clk: samsung: Add bus clock for GPU/G3D on Exynos4412
[linux/fpc-iii.git] / kernel / locking / lockdep.c
blobd06190fa5082239e00cd1be55762cf8232260652
1 /*
2 * kernel/lockdep.c
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/sched/mm.h>
34 #include <linux/delay.h>
35 #include <linux/module.h>
36 #include <linux/proc_fs.h>
37 #include <linux/seq_file.h>
38 #include <linux/spinlock.h>
39 #include <linux/kallsyms.h>
40 #include <linux/interrupt.h>
41 #include <linux/stacktrace.h>
42 #include <linux/debug_locks.h>
43 #include <linux/irqflags.h>
44 #include <linux/utsname.h>
45 #include <linux/hash.h>
46 #include <linux/ftrace.h>
47 #include <linux/stringify.h>
48 #include <linux/bitmap.h>
49 #include <linux/bitops.h>
50 #include <linux/gfp.h>
51 #include <linux/random.h>
52 #include <linux/jhash.h>
53 #include <linux/nmi.h>
54 #include <linux/rcupdate.h>
55 #include <linux/kprobes.h>
57 #include <asm/sections.h>
59 #include "lockdep_internals.h"
61 #define CREATE_TRACE_POINTS
62 #include <trace/events/lock.h>
64 #ifdef CONFIG_PROVE_LOCKING
65 int prove_locking = 1;
66 module_param(prove_locking, int, 0644);
67 #else
68 #define prove_locking 0
69 #endif
71 #ifdef CONFIG_LOCK_STAT
72 int lock_stat = 1;
73 module_param(lock_stat, int, 0644);
74 #else
75 #define lock_stat 0
76 #endif
79 * lockdep_lock: protects the lockdep graph, the hashes and the
80 * class/list/hash allocators.
82 * This is one of the rare exceptions where it's justified
83 * to use a raw spinlock - we really dont want the spinlock
84 * code to recurse back into the lockdep code...
86 static arch_spinlock_t lockdep_lock = (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
87 static struct task_struct *lockdep_selftest_task_struct;
89 static int graph_lock(void)
91 arch_spin_lock(&lockdep_lock);
93 * Make sure that if another CPU detected a bug while
94 * walking the graph we dont change it (while the other
95 * CPU is busy printing out stuff with the graph lock
96 * dropped already)
98 if (!debug_locks) {
99 arch_spin_unlock(&lockdep_lock);
100 return 0;
102 /* prevent any recursions within lockdep from causing deadlocks */
103 current->lockdep_recursion++;
104 return 1;
107 static inline int graph_unlock(void)
109 if (debug_locks && !arch_spin_is_locked(&lockdep_lock)) {
111 * The lockdep graph lock isn't locked while we expect it to
112 * be, we're confused now, bye!
114 return DEBUG_LOCKS_WARN_ON(1);
117 current->lockdep_recursion--;
118 arch_spin_unlock(&lockdep_lock);
119 return 0;
123 * Turn lock debugging off and return with 0 if it was off already,
124 * and also release the graph lock:
126 static inline int debug_locks_off_graph_unlock(void)
128 int ret = debug_locks_off();
130 arch_spin_unlock(&lockdep_lock);
132 return ret;
135 unsigned long nr_list_entries;
136 static struct lock_list list_entries[MAX_LOCKDEP_ENTRIES];
137 static DECLARE_BITMAP(list_entries_in_use, MAX_LOCKDEP_ENTRIES);
140 * All data structures here are protected by the global debug_lock.
142 * nr_lock_classes is the number of elements of lock_classes[] that is
143 * in use.
145 #define KEYHASH_BITS (MAX_LOCKDEP_KEYS_BITS - 1)
146 #define KEYHASH_SIZE (1UL << KEYHASH_BITS)
147 static struct hlist_head lock_keys_hash[KEYHASH_SIZE];
148 unsigned long nr_lock_classes;
149 #ifndef CONFIG_DEBUG_LOCKDEP
150 static
151 #endif
152 struct lock_class lock_classes[MAX_LOCKDEP_KEYS];
154 static inline struct lock_class *hlock_class(struct held_lock *hlock)
156 if (!hlock->class_idx) {
158 * Someone passed in garbage, we give up.
160 DEBUG_LOCKS_WARN_ON(1);
161 return NULL;
163 return lock_classes + hlock->class_idx - 1;
166 #ifdef CONFIG_LOCK_STAT
167 static DEFINE_PER_CPU(struct lock_class_stats[MAX_LOCKDEP_KEYS], cpu_lock_stats);
169 static inline u64 lockstat_clock(void)
171 return local_clock();
174 static int lock_point(unsigned long points[], unsigned long ip)
176 int i;
178 for (i = 0; i < LOCKSTAT_POINTS; i++) {
179 if (points[i] == 0) {
180 points[i] = ip;
181 break;
183 if (points[i] == ip)
184 break;
187 return i;
190 static void lock_time_inc(struct lock_time *lt, u64 time)
192 if (time > lt->max)
193 lt->max = time;
195 if (time < lt->min || !lt->nr)
196 lt->min = time;
198 lt->total += time;
199 lt->nr++;
202 static inline void lock_time_add(struct lock_time *src, struct lock_time *dst)
204 if (!src->nr)
205 return;
207 if (src->max > dst->max)
208 dst->max = src->max;
210 if (src->min < dst->min || !dst->nr)
211 dst->min = src->min;
213 dst->total += src->total;
214 dst->nr += src->nr;
217 struct lock_class_stats lock_stats(struct lock_class *class)
219 struct lock_class_stats stats;
220 int cpu, i;
222 memset(&stats, 0, sizeof(struct lock_class_stats));
223 for_each_possible_cpu(cpu) {
224 struct lock_class_stats *pcs =
225 &per_cpu(cpu_lock_stats, cpu)[class - lock_classes];
227 for (i = 0; i < ARRAY_SIZE(stats.contention_point); i++)
228 stats.contention_point[i] += pcs->contention_point[i];
230 for (i = 0; i < ARRAY_SIZE(stats.contending_point); i++)
231 stats.contending_point[i] += pcs->contending_point[i];
233 lock_time_add(&pcs->read_waittime, &stats.read_waittime);
234 lock_time_add(&pcs->write_waittime, &stats.write_waittime);
236 lock_time_add(&pcs->read_holdtime, &stats.read_holdtime);
237 lock_time_add(&pcs->write_holdtime, &stats.write_holdtime);
239 for (i = 0; i < ARRAY_SIZE(stats.bounces); i++)
240 stats.bounces[i] += pcs->bounces[i];
243 return stats;
246 void clear_lock_stats(struct lock_class *class)
248 int cpu;
250 for_each_possible_cpu(cpu) {
251 struct lock_class_stats *cpu_stats =
252 &per_cpu(cpu_lock_stats, cpu)[class - lock_classes];
254 memset(cpu_stats, 0, sizeof(struct lock_class_stats));
256 memset(class->contention_point, 0, sizeof(class->contention_point));
257 memset(class->contending_point, 0, sizeof(class->contending_point));
260 static struct lock_class_stats *get_lock_stats(struct lock_class *class)
262 return &this_cpu_ptr(cpu_lock_stats)[class - lock_classes];
265 static void lock_release_holdtime(struct held_lock *hlock)
267 struct lock_class_stats *stats;
268 u64 holdtime;
270 if (!lock_stat)
271 return;
273 holdtime = lockstat_clock() - hlock->holdtime_stamp;
275 stats = get_lock_stats(hlock_class(hlock));
276 if (hlock->read)
277 lock_time_inc(&stats->read_holdtime, holdtime);
278 else
279 lock_time_inc(&stats->write_holdtime, holdtime);
281 #else
282 static inline void lock_release_holdtime(struct held_lock *hlock)
285 #endif
288 * We keep a global list of all lock classes. The list is only accessed with
289 * the lockdep spinlock lock held. free_lock_classes is a list with free
290 * elements. These elements are linked together by the lock_entry member in
291 * struct lock_class.
293 LIST_HEAD(all_lock_classes);
294 static LIST_HEAD(free_lock_classes);
297 * struct pending_free - information about data structures about to be freed
298 * @zapped: Head of a list with struct lock_class elements.
299 * @lock_chains_being_freed: Bitmap that indicates which lock_chains[] elements
300 * are about to be freed.
302 struct pending_free {
303 struct list_head zapped;
304 DECLARE_BITMAP(lock_chains_being_freed, MAX_LOCKDEP_CHAINS);
308 * struct delayed_free - data structures used for delayed freeing
310 * A data structure for delayed freeing of data structures that may be
311 * accessed by RCU readers at the time these were freed.
313 * @rcu_head: Used to schedule an RCU callback for freeing data structures.
314 * @index: Index of @pf to which freed data structures are added.
315 * @scheduled: Whether or not an RCU callback has been scheduled.
316 * @pf: Array with information about data structures about to be freed.
318 static struct delayed_free {
319 struct rcu_head rcu_head;
320 int index;
321 int scheduled;
322 struct pending_free pf[2];
323 } delayed_free;
326 * The lockdep classes are in a hash-table as well, for fast lookup:
328 #define CLASSHASH_BITS (MAX_LOCKDEP_KEYS_BITS - 1)
329 #define CLASSHASH_SIZE (1UL << CLASSHASH_BITS)
330 #define __classhashfn(key) hash_long((unsigned long)key, CLASSHASH_BITS)
331 #define classhashentry(key) (classhash_table + __classhashfn((key)))
333 static struct hlist_head classhash_table[CLASSHASH_SIZE];
336 * We put the lock dependency chains into a hash-table as well, to cache
337 * their existence:
339 #define CHAINHASH_BITS (MAX_LOCKDEP_CHAINS_BITS-1)
340 #define CHAINHASH_SIZE (1UL << CHAINHASH_BITS)
341 #define __chainhashfn(chain) hash_long(chain, CHAINHASH_BITS)
342 #define chainhashentry(chain) (chainhash_table + __chainhashfn((chain)))
344 static struct hlist_head chainhash_table[CHAINHASH_SIZE];
347 * The hash key of the lock dependency chains is a hash itself too:
348 * it's a hash of all locks taken up to that lock, including that lock.
349 * It's a 64-bit hash, because it's important for the keys to be
350 * unique.
352 static inline u64 iterate_chain_key(u64 key, u32 idx)
354 u32 k0 = key, k1 = key >> 32;
356 __jhash_mix(idx, k0, k1); /* Macro that modifies arguments! */
358 return k0 | (u64)k1 << 32;
361 void lockdep_off(void)
363 current->lockdep_recursion++;
365 EXPORT_SYMBOL(lockdep_off);
367 void lockdep_on(void)
369 current->lockdep_recursion--;
371 EXPORT_SYMBOL(lockdep_on);
373 void lockdep_set_selftest_task(struct task_struct *task)
375 lockdep_selftest_task_struct = task;
379 * Debugging switches:
382 #define VERBOSE 0
383 #define VERY_VERBOSE 0
385 #if VERBOSE
386 # define HARDIRQ_VERBOSE 1
387 # define SOFTIRQ_VERBOSE 1
388 #else
389 # define HARDIRQ_VERBOSE 0
390 # define SOFTIRQ_VERBOSE 0
391 #endif
393 #if VERBOSE || HARDIRQ_VERBOSE || SOFTIRQ_VERBOSE
395 * Quick filtering for interesting events:
397 static int class_filter(struct lock_class *class)
399 #if 0
400 /* Example */
401 if (class->name_version == 1 &&
402 !strcmp(class->name, "lockname"))
403 return 1;
404 if (class->name_version == 1 &&
405 !strcmp(class->name, "&struct->lockfield"))
406 return 1;
407 #endif
408 /* Filter everything else. 1 would be to allow everything else */
409 return 0;
411 #endif
413 static int verbose(struct lock_class *class)
415 #if VERBOSE
416 return class_filter(class);
417 #endif
418 return 0;
422 * Stack-trace: tightly packed array of stack backtrace
423 * addresses. Protected by the graph_lock.
425 unsigned long nr_stack_trace_entries;
426 static unsigned long stack_trace[MAX_STACK_TRACE_ENTRIES];
428 static void print_lockdep_off(const char *bug_msg)
430 printk(KERN_DEBUG "%s\n", bug_msg);
431 printk(KERN_DEBUG "turning off the locking correctness validator.\n");
432 #ifdef CONFIG_LOCK_STAT
433 printk(KERN_DEBUG "Please attach the output of /proc/lock_stat to the bug report\n");
434 #endif
437 static int save_trace(struct lock_trace *trace)
439 unsigned long *entries = stack_trace + nr_stack_trace_entries;
440 unsigned int max_entries;
442 trace->offset = nr_stack_trace_entries;
443 max_entries = MAX_STACK_TRACE_ENTRIES - nr_stack_trace_entries;
444 trace->nr_entries = stack_trace_save(entries, max_entries, 3);
445 nr_stack_trace_entries += trace->nr_entries;
447 if (nr_stack_trace_entries >= MAX_STACK_TRACE_ENTRIES-1) {
448 if (!debug_locks_off_graph_unlock())
449 return 0;
451 print_lockdep_off("BUG: MAX_STACK_TRACE_ENTRIES too low!");
452 dump_stack();
454 return 0;
457 return 1;
460 unsigned int nr_hardirq_chains;
461 unsigned int nr_softirq_chains;
462 unsigned int nr_process_chains;
463 unsigned int max_lockdep_depth;
465 #ifdef CONFIG_DEBUG_LOCKDEP
467 * Various lockdep statistics:
469 DEFINE_PER_CPU(struct lockdep_stats, lockdep_stats);
470 #endif
473 * Locking printouts:
476 #define __USAGE(__STATE) \
477 [LOCK_USED_IN_##__STATE] = "IN-"__stringify(__STATE)"-W", \
478 [LOCK_ENABLED_##__STATE] = __stringify(__STATE)"-ON-W", \
479 [LOCK_USED_IN_##__STATE##_READ] = "IN-"__stringify(__STATE)"-R",\
480 [LOCK_ENABLED_##__STATE##_READ] = __stringify(__STATE)"-ON-R",
482 static const char *usage_str[] =
484 #define LOCKDEP_STATE(__STATE) __USAGE(__STATE)
485 #include "lockdep_states.h"
486 #undef LOCKDEP_STATE
487 [LOCK_USED] = "INITIAL USE",
490 const char * __get_key_name(struct lockdep_subclass_key *key, char *str)
492 return kallsyms_lookup((unsigned long)key, NULL, NULL, NULL, str);
495 static inline unsigned long lock_flag(enum lock_usage_bit bit)
497 return 1UL << bit;
500 static char get_usage_char(struct lock_class *class, enum lock_usage_bit bit)
502 char c = '.';
504 if (class->usage_mask & lock_flag(bit + LOCK_USAGE_DIR_MASK))
505 c = '+';
506 if (class->usage_mask & lock_flag(bit)) {
507 c = '-';
508 if (class->usage_mask & lock_flag(bit + LOCK_USAGE_DIR_MASK))
509 c = '?';
512 return c;
515 void get_usage_chars(struct lock_class *class, char usage[LOCK_USAGE_CHARS])
517 int i = 0;
519 #define LOCKDEP_STATE(__STATE) \
520 usage[i++] = get_usage_char(class, LOCK_USED_IN_##__STATE); \
521 usage[i++] = get_usage_char(class, LOCK_USED_IN_##__STATE##_READ);
522 #include "lockdep_states.h"
523 #undef LOCKDEP_STATE
525 usage[i] = '\0';
528 static void __print_lock_name(struct lock_class *class)
530 char str[KSYM_NAME_LEN];
531 const char *name;
533 name = class->name;
534 if (!name) {
535 name = __get_key_name(class->key, str);
536 printk(KERN_CONT "%s", name);
537 } else {
538 printk(KERN_CONT "%s", name);
539 if (class->name_version > 1)
540 printk(KERN_CONT "#%d", class->name_version);
541 if (class->subclass)
542 printk(KERN_CONT "/%d", class->subclass);
546 static void print_lock_name(struct lock_class *class)
548 char usage[LOCK_USAGE_CHARS];
550 get_usage_chars(class, usage);
552 printk(KERN_CONT " (");
553 __print_lock_name(class);
554 printk(KERN_CONT "){%s}", usage);
557 static void print_lockdep_cache(struct lockdep_map *lock)
559 const char *name;
560 char str[KSYM_NAME_LEN];
562 name = lock->name;
563 if (!name)
564 name = __get_key_name(lock->key->subkeys, str);
566 printk(KERN_CONT "%s", name);
569 static void print_lock(struct held_lock *hlock)
572 * We can be called locklessly through debug_show_all_locks() so be
573 * extra careful, the hlock might have been released and cleared.
575 unsigned int class_idx = hlock->class_idx;
577 /* Don't re-read hlock->class_idx, can't use READ_ONCE() on bitfields: */
578 barrier();
580 if (!class_idx || (class_idx - 1) >= MAX_LOCKDEP_KEYS) {
581 printk(KERN_CONT "<RELEASED>\n");
582 return;
585 printk(KERN_CONT "%p", hlock->instance);
586 print_lock_name(lock_classes + class_idx - 1);
587 printk(KERN_CONT ", at: %pS\n", (void *)hlock->acquire_ip);
590 static void lockdep_print_held_locks(struct task_struct *p)
592 int i, depth = READ_ONCE(p->lockdep_depth);
594 if (!depth)
595 printk("no locks held by %s/%d.\n", p->comm, task_pid_nr(p));
596 else
597 printk("%d lock%s held by %s/%d:\n", depth,
598 depth > 1 ? "s" : "", p->comm, task_pid_nr(p));
600 * It's not reliable to print a task's held locks if it's not sleeping
601 * and it's not the current task.
603 if (p->state == TASK_RUNNING && p != current)
604 return;
605 for (i = 0; i < depth; i++) {
606 printk(" #%d: ", i);
607 print_lock(p->held_locks + i);
611 static void print_kernel_ident(void)
613 printk("%s %.*s %s\n", init_utsname()->release,
614 (int)strcspn(init_utsname()->version, " "),
615 init_utsname()->version,
616 print_tainted());
619 static int very_verbose(struct lock_class *class)
621 #if VERY_VERBOSE
622 return class_filter(class);
623 #endif
624 return 0;
628 * Is this the address of a static object:
630 #ifdef __KERNEL__
631 static int static_obj(const void *obj)
633 unsigned long start = (unsigned long) &_stext,
634 end = (unsigned long) &_end,
635 addr = (unsigned long) obj;
637 if (arch_is_kernel_initmem_freed(addr))
638 return 0;
641 * static variable?
643 if ((addr >= start) && (addr < end))
644 return 1;
646 if (arch_is_kernel_data(addr))
647 return 1;
650 * in-kernel percpu var?
652 if (is_kernel_percpu_address(addr))
653 return 1;
656 * module static or percpu var?
658 return is_module_address(addr) || is_module_percpu_address(addr);
660 #endif
663 * To make lock name printouts unique, we calculate a unique
664 * class->name_version generation counter. The caller must hold the graph
665 * lock.
667 static int count_matching_names(struct lock_class *new_class)
669 struct lock_class *class;
670 int count = 0;
672 if (!new_class->name)
673 return 0;
675 list_for_each_entry(class, &all_lock_classes, lock_entry) {
676 if (new_class->key - new_class->subclass == class->key)
677 return class->name_version;
678 if (class->name && !strcmp(class->name, new_class->name))
679 count = max(count, class->name_version);
682 return count + 1;
685 static inline struct lock_class *
686 look_up_lock_class(const struct lockdep_map *lock, unsigned int subclass)
688 struct lockdep_subclass_key *key;
689 struct hlist_head *hash_head;
690 struct lock_class *class;
692 if (unlikely(subclass >= MAX_LOCKDEP_SUBCLASSES)) {
693 debug_locks_off();
694 printk(KERN_ERR
695 "BUG: looking up invalid subclass: %u\n", subclass);
696 printk(KERN_ERR
697 "turning off the locking correctness validator.\n");
698 dump_stack();
699 return NULL;
703 * If it is not initialised then it has never been locked,
704 * so it won't be present in the hash table.
706 if (unlikely(!lock->key))
707 return NULL;
710 * NOTE: the class-key must be unique. For dynamic locks, a static
711 * lock_class_key variable is passed in through the mutex_init()
712 * (or spin_lock_init()) call - which acts as the key. For static
713 * locks we use the lock object itself as the key.
715 BUILD_BUG_ON(sizeof(struct lock_class_key) >
716 sizeof(struct lockdep_map));
718 key = lock->key->subkeys + subclass;
720 hash_head = classhashentry(key);
723 * We do an RCU walk of the hash, see lockdep_free_key_range().
725 if (DEBUG_LOCKS_WARN_ON(!irqs_disabled()))
726 return NULL;
728 hlist_for_each_entry_rcu(class, hash_head, hash_entry) {
729 if (class->key == key) {
731 * Huh! same key, different name? Did someone trample
732 * on some memory? We're most confused.
734 WARN_ON_ONCE(class->name != lock->name);
735 return class;
739 return NULL;
743 * Static locks do not have their class-keys yet - for them the key is
744 * the lock object itself. If the lock is in the per cpu area, the
745 * canonical address of the lock (per cpu offset removed) is used.
747 static bool assign_lock_key(struct lockdep_map *lock)
749 unsigned long can_addr, addr = (unsigned long)lock;
751 #ifdef __KERNEL__
753 * lockdep_free_key_range() assumes that struct lock_class_key
754 * objects do not overlap. Since we use the address of lock
755 * objects as class key for static objects, check whether the
756 * size of lock_class_key objects does not exceed the size of
757 * the smallest lock object.
759 BUILD_BUG_ON(sizeof(struct lock_class_key) > sizeof(raw_spinlock_t));
760 #endif
762 if (__is_kernel_percpu_address(addr, &can_addr))
763 lock->key = (void *)can_addr;
764 else if (__is_module_percpu_address(addr, &can_addr))
765 lock->key = (void *)can_addr;
766 else if (static_obj(lock))
767 lock->key = (void *)lock;
768 else {
769 /* Debug-check: all keys must be persistent! */
770 debug_locks_off();
771 pr_err("INFO: trying to register non-static key.\n");
772 pr_err("the code is fine but needs lockdep annotation.\n");
773 pr_err("turning off the locking correctness validator.\n");
774 dump_stack();
775 return false;
778 return true;
781 #ifdef CONFIG_DEBUG_LOCKDEP
783 /* Check whether element @e occurs in list @h */
784 static bool in_list(struct list_head *e, struct list_head *h)
786 struct list_head *f;
788 list_for_each(f, h) {
789 if (e == f)
790 return true;
793 return false;
797 * Check whether entry @e occurs in any of the locks_after or locks_before
798 * lists.
800 static bool in_any_class_list(struct list_head *e)
802 struct lock_class *class;
803 int i;
805 for (i = 0; i < ARRAY_SIZE(lock_classes); i++) {
806 class = &lock_classes[i];
807 if (in_list(e, &class->locks_after) ||
808 in_list(e, &class->locks_before))
809 return true;
811 return false;
814 static bool class_lock_list_valid(struct lock_class *c, struct list_head *h)
816 struct lock_list *e;
818 list_for_each_entry(e, h, entry) {
819 if (e->links_to != c) {
820 printk(KERN_INFO "class %s: mismatch for lock entry %ld; class %s <> %s",
821 c->name ? : "(?)",
822 (unsigned long)(e - list_entries),
823 e->links_to && e->links_to->name ?
824 e->links_to->name : "(?)",
825 e->class && e->class->name ? e->class->name :
826 "(?)");
827 return false;
830 return true;
833 #ifdef CONFIG_PROVE_LOCKING
834 static u16 chain_hlocks[MAX_LOCKDEP_CHAIN_HLOCKS];
835 #endif
837 static bool check_lock_chain_key(struct lock_chain *chain)
839 #ifdef CONFIG_PROVE_LOCKING
840 u64 chain_key = 0;
841 int i;
843 for (i = chain->base; i < chain->base + chain->depth; i++)
844 chain_key = iterate_chain_key(chain_key, chain_hlocks[i] + 1);
846 * The 'unsigned long long' casts avoid that a compiler warning
847 * is reported when building tools/lib/lockdep.
849 if (chain->chain_key != chain_key) {
850 printk(KERN_INFO "chain %lld: key %#llx <> %#llx\n",
851 (unsigned long long)(chain - lock_chains),
852 (unsigned long long)chain->chain_key,
853 (unsigned long long)chain_key);
854 return false;
856 #endif
857 return true;
860 static bool in_any_zapped_class_list(struct lock_class *class)
862 struct pending_free *pf;
863 int i;
865 for (i = 0, pf = delayed_free.pf; i < ARRAY_SIZE(delayed_free.pf); i++, pf++) {
866 if (in_list(&class->lock_entry, &pf->zapped))
867 return true;
870 return false;
873 static bool __check_data_structures(void)
875 struct lock_class *class;
876 struct lock_chain *chain;
877 struct hlist_head *head;
878 struct lock_list *e;
879 int i;
881 /* Check whether all classes occur in a lock list. */
882 for (i = 0; i < ARRAY_SIZE(lock_classes); i++) {
883 class = &lock_classes[i];
884 if (!in_list(&class->lock_entry, &all_lock_classes) &&
885 !in_list(&class->lock_entry, &free_lock_classes) &&
886 !in_any_zapped_class_list(class)) {
887 printk(KERN_INFO "class %px/%s is not in any class list\n",
888 class, class->name ? : "(?)");
889 return false;
893 /* Check whether all classes have valid lock lists. */
894 for (i = 0; i < ARRAY_SIZE(lock_classes); i++) {
895 class = &lock_classes[i];
896 if (!class_lock_list_valid(class, &class->locks_before))
897 return false;
898 if (!class_lock_list_valid(class, &class->locks_after))
899 return false;
902 /* Check the chain_key of all lock chains. */
903 for (i = 0; i < ARRAY_SIZE(chainhash_table); i++) {
904 head = chainhash_table + i;
905 hlist_for_each_entry_rcu(chain, head, entry) {
906 if (!check_lock_chain_key(chain))
907 return false;
912 * Check whether all list entries that are in use occur in a class
913 * lock list.
915 for_each_set_bit(i, list_entries_in_use, ARRAY_SIZE(list_entries)) {
916 e = list_entries + i;
917 if (!in_any_class_list(&e->entry)) {
918 printk(KERN_INFO "list entry %d is not in any class list; class %s <> %s\n",
919 (unsigned int)(e - list_entries),
920 e->class->name ? : "(?)",
921 e->links_to->name ? : "(?)");
922 return false;
927 * Check whether all list entries that are not in use do not occur in
928 * a class lock list.
930 for_each_clear_bit(i, list_entries_in_use, ARRAY_SIZE(list_entries)) {
931 e = list_entries + i;
932 if (in_any_class_list(&e->entry)) {
933 printk(KERN_INFO "list entry %d occurs in a class list; class %s <> %s\n",
934 (unsigned int)(e - list_entries),
935 e->class && e->class->name ? e->class->name :
936 "(?)",
937 e->links_to && e->links_to->name ?
938 e->links_to->name : "(?)");
939 return false;
943 return true;
946 int check_consistency = 0;
947 module_param(check_consistency, int, 0644);
949 static void check_data_structures(void)
951 static bool once = false;
953 if (check_consistency && !once) {
954 if (!__check_data_structures()) {
955 once = true;
956 WARN_ON(once);
961 #else /* CONFIG_DEBUG_LOCKDEP */
963 static inline void check_data_structures(void) { }
965 #endif /* CONFIG_DEBUG_LOCKDEP */
968 * Initialize the lock_classes[] array elements, the free_lock_classes list
969 * and also the delayed_free structure.
971 static void init_data_structures_once(void)
973 static bool ds_initialized, rcu_head_initialized;
974 int i;
976 if (likely(rcu_head_initialized))
977 return;
979 if (system_state >= SYSTEM_SCHEDULING) {
980 init_rcu_head(&delayed_free.rcu_head);
981 rcu_head_initialized = true;
984 if (ds_initialized)
985 return;
987 ds_initialized = true;
989 INIT_LIST_HEAD(&delayed_free.pf[0].zapped);
990 INIT_LIST_HEAD(&delayed_free.pf[1].zapped);
992 for (i = 0; i < ARRAY_SIZE(lock_classes); i++) {
993 list_add_tail(&lock_classes[i].lock_entry, &free_lock_classes);
994 INIT_LIST_HEAD(&lock_classes[i].locks_after);
995 INIT_LIST_HEAD(&lock_classes[i].locks_before);
999 static inline struct hlist_head *keyhashentry(const struct lock_class_key *key)
1001 unsigned long hash = hash_long((uintptr_t)key, KEYHASH_BITS);
1003 return lock_keys_hash + hash;
1006 /* Register a dynamically allocated key. */
1007 void lockdep_register_key(struct lock_class_key *key)
1009 struct hlist_head *hash_head;
1010 struct lock_class_key *k;
1011 unsigned long flags;
1013 if (WARN_ON_ONCE(static_obj(key)))
1014 return;
1015 hash_head = keyhashentry(key);
1017 raw_local_irq_save(flags);
1018 if (!graph_lock())
1019 goto restore_irqs;
1020 hlist_for_each_entry_rcu(k, hash_head, hash_entry) {
1021 if (WARN_ON_ONCE(k == key))
1022 goto out_unlock;
1024 hlist_add_head_rcu(&key->hash_entry, hash_head);
1025 out_unlock:
1026 graph_unlock();
1027 restore_irqs:
1028 raw_local_irq_restore(flags);
1030 EXPORT_SYMBOL_GPL(lockdep_register_key);
1032 /* Check whether a key has been registered as a dynamic key. */
1033 static bool is_dynamic_key(const struct lock_class_key *key)
1035 struct hlist_head *hash_head;
1036 struct lock_class_key *k;
1037 bool found = false;
1039 if (WARN_ON_ONCE(static_obj(key)))
1040 return false;
1043 * If lock debugging is disabled lock_keys_hash[] may contain
1044 * pointers to memory that has already been freed. Avoid triggering
1045 * a use-after-free in that case by returning early.
1047 if (!debug_locks)
1048 return true;
1050 hash_head = keyhashentry(key);
1052 rcu_read_lock();
1053 hlist_for_each_entry_rcu(k, hash_head, hash_entry) {
1054 if (k == key) {
1055 found = true;
1056 break;
1059 rcu_read_unlock();
1061 return found;
1065 * Register a lock's class in the hash-table, if the class is not present
1066 * yet. Otherwise we look it up. We cache the result in the lock object
1067 * itself, so actual lookup of the hash should be once per lock object.
1069 static struct lock_class *
1070 register_lock_class(struct lockdep_map *lock, unsigned int subclass, int force)
1072 struct lockdep_subclass_key *key;
1073 struct hlist_head *hash_head;
1074 struct lock_class *class;
1076 DEBUG_LOCKS_WARN_ON(!irqs_disabled());
1078 class = look_up_lock_class(lock, subclass);
1079 if (likely(class))
1080 goto out_set_class_cache;
1082 if (!lock->key) {
1083 if (!assign_lock_key(lock))
1084 return NULL;
1085 } else if (!static_obj(lock->key) && !is_dynamic_key(lock->key)) {
1086 return NULL;
1089 key = lock->key->subkeys + subclass;
1090 hash_head = classhashentry(key);
1092 if (!graph_lock()) {
1093 return NULL;
1096 * We have to do the hash-walk again, to avoid races
1097 * with another CPU:
1099 hlist_for_each_entry_rcu(class, hash_head, hash_entry) {
1100 if (class->key == key)
1101 goto out_unlock_set;
1104 init_data_structures_once();
1106 /* Allocate a new lock class and add it to the hash. */
1107 class = list_first_entry_or_null(&free_lock_classes, typeof(*class),
1108 lock_entry);
1109 if (!class) {
1110 if (!debug_locks_off_graph_unlock()) {
1111 return NULL;
1114 print_lockdep_off("BUG: MAX_LOCKDEP_KEYS too low!");
1115 dump_stack();
1116 return NULL;
1118 nr_lock_classes++;
1119 debug_atomic_inc(nr_unused_locks);
1120 class->key = key;
1121 class->name = lock->name;
1122 class->subclass = subclass;
1123 WARN_ON_ONCE(!list_empty(&class->locks_before));
1124 WARN_ON_ONCE(!list_empty(&class->locks_after));
1125 class->name_version = count_matching_names(class);
1127 * We use RCU's safe list-add method to make
1128 * parallel walking of the hash-list safe:
1130 hlist_add_head_rcu(&class->hash_entry, hash_head);
1132 * Remove the class from the free list and add it to the global list
1133 * of classes.
1135 list_move_tail(&class->lock_entry, &all_lock_classes);
1137 if (verbose(class)) {
1138 graph_unlock();
1140 printk("\nnew class %px: %s", class->key, class->name);
1141 if (class->name_version > 1)
1142 printk(KERN_CONT "#%d", class->name_version);
1143 printk(KERN_CONT "\n");
1144 dump_stack();
1146 if (!graph_lock()) {
1147 return NULL;
1150 out_unlock_set:
1151 graph_unlock();
1153 out_set_class_cache:
1154 if (!subclass || force)
1155 lock->class_cache[0] = class;
1156 else if (subclass < NR_LOCKDEP_CACHING_CLASSES)
1157 lock->class_cache[subclass] = class;
1160 * Hash collision, did we smoke some? We found a class with a matching
1161 * hash but the subclass -- which is hashed in -- didn't match.
1163 if (DEBUG_LOCKS_WARN_ON(class->subclass != subclass))
1164 return NULL;
1166 return class;
1169 #ifdef CONFIG_PROVE_LOCKING
1171 * Allocate a lockdep entry. (assumes the graph_lock held, returns
1172 * with NULL on failure)
1174 static struct lock_list *alloc_list_entry(void)
1176 int idx = find_first_zero_bit(list_entries_in_use,
1177 ARRAY_SIZE(list_entries));
1179 if (idx >= ARRAY_SIZE(list_entries)) {
1180 if (!debug_locks_off_graph_unlock())
1181 return NULL;
1183 print_lockdep_off("BUG: MAX_LOCKDEP_ENTRIES too low!");
1184 dump_stack();
1185 return NULL;
1187 nr_list_entries++;
1188 __set_bit(idx, list_entries_in_use);
1189 return list_entries + idx;
1193 * Add a new dependency to the head of the list:
1195 static int add_lock_to_list(struct lock_class *this,
1196 struct lock_class *links_to, struct list_head *head,
1197 unsigned long ip, int distance,
1198 struct lock_trace *trace)
1200 struct lock_list *entry;
1202 * Lock not present yet - get a new dependency struct and
1203 * add it to the list:
1205 entry = alloc_list_entry();
1206 if (!entry)
1207 return 0;
1209 entry->class = this;
1210 entry->links_to = links_to;
1211 entry->distance = distance;
1212 entry->trace = *trace;
1214 * Both allocation and removal are done under the graph lock; but
1215 * iteration is under RCU-sched; see look_up_lock_class() and
1216 * lockdep_free_key_range().
1218 list_add_tail_rcu(&entry->entry, head);
1220 return 1;
1224 * For good efficiency of modular, we use power of 2
1226 #define MAX_CIRCULAR_QUEUE_SIZE 4096UL
1227 #define CQ_MASK (MAX_CIRCULAR_QUEUE_SIZE-1)
1230 * The circular_queue and helpers is used to implement the
1231 * breadth-first search(BFS)algorithem, by which we can build
1232 * the shortest path from the next lock to be acquired to the
1233 * previous held lock if there is a circular between them.
1235 struct circular_queue {
1236 unsigned long element[MAX_CIRCULAR_QUEUE_SIZE];
1237 unsigned int front, rear;
1240 static struct circular_queue lock_cq;
1242 unsigned int max_bfs_queue_depth;
1244 static unsigned int lockdep_dependency_gen_id;
1246 static inline void __cq_init(struct circular_queue *cq)
1248 cq->front = cq->rear = 0;
1249 lockdep_dependency_gen_id++;
1252 static inline int __cq_empty(struct circular_queue *cq)
1254 return (cq->front == cq->rear);
1257 static inline int __cq_full(struct circular_queue *cq)
1259 return ((cq->rear + 1) & CQ_MASK) == cq->front;
1262 static inline int __cq_enqueue(struct circular_queue *cq, unsigned long elem)
1264 if (__cq_full(cq))
1265 return -1;
1267 cq->element[cq->rear] = elem;
1268 cq->rear = (cq->rear + 1) & CQ_MASK;
1269 return 0;
1272 static inline int __cq_dequeue(struct circular_queue *cq, unsigned long *elem)
1274 if (__cq_empty(cq))
1275 return -1;
1277 *elem = cq->element[cq->front];
1278 cq->front = (cq->front + 1) & CQ_MASK;
1279 return 0;
1282 static inline unsigned int __cq_get_elem_count(struct circular_queue *cq)
1284 return (cq->rear - cq->front) & CQ_MASK;
1287 static inline void mark_lock_accessed(struct lock_list *lock,
1288 struct lock_list *parent)
1290 unsigned long nr;
1292 nr = lock - list_entries;
1293 WARN_ON(nr >= ARRAY_SIZE(list_entries)); /* Out-of-bounds, input fail */
1294 lock->parent = parent;
1295 lock->class->dep_gen_id = lockdep_dependency_gen_id;
1298 static inline unsigned long lock_accessed(struct lock_list *lock)
1300 unsigned long nr;
1302 nr = lock - list_entries;
1303 WARN_ON(nr >= ARRAY_SIZE(list_entries)); /* Out-of-bounds, input fail */
1304 return lock->class->dep_gen_id == lockdep_dependency_gen_id;
1307 static inline struct lock_list *get_lock_parent(struct lock_list *child)
1309 return child->parent;
1312 static inline int get_lock_depth(struct lock_list *child)
1314 int depth = 0;
1315 struct lock_list *parent;
1317 while ((parent = get_lock_parent(child))) {
1318 child = parent;
1319 depth++;
1321 return depth;
1324 static int __bfs(struct lock_list *source_entry,
1325 void *data,
1326 int (*match)(struct lock_list *entry, void *data),
1327 struct lock_list **target_entry,
1328 int forward)
1330 struct lock_list *entry;
1331 struct list_head *head;
1332 struct circular_queue *cq = &lock_cq;
1333 int ret = 1;
1335 if (match(source_entry, data)) {
1336 *target_entry = source_entry;
1337 ret = 0;
1338 goto exit;
1341 if (forward)
1342 head = &source_entry->class->locks_after;
1343 else
1344 head = &source_entry->class->locks_before;
1346 if (list_empty(head))
1347 goto exit;
1349 __cq_init(cq);
1350 __cq_enqueue(cq, (unsigned long)source_entry);
1352 while (!__cq_empty(cq)) {
1353 struct lock_list *lock;
1355 __cq_dequeue(cq, (unsigned long *)&lock);
1357 if (!lock->class) {
1358 ret = -2;
1359 goto exit;
1362 if (forward)
1363 head = &lock->class->locks_after;
1364 else
1365 head = &lock->class->locks_before;
1367 DEBUG_LOCKS_WARN_ON(!irqs_disabled());
1369 list_for_each_entry_rcu(entry, head, entry) {
1370 if (!lock_accessed(entry)) {
1371 unsigned int cq_depth;
1372 mark_lock_accessed(entry, lock);
1373 if (match(entry, data)) {
1374 *target_entry = entry;
1375 ret = 0;
1376 goto exit;
1379 if (__cq_enqueue(cq, (unsigned long)entry)) {
1380 ret = -1;
1381 goto exit;
1383 cq_depth = __cq_get_elem_count(cq);
1384 if (max_bfs_queue_depth < cq_depth)
1385 max_bfs_queue_depth = cq_depth;
1389 exit:
1390 return ret;
1393 static inline int __bfs_forwards(struct lock_list *src_entry,
1394 void *data,
1395 int (*match)(struct lock_list *entry, void *data),
1396 struct lock_list **target_entry)
1398 return __bfs(src_entry, data, match, target_entry, 1);
1402 static inline int __bfs_backwards(struct lock_list *src_entry,
1403 void *data,
1404 int (*match)(struct lock_list *entry, void *data),
1405 struct lock_list **target_entry)
1407 return __bfs(src_entry, data, match, target_entry, 0);
1412 * Recursive, forwards-direction lock-dependency checking, used for
1413 * both noncyclic checking and for hardirq-unsafe/softirq-unsafe
1414 * checking.
1417 static void print_lock_trace(struct lock_trace *trace, unsigned int spaces)
1419 unsigned long *entries = stack_trace + trace->offset;
1421 stack_trace_print(entries, trace->nr_entries, spaces);
1425 * Print a dependency chain entry (this is only done when a deadlock
1426 * has been detected):
1428 static noinline int
1429 print_circular_bug_entry(struct lock_list *target, int depth)
1431 if (debug_locks_silent)
1432 return 0;
1433 printk("\n-> #%u", depth);
1434 print_lock_name(target->class);
1435 printk(KERN_CONT ":\n");
1436 print_lock_trace(&target->trace, 6);
1437 return 0;
1440 static void
1441 print_circular_lock_scenario(struct held_lock *src,
1442 struct held_lock *tgt,
1443 struct lock_list *prt)
1445 struct lock_class *source = hlock_class(src);
1446 struct lock_class *target = hlock_class(tgt);
1447 struct lock_class *parent = prt->class;
1450 * A direct locking problem where unsafe_class lock is taken
1451 * directly by safe_class lock, then all we need to show
1452 * is the deadlock scenario, as it is obvious that the
1453 * unsafe lock is taken under the safe lock.
1455 * But if there is a chain instead, where the safe lock takes
1456 * an intermediate lock (middle_class) where this lock is
1457 * not the same as the safe lock, then the lock chain is
1458 * used to describe the problem. Otherwise we would need
1459 * to show a different CPU case for each link in the chain
1460 * from the safe_class lock to the unsafe_class lock.
1462 if (parent != source) {
1463 printk("Chain exists of:\n ");
1464 __print_lock_name(source);
1465 printk(KERN_CONT " --> ");
1466 __print_lock_name(parent);
1467 printk(KERN_CONT " --> ");
1468 __print_lock_name(target);
1469 printk(KERN_CONT "\n\n");
1472 printk(" Possible unsafe locking scenario:\n\n");
1473 printk(" CPU0 CPU1\n");
1474 printk(" ---- ----\n");
1475 printk(" lock(");
1476 __print_lock_name(target);
1477 printk(KERN_CONT ");\n");
1478 printk(" lock(");
1479 __print_lock_name(parent);
1480 printk(KERN_CONT ");\n");
1481 printk(" lock(");
1482 __print_lock_name(target);
1483 printk(KERN_CONT ");\n");
1484 printk(" lock(");
1485 __print_lock_name(source);
1486 printk(KERN_CONT ");\n");
1487 printk("\n *** DEADLOCK ***\n\n");
1491 * When a circular dependency is detected, print the
1492 * header first:
1494 static noinline int
1495 print_circular_bug_header(struct lock_list *entry, unsigned int depth,
1496 struct held_lock *check_src,
1497 struct held_lock *check_tgt)
1499 struct task_struct *curr = current;
1501 if (debug_locks_silent)
1502 return 0;
1504 pr_warn("\n");
1505 pr_warn("======================================================\n");
1506 pr_warn("WARNING: possible circular locking dependency detected\n");
1507 print_kernel_ident();
1508 pr_warn("------------------------------------------------------\n");
1509 pr_warn("%s/%d is trying to acquire lock:\n",
1510 curr->comm, task_pid_nr(curr));
1511 print_lock(check_src);
1513 pr_warn("\nbut task is already holding lock:\n");
1515 print_lock(check_tgt);
1516 pr_warn("\nwhich lock already depends on the new lock.\n\n");
1517 pr_warn("\nthe existing dependency chain (in reverse order) is:\n");
1519 print_circular_bug_entry(entry, depth);
1521 return 0;
1524 static inline int class_equal(struct lock_list *entry, void *data)
1526 return entry->class == data;
1529 static noinline int print_circular_bug(struct lock_list *this,
1530 struct lock_list *target,
1531 struct held_lock *check_src,
1532 struct held_lock *check_tgt)
1534 struct task_struct *curr = current;
1535 struct lock_list *parent;
1536 struct lock_list *first_parent;
1537 int depth;
1539 if (!debug_locks_off_graph_unlock() || debug_locks_silent)
1540 return 0;
1542 if (!save_trace(&this->trace))
1543 return 0;
1545 depth = get_lock_depth(target);
1547 print_circular_bug_header(target, depth, check_src, check_tgt);
1549 parent = get_lock_parent(target);
1550 first_parent = parent;
1552 while (parent) {
1553 print_circular_bug_entry(parent, --depth);
1554 parent = get_lock_parent(parent);
1557 printk("\nother info that might help us debug this:\n\n");
1558 print_circular_lock_scenario(check_src, check_tgt,
1559 first_parent);
1561 lockdep_print_held_locks(curr);
1563 printk("\nstack backtrace:\n");
1564 dump_stack();
1566 return 0;
1569 static noinline int print_bfs_bug(int ret)
1571 if (!debug_locks_off_graph_unlock())
1572 return 0;
1575 * Breadth-first-search failed, graph got corrupted?
1577 WARN(1, "lockdep bfs error:%d\n", ret);
1579 return 0;
1582 static int noop_count(struct lock_list *entry, void *data)
1584 (*(unsigned long *)data)++;
1585 return 0;
1588 static unsigned long __lockdep_count_forward_deps(struct lock_list *this)
1590 unsigned long count = 0;
1591 struct lock_list *uninitialized_var(target_entry);
1593 __bfs_forwards(this, (void *)&count, noop_count, &target_entry);
1595 return count;
1597 unsigned long lockdep_count_forward_deps(struct lock_class *class)
1599 unsigned long ret, flags;
1600 struct lock_list this;
1602 this.parent = NULL;
1603 this.class = class;
1605 raw_local_irq_save(flags);
1606 arch_spin_lock(&lockdep_lock);
1607 ret = __lockdep_count_forward_deps(&this);
1608 arch_spin_unlock(&lockdep_lock);
1609 raw_local_irq_restore(flags);
1611 return ret;
1614 static unsigned long __lockdep_count_backward_deps(struct lock_list *this)
1616 unsigned long count = 0;
1617 struct lock_list *uninitialized_var(target_entry);
1619 __bfs_backwards(this, (void *)&count, noop_count, &target_entry);
1621 return count;
1624 unsigned long lockdep_count_backward_deps(struct lock_class *class)
1626 unsigned long ret, flags;
1627 struct lock_list this;
1629 this.parent = NULL;
1630 this.class = class;
1632 raw_local_irq_save(flags);
1633 arch_spin_lock(&lockdep_lock);
1634 ret = __lockdep_count_backward_deps(&this);
1635 arch_spin_unlock(&lockdep_lock);
1636 raw_local_irq_restore(flags);
1638 return ret;
1642 * Prove that the dependency graph starting at <entry> can not
1643 * lead to <target>. Print an error and return 0 if it does.
1645 static noinline int
1646 check_noncircular(struct lock_list *root, struct lock_class *target,
1647 struct lock_list **target_entry)
1649 int result;
1651 debug_atomic_inc(nr_cyclic_checks);
1653 result = __bfs_forwards(root, target, class_equal, target_entry);
1655 return result;
1658 static noinline int
1659 check_redundant(struct lock_list *root, struct lock_class *target,
1660 struct lock_list **target_entry)
1662 int result;
1664 debug_atomic_inc(nr_redundant_checks);
1666 result = __bfs_forwards(root, target, class_equal, target_entry);
1668 return result;
1671 #if defined(CONFIG_TRACE_IRQFLAGS) && defined(CONFIG_PROVE_LOCKING)
1673 static inline int usage_accumulate(struct lock_list *entry, void *mask)
1675 *(unsigned long *)mask |= entry->class->usage_mask;
1677 return 0;
1681 * Forwards and backwards subgraph searching, for the purposes of
1682 * proving that two subgraphs can be connected by a new dependency
1683 * without creating any illegal irq-safe -> irq-unsafe lock dependency.
1686 static inline int usage_match(struct lock_list *entry, void *mask)
1688 return entry->class->usage_mask & *(unsigned long *)mask;
1692 * Find a node in the forwards-direction dependency sub-graph starting
1693 * at @root->class that matches @bit.
1695 * Return 0 if such a node exists in the subgraph, and put that node
1696 * into *@target_entry.
1698 * Return 1 otherwise and keep *@target_entry unchanged.
1699 * Return <0 on error.
1701 static int
1702 find_usage_forwards(struct lock_list *root, unsigned long usage_mask,
1703 struct lock_list **target_entry)
1705 int result;
1707 debug_atomic_inc(nr_find_usage_forwards_checks);
1709 result = __bfs_forwards(root, &usage_mask, usage_match, target_entry);
1711 return result;
1715 * Find a node in the backwards-direction dependency sub-graph starting
1716 * at @root->class that matches @bit.
1718 * Return 0 if such a node exists in the subgraph, and put that node
1719 * into *@target_entry.
1721 * Return 1 otherwise and keep *@target_entry unchanged.
1722 * Return <0 on error.
1724 static int
1725 find_usage_backwards(struct lock_list *root, unsigned long usage_mask,
1726 struct lock_list **target_entry)
1728 int result;
1730 debug_atomic_inc(nr_find_usage_backwards_checks);
1732 result = __bfs_backwards(root, &usage_mask, usage_match, target_entry);
1734 return result;
1737 static void print_lock_class_header(struct lock_class *class, int depth)
1739 int bit;
1741 printk("%*s->", depth, "");
1742 print_lock_name(class);
1743 #ifdef CONFIG_DEBUG_LOCKDEP
1744 printk(KERN_CONT " ops: %lu", debug_class_ops_read(class));
1745 #endif
1746 printk(KERN_CONT " {\n");
1748 for (bit = 0; bit < LOCK_USAGE_STATES; bit++) {
1749 if (class->usage_mask & (1 << bit)) {
1750 int len = depth;
1752 len += printk("%*s %s", depth, "", usage_str[bit]);
1753 len += printk(KERN_CONT " at:\n");
1754 print_lock_trace(class->usage_traces + bit, len);
1757 printk("%*s }\n", depth, "");
1759 printk("%*s ... key at: [<%px>] %pS\n",
1760 depth, "", class->key, class->key);
1764 * printk the shortest lock dependencies from @start to @end in reverse order:
1766 static void __used
1767 print_shortest_lock_dependencies(struct lock_list *leaf,
1768 struct lock_list *root)
1770 struct lock_list *entry = leaf;
1771 int depth;
1773 /*compute depth from generated tree by BFS*/
1774 depth = get_lock_depth(leaf);
1776 do {
1777 print_lock_class_header(entry->class, depth);
1778 printk("%*s ... acquired at:\n", depth, "");
1779 print_lock_trace(&entry->trace, 2);
1780 printk("\n");
1782 if (depth == 0 && (entry != root)) {
1783 printk("lockdep:%s bad path found in chain graph\n", __func__);
1784 break;
1787 entry = get_lock_parent(entry);
1788 depth--;
1789 } while (entry && (depth >= 0));
1791 return;
1794 static void
1795 print_irq_lock_scenario(struct lock_list *safe_entry,
1796 struct lock_list *unsafe_entry,
1797 struct lock_class *prev_class,
1798 struct lock_class *next_class)
1800 struct lock_class *safe_class = safe_entry->class;
1801 struct lock_class *unsafe_class = unsafe_entry->class;
1802 struct lock_class *middle_class = prev_class;
1804 if (middle_class == safe_class)
1805 middle_class = next_class;
1808 * A direct locking problem where unsafe_class lock is taken
1809 * directly by safe_class lock, then all we need to show
1810 * is the deadlock scenario, as it is obvious that the
1811 * unsafe lock is taken under the safe lock.
1813 * But if there is a chain instead, where the safe lock takes
1814 * an intermediate lock (middle_class) where this lock is
1815 * not the same as the safe lock, then the lock chain is
1816 * used to describe the problem. Otherwise we would need
1817 * to show a different CPU case for each link in the chain
1818 * from the safe_class lock to the unsafe_class lock.
1820 if (middle_class != unsafe_class) {
1821 printk("Chain exists of:\n ");
1822 __print_lock_name(safe_class);
1823 printk(KERN_CONT " --> ");
1824 __print_lock_name(middle_class);
1825 printk(KERN_CONT " --> ");
1826 __print_lock_name(unsafe_class);
1827 printk(KERN_CONT "\n\n");
1830 printk(" Possible interrupt unsafe locking scenario:\n\n");
1831 printk(" CPU0 CPU1\n");
1832 printk(" ---- ----\n");
1833 printk(" lock(");
1834 __print_lock_name(unsafe_class);
1835 printk(KERN_CONT ");\n");
1836 printk(" local_irq_disable();\n");
1837 printk(" lock(");
1838 __print_lock_name(safe_class);
1839 printk(KERN_CONT ");\n");
1840 printk(" lock(");
1841 __print_lock_name(middle_class);
1842 printk(KERN_CONT ");\n");
1843 printk(" <Interrupt>\n");
1844 printk(" lock(");
1845 __print_lock_name(safe_class);
1846 printk(KERN_CONT ");\n");
1847 printk("\n *** DEADLOCK ***\n\n");
1850 static int
1851 print_bad_irq_dependency(struct task_struct *curr,
1852 struct lock_list *prev_root,
1853 struct lock_list *next_root,
1854 struct lock_list *backwards_entry,
1855 struct lock_list *forwards_entry,
1856 struct held_lock *prev,
1857 struct held_lock *next,
1858 enum lock_usage_bit bit1,
1859 enum lock_usage_bit bit2,
1860 const char *irqclass)
1862 if (!debug_locks_off_graph_unlock() || debug_locks_silent)
1863 return 0;
1865 pr_warn("\n");
1866 pr_warn("=====================================================\n");
1867 pr_warn("WARNING: %s-safe -> %s-unsafe lock order detected\n",
1868 irqclass, irqclass);
1869 print_kernel_ident();
1870 pr_warn("-----------------------------------------------------\n");
1871 pr_warn("%s/%d [HC%u[%lu]:SC%u[%lu]:HE%u:SE%u] is trying to acquire:\n",
1872 curr->comm, task_pid_nr(curr),
1873 curr->hardirq_context, hardirq_count() >> HARDIRQ_SHIFT,
1874 curr->softirq_context, softirq_count() >> SOFTIRQ_SHIFT,
1875 curr->hardirqs_enabled,
1876 curr->softirqs_enabled);
1877 print_lock(next);
1879 pr_warn("\nand this task is already holding:\n");
1880 print_lock(prev);
1881 pr_warn("which would create a new lock dependency:\n");
1882 print_lock_name(hlock_class(prev));
1883 pr_cont(" ->");
1884 print_lock_name(hlock_class(next));
1885 pr_cont("\n");
1887 pr_warn("\nbut this new dependency connects a %s-irq-safe lock:\n",
1888 irqclass);
1889 print_lock_name(backwards_entry->class);
1890 pr_warn("\n... which became %s-irq-safe at:\n", irqclass);
1892 print_lock_trace(backwards_entry->class->usage_traces + bit1, 1);
1894 pr_warn("\nto a %s-irq-unsafe lock:\n", irqclass);
1895 print_lock_name(forwards_entry->class);
1896 pr_warn("\n... which became %s-irq-unsafe at:\n", irqclass);
1897 pr_warn("...");
1899 print_lock_trace(forwards_entry->class->usage_traces + bit2, 1);
1901 pr_warn("\nother info that might help us debug this:\n\n");
1902 print_irq_lock_scenario(backwards_entry, forwards_entry,
1903 hlock_class(prev), hlock_class(next));
1905 lockdep_print_held_locks(curr);
1907 pr_warn("\nthe dependencies between %s-irq-safe lock and the holding lock:\n", irqclass);
1908 if (!save_trace(&prev_root->trace))
1909 return 0;
1910 print_shortest_lock_dependencies(backwards_entry, prev_root);
1912 pr_warn("\nthe dependencies between the lock to be acquired");
1913 pr_warn(" and %s-irq-unsafe lock:\n", irqclass);
1914 if (!save_trace(&next_root->trace))
1915 return 0;
1916 print_shortest_lock_dependencies(forwards_entry, next_root);
1918 pr_warn("\nstack backtrace:\n");
1919 dump_stack();
1921 return 0;
1924 static const char *state_names[] = {
1925 #define LOCKDEP_STATE(__STATE) \
1926 __stringify(__STATE),
1927 #include "lockdep_states.h"
1928 #undef LOCKDEP_STATE
1931 static const char *state_rnames[] = {
1932 #define LOCKDEP_STATE(__STATE) \
1933 __stringify(__STATE)"-READ",
1934 #include "lockdep_states.h"
1935 #undef LOCKDEP_STATE
1938 static inline const char *state_name(enum lock_usage_bit bit)
1940 if (bit & LOCK_USAGE_READ_MASK)
1941 return state_rnames[bit >> LOCK_USAGE_DIR_MASK];
1942 else
1943 return state_names[bit >> LOCK_USAGE_DIR_MASK];
1947 * The bit number is encoded like:
1949 * bit0: 0 exclusive, 1 read lock
1950 * bit1: 0 used in irq, 1 irq enabled
1951 * bit2-n: state
1953 static int exclusive_bit(int new_bit)
1955 int state = new_bit & LOCK_USAGE_STATE_MASK;
1956 int dir = new_bit & LOCK_USAGE_DIR_MASK;
1959 * keep state, bit flip the direction and strip read.
1961 return state | (dir ^ LOCK_USAGE_DIR_MASK);
1965 * Observe that when given a bitmask where each bitnr is encoded as above, a
1966 * right shift of the mask transforms the individual bitnrs as -1 and
1967 * conversely, a left shift transforms into +1 for the individual bitnrs.
1969 * So for all bits whose number have LOCK_ENABLED_* set (bitnr1 == 1), we can
1970 * create the mask with those bit numbers using LOCK_USED_IN_* (bitnr1 == 0)
1971 * instead by subtracting the bit number by 2, or shifting the mask right by 2.
1973 * Similarly, bitnr1 == 0 becomes bitnr1 == 1 by adding 2, or shifting left 2.
1975 * So split the mask (note that LOCKF_ENABLED_IRQ_ALL|LOCKF_USED_IN_IRQ_ALL is
1976 * all bits set) and recompose with bitnr1 flipped.
1978 static unsigned long invert_dir_mask(unsigned long mask)
1980 unsigned long excl = 0;
1982 /* Invert dir */
1983 excl |= (mask & LOCKF_ENABLED_IRQ_ALL) >> LOCK_USAGE_DIR_MASK;
1984 excl |= (mask & LOCKF_USED_IN_IRQ_ALL) << LOCK_USAGE_DIR_MASK;
1986 return excl;
1990 * As above, we clear bitnr0 (LOCK_*_READ off) with bitmask ops. First, for all
1991 * bits with bitnr0 set (LOCK_*_READ), add those with bitnr0 cleared (LOCK_*).
1992 * And then mask out all bitnr0.
1994 static unsigned long exclusive_mask(unsigned long mask)
1996 unsigned long excl = invert_dir_mask(mask);
1998 /* Strip read */
1999 excl |= (excl & LOCKF_IRQ_READ) >> LOCK_USAGE_READ_MASK;
2000 excl &= ~LOCKF_IRQ_READ;
2002 return excl;
2006 * Retrieve the _possible_ original mask to which @mask is
2007 * exclusive. Ie: this is the opposite of exclusive_mask().
2008 * Note that 2 possible original bits can match an exclusive
2009 * bit: one has LOCK_USAGE_READ_MASK set, the other has it
2010 * cleared. So both are returned for each exclusive bit.
2012 static unsigned long original_mask(unsigned long mask)
2014 unsigned long excl = invert_dir_mask(mask);
2016 /* Include read in existing usages */
2017 excl |= (excl & LOCKF_IRQ) << LOCK_USAGE_READ_MASK;
2019 return excl;
2023 * Find the first pair of bit match between an original
2024 * usage mask and an exclusive usage mask.
2026 static int find_exclusive_match(unsigned long mask,
2027 unsigned long excl_mask,
2028 enum lock_usage_bit *bitp,
2029 enum lock_usage_bit *excl_bitp)
2031 int bit, excl;
2033 for_each_set_bit(bit, &mask, LOCK_USED) {
2034 excl = exclusive_bit(bit);
2035 if (excl_mask & lock_flag(excl)) {
2036 *bitp = bit;
2037 *excl_bitp = excl;
2038 return 0;
2041 return -1;
2045 * Prove that the new dependency does not connect a hardirq-safe(-read)
2046 * lock with a hardirq-unsafe lock - to achieve this we search
2047 * the backwards-subgraph starting at <prev>, and the
2048 * forwards-subgraph starting at <next>:
2050 static int check_irq_usage(struct task_struct *curr, struct held_lock *prev,
2051 struct held_lock *next)
2053 unsigned long usage_mask = 0, forward_mask, backward_mask;
2054 enum lock_usage_bit forward_bit = 0, backward_bit = 0;
2055 struct lock_list *uninitialized_var(target_entry1);
2056 struct lock_list *uninitialized_var(target_entry);
2057 struct lock_list this, that;
2058 int ret;
2061 * Step 1: gather all hard/soft IRQs usages backward in an
2062 * accumulated usage mask.
2064 this.parent = NULL;
2065 this.class = hlock_class(prev);
2067 ret = __bfs_backwards(&this, &usage_mask, usage_accumulate, NULL);
2068 if (ret < 0)
2069 return print_bfs_bug(ret);
2071 usage_mask &= LOCKF_USED_IN_IRQ_ALL;
2072 if (!usage_mask)
2073 return 1;
2076 * Step 2: find exclusive uses forward that match the previous
2077 * backward accumulated mask.
2079 forward_mask = exclusive_mask(usage_mask);
2081 that.parent = NULL;
2082 that.class = hlock_class(next);
2084 ret = find_usage_forwards(&that, forward_mask, &target_entry1);
2085 if (ret < 0)
2086 return print_bfs_bug(ret);
2087 if (ret == 1)
2088 return ret;
2091 * Step 3: we found a bad match! Now retrieve a lock from the backward
2092 * list whose usage mask matches the exclusive usage mask from the
2093 * lock found on the forward list.
2095 backward_mask = original_mask(target_entry1->class->usage_mask);
2097 ret = find_usage_backwards(&this, backward_mask, &target_entry);
2098 if (ret < 0)
2099 return print_bfs_bug(ret);
2100 if (DEBUG_LOCKS_WARN_ON(ret == 1))
2101 return 1;
2104 * Step 4: narrow down to a pair of incompatible usage bits
2105 * and report it.
2107 ret = find_exclusive_match(target_entry->class->usage_mask,
2108 target_entry1->class->usage_mask,
2109 &backward_bit, &forward_bit);
2110 if (DEBUG_LOCKS_WARN_ON(ret == -1))
2111 return 1;
2113 return print_bad_irq_dependency(curr, &this, &that,
2114 target_entry, target_entry1,
2115 prev, next,
2116 backward_bit, forward_bit,
2117 state_name(backward_bit));
2120 static void inc_chains(void)
2122 if (current->hardirq_context)
2123 nr_hardirq_chains++;
2124 else {
2125 if (current->softirq_context)
2126 nr_softirq_chains++;
2127 else
2128 nr_process_chains++;
2132 #else
2134 static inline int check_irq_usage(struct task_struct *curr,
2135 struct held_lock *prev, struct held_lock *next)
2137 return 1;
2140 static inline void inc_chains(void)
2142 nr_process_chains++;
2145 #endif
2147 static void
2148 print_deadlock_scenario(struct held_lock *nxt,
2149 struct held_lock *prv)
2151 struct lock_class *next = hlock_class(nxt);
2152 struct lock_class *prev = hlock_class(prv);
2154 printk(" Possible unsafe locking scenario:\n\n");
2155 printk(" CPU0\n");
2156 printk(" ----\n");
2157 printk(" lock(");
2158 __print_lock_name(prev);
2159 printk(KERN_CONT ");\n");
2160 printk(" lock(");
2161 __print_lock_name(next);
2162 printk(KERN_CONT ");\n");
2163 printk("\n *** DEADLOCK ***\n\n");
2164 printk(" May be due to missing lock nesting notation\n\n");
2167 static int
2168 print_deadlock_bug(struct task_struct *curr, struct held_lock *prev,
2169 struct held_lock *next)
2171 if (!debug_locks_off_graph_unlock() || debug_locks_silent)
2172 return 0;
2174 pr_warn("\n");
2175 pr_warn("============================================\n");
2176 pr_warn("WARNING: possible recursive locking detected\n");
2177 print_kernel_ident();
2178 pr_warn("--------------------------------------------\n");
2179 pr_warn("%s/%d is trying to acquire lock:\n",
2180 curr->comm, task_pid_nr(curr));
2181 print_lock(next);
2182 pr_warn("\nbut task is already holding lock:\n");
2183 print_lock(prev);
2185 pr_warn("\nother info that might help us debug this:\n");
2186 print_deadlock_scenario(next, prev);
2187 lockdep_print_held_locks(curr);
2189 pr_warn("\nstack backtrace:\n");
2190 dump_stack();
2192 return 0;
2196 * Check whether we are holding such a class already.
2198 * (Note that this has to be done separately, because the graph cannot
2199 * detect such classes of deadlocks.)
2201 * Returns: 0 on deadlock detected, 1 on OK, 2 on recursive read
2203 static int
2204 check_deadlock(struct task_struct *curr, struct held_lock *next,
2205 struct lockdep_map *next_instance, int read)
2207 struct held_lock *prev;
2208 struct held_lock *nest = NULL;
2209 int i;
2211 for (i = 0; i < curr->lockdep_depth; i++) {
2212 prev = curr->held_locks + i;
2214 if (prev->instance == next->nest_lock)
2215 nest = prev;
2217 if (hlock_class(prev) != hlock_class(next))
2218 continue;
2221 * Allow read-after-read recursion of the same
2222 * lock class (i.e. read_lock(lock)+read_lock(lock)):
2224 if ((read == 2) && prev->read)
2225 return 2;
2228 * We're holding the nest_lock, which serializes this lock's
2229 * nesting behaviour.
2231 if (nest)
2232 return 2;
2234 return print_deadlock_bug(curr, prev, next);
2236 return 1;
2240 * There was a chain-cache miss, and we are about to add a new dependency
2241 * to a previous lock. We recursively validate the following rules:
2243 * - would the adding of the <prev> -> <next> dependency create a
2244 * circular dependency in the graph? [== circular deadlock]
2246 * - does the new prev->next dependency connect any hardirq-safe lock
2247 * (in the full backwards-subgraph starting at <prev>) with any
2248 * hardirq-unsafe lock (in the full forwards-subgraph starting at
2249 * <next>)? [== illegal lock inversion with hardirq contexts]
2251 * - does the new prev->next dependency connect any softirq-safe lock
2252 * (in the full backwards-subgraph starting at <prev>) with any
2253 * softirq-unsafe lock (in the full forwards-subgraph starting at
2254 * <next>)? [== illegal lock inversion with softirq contexts]
2256 * any of these scenarios could lead to a deadlock.
2258 * Then if all the validations pass, we add the forwards and backwards
2259 * dependency.
2261 static int
2262 check_prev_add(struct task_struct *curr, struct held_lock *prev,
2263 struct held_lock *next, int distance, struct lock_trace *trace)
2265 struct lock_list *uninitialized_var(target_entry);
2266 struct lock_list *entry;
2267 struct lock_list this;
2268 int ret;
2270 if (!hlock_class(prev)->key || !hlock_class(next)->key) {
2272 * The warning statements below may trigger a use-after-free
2273 * of the class name. It is better to trigger a use-after free
2274 * and to have the class name most of the time instead of not
2275 * having the class name available.
2277 WARN_ONCE(!debug_locks_silent && !hlock_class(prev)->key,
2278 "Detected use-after-free of lock class %px/%s\n",
2279 hlock_class(prev),
2280 hlock_class(prev)->name);
2281 WARN_ONCE(!debug_locks_silent && !hlock_class(next)->key,
2282 "Detected use-after-free of lock class %px/%s\n",
2283 hlock_class(next),
2284 hlock_class(next)->name);
2285 return 2;
2289 * Prove that the new <prev> -> <next> dependency would not
2290 * create a circular dependency in the graph. (We do this by
2291 * forward-recursing into the graph starting at <next>, and
2292 * checking whether we can reach <prev>.)
2294 * We are using global variables to control the recursion, to
2295 * keep the stackframe size of the recursive functions low:
2297 this.class = hlock_class(next);
2298 this.parent = NULL;
2299 ret = check_noncircular(&this, hlock_class(prev), &target_entry);
2300 if (unlikely(!ret)) {
2301 if (!trace->nr_entries) {
2303 * If save_trace fails here, the printing might
2304 * trigger a WARN but because of the !nr_entries it
2305 * should not do bad things.
2307 save_trace(trace);
2309 return print_circular_bug(&this, target_entry, next, prev);
2311 else if (unlikely(ret < 0))
2312 return print_bfs_bug(ret);
2314 if (!check_irq_usage(curr, prev, next))
2315 return 0;
2318 * For recursive read-locks we do all the dependency checks,
2319 * but we dont store read-triggered dependencies (only
2320 * write-triggered dependencies). This ensures that only the
2321 * write-side dependencies matter, and that if for example a
2322 * write-lock never takes any other locks, then the reads are
2323 * equivalent to a NOP.
2325 if (next->read == 2 || prev->read == 2)
2326 return 1;
2328 * Is the <prev> -> <next> dependency already present?
2330 * (this may occur even though this is a new chain: consider
2331 * e.g. the L1 -> L2 -> L3 -> L4 and the L5 -> L1 -> L2 -> L3
2332 * chains - the second one will be new, but L1 already has
2333 * L2 added to its dependency list, due to the first chain.)
2335 list_for_each_entry(entry, &hlock_class(prev)->locks_after, entry) {
2336 if (entry->class == hlock_class(next)) {
2337 if (distance == 1)
2338 entry->distance = 1;
2339 return 1;
2344 * Is the <prev> -> <next> link redundant?
2346 this.class = hlock_class(prev);
2347 this.parent = NULL;
2348 ret = check_redundant(&this, hlock_class(next), &target_entry);
2349 if (!ret) {
2350 debug_atomic_inc(nr_redundant);
2351 return 2;
2353 if (ret < 0)
2354 return print_bfs_bug(ret);
2357 if (!trace->nr_entries && !save_trace(trace))
2358 return 0;
2361 * Ok, all validations passed, add the new lock
2362 * to the previous lock's dependency list:
2364 ret = add_lock_to_list(hlock_class(next), hlock_class(prev),
2365 &hlock_class(prev)->locks_after,
2366 next->acquire_ip, distance, trace);
2368 if (!ret)
2369 return 0;
2371 ret = add_lock_to_list(hlock_class(prev), hlock_class(next),
2372 &hlock_class(next)->locks_before,
2373 next->acquire_ip, distance, trace);
2374 if (!ret)
2375 return 0;
2377 return 2;
2381 * Add the dependency to all directly-previous locks that are 'relevant'.
2382 * The ones that are relevant are (in increasing distance from curr):
2383 * all consecutive trylock entries and the final non-trylock entry - or
2384 * the end of this context's lock-chain - whichever comes first.
2386 static int
2387 check_prevs_add(struct task_struct *curr, struct held_lock *next)
2389 struct lock_trace trace = { .nr_entries = 0 };
2390 int depth = curr->lockdep_depth;
2391 struct held_lock *hlock;
2394 * Debugging checks.
2396 * Depth must not be zero for a non-head lock:
2398 if (!depth)
2399 goto out_bug;
2401 * At least two relevant locks must exist for this
2402 * to be a head:
2404 if (curr->held_locks[depth].irq_context !=
2405 curr->held_locks[depth-1].irq_context)
2406 goto out_bug;
2408 for (;;) {
2409 int distance = curr->lockdep_depth - depth + 1;
2410 hlock = curr->held_locks + depth - 1;
2413 * Only non-recursive-read entries get new dependencies
2414 * added:
2416 if (hlock->read != 2 && hlock->check) {
2417 int ret = check_prev_add(curr, hlock, next, distance,
2418 &trace);
2419 if (!ret)
2420 return 0;
2423 * Stop after the first non-trylock entry,
2424 * as non-trylock entries have added their
2425 * own direct dependencies already, so this
2426 * lock is connected to them indirectly:
2428 if (!hlock->trylock)
2429 break;
2432 depth--;
2434 * End of lock-stack?
2436 if (!depth)
2437 break;
2439 * Stop the search if we cross into another context:
2441 if (curr->held_locks[depth].irq_context !=
2442 curr->held_locks[depth-1].irq_context)
2443 break;
2445 return 1;
2446 out_bug:
2447 if (!debug_locks_off_graph_unlock())
2448 return 0;
2451 * Clearly we all shouldn't be here, but since we made it we
2452 * can reliable say we messed up our state. See the above two
2453 * gotos for reasons why we could possibly end up here.
2455 WARN_ON(1);
2457 return 0;
2460 struct lock_chain lock_chains[MAX_LOCKDEP_CHAINS];
2461 static DECLARE_BITMAP(lock_chains_in_use, MAX_LOCKDEP_CHAINS);
2462 int nr_chain_hlocks;
2463 static u16 chain_hlocks[MAX_LOCKDEP_CHAIN_HLOCKS];
2465 struct lock_class *lock_chain_get_class(struct lock_chain *chain, int i)
2467 return lock_classes + chain_hlocks[chain->base + i];
2471 * Returns the index of the first held_lock of the current chain
2473 static inline int get_first_held_lock(struct task_struct *curr,
2474 struct held_lock *hlock)
2476 int i;
2477 struct held_lock *hlock_curr;
2479 for (i = curr->lockdep_depth - 1; i >= 0; i--) {
2480 hlock_curr = curr->held_locks + i;
2481 if (hlock_curr->irq_context != hlock->irq_context)
2482 break;
2486 return ++i;
2489 #ifdef CONFIG_DEBUG_LOCKDEP
2491 * Returns the next chain_key iteration
2493 static u64 print_chain_key_iteration(int class_idx, u64 chain_key)
2495 u64 new_chain_key = iterate_chain_key(chain_key, class_idx);
2497 printk(" class_idx:%d -> chain_key:%016Lx",
2498 class_idx,
2499 (unsigned long long)new_chain_key);
2500 return new_chain_key;
2503 static void
2504 print_chain_keys_held_locks(struct task_struct *curr, struct held_lock *hlock_next)
2506 struct held_lock *hlock;
2507 u64 chain_key = 0;
2508 int depth = curr->lockdep_depth;
2509 int i;
2511 printk("depth: %u\n", depth + 1);
2512 for (i = get_first_held_lock(curr, hlock_next); i < depth; i++) {
2513 hlock = curr->held_locks + i;
2514 chain_key = print_chain_key_iteration(hlock->class_idx, chain_key);
2516 print_lock(hlock);
2519 print_chain_key_iteration(hlock_next->class_idx, chain_key);
2520 print_lock(hlock_next);
2523 static void print_chain_keys_chain(struct lock_chain *chain)
2525 int i;
2526 u64 chain_key = 0;
2527 int class_id;
2529 printk("depth: %u\n", chain->depth);
2530 for (i = 0; i < chain->depth; i++) {
2531 class_id = chain_hlocks[chain->base + i];
2532 chain_key = print_chain_key_iteration(class_id + 1, chain_key);
2534 print_lock_name(lock_classes + class_id);
2535 printk("\n");
2539 static void print_collision(struct task_struct *curr,
2540 struct held_lock *hlock_next,
2541 struct lock_chain *chain)
2543 pr_warn("\n");
2544 pr_warn("============================\n");
2545 pr_warn("WARNING: chain_key collision\n");
2546 print_kernel_ident();
2547 pr_warn("----------------------------\n");
2548 pr_warn("%s/%d: ", current->comm, task_pid_nr(current));
2549 pr_warn("Hash chain already cached but the contents don't match!\n");
2551 pr_warn("Held locks:");
2552 print_chain_keys_held_locks(curr, hlock_next);
2554 pr_warn("Locks in cached chain:");
2555 print_chain_keys_chain(chain);
2557 pr_warn("\nstack backtrace:\n");
2558 dump_stack();
2560 #endif
2563 * Checks whether the chain and the current held locks are consistent
2564 * in depth and also in content. If they are not it most likely means
2565 * that there was a collision during the calculation of the chain_key.
2566 * Returns: 0 not passed, 1 passed
2568 static int check_no_collision(struct task_struct *curr,
2569 struct held_lock *hlock,
2570 struct lock_chain *chain)
2572 #ifdef CONFIG_DEBUG_LOCKDEP
2573 int i, j, id;
2575 i = get_first_held_lock(curr, hlock);
2577 if (DEBUG_LOCKS_WARN_ON(chain->depth != curr->lockdep_depth - (i - 1))) {
2578 print_collision(curr, hlock, chain);
2579 return 0;
2582 for (j = 0; j < chain->depth - 1; j++, i++) {
2583 id = curr->held_locks[i].class_idx - 1;
2585 if (DEBUG_LOCKS_WARN_ON(chain_hlocks[chain->base + j] != id)) {
2586 print_collision(curr, hlock, chain);
2587 return 0;
2590 #endif
2591 return 1;
2595 * Given an index that is >= -1, return the index of the next lock chain.
2596 * Return -2 if there is no next lock chain.
2598 long lockdep_next_lockchain(long i)
2600 i = find_next_bit(lock_chains_in_use, ARRAY_SIZE(lock_chains), i + 1);
2601 return i < ARRAY_SIZE(lock_chains) ? i : -2;
2604 unsigned long lock_chain_count(void)
2606 return bitmap_weight(lock_chains_in_use, ARRAY_SIZE(lock_chains));
2609 /* Must be called with the graph lock held. */
2610 static struct lock_chain *alloc_lock_chain(void)
2612 int idx = find_first_zero_bit(lock_chains_in_use,
2613 ARRAY_SIZE(lock_chains));
2615 if (unlikely(idx >= ARRAY_SIZE(lock_chains)))
2616 return NULL;
2617 __set_bit(idx, lock_chains_in_use);
2618 return lock_chains + idx;
2622 * Adds a dependency chain into chain hashtable. And must be called with
2623 * graph_lock held.
2625 * Return 0 if fail, and graph_lock is released.
2626 * Return 1 if succeed, with graph_lock held.
2628 static inline int add_chain_cache(struct task_struct *curr,
2629 struct held_lock *hlock,
2630 u64 chain_key)
2632 struct lock_class *class = hlock_class(hlock);
2633 struct hlist_head *hash_head = chainhashentry(chain_key);
2634 struct lock_chain *chain;
2635 int i, j;
2638 * The caller must hold the graph lock, ensure we've got IRQs
2639 * disabled to make this an IRQ-safe lock.. for recursion reasons
2640 * lockdep won't complain about its own locking errors.
2642 if (DEBUG_LOCKS_WARN_ON(!irqs_disabled()))
2643 return 0;
2645 chain = alloc_lock_chain();
2646 if (!chain) {
2647 if (!debug_locks_off_graph_unlock())
2648 return 0;
2650 print_lockdep_off("BUG: MAX_LOCKDEP_CHAINS too low!");
2651 dump_stack();
2652 return 0;
2654 chain->chain_key = chain_key;
2655 chain->irq_context = hlock->irq_context;
2656 i = get_first_held_lock(curr, hlock);
2657 chain->depth = curr->lockdep_depth + 1 - i;
2659 BUILD_BUG_ON((1UL << 24) <= ARRAY_SIZE(chain_hlocks));
2660 BUILD_BUG_ON((1UL << 6) <= ARRAY_SIZE(curr->held_locks));
2661 BUILD_BUG_ON((1UL << 8*sizeof(chain_hlocks[0])) <= ARRAY_SIZE(lock_classes));
2663 if (likely(nr_chain_hlocks + chain->depth <= MAX_LOCKDEP_CHAIN_HLOCKS)) {
2664 chain->base = nr_chain_hlocks;
2665 for (j = 0; j < chain->depth - 1; j++, i++) {
2666 int lock_id = curr->held_locks[i].class_idx - 1;
2667 chain_hlocks[chain->base + j] = lock_id;
2669 chain_hlocks[chain->base + j] = class - lock_classes;
2670 nr_chain_hlocks += chain->depth;
2671 } else {
2672 if (!debug_locks_off_graph_unlock())
2673 return 0;
2675 print_lockdep_off("BUG: MAX_LOCKDEP_CHAIN_HLOCKS too low!");
2676 dump_stack();
2677 return 0;
2680 hlist_add_head_rcu(&chain->entry, hash_head);
2681 debug_atomic_inc(chain_lookup_misses);
2682 inc_chains();
2684 return 1;
2688 * Look up a dependency chain. Must be called with either the graph lock or
2689 * the RCU read lock held.
2691 static inline struct lock_chain *lookup_chain_cache(u64 chain_key)
2693 struct hlist_head *hash_head = chainhashentry(chain_key);
2694 struct lock_chain *chain;
2696 hlist_for_each_entry_rcu(chain, hash_head, entry) {
2697 if (READ_ONCE(chain->chain_key) == chain_key) {
2698 debug_atomic_inc(chain_lookup_hits);
2699 return chain;
2702 return NULL;
2706 * If the key is not present yet in dependency chain cache then
2707 * add it and return 1 - in this case the new dependency chain is
2708 * validated. If the key is already hashed, return 0.
2709 * (On return with 1 graph_lock is held.)
2711 static inline int lookup_chain_cache_add(struct task_struct *curr,
2712 struct held_lock *hlock,
2713 u64 chain_key)
2715 struct lock_class *class = hlock_class(hlock);
2716 struct lock_chain *chain = lookup_chain_cache(chain_key);
2718 if (chain) {
2719 cache_hit:
2720 if (!check_no_collision(curr, hlock, chain))
2721 return 0;
2723 if (very_verbose(class)) {
2724 printk("\nhash chain already cached, key: "
2725 "%016Lx tail class: [%px] %s\n",
2726 (unsigned long long)chain_key,
2727 class->key, class->name);
2730 return 0;
2733 if (very_verbose(class)) {
2734 printk("\nnew hash chain, key: %016Lx tail class: [%px] %s\n",
2735 (unsigned long long)chain_key, class->key, class->name);
2738 if (!graph_lock())
2739 return 0;
2742 * We have to walk the chain again locked - to avoid duplicates:
2744 chain = lookup_chain_cache(chain_key);
2745 if (chain) {
2746 graph_unlock();
2747 goto cache_hit;
2750 if (!add_chain_cache(curr, hlock, chain_key))
2751 return 0;
2753 return 1;
2756 static int validate_chain(struct task_struct *curr, struct lockdep_map *lock,
2757 struct held_lock *hlock, int chain_head, u64 chain_key)
2760 * Trylock needs to maintain the stack of held locks, but it
2761 * does not add new dependencies, because trylock can be done
2762 * in any order.
2764 * We look up the chain_key and do the O(N^2) check and update of
2765 * the dependencies only if this is a new dependency chain.
2766 * (If lookup_chain_cache_add() return with 1 it acquires
2767 * graph_lock for us)
2769 if (!hlock->trylock && hlock->check &&
2770 lookup_chain_cache_add(curr, hlock, chain_key)) {
2772 * Check whether last held lock:
2774 * - is irq-safe, if this lock is irq-unsafe
2775 * - is softirq-safe, if this lock is hardirq-unsafe
2777 * And check whether the new lock's dependency graph
2778 * could lead back to the previous lock.
2780 * any of these scenarios could lead to a deadlock. If
2781 * All validations
2783 int ret = check_deadlock(curr, hlock, lock, hlock->read);
2785 if (!ret)
2786 return 0;
2788 * Mark recursive read, as we jump over it when
2789 * building dependencies (just like we jump over
2790 * trylock entries):
2792 if (ret == 2)
2793 hlock->read = 2;
2795 * Add dependency only if this lock is not the head
2796 * of the chain, and if it's not a secondary read-lock:
2798 if (!chain_head && ret != 2) {
2799 if (!check_prevs_add(curr, hlock))
2800 return 0;
2803 graph_unlock();
2804 } else {
2805 /* after lookup_chain_cache_add(): */
2806 if (unlikely(!debug_locks))
2807 return 0;
2810 return 1;
2812 #else
2813 static inline int validate_chain(struct task_struct *curr,
2814 struct lockdep_map *lock, struct held_lock *hlock,
2815 int chain_head, u64 chain_key)
2817 return 1;
2820 static void print_lock_trace(struct lock_trace *trace, unsigned int spaces)
2823 #endif
2826 * We are building curr_chain_key incrementally, so double-check
2827 * it from scratch, to make sure that it's done correctly:
2829 static void check_chain_key(struct task_struct *curr)
2831 #ifdef CONFIG_DEBUG_LOCKDEP
2832 struct held_lock *hlock, *prev_hlock = NULL;
2833 unsigned int i;
2834 u64 chain_key = 0;
2836 for (i = 0; i < curr->lockdep_depth; i++) {
2837 hlock = curr->held_locks + i;
2838 if (chain_key != hlock->prev_chain_key) {
2839 debug_locks_off();
2841 * We got mighty confused, our chain keys don't match
2842 * with what we expect, someone trample on our task state?
2844 WARN(1, "hm#1, depth: %u [%u], %016Lx != %016Lx\n",
2845 curr->lockdep_depth, i,
2846 (unsigned long long)chain_key,
2847 (unsigned long long)hlock->prev_chain_key);
2848 return;
2851 * Whoops ran out of static storage again?
2853 if (DEBUG_LOCKS_WARN_ON(hlock->class_idx > MAX_LOCKDEP_KEYS))
2854 return;
2856 if (prev_hlock && (prev_hlock->irq_context !=
2857 hlock->irq_context))
2858 chain_key = 0;
2859 chain_key = iterate_chain_key(chain_key, hlock->class_idx);
2860 prev_hlock = hlock;
2862 if (chain_key != curr->curr_chain_key) {
2863 debug_locks_off();
2865 * More smoking hash instead of calculating it, damn see these
2866 * numbers float.. I bet that a pink elephant stepped on my memory.
2868 WARN(1, "hm#2, depth: %u [%u], %016Lx != %016Lx\n",
2869 curr->lockdep_depth, i,
2870 (unsigned long long)chain_key,
2871 (unsigned long long)curr->curr_chain_key);
2873 #endif
2876 static int mark_lock(struct task_struct *curr, struct held_lock *this,
2877 enum lock_usage_bit new_bit);
2879 #if defined(CONFIG_TRACE_IRQFLAGS) && defined(CONFIG_PROVE_LOCKING)
2882 static void
2883 print_usage_bug_scenario(struct held_lock *lock)
2885 struct lock_class *class = hlock_class(lock);
2887 printk(" Possible unsafe locking scenario:\n\n");
2888 printk(" CPU0\n");
2889 printk(" ----\n");
2890 printk(" lock(");
2891 __print_lock_name(class);
2892 printk(KERN_CONT ");\n");
2893 printk(" <Interrupt>\n");
2894 printk(" lock(");
2895 __print_lock_name(class);
2896 printk(KERN_CONT ");\n");
2897 printk("\n *** DEADLOCK ***\n\n");
2900 static int
2901 print_usage_bug(struct task_struct *curr, struct held_lock *this,
2902 enum lock_usage_bit prev_bit, enum lock_usage_bit new_bit)
2904 if (!debug_locks_off_graph_unlock() || debug_locks_silent)
2905 return 0;
2907 pr_warn("\n");
2908 pr_warn("================================\n");
2909 pr_warn("WARNING: inconsistent lock state\n");
2910 print_kernel_ident();
2911 pr_warn("--------------------------------\n");
2913 pr_warn("inconsistent {%s} -> {%s} usage.\n",
2914 usage_str[prev_bit], usage_str[new_bit]);
2916 pr_warn("%s/%d [HC%u[%lu]:SC%u[%lu]:HE%u:SE%u] takes:\n",
2917 curr->comm, task_pid_nr(curr),
2918 trace_hardirq_context(curr), hardirq_count() >> HARDIRQ_SHIFT,
2919 trace_softirq_context(curr), softirq_count() >> SOFTIRQ_SHIFT,
2920 trace_hardirqs_enabled(curr),
2921 trace_softirqs_enabled(curr));
2922 print_lock(this);
2924 pr_warn("{%s} state was registered at:\n", usage_str[prev_bit]);
2925 print_lock_trace(hlock_class(this)->usage_traces + prev_bit, 1);
2927 print_irqtrace_events(curr);
2928 pr_warn("\nother info that might help us debug this:\n");
2929 print_usage_bug_scenario(this);
2931 lockdep_print_held_locks(curr);
2933 pr_warn("\nstack backtrace:\n");
2934 dump_stack();
2936 return 0;
2940 * Print out an error if an invalid bit is set:
2942 static inline int
2943 valid_state(struct task_struct *curr, struct held_lock *this,
2944 enum lock_usage_bit new_bit, enum lock_usage_bit bad_bit)
2946 if (unlikely(hlock_class(this)->usage_mask & (1 << bad_bit)))
2947 return print_usage_bug(curr, this, bad_bit, new_bit);
2948 return 1;
2953 * print irq inversion bug:
2955 static int
2956 print_irq_inversion_bug(struct task_struct *curr,
2957 struct lock_list *root, struct lock_list *other,
2958 struct held_lock *this, int forwards,
2959 const char *irqclass)
2961 struct lock_list *entry = other;
2962 struct lock_list *middle = NULL;
2963 int depth;
2965 if (!debug_locks_off_graph_unlock() || debug_locks_silent)
2966 return 0;
2968 pr_warn("\n");
2969 pr_warn("========================================================\n");
2970 pr_warn("WARNING: possible irq lock inversion dependency detected\n");
2971 print_kernel_ident();
2972 pr_warn("--------------------------------------------------------\n");
2973 pr_warn("%s/%d just changed the state of lock:\n",
2974 curr->comm, task_pid_nr(curr));
2975 print_lock(this);
2976 if (forwards)
2977 pr_warn("but this lock took another, %s-unsafe lock in the past:\n", irqclass);
2978 else
2979 pr_warn("but this lock was taken by another, %s-safe lock in the past:\n", irqclass);
2980 print_lock_name(other->class);
2981 pr_warn("\n\nand interrupts could create inverse lock ordering between them.\n\n");
2983 pr_warn("\nother info that might help us debug this:\n");
2985 /* Find a middle lock (if one exists) */
2986 depth = get_lock_depth(other);
2987 do {
2988 if (depth == 0 && (entry != root)) {
2989 pr_warn("lockdep:%s bad path found in chain graph\n", __func__);
2990 break;
2992 middle = entry;
2993 entry = get_lock_parent(entry);
2994 depth--;
2995 } while (entry && entry != root && (depth >= 0));
2996 if (forwards)
2997 print_irq_lock_scenario(root, other,
2998 middle ? middle->class : root->class, other->class);
2999 else
3000 print_irq_lock_scenario(other, root,
3001 middle ? middle->class : other->class, root->class);
3003 lockdep_print_held_locks(curr);
3005 pr_warn("\nthe shortest dependencies between 2nd lock and 1st lock:\n");
3006 if (!save_trace(&root->trace))
3007 return 0;
3008 print_shortest_lock_dependencies(other, root);
3010 pr_warn("\nstack backtrace:\n");
3011 dump_stack();
3013 return 0;
3017 * Prove that in the forwards-direction subgraph starting at <this>
3018 * there is no lock matching <mask>:
3020 static int
3021 check_usage_forwards(struct task_struct *curr, struct held_lock *this,
3022 enum lock_usage_bit bit, const char *irqclass)
3024 int ret;
3025 struct lock_list root;
3026 struct lock_list *uninitialized_var(target_entry);
3028 root.parent = NULL;
3029 root.class = hlock_class(this);
3030 ret = find_usage_forwards(&root, lock_flag(bit), &target_entry);
3031 if (ret < 0)
3032 return print_bfs_bug(ret);
3033 if (ret == 1)
3034 return ret;
3036 return print_irq_inversion_bug(curr, &root, target_entry,
3037 this, 1, irqclass);
3041 * Prove that in the backwards-direction subgraph starting at <this>
3042 * there is no lock matching <mask>:
3044 static int
3045 check_usage_backwards(struct task_struct *curr, struct held_lock *this,
3046 enum lock_usage_bit bit, const char *irqclass)
3048 int ret;
3049 struct lock_list root;
3050 struct lock_list *uninitialized_var(target_entry);
3052 root.parent = NULL;
3053 root.class = hlock_class(this);
3054 ret = find_usage_backwards(&root, lock_flag(bit), &target_entry);
3055 if (ret < 0)
3056 return print_bfs_bug(ret);
3057 if (ret == 1)
3058 return ret;
3060 return print_irq_inversion_bug(curr, &root, target_entry,
3061 this, 0, irqclass);
3064 void print_irqtrace_events(struct task_struct *curr)
3066 printk("irq event stamp: %u\n", curr->irq_events);
3067 printk("hardirqs last enabled at (%u): [<%px>] %pS\n",
3068 curr->hardirq_enable_event, (void *)curr->hardirq_enable_ip,
3069 (void *)curr->hardirq_enable_ip);
3070 printk("hardirqs last disabled at (%u): [<%px>] %pS\n",
3071 curr->hardirq_disable_event, (void *)curr->hardirq_disable_ip,
3072 (void *)curr->hardirq_disable_ip);
3073 printk("softirqs last enabled at (%u): [<%px>] %pS\n",
3074 curr->softirq_enable_event, (void *)curr->softirq_enable_ip,
3075 (void *)curr->softirq_enable_ip);
3076 printk("softirqs last disabled at (%u): [<%px>] %pS\n",
3077 curr->softirq_disable_event, (void *)curr->softirq_disable_ip,
3078 (void *)curr->softirq_disable_ip);
3081 static int HARDIRQ_verbose(struct lock_class *class)
3083 #if HARDIRQ_VERBOSE
3084 return class_filter(class);
3085 #endif
3086 return 0;
3089 static int SOFTIRQ_verbose(struct lock_class *class)
3091 #if SOFTIRQ_VERBOSE
3092 return class_filter(class);
3093 #endif
3094 return 0;
3097 #define STRICT_READ_CHECKS 1
3099 static int (*state_verbose_f[])(struct lock_class *class) = {
3100 #define LOCKDEP_STATE(__STATE) \
3101 __STATE##_verbose,
3102 #include "lockdep_states.h"
3103 #undef LOCKDEP_STATE
3106 static inline int state_verbose(enum lock_usage_bit bit,
3107 struct lock_class *class)
3109 return state_verbose_f[bit >> LOCK_USAGE_DIR_MASK](class);
3112 typedef int (*check_usage_f)(struct task_struct *, struct held_lock *,
3113 enum lock_usage_bit bit, const char *name);
3115 static int
3116 mark_lock_irq(struct task_struct *curr, struct held_lock *this,
3117 enum lock_usage_bit new_bit)
3119 int excl_bit = exclusive_bit(new_bit);
3120 int read = new_bit & LOCK_USAGE_READ_MASK;
3121 int dir = new_bit & LOCK_USAGE_DIR_MASK;
3124 * mark USED_IN has to look forwards -- to ensure no dependency
3125 * has ENABLED state, which would allow recursion deadlocks.
3127 * mark ENABLED has to look backwards -- to ensure no dependee
3128 * has USED_IN state, which, again, would allow recursion deadlocks.
3130 check_usage_f usage = dir ?
3131 check_usage_backwards : check_usage_forwards;
3134 * Validate that this particular lock does not have conflicting
3135 * usage states.
3137 if (!valid_state(curr, this, new_bit, excl_bit))
3138 return 0;
3141 * Validate that the lock dependencies don't have conflicting usage
3142 * states.
3144 if ((!read || !dir || STRICT_READ_CHECKS) &&
3145 !usage(curr, this, excl_bit, state_name(new_bit & ~LOCK_USAGE_READ_MASK)))
3146 return 0;
3149 * Check for read in write conflicts
3151 if (!read) {
3152 if (!valid_state(curr, this, new_bit, excl_bit + LOCK_USAGE_READ_MASK))
3153 return 0;
3155 if (STRICT_READ_CHECKS &&
3156 !usage(curr, this, excl_bit + LOCK_USAGE_READ_MASK,
3157 state_name(new_bit + LOCK_USAGE_READ_MASK)))
3158 return 0;
3161 if (state_verbose(new_bit, hlock_class(this)))
3162 return 2;
3164 return 1;
3168 * Mark all held locks with a usage bit:
3170 static int
3171 mark_held_locks(struct task_struct *curr, enum lock_usage_bit base_bit)
3173 struct held_lock *hlock;
3174 int i;
3176 for (i = 0; i < curr->lockdep_depth; i++) {
3177 enum lock_usage_bit hlock_bit = base_bit;
3178 hlock = curr->held_locks + i;
3180 if (hlock->read)
3181 hlock_bit += LOCK_USAGE_READ_MASK;
3183 BUG_ON(hlock_bit >= LOCK_USAGE_STATES);
3185 if (!hlock->check)
3186 continue;
3188 if (!mark_lock(curr, hlock, hlock_bit))
3189 return 0;
3192 return 1;
3196 * Hardirqs will be enabled:
3198 static void __trace_hardirqs_on_caller(unsigned long ip)
3200 struct task_struct *curr = current;
3202 /* we'll do an OFF -> ON transition: */
3203 curr->hardirqs_enabled = 1;
3206 * We are going to turn hardirqs on, so set the
3207 * usage bit for all held locks:
3209 if (!mark_held_locks(curr, LOCK_ENABLED_HARDIRQ))
3210 return;
3212 * If we have softirqs enabled, then set the usage
3213 * bit for all held locks. (disabled hardirqs prevented
3214 * this bit from being set before)
3216 if (curr->softirqs_enabled)
3217 if (!mark_held_locks(curr, LOCK_ENABLED_SOFTIRQ))
3218 return;
3220 curr->hardirq_enable_ip = ip;
3221 curr->hardirq_enable_event = ++curr->irq_events;
3222 debug_atomic_inc(hardirqs_on_events);
3225 void lockdep_hardirqs_on(unsigned long ip)
3227 if (unlikely(!debug_locks || current->lockdep_recursion))
3228 return;
3230 if (unlikely(current->hardirqs_enabled)) {
3232 * Neither irq nor preemption are disabled here
3233 * so this is racy by nature but losing one hit
3234 * in a stat is not a big deal.
3236 __debug_atomic_inc(redundant_hardirqs_on);
3237 return;
3241 * We're enabling irqs and according to our state above irqs weren't
3242 * already enabled, yet we find the hardware thinks they are in fact
3243 * enabled.. someone messed up their IRQ state tracing.
3245 if (DEBUG_LOCKS_WARN_ON(!irqs_disabled()))
3246 return;
3249 * See the fine text that goes along with this variable definition.
3251 if (DEBUG_LOCKS_WARN_ON(early_boot_irqs_disabled))
3252 return;
3255 * Can't allow enabling interrupts while in an interrupt handler,
3256 * that's general bad form and such. Recursion, limited stack etc..
3258 if (DEBUG_LOCKS_WARN_ON(current->hardirq_context))
3259 return;
3261 current->lockdep_recursion = 1;
3262 __trace_hardirqs_on_caller(ip);
3263 current->lockdep_recursion = 0;
3265 NOKPROBE_SYMBOL(lockdep_hardirqs_on);
3268 * Hardirqs were disabled:
3270 void lockdep_hardirqs_off(unsigned long ip)
3272 struct task_struct *curr = current;
3274 if (unlikely(!debug_locks || current->lockdep_recursion))
3275 return;
3278 * So we're supposed to get called after you mask local IRQs, but for
3279 * some reason the hardware doesn't quite think you did a proper job.
3281 if (DEBUG_LOCKS_WARN_ON(!irqs_disabled()))
3282 return;
3284 if (curr->hardirqs_enabled) {
3286 * We have done an ON -> OFF transition:
3288 curr->hardirqs_enabled = 0;
3289 curr->hardirq_disable_ip = ip;
3290 curr->hardirq_disable_event = ++curr->irq_events;
3291 debug_atomic_inc(hardirqs_off_events);
3292 } else
3293 debug_atomic_inc(redundant_hardirqs_off);
3295 NOKPROBE_SYMBOL(lockdep_hardirqs_off);
3298 * Softirqs will be enabled:
3300 void trace_softirqs_on(unsigned long ip)
3302 struct task_struct *curr = current;
3304 if (unlikely(!debug_locks || current->lockdep_recursion))
3305 return;
3308 * We fancy IRQs being disabled here, see softirq.c, avoids
3309 * funny state and nesting things.
3311 if (DEBUG_LOCKS_WARN_ON(!irqs_disabled()))
3312 return;
3314 if (curr->softirqs_enabled) {
3315 debug_atomic_inc(redundant_softirqs_on);
3316 return;
3319 current->lockdep_recursion = 1;
3321 * We'll do an OFF -> ON transition:
3323 curr->softirqs_enabled = 1;
3324 curr->softirq_enable_ip = ip;
3325 curr->softirq_enable_event = ++curr->irq_events;
3326 debug_atomic_inc(softirqs_on_events);
3328 * We are going to turn softirqs on, so set the
3329 * usage bit for all held locks, if hardirqs are
3330 * enabled too:
3332 if (curr->hardirqs_enabled)
3333 mark_held_locks(curr, LOCK_ENABLED_SOFTIRQ);
3334 current->lockdep_recursion = 0;
3338 * Softirqs were disabled:
3340 void trace_softirqs_off(unsigned long ip)
3342 struct task_struct *curr = current;
3344 if (unlikely(!debug_locks || current->lockdep_recursion))
3345 return;
3348 * We fancy IRQs being disabled here, see softirq.c
3350 if (DEBUG_LOCKS_WARN_ON(!irqs_disabled()))
3351 return;
3353 if (curr->softirqs_enabled) {
3355 * We have done an ON -> OFF transition:
3357 curr->softirqs_enabled = 0;
3358 curr->softirq_disable_ip = ip;
3359 curr->softirq_disable_event = ++curr->irq_events;
3360 debug_atomic_inc(softirqs_off_events);
3362 * Whoops, we wanted softirqs off, so why aren't they?
3364 DEBUG_LOCKS_WARN_ON(!softirq_count());
3365 } else
3366 debug_atomic_inc(redundant_softirqs_off);
3369 static int mark_irqflags(struct task_struct *curr, struct held_lock *hlock)
3372 * If non-trylock use in a hardirq or softirq context, then
3373 * mark the lock as used in these contexts:
3375 if (!hlock->trylock) {
3376 if (hlock->read) {
3377 if (curr->hardirq_context)
3378 if (!mark_lock(curr, hlock,
3379 LOCK_USED_IN_HARDIRQ_READ))
3380 return 0;
3381 if (curr->softirq_context)
3382 if (!mark_lock(curr, hlock,
3383 LOCK_USED_IN_SOFTIRQ_READ))
3384 return 0;
3385 } else {
3386 if (curr->hardirq_context)
3387 if (!mark_lock(curr, hlock, LOCK_USED_IN_HARDIRQ))
3388 return 0;
3389 if (curr->softirq_context)
3390 if (!mark_lock(curr, hlock, LOCK_USED_IN_SOFTIRQ))
3391 return 0;
3394 if (!hlock->hardirqs_off) {
3395 if (hlock->read) {
3396 if (!mark_lock(curr, hlock,
3397 LOCK_ENABLED_HARDIRQ_READ))
3398 return 0;
3399 if (curr->softirqs_enabled)
3400 if (!mark_lock(curr, hlock,
3401 LOCK_ENABLED_SOFTIRQ_READ))
3402 return 0;
3403 } else {
3404 if (!mark_lock(curr, hlock,
3405 LOCK_ENABLED_HARDIRQ))
3406 return 0;
3407 if (curr->softirqs_enabled)
3408 if (!mark_lock(curr, hlock,
3409 LOCK_ENABLED_SOFTIRQ))
3410 return 0;
3414 return 1;
3417 static inline unsigned int task_irq_context(struct task_struct *task)
3419 return 2 * !!task->hardirq_context + !!task->softirq_context;
3422 static int separate_irq_context(struct task_struct *curr,
3423 struct held_lock *hlock)
3425 unsigned int depth = curr->lockdep_depth;
3428 * Keep track of points where we cross into an interrupt context:
3430 if (depth) {
3431 struct held_lock *prev_hlock;
3433 prev_hlock = curr->held_locks + depth-1;
3435 * If we cross into another context, reset the
3436 * hash key (this also prevents the checking and the
3437 * adding of the dependency to 'prev'):
3439 if (prev_hlock->irq_context != hlock->irq_context)
3440 return 1;
3442 return 0;
3445 #else /* defined(CONFIG_TRACE_IRQFLAGS) && defined(CONFIG_PROVE_LOCKING) */
3447 static inline
3448 int mark_lock_irq(struct task_struct *curr, struct held_lock *this,
3449 enum lock_usage_bit new_bit)
3451 WARN_ON(1); /* Impossible innit? when we don't have TRACE_IRQFLAG */
3452 return 1;
3455 static inline int mark_irqflags(struct task_struct *curr,
3456 struct held_lock *hlock)
3458 return 1;
3461 static inline unsigned int task_irq_context(struct task_struct *task)
3463 return 0;
3466 static inline int separate_irq_context(struct task_struct *curr,
3467 struct held_lock *hlock)
3469 return 0;
3472 #endif /* defined(CONFIG_TRACE_IRQFLAGS) && defined(CONFIG_PROVE_LOCKING) */
3475 * Mark a lock with a usage bit, and validate the state transition:
3477 static int mark_lock(struct task_struct *curr, struct held_lock *this,
3478 enum lock_usage_bit new_bit)
3480 unsigned int new_mask = 1 << new_bit, ret = 1;
3483 * If already set then do not dirty the cacheline,
3484 * nor do any checks:
3486 if (likely(hlock_class(this)->usage_mask & new_mask))
3487 return 1;
3489 if (!graph_lock())
3490 return 0;
3492 * Make sure we didn't race:
3494 if (unlikely(hlock_class(this)->usage_mask & new_mask)) {
3495 graph_unlock();
3496 return 1;
3499 hlock_class(this)->usage_mask |= new_mask;
3501 if (!save_trace(hlock_class(this)->usage_traces + new_bit))
3502 return 0;
3504 switch (new_bit) {
3505 #define LOCKDEP_STATE(__STATE) \
3506 case LOCK_USED_IN_##__STATE: \
3507 case LOCK_USED_IN_##__STATE##_READ: \
3508 case LOCK_ENABLED_##__STATE: \
3509 case LOCK_ENABLED_##__STATE##_READ:
3510 #include "lockdep_states.h"
3511 #undef LOCKDEP_STATE
3512 ret = mark_lock_irq(curr, this, new_bit);
3513 if (!ret)
3514 return 0;
3515 break;
3516 case LOCK_USED:
3517 debug_atomic_dec(nr_unused_locks);
3518 break;
3519 default:
3520 if (!debug_locks_off_graph_unlock())
3521 return 0;
3522 WARN_ON(1);
3523 return 0;
3526 graph_unlock();
3529 * We must printk outside of the graph_lock:
3531 if (ret == 2) {
3532 printk("\nmarked lock as {%s}:\n", usage_str[new_bit]);
3533 print_lock(this);
3534 print_irqtrace_events(curr);
3535 dump_stack();
3538 return ret;
3542 * Initialize a lock instance's lock-class mapping info:
3544 void lockdep_init_map(struct lockdep_map *lock, const char *name,
3545 struct lock_class_key *key, int subclass)
3547 int i;
3549 for (i = 0; i < NR_LOCKDEP_CACHING_CLASSES; i++)
3550 lock->class_cache[i] = NULL;
3552 #ifdef CONFIG_LOCK_STAT
3553 lock->cpu = raw_smp_processor_id();
3554 #endif
3557 * Can't be having no nameless bastards around this place!
3559 if (DEBUG_LOCKS_WARN_ON(!name)) {
3560 lock->name = "NULL";
3561 return;
3564 lock->name = name;
3567 * No key, no joy, we need to hash something.
3569 if (DEBUG_LOCKS_WARN_ON(!key))
3570 return;
3572 * Sanity check, the lock-class key must either have been allocated
3573 * statically or must have been registered as a dynamic key.
3575 if (!static_obj(key) && !is_dynamic_key(key)) {
3576 if (debug_locks)
3577 printk(KERN_ERR "BUG: key %px has not been registered!\n", key);
3578 DEBUG_LOCKS_WARN_ON(1);
3579 return;
3581 lock->key = key;
3583 if (unlikely(!debug_locks))
3584 return;
3586 if (subclass) {
3587 unsigned long flags;
3589 if (DEBUG_LOCKS_WARN_ON(current->lockdep_recursion))
3590 return;
3592 raw_local_irq_save(flags);
3593 current->lockdep_recursion = 1;
3594 register_lock_class(lock, subclass, 1);
3595 current->lockdep_recursion = 0;
3596 raw_local_irq_restore(flags);
3599 EXPORT_SYMBOL_GPL(lockdep_init_map);
3601 struct lock_class_key __lockdep_no_validate__;
3602 EXPORT_SYMBOL_GPL(__lockdep_no_validate__);
3604 static int
3605 print_lock_nested_lock_not_held(struct task_struct *curr,
3606 struct held_lock *hlock,
3607 unsigned long ip)
3609 if (!debug_locks_off())
3610 return 0;
3611 if (debug_locks_silent)
3612 return 0;
3614 pr_warn("\n");
3615 pr_warn("==================================\n");
3616 pr_warn("WARNING: Nested lock was not taken\n");
3617 print_kernel_ident();
3618 pr_warn("----------------------------------\n");
3620 pr_warn("%s/%d is trying to lock:\n", curr->comm, task_pid_nr(curr));
3621 print_lock(hlock);
3623 pr_warn("\nbut this task is not holding:\n");
3624 pr_warn("%s\n", hlock->nest_lock->name);
3626 pr_warn("\nstack backtrace:\n");
3627 dump_stack();
3629 pr_warn("\nother info that might help us debug this:\n");
3630 lockdep_print_held_locks(curr);
3632 pr_warn("\nstack backtrace:\n");
3633 dump_stack();
3635 return 0;
3638 static int __lock_is_held(const struct lockdep_map *lock, int read);
3641 * This gets called for every mutex_lock*()/spin_lock*() operation.
3642 * We maintain the dependency maps and validate the locking attempt:
3644 * The callers must make sure that IRQs are disabled before calling it,
3645 * otherwise we could get an interrupt which would want to take locks,
3646 * which would end up in lockdep again.
3648 static int __lock_acquire(struct lockdep_map *lock, unsigned int subclass,
3649 int trylock, int read, int check, int hardirqs_off,
3650 struct lockdep_map *nest_lock, unsigned long ip,
3651 int references, int pin_count)
3653 struct task_struct *curr = current;
3654 struct lock_class *class = NULL;
3655 struct held_lock *hlock;
3656 unsigned int depth;
3657 int chain_head = 0;
3658 int class_idx;
3659 u64 chain_key;
3661 if (unlikely(!debug_locks))
3662 return 0;
3664 if (!prove_locking || lock->key == &__lockdep_no_validate__)
3665 check = 0;
3667 if (subclass < NR_LOCKDEP_CACHING_CLASSES)
3668 class = lock->class_cache[subclass];
3670 * Not cached?
3672 if (unlikely(!class)) {
3673 class = register_lock_class(lock, subclass, 0);
3674 if (!class)
3675 return 0;
3678 debug_class_ops_inc(class);
3680 if (very_verbose(class)) {
3681 printk("\nacquire class [%px] %s", class->key, class->name);
3682 if (class->name_version > 1)
3683 printk(KERN_CONT "#%d", class->name_version);
3684 printk(KERN_CONT "\n");
3685 dump_stack();
3689 * Add the lock to the list of currently held locks.
3690 * (we dont increase the depth just yet, up until the
3691 * dependency checks are done)
3693 depth = curr->lockdep_depth;
3695 * Ran out of static storage for our per-task lock stack again have we?
3697 if (DEBUG_LOCKS_WARN_ON(depth >= MAX_LOCK_DEPTH))
3698 return 0;
3700 class_idx = class - lock_classes + 1;
3702 if (depth) {
3703 hlock = curr->held_locks + depth - 1;
3704 if (hlock->class_idx == class_idx && nest_lock) {
3705 if (hlock->references) {
3707 * Check: unsigned int references:12, overflow.
3709 if (DEBUG_LOCKS_WARN_ON(hlock->references == (1 << 12)-1))
3710 return 0;
3712 hlock->references++;
3713 } else {
3714 hlock->references = 2;
3717 return 1;
3721 hlock = curr->held_locks + depth;
3723 * Plain impossible, we just registered it and checked it weren't no
3724 * NULL like.. I bet this mushroom I ate was good!
3726 if (DEBUG_LOCKS_WARN_ON(!class))
3727 return 0;
3728 hlock->class_idx = class_idx;
3729 hlock->acquire_ip = ip;
3730 hlock->instance = lock;
3731 hlock->nest_lock = nest_lock;
3732 hlock->irq_context = task_irq_context(curr);
3733 hlock->trylock = trylock;
3734 hlock->read = read;
3735 hlock->check = check;
3736 hlock->hardirqs_off = !!hardirqs_off;
3737 hlock->references = references;
3738 #ifdef CONFIG_LOCK_STAT
3739 hlock->waittime_stamp = 0;
3740 hlock->holdtime_stamp = lockstat_clock();
3741 #endif
3742 hlock->pin_count = pin_count;
3744 if (check && !mark_irqflags(curr, hlock))
3745 return 0;
3747 /* mark it as used: */
3748 if (!mark_lock(curr, hlock, LOCK_USED))
3749 return 0;
3752 * Calculate the chain hash: it's the combined hash of all the
3753 * lock keys along the dependency chain. We save the hash value
3754 * at every step so that we can get the current hash easily
3755 * after unlock. The chain hash is then used to cache dependency
3756 * results.
3758 * The 'key ID' is what is the most compact key value to drive
3759 * the hash, not class->key.
3762 * Whoops, we did it again.. ran straight out of our static allocation.
3764 if (DEBUG_LOCKS_WARN_ON(class_idx > MAX_LOCKDEP_KEYS))
3765 return 0;
3767 chain_key = curr->curr_chain_key;
3768 if (!depth) {
3770 * How can we have a chain hash when we ain't got no keys?!
3772 if (DEBUG_LOCKS_WARN_ON(chain_key != 0))
3773 return 0;
3774 chain_head = 1;
3777 hlock->prev_chain_key = chain_key;
3778 if (separate_irq_context(curr, hlock)) {
3779 chain_key = 0;
3780 chain_head = 1;
3782 chain_key = iterate_chain_key(chain_key, class_idx);
3784 if (nest_lock && !__lock_is_held(nest_lock, -1))
3785 return print_lock_nested_lock_not_held(curr, hlock, ip);
3787 if (!debug_locks_silent) {
3788 WARN_ON_ONCE(depth && !hlock_class(hlock - 1)->key);
3789 WARN_ON_ONCE(!hlock_class(hlock)->key);
3792 if (!validate_chain(curr, lock, hlock, chain_head, chain_key))
3793 return 0;
3795 curr->curr_chain_key = chain_key;
3796 curr->lockdep_depth++;
3797 check_chain_key(curr);
3798 #ifdef CONFIG_DEBUG_LOCKDEP
3799 if (unlikely(!debug_locks))
3800 return 0;
3801 #endif
3802 if (unlikely(curr->lockdep_depth >= MAX_LOCK_DEPTH)) {
3803 debug_locks_off();
3804 print_lockdep_off("BUG: MAX_LOCK_DEPTH too low!");
3805 printk(KERN_DEBUG "depth: %i max: %lu!\n",
3806 curr->lockdep_depth, MAX_LOCK_DEPTH);
3808 lockdep_print_held_locks(current);
3809 debug_show_all_locks();
3810 dump_stack();
3812 return 0;
3815 if (unlikely(curr->lockdep_depth > max_lockdep_depth))
3816 max_lockdep_depth = curr->lockdep_depth;
3818 return 1;
3821 static int
3822 print_unlock_imbalance_bug(struct task_struct *curr, struct lockdep_map *lock,
3823 unsigned long ip)
3825 if (!debug_locks_off())
3826 return 0;
3827 if (debug_locks_silent)
3828 return 0;
3830 pr_warn("\n");
3831 pr_warn("=====================================\n");
3832 pr_warn("WARNING: bad unlock balance detected!\n");
3833 print_kernel_ident();
3834 pr_warn("-------------------------------------\n");
3835 pr_warn("%s/%d is trying to release lock (",
3836 curr->comm, task_pid_nr(curr));
3837 print_lockdep_cache(lock);
3838 pr_cont(") at:\n");
3839 print_ip_sym(ip);
3840 pr_warn("but there are no more locks to release!\n");
3841 pr_warn("\nother info that might help us debug this:\n");
3842 lockdep_print_held_locks(curr);
3844 pr_warn("\nstack backtrace:\n");
3845 dump_stack();
3847 return 0;
3850 static int match_held_lock(const struct held_lock *hlock,
3851 const struct lockdep_map *lock)
3853 if (hlock->instance == lock)
3854 return 1;
3856 if (hlock->references) {
3857 const struct lock_class *class = lock->class_cache[0];
3859 if (!class)
3860 class = look_up_lock_class(lock, 0);
3863 * If look_up_lock_class() failed to find a class, we're trying
3864 * to test if we hold a lock that has never yet been acquired.
3865 * Clearly if the lock hasn't been acquired _ever_, we're not
3866 * holding it either, so report failure.
3868 if (!class)
3869 return 0;
3872 * References, but not a lock we're actually ref-counting?
3873 * State got messed up, follow the sites that change ->references
3874 * and try to make sense of it.
3876 if (DEBUG_LOCKS_WARN_ON(!hlock->nest_lock))
3877 return 0;
3879 if (hlock->class_idx == class - lock_classes + 1)
3880 return 1;
3883 return 0;
3886 /* @depth must not be zero */
3887 static struct held_lock *find_held_lock(struct task_struct *curr,
3888 struct lockdep_map *lock,
3889 unsigned int depth, int *idx)
3891 struct held_lock *ret, *hlock, *prev_hlock;
3892 int i;
3894 i = depth - 1;
3895 hlock = curr->held_locks + i;
3896 ret = hlock;
3897 if (match_held_lock(hlock, lock))
3898 goto out;
3900 ret = NULL;
3901 for (i--, prev_hlock = hlock--;
3902 i >= 0;
3903 i--, prev_hlock = hlock--) {
3905 * We must not cross into another context:
3907 if (prev_hlock->irq_context != hlock->irq_context) {
3908 ret = NULL;
3909 break;
3911 if (match_held_lock(hlock, lock)) {
3912 ret = hlock;
3913 break;
3917 out:
3918 *idx = i;
3919 return ret;
3922 static int reacquire_held_locks(struct task_struct *curr, unsigned int depth,
3923 int idx)
3925 struct held_lock *hlock;
3927 if (DEBUG_LOCKS_WARN_ON(!irqs_disabled()))
3928 return 0;
3930 for (hlock = curr->held_locks + idx; idx < depth; idx++, hlock++) {
3931 if (!__lock_acquire(hlock->instance,
3932 hlock_class(hlock)->subclass,
3933 hlock->trylock,
3934 hlock->read, hlock->check,
3935 hlock->hardirqs_off,
3936 hlock->nest_lock, hlock->acquire_ip,
3937 hlock->references, hlock->pin_count))
3938 return 1;
3940 return 0;
3943 static int
3944 __lock_set_class(struct lockdep_map *lock, const char *name,
3945 struct lock_class_key *key, unsigned int subclass,
3946 unsigned long ip)
3948 struct task_struct *curr = current;
3949 struct held_lock *hlock;
3950 struct lock_class *class;
3951 unsigned int depth;
3952 int i;
3954 if (unlikely(!debug_locks))
3955 return 0;
3957 depth = curr->lockdep_depth;
3959 * This function is about (re)setting the class of a held lock,
3960 * yet we're not actually holding any locks. Naughty user!
3962 if (DEBUG_LOCKS_WARN_ON(!depth))
3963 return 0;
3965 hlock = find_held_lock(curr, lock, depth, &i);
3966 if (!hlock)
3967 return print_unlock_imbalance_bug(curr, lock, ip);
3969 lockdep_init_map(lock, name, key, 0);
3970 class = register_lock_class(lock, subclass, 0);
3971 hlock->class_idx = class - lock_classes + 1;
3973 curr->lockdep_depth = i;
3974 curr->curr_chain_key = hlock->prev_chain_key;
3976 if (reacquire_held_locks(curr, depth, i))
3977 return 0;
3980 * I took it apart and put it back together again, except now I have
3981 * these 'spare' parts.. where shall I put them.
3983 if (DEBUG_LOCKS_WARN_ON(curr->lockdep_depth != depth))
3984 return 0;
3985 return 1;
3988 static int __lock_downgrade(struct lockdep_map *lock, unsigned long ip)
3990 struct task_struct *curr = current;
3991 struct held_lock *hlock;
3992 unsigned int depth;
3993 int i;
3995 if (unlikely(!debug_locks))
3996 return 0;
3998 depth = curr->lockdep_depth;
4000 * This function is about (re)setting the class of a held lock,
4001 * yet we're not actually holding any locks. Naughty user!
4003 if (DEBUG_LOCKS_WARN_ON(!depth))
4004 return 0;
4006 hlock = find_held_lock(curr, lock, depth, &i);
4007 if (!hlock)
4008 return print_unlock_imbalance_bug(curr, lock, ip);
4010 curr->lockdep_depth = i;
4011 curr->curr_chain_key = hlock->prev_chain_key;
4013 WARN(hlock->read, "downgrading a read lock");
4014 hlock->read = 1;
4015 hlock->acquire_ip = ip;
4017 if (reacquire_held_locks(curr, depth, i))
4018 return 0;
4021 * I took it apart and put it back together again, except now I have
4022 * these 'spare' parts.. where shall I put them.
4024 if (DEBUG_LOCKS_WARN_ON(curr->lockdep_depth != depth))
4025 return 0;
4026 return 1;
4030 * Remove the lock to the list of currently held locks - this gets
4031 * called on mutex_unlock()/spin_unlock*() (or on a failed
4032 * mutex_lock_interruptible()).
4034 * @nested is an hysterical artifact, needs a tree wide cleanup.
4036 static int
4037 __lock_release(struct lockdep_map *lock, int nested, unsigned long ip)
4039 struct task_struct *curr = current;
4040 struct held_lock *hlock;
4041 unsigned int depth;
4042 int i;
4044 if (unlikely(!debug_locks))
4045 return 0;
4047 depth = curr->lockdep_depth;
4049 * So we're all set to release this lock.. wait what lock? We don't
4050 * own any locks, you've been drinking again?
4052 if (DEBUG_LOCKS_WARN_ON(depth <= 0))
4053 return print_unlock_imbalance_bug(curr, lock, ip);
4056 * Check whether the lock exists in the current stack
4057 * of held locks:
4059 hlock = find_held_lock(curr, lock, depth, &i);
4060 if (!hlock)
4061 return print_unlock_imbalance_bug(curr, lock, ip);
4063 if (hlock->instance == lock)
4064 lock_release_holdtime(hlock);
4066 WARN(hlock->pin_count, "releasing a pinned lock\n");
4068 if (hlock->references) {
4069 hlock->references--;
4070 if (hlock->references) {
4072 * We had, and after removing one, still have
4073 * references, the current lock stack is still
4074 * valid. We're done!
4076 return 1;
4081 * We have the right lock to unlock, 'hlock' points to it.
4082 * Now we remove it from the stack, and add back the other
4083 * entries (if any), recalculating the hash along the way:
4086 curr->lockdep_depth = i;
4087 curr->curr_chain_key = hlock->prev_chain_key;
4090 * The most likely case is when the unlock is on the innermost
4091 * lock. In this case, we are done!
4093 if (i == depth-1)
4094 return 1;
4096 if (reacquire_held_locks(curr, depth, i + 1))
4097 return 0;
4100 * We had N bottles of beer on the wall, we drank one, but now
4101 * there's not N-1 bottles of beer left on the wall...
4103 DEBUG_LOCKS_WARN_ON(curr->lockdep_depth != depth-1);
4106 * Since reacquire_held_locks() would have called check_chain_key()
4107 * indirectly via __lock_acquire(), we don't need to do it again
4108 * on return.
4110 return 0;
4113 static nokprobe_inline
4114 int __lock_is_held(const struct lockdep_map *lock, int read)
4116 struct task_struct *curr = current;
4117 int i;
4119 for (i = 0; i < curr->lockdep_depth; i++) {
4120 struct held_lock *hlock = curr->held_locks + i;
4122 if (match_held_lock(hlock, lock)) {
4123 if (read == -1 || hlock->read == read)
4124 return 1;
4126 return 0;
4130 return 0;
4133 static struct pin_cookie __lock_pin_lock(struct lockdep_map *lock)
4135 struct pin_cookie cookie = NIL_COOKIE;
4136 struct task_struct *curr = current;
4137 int i;
4139 if (unlikely(!debug_locks))
4140 return cookie;
4142 for (i = 0; i < curr->lockdep_depth; i++) {
4143 struct held_lock *hlock = curr->held_locks + i;
4145 if (match_held_lock(hlock, lock)) {
4147 * Grab 16bits of randomness; this is sufficient to not
4148 * be guessable and still allows some pin nesting in
4149 * our u32 pin_count.
4151 cookie.val = 1 + (prandom_u32() >> 16);
4152 hlock->pin_count += cookie.val;
4153 return cookie;
4157 WARN(1, "pinning an unheld lock\n");
4158 return cookie;
4161 static void __lock_repin_lock(struct lockdep_map *lock, struct pin_cookie cookie)
4163 struct task_struct *curr = current;
4164 int i;
4166 if (unlikely(!debug_locks))
4167 return;
4169 for (i = 0; i < curr->lockdep_depth; i++) {
4170 struct held_lock *hlock = curr->held_locks + i;
4172 if (match_held_lock(hlock, lock)) {
4173 hlock->pin_count += cookie.val;
4174 return;
4178 WARN(1, "pinning an unheld lock\n");
4181 static void __lock_unpin_lock(struct lockdep_map *lock, struct pin_cookie cookie)
4183 struct task_struct *curr = current;
4184 int i;
4186 if (unlikely(!debug_locks))
4187 return;
4189 for (i = 0; i < curr->lockdep_depth; i++) {
4190 struct held_lock *hlock = curr->held_locks + i;
4192 if (match_held_lock(hlock, lock)) {
4193 if (WARN(!hlock->pin_count, "unpinning an unpinned lock\n"))
4194 return;
4196 hlock->pin_count -= cookie.val;
4198 if (WARN((int)hlock->pin_count < 0, "pin count corrupted\n"))
4199 hlock->pin_count = 0;
4201 return;
4205 WARN(1, "unpinning an unheld lock\n");
4209 * Check whether we follow the irq-flags state precisely:
4211 static void check_flags(unsigned long flags)
4213 #if defined(CONFIG_PROVE_LOCKING) && defined(CONFIG_DEBUG_LOCKDEP) && \
4214 defined(CONFIG_TRACE_IRQFLAGS)
4215 if (!debug_locks)
4216 return;
4218 if (irqs_disabled_flags(flags)) {
4219 if (DEBUG_LOCKS_WARN_ON(current->hardirqs_enabled)) {
4220 printk("possible reason: unannotated irqs-off.\n");
4222 } else {
4223 if (DEBUG_LOCKS_WARN_ON(!current->hardirqs_enabled)) {
4224 printk("possible reason: unannotated irqs-on.\n");
4229 * We dont accurately track softirq state in e.g.
4230 * hardirq contexts (such as on 4KSTACKS), so only
4231 * check if not in hardirq contexts:
4233 if (!hardirq_count()) {
4234 if (softirq_count()) {
4235 /* like the above, but with softirqs */
4236 DEBUG_LOCKS_WARN_ON(current->softirqs_enabled);
4237 } else {
4238 /* lick the above, does it taste good? */
4239 DEBUG_LOCKS_WARN_ON(!current->softirqs_enabled);
4243 if (!debug_locks)
4244 print_irqtrace_events(current);
4245 #endif
4248 void lock_set_class(struct lockdep_map *lock, const char *name,
4249 struct lock_class_key *key, unsigned int subclass,
4250 unsigned long ip)
4252 unsigned long flags;
4254 if (unlikely(current->lockdep_recursion))
4255 return;
4257 raw_local_irq_save(flags);
4258 current->lockdep_recursion = 1;
4259 check_flags(flags);
4260 if (__lock_set_class(lock, name, key, subclass, ip))
4261 check_chain_key(current);
4262 current->lockdep_recursion = 0;
4263 raw_local_irq_restore(flags);
4265 EXPORT_SYMBOL_GPL(lock_set_class);
4267 void lock_downgrade(struct lockdep_map *lock, unsigned long ip)
4269 unsigned long flags;
4271 if (unlikely(current->lockdep_recursion))
4272 return;
4274 raw_local_irq_save(flags);
4275 current->lockdep_recursion = 1;
4276 check_flags(flags);
4277 if (__lock_downgrade(lock, ip))
4278 check_chain_key(current);
4279 current->lockdep_recursion = 0;
4280 raw_local_irq_restore(flags);
4282 EXPORT_SYMBOL_GPL(lock_downgrade);
4285 * We are not always called with irqs disabled - do that here,
4286 * and also avoid lockdep recursion:
4288 void lock_acquire(struct lockdep_map *lock, unsigned int subclass,
4289 int trylock, int read, int check,
4290 struct lockdep_map *nest_lock, unsigned long ip)
4292 unsigned long flags;
4294 if (unlikely(current->lockdep_recursion))
4295 return;
4297 raw_local_irq_save(flags);
4298 check_flags(flags);
4300 current->lockdep_recursion = 1;
4301 trace_lock_acquire(lock, subclass, trylock, read, check, nest_lock, ip);
4302 __lock_acquire(lock, subclass, trylock, read, check,
4303 irqs_disabled_flags(flags), nest_lock, ip, 0, 0);
4304 current->lockdep_recursion = 0;
4305 raw_local_irq_restore(flags);
4307 EXPORT_SYMBOL_GPL(lock_acquire);
4309 void lock_release(struct lockdep_map *lock, int nested,
4310 unsigned long ip)
4312 unsigned long flags;
4314 if (unlikely(current->lockdep_recursion))
4315 return;
4317 raw_local_irq_save(flags);
4318 check_flags(flags);
4319 current->lockdep_recursion = 1;
4320 trace_lock_release(lock, ip);
4321 if (__lock_release(lock, nested, ip))
4322 check_chain_key(current);
4323 current->lockdep_recursion = 0;
4324 raw_local_irq_restore(flags);
4326 EXPORT_SYMBOL_GPL(lock_release);
4328 int lock_is_held_type(const struct lockdep_map *lock, int read)
4330 unsigned long flags;
4331 int ret = 0;
4333 if (unlikely(current->lockdep_recursion))
4334 return 1; /* avoid false negative lockdep_assert_held() */
4336 raw_local_irq_save(flags);
4337 check_flags(flags);
4339 current->lockdep_recursion = 1;
4340 ret = __lock_is_held(lock, read);
4341 current->lockdep_recursion = 0;
4342 raw_local_irq_restore(flags);
4344 return ret;
4346 EXPORT_SYMBOL_GPL(lock_is_held_type);
4347 NOKPROBE_SYMBOL(lock_is_held_type);
4349 struct pin_cookie lock_pin_lock(struct lockdep_map *lock)
4351 struct pin_cookie cookie = NIL_COOKIE;
4352 unsigned long flags;
4354 if (unlikely(current->lockdep_recursion))
4355 return cookie;
4357 raw_local_irq_save(flags);
4358 check_flags(flags);
4360 current->lockdep_recursion = 1;
4361 cookie = __lock_pin_lock(lock);
4362 current->lockdep_recursion = 0;
4363 raw_local_irq_restore(flags);
4365 return cookie;
4367 EXPORT_SYMBOL_GPL(lock_pin_lock);
4369 void lock_repin_lock(struct lockdep_map *lock, struct pin_cookie cookie)
4371 unsigned long flags;
4373 if (unlikely(current->lockdep_recursion))
4374 return;
4376 raw_local_irq_save(flags);
4377 check_flags(flags);
4379 current->lockdep_recursion = 1;
4380 __lock_repin_lock(lock, cookie);
4381 current->lockdep_recursion = 0;
4382 raw_local_irq_restore(flags);
4384 EXPORT_SYMBOL_GPL(lock_repin_lock);
4386 void lock_unpin_lock(struct lockdep_map *lock, struct pin_cookie cookie)
4388 unsigned long flags;
4390 if (unlikely(current->lockdep_recursion))
4391 return;
4393 raw_local_irq_save(flags);
4394 check_flags(flags);
4396 current->lockdep_recursion = 1;
4397 __lock_unpin_lock(lock, cookie);
4398 current->lockdep_recursion = 0;
4399 raw_local_irq_restore(flags);
4401 EXPORT_SYMBOL_GPL(lock_unpin_lock);
4403 #ifdef CONFIG_LOCK_STAT
4404 static int
4405 print_lock_contention_bug(struct task_struct *curr, struct lockdep_map *lock,
4406 unsigned long ip)
4408 if (!debug_locks_off())
4409 return 0;
4410 if (debug_locks_silent)
4411 return 0;
4413 pr_warn("\n");
4414 pr_warn("=================================\n");
4415 pr_warn("WARNING: bad contention detected!\n");
4416 print_kernel_ident();
4417 pr_warn("---------------------------------\n");
4418 pr_warn("%s/%d is trying to contend lock (",
4419 curr->comm, task_pid_nr(curr));
4420 print_lockdep_cache(lock);
4421 pr_cont(") at:\n");
4422 print_ip_sym(ip);
4423 pr_warn("but there are no locks held!\n");
4424 pr_warn("\nother info that might help us debug this:\n");
4425 lockdep_print_held_locks(curr);
4427 pr_warn("\nstack backtrace:\n");
4428 dump_stack();
4430 return 0;
4433 static void
4434 __lock_contended(struct lockdep_map *lock, unsigned long ip)
4436 struct task_struct *curr = current;
4437 struct held_lock *hlock;
4438 struct lock_class_stats *stats;
4439 unsigned int depth;
4440 int i, contention_point, contending_point;
4442 depth = curr->lockdep_depth;
4444 * Whee, we contended on this lock, except it seems we're not
4445 * actually trying to acquire anything much at all..
4447 if (DEBUG_LOCKS_WARN_ON(!depth))
4448 return;
4450 hlock = find_held_lock(curr, lock, depth, &i);
4451 if (!hlock) {
4452 print_lock_contention_bug(curr, lock, ip);
4453 return;
4456 if (hlock->instance != lock)
4457 return;
4459 hlock->waittime_stamp = lockstat_clock();
4461 contention_point = lock_point(hlock_class(hlock)->contention_point, ip);
4462 contending_point = lock_point(hlock_class(hlock)->contending_point,
4463 lock->ip);
4465 stats = get_lock_stats(hlock_class(hlock));
4466 if (contention_point < LOCKSTAT_POINTS)
4467 stats->contention_point[contention_point]++;
4468 if (contending_point < LOCKSTAT_POINTS)
4469 stats->contending_point[contending_point]++;
4470 if (lock->cpu != smp_processor_id())
4471 stats->bounces[bounce_contended + !!hlock->read]++;
4474 static void
4475 __lock_acquired(struct lockdep_map *lock, unsigned long ip)
4477 struct task_struct *curr = current;
4478 struct held_lock *hlock;
4479 struct lock_class_stats *stats;
4480 unsigned int depth;
4481 u64 now, waittime = 0;
4482 int i, cpu;
4484 depth = curr->lockdep_depth;
4486 * Yay, we acquired ownership of this lock we didn't try to
4487 * acquire, how the heck did that happen?
4489 if (DEBUG_LOCKS_WARN_ON(!depth))
4490 return;
4492 hlock = find_held_lock(curr, lock, depth, &i);
4493 if (!hlock) {
4494 print_lock_contention_bug(curr, lock, _RET_IP_);
4495 return;
4498 if (hlock->instance != lock)
4499 return;
4501 cpu = smp_processor_id();
4502 if (hlock->waittime_stamp) {
4503 now = lockstat_clock();
4504 waittime = now - hlock->waittime_stamp;
4505 hlock->holdtime_stamp = now;
4508 trace_lock_acquired(lock, ip);
4510 stats = get_lock_stats(hlock_class(hlock));
4511 if (waittime) {
4512 if (hlock->read)
4513 lock_time_inc(&stats->read_waittime, waittime);
4514 else
4515 lock_time_inc(&stats->write_waittime, waittime);
4517 if (lock->cpu != cpu)
4518 stats->bounces[bounce_acquired + !!hlock->read]++;
4520 lock->cpu = cpu;
4521 lock->ip = ip;
4524 void lock_contended(struct lockdep_map *lock, unsigned long ip)
4526 unsigned long flags;
4528 if (unlikely(!lock_stat || !debug_locks))
4529 return;
4531 if (unlikely(current->lockdep_recursion))
4532 return;
4534 raw_local_irq_save(flags);
4535 check_flags(flags);
4536 current->lockdep_recursion = 1;
4537 trace_lock_contended(lock, ip);
4538 __lock_contended(lock, ip);
4539 current->lockdep_recursion = 0;
4540 raw_local_irq_restore(flags);
4542 EXPORT_SYMBOL_GPL(lock_contended);
4544 void lock_acquired(struct lockdep_map *lock, unsigned long ip)
4546 unsigned long flags;
4548 if (unlikely(!lock_stat || !debug_locks))
4549 return;
4551 if (unlikely(current->lockdep_recursion))
4552 return;
4554 raw_local_irq_save(flags);
4555 check_flags(flags);
4556 current->lockdep_recursion = 1;
4557 __lock_acquired(lock, ip);
4558 current->lockdep_recursion = 0;
4559 raw_local_irq_restore(flags);
4561 EXPORT_SYMBOL_GPL(lock_acquired);
4562 #endif
4565 * Used by the testsuite, sanitize the validator state
4566 * after a simulated failure:
4569 void lockdep_reset(void)
4571 unsigned long flags;
4572 int i;
4574 raw_local_irq_save(flags);
4575 current->curr_chain_key = 0;
4576 current->lockdep_depth = 0;
4577 current->lockdep_recursion = 0;
4578 memset(current->held_locks, 0, MAX_LOCK_DEPTH*sizeof(struct held_lock));
4579 nr_hardirq_chains = 0;
4580 nr_softirq_chains = 0;
4581 nr_process_chains = 0;
4582 debug_locks = 1;
4583 for (i = 0; i < CHAINHASH_SIZE; i++)
4584 INIT_HLIST_HEAD(chainhash_table + i);
4585 raw_local_irq_restore(flags);
4588 /* Remove a class from a lock chain. Must be called with the graph lock held. */
4589 static void remove_class_from_lock_chain(struct pending_free *pf,
4590 struct lock_chain *chain,
4591 struct lock_class *class)
4593 #ifdef CONFIG_PROVE_LOCKING
4594 struct lock_chain *new_chain;
4595 u64 chain_key;
4596 int i;
4598 for (i = chain->base; i < chain->base + chain->depth; i++) {
4599 if (chain_hlocks[i] != class - lock_classes)
4600 continue;
4601 /* The code below leaks one chain_hlock[] entry. */
4602 if (--chain->depth > 0) {
4603 memmove(&chain_hlocks[i], &chain_hlocks[i + 1],
4604 (chain->base + chain->depth - i) *
4605 sizeof(chain_hlocks[0]));
4608 * Each lock class occurs at most once in a lock chain so once
4609 * we found a match we can break out of this loop.
4611 goto recalc;
4613 /* Since the chain has not been modified, return. */
4614 return;
4616 recalc:
4617 chain_key = 0;
4618 for (i = chain->base; i < chain->base + chain->depth; i++)
4619 chain_key = iterate_chain_key(chain_key, chain_hlocks[i] + 1);
4620 if (chain->depth && chain->chain_key == chain_key)
4621 return;
4622 /* Overwrite the chain key for concurrent RCU readers. */
4623 WRITE_ONCE(chain->chain_key, chain_key);
4625 * Note: calling hlist_del_rcu() from inside a
4626 * hlist_for_each_entry_rcu() loop is safe.
4628 hlist_del_rcu(&chain->entry);
4629 __set_bit(chain - lock_chains, pf->lock_chains_being_freed);
4630 if (chain->depth == 0)
4631 return;
4633 * If the modified lock chain matches an existing lock chain, drop
4634 * the modified lock chain.
4636 if (lookup_chain_cache(chain_key))
4637 return;
4638 new_chain = alloc_lock_chain();
4639 if (WARN_ON_ONCE(!new_chain)) {
4640 debug_locks_off();
4641 return;
4643 *new_chain = *chain;
4644 hlist_add_head_rcu(&new_chain->entry, chainhashentry(chain_key));
4645 #endif
4648 /* Must be called with the graph lock held. */
4649 static void remove_class_from_lock_chains(struct pending_free *pf,
4650 struct lock_class *class)
4652 struct lock_chain *chain;
4653 struct hlist_head *head;
4654 int i;
4656 for (i = 0; i < ARRAY_SIZE(chainhash_table); i++) {
4657 head = chainhash_table + i;
4658 hlist_for_each_entry_rcu(chain, head, entry) {
4659 remove_class_from_lock_chain(pf, chain, class);
4665 * Remove all references to a lock class. The caller must hold the graph lock.
4667 static void zap_class(struct pending_free *pf, struct lock_class *class)
4669 struct lock_list *entry;
4670 int i;
4672 WARN_ON_ONCE(!class->key);
4675 * Remove all dependencies this lock is
4676 * involved in:
4678 for_each_set_bit(i, list_entries_in_use, ARRAY_SIZE(list_entries)) {
4679 entry = list_entries + i;
4680 if (entry->class != class && entry->links_to != class)
4681 continue;
4682 __clear_bit(i, list_entries_in_use);
4683 nr_list_entries--;
4684 list_del_rcu(&entry->entry);
4686 if (list_empty(&class->locks_after) &&
4687 list_empty(&class->locks_before)) {
4688 list_move_tail(&class->lock_entry, &pf->zapped);
4689 hlist_del_rcu(&class->hash_entry);
4690 WRITE_ONCE(class->key, NULL);
4691 WRITE_ONCE(class->name, NULL);
4692 nr_lock_classes--;
4693 } else {
4694 WARN_ONCE(true, "%s() failed for class %s\n", __func__,
4695 class->name);
4698 remove_class_from_lock_chains(pf, class);
4701 static void reinit_class(struct lock_class *class)
4703 void *const p = class;
4704 const unsigned int offset = offsetof(struct lock_class, key);
4706 WARN_ON_ONCE(!class->lock_entry.next);
4707 WARN_ON_ONCE(!list_empty(&class->locks_after));
4708 WARN_ON_ONCE(!list_empty(&class->locks_before));
4709 memset(p + offset, 0, sizeof(*class) - offset);
4710 WARN_ON_ONCE(!class->lock_entry.next);
4711 WARN_ON_ONCE(!list_empty(&class->locks_after));
4712 WARN_ON_ONCE(!list_empty(&class->locks_before));
4715 static inline int within(const void *addr, void *start, unsigned long size)
4717 return addr >= start && addr < start + size;
4720 static bool inside_selftest(void)
4722 return current == lockdep_selftest_task_struct;
4725 /* The caller must hold the graph lock. */
4726 static struct pending_free *get_pending_free(void)
4728 return delayed_free.pf + delayed_free.index;
4731 static void free_zapped_rcu(struct rcu_head *cb);
4734 * Schedule an RCU callback if no RCU callback is pending. Must be called with
4735 * the graph lock held.
4737 static void call_rcu_zapped(struct pending_free *pf)
4739 WARN_ON_ONCE(inside_selftest());
4741 if (list_empty(&pf->zapped))
4742 return;
4744 if (delayed_free.scheduled)
4745 return;
4747 delayed_free.scheduled = true;
4749 WARN_ON_ONCE(delayed_free.pf + delayed_free.index != pf);
4750 delayed_free.index ^= 1;
4752 call_rcu(&delayed_free.rcu_head, free_zapped_rcu);
4755 /* The caller must hold the graph lock. May be called from RCU context. */
4756 static void __free_zapped_classes(struct pending_free *pf)
4758 struct lock_class *class;
4760 check_data_structures();
4762 list_for_each_entry(class, &pf->zapped, lock_entry)
4763 reinit_class(class);
4765 list_splice_init(&pf->zapped, &free_lock_classes);
4767 #ifdef CONFIG_PROVE_LOCKING
4768 bitmap_andnot(lock_chains_in_use, lock_chains_in_use,
4769 pf->lock_chains_being_freed, ARRAY_SIZE(lock_chains));
4770 bitmap_clear(pf->lock_chains_being_freed, 0, ARRAY_SIZE(lock_chains));
4771 #endif
4774 static void free_zapped_rcu(struct rcu_head *ch)
4776 struct pending_free *pf;
4777 unsigned long flags;
4779 if (WARN_ON_ONCE(ch != &delayed_free.rcu_head))
4780 return;
4782 raw_local_irq_save(flags);
4783 arch_spin_lock(&lockdep_lock);
4784 current->lockdep_recursion = 1;
4786 /* closed head */
4787 pf = delayed_free.pf + (delayed_free.index ^ 1);
4788 __free_zapped_classes(pf);
4789 delayed_free.scheduled = false;
4792 * If there's anything on the open list, close and start a new callback.
4794 call_rcu_zapped(delayed_free.pf + delayed_free.index);
4796 current->lockdep_recursion = 0;
4797 arch_spin_unlock(&lockdep_lock);
4798 raw_local_irq_restore(flags);
4802 * Remove all lock classes from the class hash table and from the
4803 * all_lock_classes list whose key or name is in the address range [start,
4804 * start + size). Move these lock classes to the zapped_classes list. Must
4805 * be called with the graph lock held.
4807 static void __lockdep_free_key_range(struct pending_free *pf, void *start,
4808 unsigned long size)
4810 struct lock_class *class;
4811 struct hlist_head *head;
4812 int i;
4814 /* Unhash all classes that were created by a module. */
4815 for (i = 0; i < CLASSHASH_SIZE; i++) {
4816 head = classhash_table + i;
4817 hlist_for_each_entry_rcu(class, head, hash_entry) {
4818 if (!within(class->key, start, size) &&
4819 !within(class->name, start, size))
4820 continue;
4821 zap_class(pf, class);
4827 * Used in module.c to remove lock classes from memory that is going to be
4828 * freed; and possibly re-used by other modules.
4830 * We will have had one synchronize_rcu() before getting here, so we're
4831 * guaranteed nobody will look up these exact classes -- they're properly dead
4832 * but still allocated.
4834 static void lockdep_free_key_range_reg(void *start, unsigned long size)
4836 struct pending_free *pf;
4837 unsigned long flags;
4839 init_data_structures_once();
4841 raw_local_irq_save(flags);
4842 arch_spin_lock(&lockdep_lock);
4843 current->lockdep_recursion = 1;
4844 pf = get_pending_free();
4845 __lockdep_free_key_range(pf, start, size);
4846 call_rcu_zapped(pf);
4847 current->lockdep_recursion = 0;
4848 arch_spin_unlock(&lockdep_lock);
4849 raw_local_irq_restore(flags);
4852 * Wait for any possible iterators from look_up_lock_class() to pass
4853 * before continuing to free the memory they refer to.
4855 synchronize_rcu();
4859 * Free all lockdep keys in the range [start, start+size). Does not sleep.
4860 * Ignores debug_locks. Must only be used by the lockdep selftests.
4862 static void lockdep_free_key_range_imm(void *start, unsigned long size)
4864 struct pending_free *pf = delayed_free.pf;
4865 unsigned long flags;
4867 init_data_structures_once();
4869 raw_local_irq_save(flags);
4870 arch_spin_lock(&lockdep_lock);
4871 __lockdep_free_key_range(pf, start, size);
4872 __free_zapped_classes(pf);
4873 arch_spin_unlock(&lockdep_lock);
4874 raw_local_irq_restore(flags);
4877 void lockdep_free_key_range(void *start, unsigned long size)
4879 init_data_structures_once();
4881 if (inside_selftest())
4882 lockdep_free_key_range_imm(start, size);
4883 else
4884 lockdep_free_key_range_reg(start, size);
4888 * Check whether any element of the @lock->class_cache[] array refers to a
4889 * registered lock class. The caller must hold either the graph lock or the
4890 * RCU read lock.
4892 static bool lock_class_cache_is_registered(struct lockdep_map *lock)
4894 struct lock_class *class;
4895 struct hlist_head *head;
4896 int i, j;
4898 for (i = 0; i < CLASSHASH_SIZE; i++) {
4899 head = classhash_table + i;
4900 hlist_for_each_entry_rcu(class, head, hash_entry) {
4901 for (j = 0; j < NR_LOCKDEP_CACHING_CLASSES; j++)
4902 if (lock->class_cache[j] == class)
4903 return true;
4906 return false;
4909 /* The caller must hold the graph lock. Does not sleep. */
4910 static void __lockdep_reset_lock(struct pending_free *pf,
4911 struct lockdep_map *lock)
4913 struct lock_class *class;
4914 int j;
4917 * Remove all classes this lock might have:
4919 for (j = 0; j < MAX_LOCKDEP_SUBCLASSES; j++) {
4921 * If the class exists we look it up and zap it:
4923 class = look_up_lock_class(lock, j);
4924 if (class)
4925 zap_class(pf, class);
4928 * Debug check: in the end all mapped classes should
4929 * be gone.
4931 if (WARN_ON_ONCE(lock_class_cache_is_registered(lock)))
4932 debug_locks_off();
4936 * Remove all information lockdep has about a lock if debug_locks == 1. Free
4937 * released data structures from RCU context.
4939 static void lockdep_reset_lock_reg(struct lockdep_map *lock)
4941 struct pending_free *pf;
4942 unsigned long flags;
4943 int locked;
4945 raw_local_irq_save(flags);
4946 locked = graph_lock();
4947 if (!locked)
4948 goto out_irq;
4950 pf = get_pending_free();
4951 __lockdep_reset_lock(pf, lock);
4952 call_rcu_zapped(pf);
4954 graph_unlock();
4955 out_irq:
4956 raw_local_irq_restore(flags);
4960 * Reset a lock. Does not sleep. Ignores debug_locks. Must only be used by the
4961 * lockdep selftests.
4963 static void lockdep_reset_lock_imm(struct lockdep_map *lock)
4965 struct pending_free *pf = delayed_free.pf;
4966 unsigned long flags;
4968 raw_local_irq_save(flags);
4969 arch_spin_lock(&lockdep_lock);
4970 __lockdep_reset_lock(pf, lock);
4971 __free_zapped_classes(pf);
4972 arch_spin_unlock(&lockdep_lock);
4973 raw_local_irq_restore(flags);
4976 void lockdep_reset_lock(struct lockdep_map *lock)
4978 init_data_structures_once();
4980 if (inside_selftest())
4981 lockdep_reset_lock_imm(lock);
4982 else
4983 lockdep_reset_lock_reg(lock);
4986 /* Unregister a dynamically allocated key. */
4987 void lockdep_unregister_key(struct lock_class_key *key)
4989 struct hlist_head *hash_head = keyhashentry(key);
4990 struct lock_class_key *k;
4991 struct pending_free *pf;
4992 unsigned long flags;
4993 bool found = false;
4995 might_sleep();
4997 if (WARN_ON_ONCE(static_obj(key)))
4998 return;
5000 raw_local_irq_save(flags);
5001 if (!graph_lock())
5002 goto out_irq;
5004 pf = get_pending_free();
5005 hlist_for_each_entry_rcu(k, hash_head, hash_entry) {
5006 if (k == key) {
5007 hlist_del_rcu(&k->hash_entry);
5008 found = true;
5009 break;
5012 WARN_ON_ONCE(!found);
5013 __lockdep_free_key_range(pf, key, 1);
5014 call_rcu_zapped(pf);
5015 graph_unlock();
5016 out_irq:
5017 raw_local_irq_restore(flags);
5019 /* Wait until is_dynamic_key() has finished accessing k->hash_entry. */
5020 synchronize_rcu();
5022 EXPORT_SYMBOL_GPL(lockdep_unregister_key);
5024 void __init lockdep_init(void)
5026 printk("Lock dependency validator: Copyright (c) 2006 Red Hat, Inc., Ingo Molnar\n");
5028 printk("... MAX_LOCKDEP_SUBCLASSES: %lu\n", MAX_LOCKDEP_SUBCLASSES);
5029 printk("... MAX_LOCK_DEPTH: %lu\n", MAX_LOCK_DEPTH);
5030 printk("... MAX_LOCKDEP_KEYS: %lu\n", MAX_LOCKDEP_KEYS);
5031 printk("... CLASSHASH_SIZE: %lu\n", CLASSHASH_SIZE);
5032 printk("... MAX_LOCKDEP_ENTRIES: %lu\n", MAX_LOCKDEP_ENTRIES);
5033 printk("... MAX_LOCKDEP_CHAINS: %lu\n", MAX_LOCKDEP_CHAINS);
5034 printk("... CHAINHASH_SIZE: %lu\n", CHAINHASH_SIZE);
5036 printk(" memory used by lock dependency info: %zu kB\n",
5037 (sizeof(lock_classes) +
5038 sizeof(classhash_table) +
5039 sizeof(list_entries) +
5040 sizeof(list_entries_in_use) +
5041 sizeof(chainhash_table) +
5042 sizeof(delayed_free)
5043 #ifdef CONFIG_PROVE_LOCKING
5044 + sizeof(lock_cq)
5045 + sizeof(lock_chains)
5046 + sizeof(lock_chains_in_use)
5047 + sizeof(chain_hlocks)
5048 #endif
5049 ) / 1024
5052 printk(" per task-struct memory footprint: %zu bytes\n",
5053 sizeof(((struct task_struct *)NULL)->held_locks));
5056 static void
5057 print_freed_lock_bug(struct task_struct *curr, const void *mem_from,
5058 const void *mem_to, struct held_lock *hlock)
5060 if (!debug_locks_off())
5061 return;
5062 if (debug_locks_silent)
5063 return;
5065 pr_warn("\n");
5066 pr_warn("=========================\n");
5067 pr_warn("WARNING: held lock freed!\n");
5068 print_kernel_ident();
5069 pr_warn("-------------------------\n");
5070 pr_warn("%s/%d is freeing memory %px-%px, with a lock still held there!\n",
5071 curr->comm, task_pid_nr(curr), mem_from, mem_to-1);
5072 print_lock(hlock);
5073 lockdep_print_held_locks(curr);
5075 pr_warn("\nstack backtrace:\n");
5076 dump_stack();
5079 static inline int not_in_range(const void* mem_from, unsigned long mem_len,
5080 const void* lock_from, unsigned long lock_len)
5082 return lock_from + lock_len <= mem_from ||
5083 mem_from + mem_len <= lock_from;
5087 * Called when kernel memory is freed (or unmapped), or if a lock
5088 * is destroyed or reinitialized - this code checks whether there is
5089 * any held lock in the memory range of <from> to <to>:
5091 void debug_check_no_locks_freed(const void *mem_from, unsigned long mem_len)
5093 struct task_struct *curr = current;
5094 struct held_lock *hlock;
5095 unsigned long flags;
5096 int i;
5098 if (unlikely(!debug_locks))
5099 return;
5101 raw_local_irq_save(flags);
5102 for (i = 0; i < curr->lockdep_depth; i++) {
5103 hlock = curr->held_locks + i;
5105 if (not_in_range(mem_from, mem_len, hlock->instance,
5106 sizeof(*hlock->instance)))
5107 continue;
5109 print_freed_lock_bug(curr, mem_from, mem_from + mem_len, hlock);
5110 break;
5112 raw_local_irq_restore(flags);
5114 EXPORT_SYMBOL_GPL(debug_check_no_locks_freed);
5116 static void print_held_locks_bug(void)
5118 if (!debug_locks_off())
5119 return;
5120 if (debug_locks_silent)
5121 return;
5123 pr_warn("\n");
5124 pr_warn("====================================\n");
5125 pr_warn("WARNING: %s/%d still has locks held!\n",
5126 current->comm, task_pid_nr(current));
5127 print_kernel_ident();
5128 pr_warn("------------------------------------\n");
5129 lockdep_print_held_locks(current);
5130 pr_warn("\nstack backtrace:\n");
5131 dump_stack();
5134 void debug_check_no_locks_held(void)
5136 if (unlikely(current->lockdep_depth > 0))
5137 print_held_locks_bug();
5139 EXPORT_SYMBOL_GPL(debug_check_no_locks_held);
5141 #ifdef __KERNEL__
5142 void debug_show_all_locks(void)
5144 struct task_struct *g, *p;
5146 if (unlikely(!debug_locks)) {
5147 pr_warn("INFO: lockdep is turned off.\n");
5148 return;
5150 pr_warn("\nShowing all locks held in the system:\n");
5152 rcu_read_lock();
5153 for_each_process_thread(g, p) {
5154 if (!p->lockdep_depth)
5155 continue;
5156 lockdep_print_held_locks(p);
5157 touch_nmi_watchdog();
5158 touch_all_softlockup_watchdogs();
5160 rcu_read_unlock();
5162 pr_warn("\n");
5163 pr_warn("=============================================\n\n");
5165 EXPORT_SYMBOL_GPL(debug_show_all_locks);
5166 #endif
5169 * Careful: only use this function if you are sure that
5170 * the task cannot run in parallel!
5172 void debug_show_held_locks(struct task_struct *task)
5174 if (unlikely(!debug_locks)) {
5175 printk("INFO: lockdep is turned off.\n");
5176 return;
5178 lockdep_print_held_locks(task);
5180 EXPORT_SYMBOL_GPL(debug_show_held_locks);
5182 asmlinkage __visible void lockdep_sys_exit(void)
5184 struct task_struct *curr = current;
5186 if (unlikely(curr->lockdep_depth)) {
5187 if (!debug_locks_off())
5188 return;
5189 pr_warn("\n");
5190 pr_warn("================================================\n");
5191 pr_warn("WARNING: lock held when returning to user space!\n");
5192 print_kernel_ident();
5193 pr_warn("------------------------------------------------\n");
5194 pr_warn("%s/%d is leaving the kernel with locks still held!\n",
5195 curr->comm, curr->pid);
5196 lockdep_print_held_locks(curr);
5200 * The lock history for each syscall should be independent. So wipe the
5201 * slate clean on return to userspace.
5203 lockdep_invariant_state(false);
5206 void lockdep_rcu_suspicious(const char *file, const int line, const char *s)
5208 struct task_struct *curr = current;
5210 /* Note: the following can be executed concurrently, so be careful. */
5211 pr_warn("\n");
5212 pr_warn("=============================\n");
5213 pr_warn("WARNING: suspicious RCU usage\n");
5214 print_kernel_ident();
5215 pr_warn("-----------------------------\n");
5216 pr_warn("%s:%d %s!\n", file, line, s);
5217 pr_warn("\nother info that might help us debug this:\n\n");
5218 pr_warn("\n%srcu_scheduler_active = %d, debug_locks = %d\n",
5219 !rcu_lockdep_current_cpu_online()
5220 ? "RCU used illegally from offline CPU!\n"
5221 : !rcu_is_watching()
5222 ? "RCU used illegally from idle CPU!\n"
5223 : "",
5224 rcu_scheduler_active, debug_locks);
5227 * If a CPU is in the RCU-free window in idle (ie: in the section
5228 * between rcu_idle_enter() and rcu_idle_exit(), then RCU
5229 * considers that CPU to be in an "extended quiescent state",
5230 * which means that RCU will be completely ignoring that CPU.
5231 * Therefore, rcu_read_lock() and friends have absolutely no
5232 * effect on a CPU running in that state. In other words, even if
5233 * such an RCU-idle CPU has called rcu_read_lock(), RCU might well
5234 * delete data structures out from under it. RCU really has no
5235 * choice here: we need to keep an RCU-free window in idle where
5236 * the CPU may possibly enter into low power mode. This way we can
5237 * notice an extended quiescent state to other CPUs that started a grace
5238 * period. Otherwise we would delay any grace period as long as we run
5239 * in the idle task.
5241 * So complain bitterly if someone does call rcu_read_lock(),
5242 * rcu_read_lock_bh() and so on from extended quiescent states.
5244 if (!rcu_is_watching())
5245 pr_warn("RCU used illegally from extended quiescent state!\n");
5247 lockdep_print_held_locks(curr);
5248 pr_warn("\nstack backtrace:\n");
5249 dump_stack();
5251 EXPORT_SYMBOL_GPL(lockdep_rcu_suspicious);