 curr->softirqs_enabled = 1;
1943 curr->softirq_enable_ip = ip;
1944 curr->softirq_enable_event = ++curr->irq_events;
1945 debug_atomic_inc(&softirqs_on_events);
1947 * We are going to turn softirqs on, so set the
1948 * usage bit for all held locks, if hardirqs are
1951 if (curr->hardirqs_enabled)
1952 mark_held_locks(curr, 0, ip);
1956 * Softirqs were disabled:
1958 void trace_softirqs_off(unsigned long ip)
1960 struct task_struct *curr = current;
1962 if (unlikely(!debug_locks))
1965 if (DEBUG_LOCKS_WARN_ON(!irqs_disabled()))
1968 if (curr->softirqs_enabled) {
1970 * We have done an ON -> OFF transition:
1972 curr->softirqs_enabled = 0;
1973 curr->softirq_disable_ip = ip;
1974 curr->softirq_disable_event = ++curr->irq_events;
1975 debug_atomic_inc(&softirqs_off_events);
1976 DEBUG_LOCKS_WARN_ON(!softirq_count());
1978 debug_atomic_inc(&redundant_softirqs_off);
1984 * Initialize a lock instance's lock-class mapping info:
1986 void lockdep_init_map(struct lockdep_map *lock, const char *name,
1987 struct lock_class_key *key, int subclass)
1989 if (unlikely(!debug_locks))
1992 if (DEBUG_LOCKS_WARN_ON(!key))
1994 if (DEBUG_LOCKS_WARN_ON(!name))
1997 * Sanity check, the lock-class key must be persistent:
1999 if (!static_obj(key)) {
2000 printk("BUG: key %p not in .data!\n", key);
2001 DEBUG_LOCKS_WARN_ON(1);
2006 lock->class_cache = NULL;
2008 register_lock_class(lock, subclass, 1);
2011 EXPORT_SYMBOL_GPL(lockdep_init_map);
2014 * This gets called for every mutex_lock*()/spin_lock*() operation.
2015 * We maintain the dependency maps and validate the locking attempt:
2017 static int __lock_acquire(struct lockdep_map *lock, unsigned int subclass,
2018 int trylock, int read, int check, int hardirqs_off,
2021 struct task_struct *curr = current;
2022 struct lock_class *class = NULL;
2023 struct held_lock *hlock;
2024 unsigned int depth, id;
2028 if (unlikely(!debug_locks))
2031 if (DEBUG_LOCKS_WARN_ON(!irqs_disabled()))
2034 if (unlikely(subclass >= MAX_LOCKDEP_SUBCLASSES)) {
2036 printk("BUG: MAX_LOCKDEP_SUBCLASSES too low!\n");
2037 printk("turning off the locking correctness validator.\n");
2042 class = lock->class_cache;
2044 * Not cached yet or subclass?
2046 if (unlikely(!class)) {
2047 class = register_lock_class(lock, subclass, 0);
2051 debug_atomic_inc((atomic_t *)&class->ops);
2052 if (very_verbose(class)) {
2053 printk("\nacquire class [%p] %s", class->key, class->name);
2054 if (class->name_version > 1)
2055 printk("#%d", class->name_version);
2061 * Add the lock to the list of currently held locks.
2062 * (we dont increase the depth just yet, up until the
2063 * dependency checks are done)
2065 depth = curr->lockdep_depth;
2066 if (DEBUG_LOCKS_WARN_ON(depth >= MAX_LOCK_DEPTH))
2069 hlock = curr->held_locks + depth;
2071 hlock->class = class;
2072 hlock->acquire_ip = ip;
2073 hlock->instance = lock;
2074 hlock->trylock = trylock;
2076 hlock->check = check;
2077 hlock->hardirqs_off = hardirqs_off;
2081 #ifdef CONFIG_TRACE_IRQFLAGS
2083 * If non-trylock use in a hardirq or softirq context, then
2084 * mark the lock as used in these contexts:
2088 if (curr->hardirq_context)
2089 if (!mark_lock(curr, hlock,
2090 LOCK_USED_IN_HARDIRQ_READ, ip))
2092 if (curr->softirq_context)
2093 if (!mark_lock(curr, hlock,
2094 LOCK_USED_IN_SOFTIRQ_READ, ip))
2097 if (curr->hardirq_context)
2098 if (!mark_lock(curr, hlock, LOCK_USED_IN_HARDIRQ, ip))
2100 if (curr->softirq_context)
2101 if (!mark_lock(curr, hlock, LOCK_USED_IN_SOFTIRQ, ip))
2105 if (!hardirqs_off) {
2107 if (!mark_lock(curr, hlock,
2108 LOCK_ENABLED_HARDIRQS_READ, ip))
2110 if (curr->softirqs_enabled)
2111 if (!mark_lock(curr, hlock,
2112 LOCK_ENABLED_SOFTIRQS_READ, ip))
2115 if (!mark_lock(curr, hlock,
2116 LOCK_ENABLED_HARDIRQS, ip))
2118 if (curr->softirqs_enabled)
2119 if (!mark_lock(curr, hlock,
2120 LOCK_ENABLED_SOFTIRQS, ip))
2125 /* mark it as used: */
2126 if (!mark_lock(curr, hlock, LOCK_USED, ip))
2130 * Calculate the chain hash: it's the combined has of all the
2131 * lock keys along the dependency chain. We save the hash value
2132 * at every step so that we can get the current hash easily
2133 * after unlock. The chain hash is then used to cache dependency
2136 * The 'key ID' is what is the most compact key value to drive
2137 * the hash, not class->key.
2139 id = class - lock_classes;
2140 if (DEBUG_LOCKS_WARN_ON(id >= MAX_LOCKDEP_KEYS))
2143 chain_key = curr->curr_chain_key;
2145 if (DEBUG_LOCKS_WARN_ON(chain_key != 0))
2150 hlock->prev_chain_key = chain_key;
2152 #ifdef CONFIG_TRACE_IRQFLAGS
2154 * Keep track of points where we cross into an interrupt context:
2156 hlock->irq_context = 2*(curr->hardirq_context ? 1 : 0) +
2157 curr->softirq_context;
2159 struct held_lock *prev_hlock;
2161 prev_hlock = curr->held_locks + depth-1;
2163 * If we cross into another context, reset the
2164 * hash key (this also prevents the checking and the
2165 * adding of the dependency to 'prev'):
2167 if (prev_hlock->irq_context != hlock->irq_context) {
2173 chain_key = iterate_chain_key(chain_key, id);
2174 curr->curr_chain_key = chain_key;
2177 * Trylock needs to maintain the stack of held locks, but it
2178 * does not add new dependencies, because trylock can be done
2181 * We look up the chain_key and do the O(N^2) check and update of
2182 * the dependencies only if this is a new dependency chain.
2183 * (If lookup_chain_cache() returns with 1 it acquires
2184 * graph_lock for us)
2186 if (!trylock && (check == 2) && lookup_chain_cache(chain_key, class)) {
2188 * Check whether last held lock:
2190 * - is irq-safe, if this lock is irq-unsafe
2191 * - is softirq-safe, if this lock is hardirq-unsafe
2193 * And check whether the new lock's dependency graph
2194 * could lead back to the previous lock.
2196 * any of these scenarios could lead to a deadlock. If
2199 int ret = check_deadlock(curr, hlock, lock, read);
2204 * Mark recursive read, as we jump over it when
2205 * building dependencies (just like we jump over
2211 * Add dependency only if this lock is not the head
2212 * of the chain, and if it's not a secondary read-lock:
2214 if (!chain_head && ret != 2)
2215 if (!check_prevs_add(curr, hlock))
2219 /* after lookup_chain_cache(): */
2220 if (unlikely(!debug_locks))
2223 curr->lockdep_depth++;
2224 check_chain_key(curr);
2225 if (unlikely(curr->lockdep_depth >= MAX_LOCK_DEPTH)) {
2227 printk("BUG: MAX_LOCK_DEPTH too low!\n");
2228 printk("turning off the locking correctness validator.\n");
2232 if (unlikely(curr->lockdep_depth > max_lockdep_depth))
2233 max_lockdep_depth = curr->lockdep_depth;
2239 print_unlock_inbalance_bug(struct task_struct *curr, struct lockdep_map *lock,
2242 if (!debug_locks_off())
2244 if (debug_locks_silent)
2247 printk("\n=====================================\n");
2248 printk( "[ BUG: bad unlock balance detected! ]\n");
2249 printk( "-------------------------------------\n");
2250 printk("%s/%d is trying to release lock (",
2251 curr->comm, curr->pid);
2252 print_lockdep_cache(lock);
2255 printk("but there are no more locks to release!\n");
2256 printk("\nother info that might help us debug this:\n");
2257 lockdep_print_held_locks(curr);
2259 printk("\nstack backtrace:\n");
2266 * Common debugging checks for both nested and non-nested unlock:
2268 static int check_unlock(struct task_struct *curr, struct lockdep_map *lock,
2271 if (unlikely(!debug_locks))
2273 if (DEBUG_LOCKS_WARN_ON(!irqs_disabled()))
2276 if (curr->lockdep_depth <= 0)
2277 return print_unlock_inbalance_bug(curr, lock, ip);
2283 * Remove the lock to the list of currently held locks in a
2284 * potentially non-nested (out of order) manner. This is a
2285 * relatively rare operation, as all the unlock APIs default
2286 * to nested mode (which uses lock_release()):
2289 lock_release_non_nested(struct task_struct *curr,
2290 struct lockdep_map *lock, unsigned long ip)
2292 struct held_lock *hlock, *prev_hlock;
2297 * Check whether the lock exists in the current stack
2300 depth = curr->lockdep_depth;
2301 if (DEBUG_LOCKS_WARN_ON(!depth))
2305 for (i = depth-1; i >= 0; i--) {
2306 hlock = curr->held_locks + i;
2308 * We must not cross into another context:
2310 if (prev_hlock && prev_hlock->irq_context != hlock->irq_context)
2312 if (hlock->instance == lock)
2316 return print_unlock_inbalance_bug(curr, lock, ip);
2320 * We have the right lock to unlock, 'hlock' points to it.
2321 * Now we remove it from the stack, and add back the other
2322 * entries (if any), recalculating the hash along the way:
2324 curr->lockdep_depth = i;
2325 curr->curr_chain_key = hlock->prev_chain_key;
2327 for (i++; i < depth; i++) {
2328 hlock = curr->held_locks + i;
2329 if (!__lock_acquire(hlock->instance,
2330 hlock->class->subclass, hlock->trylock,
2331 hlock->read, hlock->check, hlock->hardirqs_off,
2336 if (DEBUG_LOCKS_WARN_ON(curr->lockdep_depth != depth - 1))
2342 * Remove the lock to the list of currently held locks - this gets
2343 * called on mutex_unlock()/spin_unlock*() (or on a failed
2344 * mutex_lock_interruptible()). This is done for unlocks that nest
2345 * perfectly. (i.e. the current top of the lock-stack is unlocked)
2347 static int lock_release_nested(struct task_struct *curr,
2348 struct lockdep_map *lock, unsigned long ip)
2350 struct held_lock *hlock;
2354 * Pop off the top of the lock stack:
2356 depth = curr->lockdep_depth - 1;
2357 hlock = curr->held_locks + depth;
2360 * Is the unlock non-nested:
2362 if (hlock->instance != lock)
2363 return lock_release_non_nested(curr, lock, ip);
2364 curr->lockdep_depth--;
2366 if (DEBUG_LOCKS_WARN_ON(!depth && (hlock->prev_chain_key != 0)))
2369 curr->curr_chain_key = hlock->prev_chain_key;
2371 #ifdef CONFIG_DEBUG_LOCKDEP
2372 hlock->prev_chain_key = 0;
2373 hlock->class = NULL;
2374 hlock->acquire_ip = 0;
2375 hlock->irq_context = 0;
2381 * Remove the lock to the list of currently held locks - this gets
2382 * called on mutex_unlock()/spin_unlock*() (or on a failed
2383 * mutex_lock_interruptible()). This is done for unlocks that nest
2384 * perfectly. (i.e. the current top of the lock-stack is unlocked)
2387 __lock_release(struct lockdep_map *lock, int nested, unsigned long ip)
2389 struct task_struct *curr = current;
2391 if (!check_unlock(curr, lock, ip))
2395 if (!lock_release_nested(curr, lock, ip))
2398 if (!lock_release_non_nested(curr, lock, ip))
2402 check_chain_key(curr);
2406 * Check whether we follow the irq-flags state precisely:
2408 static void check_flags(unsigned long flags)
2410 #if defined(CONFIG_DEBUG_LOCKDEP) && defined(CONFIG_TRACE_IRQFLAGS)
2414 if (irqs_disabled_flags(flags))
2415 DEBUG_LOCKS_WARN_ON(current->hardirqs_enabled);
2417 DEBUG_LOCKS_WARN_ON(!current->hardirqs_enabled);
2420 * We dont accurately track softirq state in e.g.
2421 * hardirq contexts (such as on 4KSTACKS), so only
2422 * check if not in hardirq contexts:
2424 if (!hardirq_count()) {
2425 if (softirq_count())
2426 DEBUG_LOCKS_WARN_ON(current->softirqs_enabled);
2428 DEBUG_LOCKS_WARN_ON(!current->softirqs_enabled);
2432 print_irqtrace_events(current);
2437 * We are not always called with irqs disabled - do that here,
2438 * and also avoid lockdep recursion:
2440 void lock_acquire(struct lockdep_map *lock, unsigned int subclass,
2441 int trylock, int read, int check, unsigned long ip)
2443 unsigned long flags;
2445 if (unlikely(current->lockdep_recursion))
2448 raw_local_irq_save(flags);
2451 current->lockdep_recursion = 1;
2452 __lock_acquire(lock, subclass, trylock, read, check,
2453 irqs_disabled_flags(flags), ip);
2454 current->lockdep_recursion = 0;
2455 raw_local_irq_restore(flags);
2458 EXPORT_SYMBOL_GPL(lock_acquire);
2460 void lock_release(struct lockdep_map *lock, int nested, unsigned long ip)
2462 unsigned long flags;
2464 if (unlikely(current->lockdep_recursion))
2467 raw_local_irq_save(flags);
2469 current->lockdep_recursion = 1;
2470 __lock_release(lock, nested, ip);
2471 current->lockdep_recursion = 0;
2472 raw_local_irq_restore(flags);
2475 EXPORT_SYMBOL_GPL(lock_release);
2478 * Used by the testsuite, sanitize the validator state
2479 * after a simulated failure:
2482 void lockdep_reset(void)
2484 unsigned long flags;
2487 raw_local_irq_save(flags);
2488 current->curr_chain_key = 0;
2489 current->lockdep_depth = 0;
2490 current->lockdep_recursion = 0;
2491 memset(current->held_locks, 0, MAX_LOCK_DEPTH*sizeof(struct held_lock));
2492 nr_hardirq_chains = 0;
2493 nr_softirq_chains = 0;
2494 nr_process_chains = 0;
2496 for (i = 0; i < CHAINHASH_SIZE; i++)
2497 INIT_LIST_HEAD(chainhash_table + i);
2498 raw_local_irq_restore(flags);
2501 static void zap_class(struct lock_class *class)
2506 * Remove all dependencies this lock is
2509 for (i = 0; i < nr_list_entries; i++) {
2510 if (list_entries[i].class == class)
2511 list_del_rcu(&list_entries[i].entry);
2514 * Unhash the class and remove it from the all_lock_classes list:
2516 list_del_rcu(&class->hash_entry);
2517 list_del_rcu(&class->lock_entry);
2521 static inline int within(void *addr, void *start, unsigned long size)
2523 return addr >= start && addr < start + size;
2526 void lockdep_free_key_range(void *start, unsigned long size)
2528 struct lock_class *class, *next;
2529 struct list_head *head;
2530 unsigned long flags;
2533 raw_local_irq_save(flags);
2537 * Unhash all classes that were created by this module:
2539 for (i = 0; i < CLASSHASH_SIZE; i++) {
2540 head = classhash_table + i;
2541 if (list_empty(head))
2543 list_for_each_entry_safe(class, next, head, hash_entry)
2544 if (within(class->key, start, size))
2549 raw_local_irq_restore(flags);
2552 void lockdep_reset_lock(struct lockdep_map *lock)
2554 struct lock_class *class, *next;
2555 struct list_head *head;
2556 unsigned long flags;
2559 raw_local_irq_save(flags);
2562 * Remove all classes this lock might have:
2564 for (j = 0; j < MAX_LOCKDEP_SUBCLASSES; j++) {
2566 * If the class exists we look it up and zap it:
2568 class = look_up_lock_class(lock, j);
2573 * Debug check: in the end all mapped classes should
2577 for (i = 0; i < CLASSHASH_SIZE; i++) {
2578 head = classhash_table + i;
2579 if (list_empty(head))
2581 list_for_each_entry_safe(class, next, head, hash_entry) {
2582 if (unlikely(class == lock->class_cache)) {
2583 if (debug_locks_off_graph_unlock())
2592 raw_local_irq_restore(flags);
2595 void __init lockdep_init(void)
2600 * Some architectures have their own start_kernel()
2601 * code which calls lockdep_init(), while we also
2602 * call lockdep_init() from the start_kernel() itself,
2603 * and we want to initialize the hashes only once:
2605 if (lockdep_initialized)
2608 for (i = 0; i < CLASSHASH_SIZE; i++)
2609 INIT_LIST_HEAD(classhash_table + i);
2611 for (i = 0; i < CHAINHASH_SIZE; i++)
2612 INIT_LIST_HEAD(chainhash_table + i);
2614 lockdep_initialized = 1;
2617 void __init lockdep_info(void)
2619 printk("Lock dependency validator: Copyright (c) 2006 Red Hat, Inc., Ingo Molnar\n");
2621 printk("... MAX_LOCKDEP_SUBCLASSES: %lu\n", MAX_LOCKDEP_SUBCLASSES);
2622 printk("... MAX_LOCK_DEPTH: %lu\n", MAX_LOCK_DEPTH);
2623 printk("... MAX_LOCKDEP_KEYS: %lu\n", MAX_LOCKDEP_KEYS);
2624 printk("... CLASSHASH_SIZE: %lu\n", CLASSHASH_SIZE);
2625 printk("... MAX_LOCKDEP_ENTRIES: %lu\n", MAX_LOCKDEP_ENTRIES);
2626 printk("... MAX_LOCKDEP_CHAINS: %lu\n", MAX_LOCKDEP_CHAINS);
2627 printk("... CHAINHASH_SIZE: %lu\n", CHAINHASH_SIZE);
2629 printk(" memory used by lock dependency info: %lu kB\n",
2630 (sizeof(struct lock_class) * MAX_LOCKDEP_KEYS +
2631 sizeof(struct list_head) * CLASSHASH_SIZE +
2632 sizeof(struct lock_list) * MAX_LOCKDEP_ENTRIES +
2633 sizeof(struct lock_chain) * MAX_LOCKDEP_CHAINS +
2634 sizeof(struct list_head) * CHAINHASH_SIZE) / 1024);
2636 printk(" per task-struct memory footprint: %lu bytes\n",
2637 sizeof(struct held_lock) * MAX_LOCK_DEPTH);
2639 #ifdef CONFIG_DEBUG_LOCKDEP
2640 if (lockdep_init_error)
2641 printk("WARNING: lockdep init error! Arch code didnt call lockdep_init() early enough?\n");
2645 static inline int in_range(const void *start, const void *addr, const void *end)
2647 return addr >= start && addr <= end;
2651 print_freed_lock_bug(struct task_struct *curr, const void *mem_from,
2652 const void *mem_to, struct held_lock *hlock)
2654 if (!debug_locks_off())
2656 if (debug_locks_silent)
2659 printk("\n=========================\n");
2660 printk( "[ BUG: held lock freed! ]\n");
2661 printk( "-------------------------\n");
2662 printk("%s/%d is freeing memory %p-%p, with a lock still held there!\n",
2663 curr->comm, curr->pid, mem_from, mem_to-1);
2665 lockdep_print_held_locks(curr);
2667 printk("\nstack backtrace:\n");
2672 * Called when kernel memory is freed (or unmapped), or if a lock
2673 * is destroyed or reinitialized - this code checks whether there is
2674 * any held lock in the memory range of <from> to <to>:
2676 void debug_check_no_locks_freed(const void *mem_from, unsigned long mem_len)
2678 const void *mem_to = mem_from + mem_len, *lock_from, *lock_to;
2679 struct task_struct *curr = current;
2680 struct held_lock *hlock;
2681 unsigned long flags;
2684 if (unlikely(!debug_locks))
2687 local_irq_save(flags);
2688 for (i = 0; i < curr->lockdep_depth; i++) {
2689 hlock = curr->held_locks + i;
2691 lock_from = (void *)hlock->instance;
2692 lock_to = (void *)(hlock->instance + 1);
2694 if (!in_range(mem_from, lock_from, mem_to) &&
2695 !in_range(mem_from, lock_to, mem_to))
2698 print_freed_lock_bug(curr, mem_from, mem_to, hlock);
2701 local_irq_restore(flags);
2703 EXPORT_SYMBOL_GPL(debug_check_no_locks_freed);
2705 static void print_held_locks_bug(struct task_struct *curr)
2707 if (!debug_locks_off())
2709 if (debug_locks_silent)
2712 printk("\n=====================================\n");
2713 printk( "[ BUG: lock held at task exit time! ]\n");
2714 printk( "-------------------------------------\n");
2715 printk("%s/%d is exiting with locks still held!\n",
2716 curr->comm, curr->pid);
2717 lockdep_print_held_locks(curr);
2719 printk("\nstack backtrace:\n");
2723 void debug_check_no_locks_held(struct task_struct *task)
2725 if (unlikely(task->lockdep_depth > 0))
2726 print_held_locks_bug(task);
2729 void debug_show_all_locks(void)
2731 struct task_struct *g, *p;
2735 printk("\nShowing all locks held in the system:\n");
2738 * Here we try to get the tasklist_lock as hard as possible,
2739 * if not successful after 2 seconds we ignore it (but keep
2740 * trying). This is to enable a debug printout even if a
2741 * tasklist_lock-holding task deadlocks or crashes.
2744 if (!read_trylock(&tasklist_lock)) {
2746 printk("hm, tasklist_lock locked, retrying... ");
2749 printk(" #%d", 10-count);
2753 printk(" ignoring it.\n");
2757 printk(" locked it.\n");
2759 do_each_thread(g, p) {
2760 if (p->lockdep_depth)
2761 lockdep_print_held_locks(p);
2763 if (read_trylock(&tasklist_lock))
2765 } while_each_thread(g, p);
2768 printk("=============================================\n\n");
2771 read_unlock(&tasklist_lock);
2774 EXPORT_SYMBOL_GPL(debug_show_all_locks);
2776 void debug_show_held_locks(struct task_struct *task)
2778 lockdep_print_held_locks(task);
2781 EXPORT_SYMBOL_GPL(debug_show_held_locks);