Linux Audio

Check our new training course

Loading...
Note: File does not exist in v3.1.
   1// SPDX-License-Identifier: GPL-2.0-only
   2/*
   3 * kernel/lockdep.c
   4 *
   5 * Runtime locking correctness validator
   6 *
   7 * Started by Ingo Molnar:
   8 *
   9 *  Copyright (C) 2006,2007 Red Hat, Inc., Ingo Molnar <mingo@redhat.com>
  10 *  Copyright (C) 2007 Red Hat, Inc., Peter Zijlstra
  11 *
  12 * this code maps all the lock dependencies as they occur in a live kernel
  13 * and will warn about the following classes of locking bugs:
  14 *
  15 * - lock inversion scenarios
  16 * - circular lock dependencies
  17 * - hardirq/softirq safe/unsafe locking bugs
  18 *
  19 * Bugs are reported even if the current locking scenario does not cause
  20 * any deadlock at this point.
  21 *
  22 * I.e. if anytime in the past two locks were taken in a different order,
  23 * even if it happened for another task, even if those were different
  24 * locks (but of the same class as this lock), this code will detect it.
  25 *
  26 * Thanks to Arjan van de Ven for coming up with the initial idea of
  27 * mapping lock dependencies runtime.
  28 */
  29#define DISABLE_BRANCH_PROFILING
  30#include <linux/mutex.h>
  31#include <linux/sched.h>
  32#include <linux/sched/clock.h>
  33#include <linux/sched/task.h>
  34#include <linux/sched/mm.h>
  35#include <linux/delay.h>
  36#include <linux/module.h>
  37#include <linux/proc_fs.h>
  38#include <linux/seq_file.h>
  39#include <linux/spinlock.h>
  40#include <linux/kallsyms.h>
  41#include <linux/interrupt.h>
  42#include <linux/stacktrace.h>
  43#include <linux/debug_locks.h>
  44#include <linux/irqflags.h>
  45#include <linux/utsname.h>
  46#include <linux/hash.h>
  47#include <linux/ftrace.h>
  48#include <linux/stringify.h>
  49#include <linux/bitmap.h>
  50#include <linux/bitops.h>
  51#include <linux/gfp.h>
  52#include <linux/random.h>
  53#include <linux/jhash.h>
  54#include <linux/nmi.h>
  55#include <linux/rcupdate.h>
  56#include <linux/kprobes.h>
  57#include <linux/lockdep.h>
  58#include <linux/context_tracking.h>
  59#include <linux/console.h>
  60
  61#include <asm/sections.h>
  62
  63#include "lockdep_internals.h"
  64
  65#include <trace/events/lock.h>
  66
  67#ifdef CONFIG_PROVE_LOCKING
  68static int prove_locking = 1;
  69module_param(prove_locking, int, 0644);
  70#else
  71#define prove_locking 0
  72#endif
  73
  74#ifdef CONFIG_LOCK_STAT
  75static int lock_stat = 1;
  76module_param(lock_stat, int, 0644);
  77#else
  78#define lock_stat 0
  79#endif
  80
  81#ifdef CONFIG_SYSCTL
  82static struct ctl_table kern_lockdep_table[] = {
  83#ifdef CONFIG_PROVE_LOCKING
  84	{
  85		.procname       = "prove_locking",
  86		.data           = &prove_locking,
  87		.maxlen         = sizeof(int),
  88		.mode           = 0644,
  89		.proc_handler   = proc_dointvec,
  90	},
  91#endif /* CONFIG_PROVE_LOCKING */
  92#ifdef CONFIG_LOCK_STAT
  93	{
  94		.procname       = "lock_stat",
  95		.data           = &lock_stat,
  96		.maxlen         = sizeof(int),
  97		.mode           = 0644,
  98		.proc_handler   = proc_dointvec,
  99	},
 100#endif /* CONFIG_LOCK_STAT */
 101};
 102
 103static __init int kernel_lockdep_sysctls_init(void)
 104{
 105	register_sysctl_init("kernel", kern_lockdep_table);
 106	return 0;
 107}
 108late_initcall(kernel_lockdep_sysctls_init);
 109#endif /* CONFIG_SYSCTL */
 110
 111DEFINE_PER_CPU(unsigned int, lockdep_recursion);
 112EXPORT_PER_CPU_SYMBOL_GPL(lockdep_recursion);
 113
 114static __always_inline bool lockdep_enabled(void)
 115{
 116	if (!debug_locks)
 117		return false;
 118
 119	if (this_cpu_read(lockdep_recursion))
 120		return false;
 121
 122	if (current->lockdep_recursion)
 123		return false;
 124
 125	return true;
 126}
 127
 128/*
 129 * lockdep_lock: protects the lockdep graph, the hashes and the
 130 *               class/list/hash allocators.
 131 *
 132 * This is one of the rare exceptions where it's justified
 133 * to use a raw spinlock - we really dont want the spinlock
 134 * code to recurse back into the lockdep code...
 135 */
 136static arch_spinlock_t __lock = (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
 137static struct task_struct *__owner;
 138
 139static inline void lockdep_lock(void)
 140{
 141	DEBUG_LOCKS_WARN_ON(!irqs_disabled());
 142
 143	__this_cpu_inc(lockdep_recursion);
 144	arch_spin_lock(&__lock);
 145	__owner = current;
 146}
 147
 148static inline void lockdep_unlock(void)
 149{
 150	DEBUG_LOCKS_WARN_ON(!irqs_disabled());
 151
 152	if (debug_locks && DEBUG_LOCKS_WARN_ON(__owner != current))
 153		return;
 154
 155	__owner = NULL;
 156	arch_spin_unlock(&__lock);
 157	__this_cpu_dec(lockdep_recursion);
 158}
 159
 160static inline bool lockdep_assert_locked(void)
 161{
 162	return DEBUG_LOCKS_WARN_ON(__owner != current);
 163}
 164
 165static struct task_struct *lockdep_selftest_task_struct;
 166
 167
 168static int graph_lock(void)
 169{
 170	lockdep_lock();
 171	/*
 172	 * Make sure that if another CPU detected a bug while
 173	 * walking the graph we dont change it (while the other
 174	 * CPU is busy printing out stuff with the graph lock
 175	 * dropped already)
 176	 */
 177	if (!debug_locks) {
 178		lockdep_unlock();
 179		return 0;
 180	}
 181	return 1;
 182}
 183
 184static inline void graph_unlock(void)
 185{
 186	lockdep_unlock();
 187}
 188
 189/*
 190 * Turn lock debugging off and return with 0 if it was off already,
 191 * and also release the graph lock:
 192 */
 193static inline int debug_locks_off_graph_unlock(void)
 194{
 195	int ret = debug_locks_off();
 196
 197	lockdep_unlock();
 198
 199	return ret;
 200}
 201
 202unsigned long nr_list_entries;
 203static struct lock_list list_entries[MAX_LOCKDEP_ENTRIES];
 204static DECLARE_BITMAP(list_entries_in_use, MAX_LOCKDEP_ENTRIES);
 205
 206/*
 207 * All data structures here are protected by the global debug_lock.
 208 *
 209 * nr_lock_classes is the number of elements of lock_classes[] that is
 210 * in use.
 211 */
 212#define KEYHASH_BITS		(MAX_LOCKDEP_KEYS_BITS - 1)
 213#define KEYHASH_SIZE		(1UL << KEYHASH_BITS)
 214static struct hlist_head lock_keys_hash[KEYHASH_SIZE];
 215unsigned long nr_lock_classes;
 216unsigned long nr_zapped_classes;
 217unsigned long max_lock_class_idx;
 218struct lock_class lock_classes[MAX_LOCKDEP_KEYS];
 219DECLARE_BITMAP(lock_classes_in_use, MAX_LOCKDEP_KEYS);
 220
 221static inline struct lock_class *hlock_class(struct held_lock *hlock)
 222{
 223	unsigned int class_idx = hlock->class_idx;
 224
 225	/* Don't re-read hlock->class_idx, can't use READ_ONCE() on bitfield */
 226	barrier();
 227
 228	if (!test_bit(class_idx, lock_classes_in_use)) {
 229		/*
 230		 * Someone passed in garbage, we give up.
 231		 */
 232		DEBUG_LOCKS_WARN_ON(1);
 233		return NULL;
 234	}
 235
 236	/*
 237	 * At this point, if the passed hlock->class_idx is still garbage,
 238	 * we just have to live with it
 239	 */
 240	return lock_classes + class_idx;
 241}
 242
 243#ifdef CONFIG_LOCK_STAT
 244static DEFINE_PER_CPU(struct lock_class_stats[MAX_LOCKDEP_KEYS], cpu_lock_stats);
 245
 246static inline u64 lockstat_clock(void)
 247{
 248	return local_clock();
 249}
 250
 251static int lock_point(unsigned long points[], unsigned long ip)
 252{
 253	int i;
 254
 255	for (i = 0; i < LOCKSTAT_POINTS; i++) {
 256		if (points[i] == 0) {
 257			points[i] = ip;
 258			break;
 259		}
 260		if (points[i] == ip)
 261			break;
 262	}
 263
 264	return i;
 265}
 266
 267static void lock_time_inc(struct lock_time *lt, u64 time)
 268{
 269	if (time > lt->max)
 270		lt->max = time;
 271
 272	if (time < lt->min || !lt->nr)
 273		lt->min = time;
 274
 275	lt->total += time;
 276	lt->nr++;
 277}
 278
 279static inline void lock_time_add(struct lock_time *src, struct lock_time *dst)
 280{
 281	if (!src->nr)
 282		return;
 283
 284	if (src->max > dst->max)
 285		dst->max = src->max;
 286
 287	if (src->min < dst->min || !dst->nr)
 288		dst->min = src->min;
 289
 290	dst->total += src->total;
 291	dst->nr += src->nr;
 292}
 293
 294struct lock_class_stats lock_stats(struct lock_class *class)
 295{
 296	struct lock_class_stats stats;
 297	int cpu, i;
 298
 299	memset(&stats, 0, sizeof(struct lock_class_stats));
 300	for_each_possible_cpu(cpu) {
 301		struct lock_class_stats *pcs =
 302			&per_cpu(cpu_lock_stats, cpu)[class - lock_classes];
 303
 304		for (i = 0; i < ARRAY_SIZE(stats.contention_point); i++)
 305			stats.contention_point[i] += pcs->contention_point[i];
 306
 307		for (i = 0; i < ARRAY_SIZE(stats.contending_point); i++)
 308			stats.contending_point[i] += pcs->contending_point[i];
 309
 310		lock_time_add(&pcs->read_waittime, &stats.read_waittime);
 311		lock_time_add(&pcs->write_waittime, &stats.write_waittime);
 312
 313		lock_time_add(&pcs->read_holdtime, &stats.read_holdtime);
 314		lock_time_add(&pcs->write_holdtime, &stats.write_holdtime);
 315
 316		for (i = 0; i < ARRAY_SIZE(stats.bounces); i++)
 317			stats.bounces[i] += pcs->bounces[i];
 318	}
 319
 320	return stats;
 321}
 322
 323void clear_lock_stats(struct lock_class *class)
 324{
 325	int cpu;
 326
 327	for_each_possible_cpu(cpu) {
 328		struct lock_class_stats *cpu_stats =
 329			&per_cpu(cpu_lock_stats, cpu)[class - lock_classes];
 330
 331		memset(cpu_stats, 0, sizeof(struct lock_class_stats));
 332	}
 333	memset(class->contention_point, 0, sizeof(class->contention_point));
 334	memset(class->contending_point, 0, sizeof(class->contending_point));
 335}
 336
 337static struct lock_class_stats *get_lock_stats(struct lock_class *class)
 338{
 339	return &this_cpu_ptr(cpu_lock_stats)[class - lock_classes];
 340}
 341
 342static void lock_release_holdtime(struct held_lock *hlock)
 343{
 344	struct lock_class_stats *stats;
 345	u64 holdtime;
 346
 347	if (!lock_stat)
 348		return;
 349
 350	holdtime = lockstat_clock() - hlock->holdtime_stamp;
 351
 352	stats = get_lock_stats(hlock_class(hlock));
 353	if (hlock->read)
 354		lock_time_inc(&stats->read_holdtime, holdtime);
 355	else
 356		lock_time_inc(&stats->write_holdtime, holdtime);
 357}
 358#else
 359static inline void lock_release_holdtime(struct held_lock *hlock)
 360{
 361}
 362#endif
 363
 364/*
 365 * We keep a global list of all lock classes. The list is only accessed with
 366 * the lockdep spinlock lock held. free_lock_classes is a list with free
 367 * elements. These elements are linked together by the lock_entry member in
 368 * struct lock_class.
 369 */
 370static LIST_HEAD(all_lock_classes);
 371static LIST_HEAD(free_lock_classes);
 372
 373/**
 374 * struct pending_free - information about data structures about to be freed
 375 * @zapped: Head of a list with struct lock_class elements.
 376 * @lock_chains_being_freed: Bitmap that indicates which lock_chains[] elements
 377 *	are about to be freed.
 378 */
 379struct pending_free {
 380	struct list_head zapped;
 381	DECLARE_BITMAP(lock_chains_being_freed, MAX_LOCKDEP_CHAINS);
 382};
 383
 384/**
 385 * struct delayed_free - data structures used for delayed freeing
 386 *
 387 * A data structure for delayed freeing of data structures that may be
 388 * accessed by RCU readers at the time these were freed.
 389 *
 390 * @rcu_head:  Used to schedule an RCU callback for freeing data structures.
 391 * @index:     Index of @pf to which freed data structures are added.
 392 * @scheduled: Whether or not an RCU callback has been scheduled.
 393 * @pf:        Array with information about data structures about to be freed.
 394 */
 395static struct delayed_free {
 396	struct rcu_head		rcu_head;
 397	int			index;
 398	int			scheduled;
 399	struct pending_free	pf[2];
 400} delayed_free;
 401
 402/*
 403 * The lockdep classes are in a hash-table as well, for fast lookup:
 404 */
 405#define CLASSHASH_BITS		(MAX_LOCKDEP_KEYS_BITS - 1)
 406#define CLASSHASH_SIZE		(1UL << CLASSHASH_BITS)
 407#define __classhashfn(key)	hash_long((unsigned long)key, CLASSHASH_BITS)
 408#define classhashentry(key)	(classhash_table + __classhashfn((key)))
 409
 410static struct hlist_head classhash_table[CLASSHASH_SIZE];
 411
 412/*
 413 * We put the lock dependency chains into a hash-table as well, to cache
 414 * their existence:
 415 */
 416#define CHAINHASH_BITS		(MAX_LOCKDEP_CHAINS_BITS-1)
 417#define CHAINHASH_SIZE		(1UL << CHAINHASH_BITS)
 418#define __chainhashfn(chain)	hash_long(chain, CHAINHASH_BITS)
 419#define chainhashentry(chain)	(chainhash_table + __chainhashfn((chain)))
 420
 421static struct hlist_head chainhash_table[CHAINHASH_SIZE];
 422
 423/*
 424 * the id of held_lock
 425 */
 426static inline u16 hlock_id(struct held_lock *hlock)
 427{
 428	BUILD_BUG_ON(MAX_LOCKDEP_KEYS_BITS + 2 > 16);
 429
 430	return (hlock->class_idx | (hlock->read << MAX_LOCKDEP_KEYS_BITS));
 431}
 432
 433static inline unsigned int chain_hlock_class_idx(u16 hlock_id)
 434{
 435	return hlock_id & (MAX_LOCKDEP_KEYS - 1);
 436}
 437
 438/*
 439 * The hash key of the lock dependency chains is a hash itself too:
 440 * it's a hash of all locks taken up to that lock, including that lock.
 441 * It's a 64-bit hash, because it's important for the keys to be
 442 * unique.
 443 */
 444static inline u64 iterate_chain_key(u64 key, u32 idx)
 445{
 446	u32 k0 = key, k1 = key >> 32;
 447
 448	__jhash_mix(idx, k0, k1); /* Macro that modifies arguments! */
 449
 450	return k0 | (u64)k1 << 32;
 451}
 452
 453void lockdep_init_task(struct task_struct *task)
 454{
 455	task->lockdep_depth = 0; /* no locks held yet */
 456	task->curr_chain_key = INITIAL_CHAIN_KEY;
 457	task->lockdep_recursion = 0;
 458}
 459
 460static __always_inline void lockdep_recursion_inc(void)
 461{
 462	__this_cpu_inc(lockdep_recursion);
 463}
 464
 465static __always_inline void lockdep_recursion_finish(void)
 466{
 467	if (WARN_ON_ONCE(__this_cpu_dec_return(lockdep_recursion)))
 468		__this_cpu_write(lockdep_recursion, 0);
 469}
 470
 471void lockdep_set_selftest_task(struct task_struct *task)
 472{
 473	lockdep_selftest_task_struct = task;
 474}
 475
 476/*
 477 * Debugging switches:
 478 */
 479
 480#define VERBOSE			0
 481#define VERY_VERBOSE		0
 482
 483#if VERBOSE
 484# define HARDIRQ_VERBOSE	1
 485# define SOFTIRQ_VERBOSE	1
 486#else
 487# define HARDIRQ_VERBOSE	0
 488# define SOFTIRQ_VERBOSE	0
 489#endif
 490
 491#if VERBOSE || HARDIRQ_VERBOSE || SOFTIRQ_VERBOSE
 492/*
 493 * Quick filtering for interesting events:
 494 */
 495static int class_filter(struct lock_class *class)
 496{
 497#if 0
 498	/* Example */
 499	if (class->name_version == 1 &&
 500			!strcmp(class->name, "lockname"))
 501		return 1;
 502	if (class->name_version == 1 &&
 503			!strcmp(class->name, "&struct->lockfield"))
 504		return 1;
 505#endif
 506	/* Filter everything else. 1 would be to allow everything else */
 507	return 0;
 508}
 509#endif
 510
 511static int verbose(struct lock_class *class)
 512{
 513#if VERBOSE
 514	return class_filter(class);
 515#endif
 516	return 0;
 517}
 518
 519static void print_lockdep_off(const char *bug_msg)
 520{
 521	printk(KERN_DEBUG "%s\n", bug_msg);
 522	printk(KERN_DEBUG "turning off the locking correctness validator.\n");
 523#ifdef CONFIG_LOCK_STAT
 524	printk(KERN_DEBUG "Please attach the output of /proc/lock_stat to the bug report\n");
 525#endif
 526}
 527
 528unsigned long nr_stack_trace_entries;
 529
 530#ifdef CONFIG_PROVE_LOCKING
 531/**
 532 * struct lock_trace - single stack backtrace
 533 * @hash_entry:	Entry in a stack_trace_hash[] list.
 534 * @hash:	jhash() of @entries.
 535 * @nr_entries:	Number of entries in @entries.
 536 * @entries:	Actual stack backtrace.
 537 */
 538struct lock_trace {
 539	struct hlist_node	hash_entry;
 540	u32			hash;
 541	u32			nr_entries;
 542	unsigned long		entries[] __aligned(sizeof(unsigned long));
 543};
 544#define LOCK_TRACE_SIZE_IN_LONGS				\
 545	(sizeof(struct lock_trace) / sizeof(unsigned long))
 546/*
 547 * Stack-trace: sequence of lock_trace structures. Protected by the graph_lock.
 548 */
 549static unsigned long stack_trace[MAX_STACK_TRACE_ENTRIES];
 550static struct hlist_head stack_trace_hash[STACK_TRACE_HASH_SIZE];
 551
 552static bool traces_identical(struct lock_trace *t1, struct lock_trace *t2)
 553{
 554	return t1->hash == t2->hash && t1->nr_entries == t2->nr_entries &&
 555		memcmp(t1->entries, t2->entries,
 556		       t1->nr_entries * sizeof(t1->entries[0])) == 0;
 557}
 558
 559static struct lock_trace *save_trace(void)
 560{
 561	struct lock_trace *trace, *t2;
 562	struct hlist_head *hash_head;
 563	u32 hash;
 564	int max_entries;
 565
 566	BUILD_BUG_ON_NOT_POWER_OF_2(STACK_TRACE_HASH_SIZE);
 567	BUILD_BUG_ON(LOCK_TRACE_SIZE_IN_LONGS >= MAX_STACK_TRACE_ENTRIES);
 568
 569	trace = (struct lock_trace *)(stack_trace + nr_stack_trace_entries);
 570	max_entries = MAX_STACK_TRACE_ENTRIES - nr_stack_trace_entries -
 571		LOCK_TRACE_SIZE_IN_LONGS;
 572
 573	if (max_entries <= 0) {
 574		if (!debug_locks_off_graph_unlock())
 575			return NULL;
 576
 577		nbcon_cpu_emergency_enter();
 578		print_lockdep_off("BUG: MAX_STACK_TRACE_ENTRIES too low!");
 579		dump_stack();
 580		nbcon_cpu_emergency_exit();
 581
 582		return NULL;
 583	}
 584	trace->nr_entries = stack_trace_save(trace->entries, max_entries, 3);
 585
 586	hash = jhash(trace->entries, trace->nr_entries *
 587		     sizeof(trace->entries[0]), 0);
 588	trace->hash = hash;
 589	hash_head = stack_trace_hash + (hash & (STACK_TRACE_HASH_SIZE - 1));
 590	hlist_for_each_entry(t2, hash_head, hash_entry) {
 591		if (traces_identical(trace, t2))
 592			return t2;
 593	}
 594	nr_stack_trace_entries += LOCK_TRACE_SIZE_IN_LONGS + trace->nr_entries;
 595	hlist_add_head(&trace->hash_entry, hash_head);
 596
 597	return trace;
 598}
 599
 600/* Return the number of stack traces in the stack_trace[] array. */
 601u64 lockdep_stack_trace_count(void)
 602{
 603	struct lock_trace *trace;
 604	u64 c = 0;
 605	int i;
 606
 607	for (i = 0; i < ARRAY_SIZE(stack_trace_hash); i++) {
 608		hlist_for_each_entry(trace, &stack_trace_hash[i], hash_entry) {
 609			c++;
 610		}
 611	}
 612
 613	return c;
 614}
 615
 616/* Return the number of stack hash chains that have at least one stack trace. */
 617u64 lockdep_stack_hash_count(void)
 618{
 619	u64 c = 0;
 620	int i;
 621
 622	for (i = 0; i < ARRAY_SIZE(stack_trace_hash); i++)
 623		if (!hlist_empty(&stack_trace_hash[i]))
 624			c++;
 625
 626	return c;
 627}
 628#endif
 629
 630unsigned int nr_hardirq_chains;
 631unsigned int nr_softirq_chains;
 632unsigned int nr_process_chains;
 633unsigned int max_lockdep_depth;
 634
 635#ifdef CONFIG_DEBUG_LOCKDEP
 636/*
 637 * Various lockdep statistics:
 638 */
 639DEFINE_PER_CPU(struct lockdep_stats, lockdep_stats);
 640#endif
 641
 642#ifdef CONFIG_PROVE_LOCKING
 643/*
 644 * Locking printouts:
 645 */
 646
 647#define __USAGE(__STATE)						\
 648	[LOCK_USED_IN_##__STATE] = "IN-"__stringify(__STATE)"-W",	\
 649	[LOCK_ENABLED_##__STATE] = __stringify(__STATE)"-ON-W",		\
 650	[LOCK_USED_IN_##__STATE##_READ] = "IN-"__stringify(__STATE)"-R",\
 651	[LOCK_ENABLED_##__STATE##_READ] = __stringify(__STATE)"-ON-R",
 652
 653static const char *usage_str[] =
 654{
 655#define LOCKDEP_STATE(__STATE) __USAGE(__STATE)
 656#include "lockdep_states.h"
 657#undef LOCKDEP_STATE
 658	[LOCK_USED] = "INITIAL USE",
 659	[LOCK_USED_READ] = "INITIAL READ USE",
 660	/* abused as string storage for verify_lock_unused() */
 661	[LOCK_USAGE_STATES] = "IN-NMI",
 662};
 663#endif
 664
 665const char *__get_key_name(const struct lockdep_subclass_key *key, char *str)
 666{
 667	return kallsyms_lookup((unsigned long)key, NULL, NULL, NULL, str);
 668}
 669
 670static inline unsigned long lock_flag(enum lock_usage_bit bit)
 671{
 672	return 1UL << bit;
 673}
 674
 675static char get_usage_char(struct lock_class *class, enum lock_usage_bit bit)
 676{
 677	/*
 678	 * The usage character defaults to '.' (i.e., irqs disabled and not in
 679	 * irq context), which is the safest usage category.
 680	 */
 681	char c = '.';
 682
 683	/*
 684	 * The order of the following usage checks matters, which will
 685	 * result in the outcome character as follows:
 686	 *
 687	 * - '+': irq is enabled and not in irq context
 688	 * - '-': in irq context and irq is disabled
 689	 * - '?': in irq context and irq is enabled
 690	 */
 691	if (class->usage_mask & lock_flag(bit + LOCK_USAGE_DIR_MASK)) {
 692		c = '+';
 693		if (class->usage_mask & lock_flag(bit))
 694			c = '?';
 695	} else if (class->usage_mask & lock_flag(bit))
 696		c = '-';
 697
 698	return c;
 699}
 700
 701void get_usage_chars(struct lock_class *class, char usage[LOCK_USAGE_CHARS])
 702{
 703	int i = 0;
 704
 705#define LOCKDEP_STATE(__STATE) 						\
 706	usage[i++] = get_usage_char(class, LOCK_USED_IN_##__STATE);	\
 707	usage[i++] = get_usage_char(class, LOCK_USED_IN_##__STATE##_READ);
 708#include "lockdep_states.h"
 709#undef LOCKDEP_STATE
 710
 711	usage[i] = '\0';
 712}
 713
 714static void __print_lock_name(struct held_lock *hlock, struct lock_class *class)
 715{
 716	char str[KSYM_NAME_LEN];
 717	const char *name;
 718
 719	name = class->name;
 720	if (!name) {
 721		name = __get_key_name(class->key, str);
 722		printk(KERN_CONT "%s", name);
 723	} else {
 724		printk(KERN_CONT "%s", name);
 725		if (class->name_version > 1)
 726			printk(KERN_CONT "#%d", class->name_version);
 727		if (class->subclass)
 728			printk(KERN_CONT "/%d", class->subclass);
 729		if (hlock && class->print_fn)
 730			class->print_fn(hlock->instance);
 731	}
 732}
 733
 734static void print_lock_name(struct held_lock *hlock, struct lock_class *class)
 735{
 736	char usage[LOCK_USAGE_CHARS];
 737
 738	get_usage_chars(class, usage);
 739
 740	printk(KERN_CONT " (");
 741	__print_lock_name(hlock, class);
 742	printk(KERN_CONT "){%s}-{%d:%d}", usage,
 743			class->wait_type_outer ?: class->wait_type_inner,
 744			class->wait_type_inner);
 745}
 746
 747static void print_lockdep_cache(struct lockdep_map *lock)
 748{
 749	const char *name;
 750	char str[KSYM_NAME_LEN];
 751
 752	name = lock->name;
 753	if (!name)
 754		name = __get_key_name(lock->key->subkeys, str);
 755
 756	printk(KERN_CONT "%s", name);
 757}
 758
 759static void print_lock(struct held_lock *hlock)
 760{
 761	/*
 762	 * We can be called locklessly through debug_show_all_locks() so be
 763	 * extra careful, the hlock might have been released and cleared.
 764	 *
 765	 * If this indeed happens, lets pretend it does not hurt to continue
 766	 * to print the lock unless the hlock class_idx does not point to a
 767	 * registered class. The rationale here is: since we don't attempt
 768	 * to distinguish whether we are in this situation, if it just
 769	 * happened we can't count on class_idx to tell either.
 770	 */
 771	struct lock_class *lock = hlock_class(hlock);
 772
 773	if (!lock) {
 774		printk(KERN_CONT "<RELEASED>\n");
 775		return;
 776	}
 777
 778	printk(KERN_CONT "%px", hlock->instance);
 779	print_lock_name(hlock, lock);
 780	printk(KERN_CONT ", at: %pS\n", (void *)hlock->acquire_ip);
 781}
 782
 783static void lockdep_print_held_locks(struct task_struct *p)
 784{
 785	int i, depth = READ_ONCE(p->lockdep_depth);
 786
 787	if (!depth)
 788		printk("no locks held by %s/%d.\n", p->comm, task_pid_nr(p));
 789	else
 790		printk("%d lock%s held by %s/%d:\n", depth,
 791		       str_plural(depth), p->comm, task_pid_nr(p));
 792	/*
 793	 * It's not reliable to print a task's held locks if it's not sleeping
 794	 * and it's not the current task.
 795	 */
 796	if (p != current && task_is_running(p))
 797		return;
 798	for (i = 0; i < depth; i++) {
 799		printk(" #%d: ", i);
 800		print_lock(p->held_locks + i);
 801	}
 802}
 803
 804static void print_kernel_ident(void)
 805{
 806	printk("%s %.*s %s\n", init_utsname()->release,
 807		(int)strcspn(init_utsname()->version, " "),
 808		init_utsname()->version,
 809		print_tainted());
 810}
 811
 812static int very_verbose(struct lock_class *class)
 813{
 814#if VERY_VERBOSE
 815	return class_filter(class);
 816#endif
 817	return 0;
 818}
 819
 820/*
 821 * Is this the address of a static object:
 822 */
 823#ifdef __KERNEL__
 824static int static_obj(const void *obj)
 825{
 826	unsigned long addr = (unsigned long) obj;
 827
 828	if (is_kernel_core_data(addr))
 829		return 1;
 830
 831	/*
 832	 * keys are allowed in the __ro_after_init section.
 833	 */
 834	if (is_kernel_rodata(addr))
 835		return 1;
 836
 837	/*
 838	 * in initdata section and used during bootup only?
 839	 * NOTE: On some platforms the initdata section is
 840	 * outside of the _stext ... _end range.
 841	 */
 842	if (system_state < SYSTEM_FREEING_INITMEM &&
 843		init_section_contains((void *)addr, 1))
 844		return 1;
 845
 846	/*
 847	 * in-kernel percpu var?
 848	 */
 849	if (is_kernel_percpu_address(addr))
 850		return 1;
 851
 852	/*
 853	 * module static or percpu var?
 854	 */
 855	return is_module_address(addr) || is_module_percpu_address(addr);
 856}
 857#endif
 858
 859/*
 860 * To make lock name printouts unique, we calculate a unique
 861 * class->name_version generation counter. The caller must hold the graph
 862 * lock.
 863 */
 864static int count_matching_names(struct lock_class *new_class)
 865{
 866	struct lock_class *class;
 867	int count = 0;
 868
 869	if (!new_class->name)
 870		return 0;
 871
 872	list_for_each_entry(class, &all_lock_classes, lock_entry) {
 873		if (new_class->key - new_class->subclass == class->key)
 874			return class->name_version;
 875		if (class->name && !strcmp(class->name, new_class->name))
 876			count = max(count, class->name_version);
 877	}
 878
 879	return count + 1;
 880}
 881
 882/* used from NMI context -- must be lockless */
 883static noinstr struct lock_class *
 884look_up_lock_class(const struct lockdep_map *lock, unsigned int subclass)
 885{
 886	struct lockdep_subclass_key *key;
 887	struct hlist_head *hash_head;
 888	struct lock_class *class;
 889
 890	if (unlikely(subclass >= MAX_LOCKDEP_SUBCLASSES)) {
 891		instrumentation_begin();
 892		debug_locks_off();
 893		nbcon_cpu_emergency_enter();
 894		printk(KERN_ERR
 895			"BUG: looking up invalid subclass: %u\n", subclass);
 896		printk(KERN_ERR
 897			"turning off the locking correctness validator.\n");
 898		dump_stack();
 899		nbcon_cpu_emergency_exit();
 900		instrumentation_end();
 901		return NULL;
 902	}
 903
 904	/*
 905	 * If it is not initialised then it has never been locked,
 906	 * so it won't be present in the hash table.
 907	 */
 908	if (unlikely(!lock->key))
 909		return NULL;
 910
 911	/*
 912	 * NOTE: the class-key must be unique. For dynamic locks, a static
 913	 * lock_class_key variable is passed in through the mutex_init()
 914	 * (or spin_lock_init()) call - which acts as the key. For static
 915	 * locks we use the lock object itself as the key.
 916	 */
 917	BUILD_BUG_ON(sizeof(struct lock_class_key) >
 918			sizeof(struct lockdep_map));
 919
 920	key = lock->key->subkeys + subclass;
 921
 922	hash_head = classhashentry(key);
 923
 924	/*
 925	 * We do an RCU walk of the hash, see lockdep_free_key_range().
 926	 */
 927	if (DEBUG_LOCKS_WARN_ON(!irqs_disabled()))
 928		return NULL;
 929
 930	hlist_for_each_entry_rcu_notrace(class, hash_head, hash_entry) {
 931		if (class->key == key) {
 932			/*
 933			 * Huh! same key, different name? Did someone trample
 934			 * on some memory? We're most confused.
 935			 */
 936			WARN_ONCE(class->name != lock->name &&
 937				  lock->key != &__lockdep_no_validate__,
 938				  "Looking for class \"%s\" with key %ps, but found a different class \"%s\" with the same key\n",
 939				  lock->name, lock->key, class->name);
 940			return class;
 941		}
 942	}
 943
 944	return NULL;
 945}
 946
 947/*
 948 * Static locks do not have their class-keys yet - for them the key is
 949 * the lock object itself. If the lock is in the per cpu area, the
 950 * canonical address of the lock (per cpu offset removed) is used.
 951 */
 952static bool assign_lock_key(struct lockdep_map *lock)
 953{
 954	unsigned long can_addr, addr = (unsigned long)lock;
 955
 956#ifdef __KERNEL__
 957	/*
 958	 * lockdep_free_key_range() assumes that struct lock_class_key
 959	 * objects do not overlap. Since we use the address of lock
 960	 * objects as class key for static objects, check whether the
 961	 * size of lock_class_key objects does not exceed the size of
 962	 * the smallest lock object.
 963	 */
 964	BUILD_BUG_ON(sizeof(struct lock_class_key) > sizeof(raw_spinlock_t));
 965#endif
 966
 967	if (__is_kernel_percpu_address(addr, &can_addr))
 968		lock->key = (void *)can_addr;
 969	else if (__is_module_percpu_address(addr, &can_addr))
 970		lock->key = (void *)can_addr;
 971	else if (static_obj(lock))
 972		lock->key = (void *)lock;
 973	else {
 974		/* Debug-check: all keys must be persistent! */
 975		debug_locks_off();
 976		nbcon_cpu_emergency_enter();
 977		pr_err("INFO: trying to register non-static key.\n");
 978		pr_err("The code is fine but needs lockdep annotation, or maybe\n");
 979		pr_err("you didn't initialize this object before use?\n");
 980		pr_err("turning off the locking correctness validator.\n");
 981		dump_stack();
 982		nbcon_cpu_emergency_exit();
 983		return false;
 984	}
 985
 986	return true;
 987}
 988
 989#ifdef CONFIG_DEBUG_LOCKDEP
 990
 991/* Check whether element @e occurs in list @h */
 992static bool in_list(struct list_head *e, struct list_head *h)
 993{
 994	struct list_head *f;
 995
 996	list_for_each(f, h) {
 997		if (e == f)
 998			return true;
 999	}
1000
1001	return false;
1002}
1003
1004/*
1005 * Check whether entry @e occurs in any of the locks_after or locks_before
1006 * lists.
1007 */
1008static bool in_any_class_list(struct list_head *e)
1009{
1010	struct lock_class *class;
1011	int i;
1012
1013	for (i = 0; i < ARRAY_SIZE(lock_classes); i++) {
1014		class = &lock_classes[i];
1015		if (in_list(e, &class->locks_after) ||
1016		    in_list(e, &class->locks_before))
1017			return true;
1018	}
1019	return false;
1020}
1021
1022static bool class_lock_list_valid(struct lock_class *c, struct list_head *h)
1023{
1024	struct lock_list *e;
1025
1026	list_for_each_entry(e, h, entry) {
1027		if (e->links_to != c) {
1028			printk(KERN_INFO "class %s: mismatch for lock entry %ld; class %s <> %s",
1029			       c->name ? : "(?)",
1030			       (unsigned long)(e - list_entries),
1031			       e->links_to && e->links_to->name ?
1032			       e->links_to->name : "(?)",
1033			       e->class && e->class->name ? e->class->name :
1034			       "(?)");
1035			return false;
1036		}
1037	}
1038	return true;
1039}
1040
1041#ifdef CONFIG_PROVE_LOCKING
1042static u16 chain_hlocks[MAX_LOCKDEP_CHAIN_HLOCKS];
1043#endif
1044
1045static bool check_lock_chain_key(struct lock_chain *chain)
1046{
1047#ifdef CONFIG_PROVE_LOCKING
1048	u64 chain_key = INITIAL_CHAIN_KEY;
1049	int i;
1050
1051	for (i = chain->base; i < chain->base + chain->depth; i++)
1052		chain_key = iterate_chain_key(chain_key, chain_hlocks[i]);
1053	/*
1054	 * The 'unsigned long long' casts avoid that a compiler warning
1055	 * is reported when building tools/lib/lockdep.
1056	 */
1057	if (chain->chain_key != chain_key) {
1058		printk(KERN_INFO "chain %lld: key %#llx <> %#llx\n",
1059		       (unsigned long long)(chain - lock_chains),
1060		       (unsigned long long)chain->chain_key,
1061		       (unsigned long long)chain_key);
1062		return false;
1063	}
1064#endif
1065	return true;
1066}
1067
1068static bool in_any_zapped_class_list(struct lock_class *class)
1069{
1070	struct pending_free *pf;
1071	int i;
1072
1073	for (i = 0, pf = delayed_free.pf; i < ARRAY_SIZE(delayed_free.pf); i++, pf++) {
1074		if (in_list(&class->lock_entry, &pf->zapped))
1075			return true;
1076	}
1077
1078	return false;
1079}
1080
1081static bool __check_data_structures(void)
1082{
1083	struct lock_class *class;
1084	struct lock_chain *chain;
1085	struct hlist_head *head;
1086	struct lock_list *e;
1087	int i;
1088
1089	/* Check whether all classes occur in a lock list. */
1090	for (i = 0; i < ARRAY_SIZE(lock_classes); i++) {
1091		class = &lock_classes[i];
1092		if (!in_list(&class->lock_entry, &all_lock_classes) &&
1093		    !in_list(&class->lock_entry, &free_lock_classes) &&
1094		    !in_any_zapped_class_list(class)) {
1095			printk(KERN_INFO "class %px/%s is not in any class list\n",
1096			       class, class->name ? : "(?)");
1097			return false;
1098		}
1099	}
1100
1101	/* Check whether all classes have valid lock lists. */
1102	for (i = 0; i < ARRAY_SIZE(lock_classes); i++) {
1103		class = &lock_classes[i];
1104		if (!class_lock_list_valid(class, &class->locks_before))
1105			return false;
1106		if (!class_lock_list_valid(class, &class->locks_after))
1107			return false;
1108	}
1109
1110	/* Check the chain_key of all lock chains. */
1111	for (i = 0; i < ARRAY_SIZE(chainhash_table); i++) {
1112		head = chainhash_table + i;
1113		hlist_for_each_entry_rcu(chain, head, entry) {
1114			if (!check_lock_chain_key(chain))
1115				return false;
1116		}
1117	}
1118
1119	/*
1120	 * Check whether all list entries that are in use occur in a class
1121	 * lock list.
1122	 */
1123	for_each_set_bit(i, list_entries_in_use, ARRAY_SIZE(list_entries)) {
1124		e = list_entries + i;
1125		if (!in_any_class_list(&e->entry)) {
1126			printk(KERN_INFO "list entry %d is not in any class list; class %s <> %s\n",
1127			       (unsigned int)(e - list_entries),
1128			       e->class->name ? : "(?)",
1129			       e->links_to->name ? : "(?)");
1130			return false;
1131		}
1132	}
1133
1134	/*
1135	 * Check whether all list entries that are not in use do not occur in
1136	 * a class lock list.
1137	 */
1138	for_each_clear_bit(i, list_entries_in_use, ARRAY_SIZE(list_entries)) {
1139		e = list_entries + i;
1140		if (in_any_class_list(&e->entry)) {
1141			printk(KERN_INFO "list entry %d occurs in a class list; class %s <> %s\n",
1142			       (unsigned int)(e - list_entries),
1143			       e->class && e->class->name ? e->class->name :
1144			       "(?)",
1145			       e->links_to && e->links_to->name ?
1146			       e->links_to->name : "(?)");
1147			return false;
1148		}
1149	}
1150
1151	return true;
1152}
1153
1154int check_consistency = 0;
1155module_param(check_consistency, int, 0644);
1156
1157static void check_data_structures(void)
1158{
1159	static bool once = false;
1160
1161	if (check_consistency && !once) {
1162		if (!__check_data_structures()) {
1163			once = true;
1164			WARN_ON(once);
1165		}
1166	}
1167}
1168
1169#else /* CONFIG_DEBUG_LOCKDEP */
1170
1171static inline void check_data_structures(void) { }
1172
1173#endif /* CONFIG_DEBUG_LOCKDEP */
1174
1175static void init_chain_block_buckets(void);
1176
1177/*
1178 * Initialize the lock_classes[] array elements, the free_lock_classes list
1179 * and also the delayed_free structure.
1180 */
1181static void init_data_structures_once(void)
1182{
1183	static bool __read_mostly ds_initialized, rcu_head_initialized;
1184	int i;
1185
1186	if (likely(rcu_head_initialized))
1187		return;
1188
1189	if (system_state >= SYSTEM_SCHEDULING) {
1190		init_rcu_head(&delayed_free.rcu_head);
1191		rcu_head_initialized = true;
1192	}
1193
1194	if (ds_initialized)
1195		return;
1196
1197	ds_initialized = true;
1198
1199	INIT_LIST_HEAD(&delayed_free.pf[0].zapped);
1200	INIT_LIST_HEAD(&delayed_free.pf[1].zapped);
1201
1202	for (i = 0; i < ARRAY_SIZE(lock_classes); i++) {
1203		list_add_tail(&lock_classes[i].lock_entry, &free_lock_classes);
1204		INIT_LIST_HEAD(&lock_classes[i].locks_after);
1205		INIT_LIST_HEAD(&lock_classes[i].locks_before);
1206	}
1207	init_chain_block_buckets();
1208}
1209
1210static inline struct hlist_head *keyhashentry(const struct lock_class_key *key)
1211{
1212	unsigned long hash = hash_long((uintptr_t)key, KEYHASH_BITS);
1213
1214	return lock_keys_hash + hash;
1215}
1216
1217/* Register a dynamically allocated key. */
1218void lockdep_register_key(struct lock_class_key *key)
1219{
1220	struct hlist_head *hash_head;
1221	struct lock_class_key *k;
1222	unsigned long flags;
1223
1224	if (WARN_ON_ONCE(static_obj(key)))
1225		return;
1226	hash_head = keyhashentry(key);
1227
1228	raw_local_irq_save(flags);
1229	if (!graph_lock())
1230		goto restore_irqs;
1231	hlist_for_each_entry_rcu(k, hash_head, hash_entry) {
1232		if (WARN_ON_ONCE(k == key))
1233			goto out_unlock;
1234	}
1235	hlist_add_head_rcu(&key->hash_entry, hash_head);
1236out_unlock:
1237	graph_unlock();
1238restore_irqs:
1239	raw_local_irq_restore(flags);
1240}
1241EXPORT_SYMBOL_GPL(lockdep_register_key);
1242
1243/* Check whether a key has been registered as a dynamic key. */
1244static bool is_dynamic_key(const struct lock_class_key *key)
1245{
1246	struct hlist_head *hash_head;
1247	struct lock_class_key *k;
1248	bool found = false;
1249
1250	if (WARN_ON_ONCE(static_obj(key)))
1251		return false;
1252
1253	/*
1254	 * If lock debugging is disabled lock_keys_hash[] may contain
1255	 * pointers to memory that has already been freed. Avoid triggering
1256	 * a use-after-free in that case by returning early.
1257	 */
1258	if (!debug_locks)
1259		return true;
1260
1261	hash_head = keyhashentry(key);
1262
1263	rcu_read_lock();
1264	hlist_for_each_entry_rcu(k, hash_head, hash_entry) {
1265		if (k == key) {
1266			found = true;
1267			break;
1268		}
1269	}
1270	rcu_read_unlock();
1271
1272	return found;
1273}
1274
1275/*
1276 * Register a lock's class in the hash-table, if the class is not present
1277 * yet. Otherwise we look it up. We cache the result in the lock object
1278 * itself, so actual lookup of the hash should be once per lock object.
1279 */
1280static struct lock_class *
1281register_lock_class(struct lockdep_map *lock, unsigned int subclass, int force)
1282{
1283	struct lockdep_subclass_key *key;
1284	struct hlist_head *hash_head;
1285	struct lock_class *class;
1286	int idx;
1287
1288	DEBUG_LOCKS_WARN_ON(!irqs_disabled());
1289
1290	class = look_up_lock_class(lock, subclass);
1291	if (likely(class))
1292		goto out_set_class_cache;
1293
1294	if (!lock->key) {
1295		if (!assign_lock_key(lock))
1296			return NULL;
1297	} else if (!static_obj(lock->key) && !is_dynamic_key(lock->key)) {
1298		return NULL;
1299	}
1300
1301	key = lock->key->subkeys + subclass;
1302	hash_head = classhashentry(key);
1303
1304	if (!graph_lock()) {
1305		return NULL;
1306	}
1307	/*
1308	 * We have to do the hash-walk again, to avoid races
1309	 * with another CPU:
1310	 */
1311	hlist_for_each_entry_rcu(class, hash_head, hash_entry) {
1312		if (class->key == key)
1313			goto out_unlock_set;
1314	}
1315
1316	init_data_structures_once();
1317
1318	/* Allocate a new lock class and add it to the hash. */
1319	class = list_first_entry_or_null(&free_lock_classes, typeof(*class),
1320					 lock_entry);
1321	if (!class) {
1322		if (!debug_locks_off_graph_unlock()) {
1323			return NULL;
1324		}
1325
1326		nbcon_cpu_emergency_enter();
1327		print_lockdep_off("BUG: MAX_LOCKDEP_KEYS too low!");
1328		dump_stack();
1329		nbcon_cpu_emergency_exit();
1330		return NULL;
1331	}
1332	nr_lock_classes++;
1333	__set_bit(class - lock_classes, lock_classes_in_use);
1334	debug_atomic_inc(nr_unused_locks);
1335	class->key = key;
1336	class->name = lock->name;
1337	class->subclass = subclass;
1338	WARN_ON_ONCE(!list_empty(&class->locks_before));
1339	WARN_ON_ONCE(!list_empty(&class->locks_after));
1340	class->name_version = count_matching_names(class);
1341	class->wait_type_inner = lock->wait_type_inner;
1342	class->wait_type_outer = lock->wait_type_outer;
1343	class->lock_type = lock->lock_type;
1344	/*
1345	 * We use RCU's safe list-add method to make
1346	 * parallel walking of the hash-list safe:
1347	 */
1348	hlist_add_head_rcu(&class->hash_entry, hash_head);
1349	/*
1350	 * Remove the class from the free list and add it to the global list
1351	 * of classes.
1352	 */
1353	list_move_tail(&class->lock_entry, &all_lock_classes);
1354	idx = class - lock_classes;
1355	if (idx > max_lock_class_idx)
1356		max_lock_class_idx = idx;
1357
1358	if (verbose(class)) {
1359		graph_unlock();
1360
1361		nbcon_cpu_emergency_enter();
1362		printk("\nnew class %px: %s", class->key, class->name);
1363		if (class->name_version > 1)
1364			printk(KERN_CONT "#%d", class->name_version);
1365		printk(KERN_CONT "\n");
1366		dump_stack();
1367		nbcon_cpu_emergency_exit();
1368
1369		if (!graph_lock()) {
1370			return NULL;
1371		}
1372	}
1373out_unlock_set:
1374	graph_unlock();
1375
1376out_set_class_cache:
1377	if (!subclass || force)
1378		lock->class_cache[0] = class;
1379	else if (subclass < NR_LOCKDEP_CACHING_CLASSES)
1380		lock->class_cache[subclass] = class;
1381
1382	/*
1383	 * Hash collision, did we smoke some? We found a class with a matching
1384	 * hash but the subclass -- which is hashed in -- didn't match.
1385	 */
1386	if (DEBUG_LOCKS_WARN_ON(class->subclass != subclass))
1387		return NULL;
1388
1389	return class;
1390}
1391
1392#ifdef CONFIG_PROVE_LOCKING
1393/*
1394 * Allocate a lockdep entry. (assumes the graph_lock held, returns
1395 * with NULL on failure)
1396 */
1397static struct lock_list *alloc_list_entry(void)
1398{
1399	int idx = find_first_zero_bit(list_entries_in_use,
1400				      ARRAY_SIZE(list_entries));
1401
1402	if (idx >= ARRAY_SIZE(list_entries)) {
1403		if (!debug_locks_off_graph_unlock())
1404			return NULL;
1405
1406		nbcon_cpu_emergency_enter();
1407		print_lockdep_off("BUG: MAX_LOCKDEP_ENTRIES too low!");
1408		dump_stack();
1409		nbcon_cpu_emergency_exit();
1410		return NULL;
1411	}
1412	nr_list_entries++;
1413	__set_bit(idx, list_entries_in_use);
1414	return list_entries + idx;
1415}
1416
1417/*
1418 * Add a new dependency to the head of the list:
1419 */
1420static int add_lock_to_list(struct lock_class *this,
1421			    struct lock_class *links_to, struct list_head *head,
1422			    u16 distance, u8 dep,
1423			    const struct lock_trace *trace)
1424{
1425	struct lock_list *entry;
1426	/*
1427	 * Lock not present yet - get a new dependency struct and
1428	 * add it to the list:
1429	 */
1430	entry = alloc_list_entry();
1431	if (!entry)
1432		return 0;
1433
1434	entry->class = this;
1435	entry->links_to = links_to;
1436	entry->dep = dep;
1437	entry->distance = distance;
1438	entry->trace = trace;
1439	/*
1440	 * Both allocation and removal are done under the graph lock; but
1441	 * iteration is under RCU-sched; see look_up_lock_class() and
1442	 * lockdep_free_key_range().
1443	 */
1444	list_add_tail_rcu(&entry->entry, head);
1445
1446	return 1;
1447}
1448
1449/*
1450 * For good efficiency of modular, we use power of 2
1451 */
1452#define MAX_CIRCULAR_QUEUE_SIZE		(1UL << CONFIG_LOCKDEP_CIRCULAR_QUEUE_BITS)
1453#define CQ_MASK				(MAX_CIRCULAR_QUEUE_SIZE-1)
1454
1455/*
1456 * The circular_queue and helpers are used to implement graph
1457 * breadth-first search (BFS) algorithm, by which we can determine
1458 * whether there is a path from a lock to another. In deadlock checks,
1459 * a path from the next lock to be acquired to a previous held lock
1460 * indicates that adding the <prev> -> <next> lock dependency will
1461 * produce a circle in the graph. Breadth-first search instead of
1462 * depth-first search is used in order to find the shortest (circular)
1463 * path.
1464 */
1465struct circular_queue {
1466	struct lock_list *element[MAX_CIRCULAR_QUEUE_SIZE];
1467	unsigned int  front, rear;
1468};
1469
1470static struct circular_queue lock_cq;
1471
1472unsigned int max_bfs_queue_depth;
1473
1474static unsigned int lockdep_dependency_gen_id;
1475
1476static inline void __cq_init(struct circular_queue *cq)
1477{
1478	cq->front = cq->rear = 0;
1479	lockdep_dependency_gen_id++;
1480}
1481
1482static inline int __cq_empty(struct circular_queue *cq)
1483{
1484	return (cq->front == cq->rear);
1485}
1486
1487static inline int __cq_full(struct circular_queue *cq)
1488{
1489	return ((cq->rear + 1) & CQ_MASK) == cq->front;
1490}
1491
1492static inline int __cq_enqueue(struct circular_queue *cq, struct lock_list *elem)
1493{
1494	if (__cq_full(cq))
1495		return -1;
1496
1497	cq->element[cq->rear] = elem;
1498	cq->rear = (cq->rear + 1) & CQ_MASK;
1499	return 0;
1500}
1501
1502/*
1503 * Dequeue an element from the circular_queue, return a lock_list if
1504 * the queue is not empty, or NULL if otherwise.
1505 */
1506static inline struct lock_list * __cq_dequeue(struct circular_queue *cq)
1507{
1508	struct lock_list * lock;
1509
1510	if (__cq_empty(cq))
1511		return NULL;
1512
1513	lock = cq->element[cq->front];
1514	cq->front = (cq->front + 1) & CQ_MASK;
1515
1516	return lock;
1517}
1518
1519static inline unsigned int  __cq_get_elem_count(struct circular_queue *cq)
1520{
1521	return (cq->rear - cq->front) & CQ_MASK;
1522}
1523
1524static inline void mark_lock_accessed(struct lock_list *lock)
1525{
1526	lock->class->dep_gen_id = lockdep_dependency_gen_id;
1527}
1528
1529static inline void visit_lock_entry(struct lock_list *lock,
1530				    struct lock_list *parent)
1531{
1532	lock->parent = parent;
1533}
1534
1535static inline unsigned long lock_accessed(struct lock_list *lock)
1536{
1537	return lock->class->dep_gen_id == lockdep_dependency_gen_id;
1538}
1539
1540static inline struct lock_list *get_lock_parent(struct lock_list *child)
1541{
1542	return child->parent;
1543}
1544
1545static inline int get_lock_depth(struct lock_list *child)
1546{
1547	int depth = 0;
1548	struct lock_list *parent;
1549
1550	while ((parent = get_lock_parent(child))) {
1551		child = parent;
1552		depth++;
1553	}
1554	return depth;
1555}
1556
1557/*
1558 * Return the forward or backward dependency list.
1559 *
1560 * @lock:   the lock_list to get its class's dependency list
1561 * @offset: the offset to struct lock_class to determine whether it is
1562 *          locks_after or locks_before
1563 */
1564static inline struct list_head *get_dep_list(struct lock_list *lock, int offset)
1565{
1566	void *lock_class = lock->class;
1567
1568	return lock_class + offset;
1569}
1570/*
1571 * Return values of a bfs search:
1572 *
1573 * BFS_E* indicates an error
1574 * BFS_R* indicates a result (match or not)
1575 *
1576 * BFS_EINVALIDNODE: Find a invalid node in the graph.
1577 *
1578 * BFS_EQUEUEFULL: The queue is full while doing the bfs.
1579 *
1580 * BFS_RMATCH: Find the matched node in the graph, and put that node into
1581 *             *@target_entry.
1582 *
1583 * BFS_RNOMATCH: Haven't found the matched node and keep *@target_entry
1584 *               _unchanged_.
1585 */
1586enum bfs_result {
1587	BFS_EINVALIDNODE = -2,
1588	BFS_EQUEUEFULL = -1,
1589	BFS_RMATCH = 0,
1590	BFS_RNOMATCH = 1,
1591};
1592
1593/*
1594 * bfs_result < 0 means error
1595 */
1596static inline bool bfs_error(enum bfs_result res)
1597{
1598	return res < 0;
1599}
1600
1601/*
1602 * DEP_*_BIT in lock_list::dep
1603 *
1604 * For dependency @prev -> @next:
1605 *
1606 *   SR: @prev is shared reader (->read != 0) and @next is recursive reader
1607 *       (->read == 2)
1608 *   ER: @prev is exclusive locker (->read == 0) and @next is recursive reader
1609 *   SN: @prev is shared reader and @next is non-recursive locker (->read != 2)
1610 *   EN: @prev is exclusive locker and @next is non-recursive locker
1611 *
1612 * Note that we define the value of DEP_*_BITs so that:
1613 *   bit0 is prev->read == 0
1614 *   bit1 is next->read != 2
1615 */
1616#define DEP_SR_BIT (0 + (0 << 1)) /* 0 */
1617#define DEP_ER_BIT (1 + (0 << 1)) /* 1 */
1618#define DEP_SN_BIT (0 + (1 << 1)) /* 2 */
1619#define DEP_EN_BIT (1 + (1 << 1)) /* 3 */
1620
1621#define DEP_SR_MASK (1U << (DEP_SR_BIT))
1622#define DEP_ER_MASK (1U << (DEP_ER_BIT))
1623#define DEP_SN_MASK (1U << (DEP_SN_BIT))
1624#define DEP_EN_MASK (1U << (DEP_EN_BIT))
1625
1626static inline unsigned int
1627__calc_dep_bit(struct held_lock *prev, struct held_lock *next)
1628{
1629	return (prev->read == 0) + ((next->read != 2) << 1);
1630}
1631
1632static inline u8 calc_dep(struct held_lock *prev, struct held_lock *next)
1633{
1634	return 1U << __calc_dep_bit(prev, next);
1635}
1636
1637/*
1638 * calculate the dep_bit for backwards edges. We care about whether @prev is
1639 * shared and whether @next is recursive.
1640 */
1641static inline unsigned int
1642__calc_dep_bitb(struct held_lock *prev, struct held_lock *next)
1643{
1644	return (next->read != 2) + ((prev->read == 0) << 1);
1645}
1646
1647static inline u8 calc_depb(struct held_lock *prev, struct held_lock *next)
1648{
1649	return 1U << __calc_dep_bitb(prev, next);
1650}
1651
1652/*
1653 * Initialize a lock_list entry @lock belonging to @class as the root for a BFS
1654 * search.
1655 */
1656static inline void __bfs_init_root(struct lock_list *lock,
1657				   struct lock_class *class)
1658{
1659	lock->class = class;
1660	lock->parent = NULL;
1661	lock->only_xr = 0;
1662}
1663
1664/*
1665 * Initialize a lock_list entry @lock based on a lock acquisition @hlock as the
1666 * root for a BFS search.
1667 *
1668 * ->only_xr of the initial lock node is set to @hlock->read == 2, to make sure
1669 * that <prev> -> @hlock and @hlock -> <whatever __bfs() found> is not -(*R)->
1670 * and -(S*)->.
1671 */
1672static inline void bfs_init_root(struct lock_list *lock,
1673				 struct held_lock *hlock)
1674{
1675	__bfs_init_root(lock, hlock_class(hlock));
1676	lock->only_xr = (hlock->read == 2);
1677}
1678
1679/*
1680 * Similar to bfs_init_root() but initialize the root for backwards BFS.
1681 *
1682 * ->only_xr of the initial lock node is set to @hlock->read != 0, to make sure
1683 * that <next> -> @hlock and @hlock -> <whatever backwards BFS found> is not
1684 * -(*S)-> and -(R*)-> (reverse order of -(*R)-> and -(S*)->).
1685 */
1686static inline void bfs_init_rootb(struct lock_list *lock,
1687				  struct held_lock *hlock)
1688{
1689	__bfs_init_root(lock, hlock_class(hlock));
1690	lock->only_xr = (hlock->read != 0);
1691}
1692
1693static inline struct lock_list *__bfs_next(struct lock_list *lock, int offset)
1694{
1695	if (!lock || !lock->parent)
1696		return NULL;
1697
1698	return list_next_or_null_rcu(get_dep_list(lock->parent, offset),
1699				     &lock->entry, struct lock_list, entry);
1700}
1701
1702/*
1703 * Breadth-First Search to find a strong path in the dependency graph.
1704 *
1705 * @source_entry: the source of the path we are searching for.
1706 * @data: data used for the second parameter of @match function
1707 * @match: match function for the search
1708 * @target_entry: pointer to the target of a matched path
1709 * @offset: the offset to struct lock_class to determine whether it is
1710 *          locks_after or locks_before
1711 *
1712 * We may have multiple edges (considering different kinds of dependencies,
1713 * e.g. ER and SN) between two nodes in the dependency graph. But
1714 * only the strong dependency path in the graph is relevant to deadlocks. A
1715 * strong dependency path is a dependency path that doesn't have two adjacent
1716 * dependencies as -(*R)-> -(S*)->, please see:
1717 *
1718 *         Documentation/locking/lockdep-design.rst
1719 *
1720 * for more explanation of the definition of strong dependency paths
1721 *
1722 * In __bfs(), we only traverse in the strong dependency path:
1723 *
1724 *     In lock_list::only_xr, we record whether the previous dependency only
1725 *     has -(*R)-> in the search, and if it does (prev only has -(*R)->), we
1726 *     filter out any -(S*)-> in the current dependency and after that, the
1727 *     ->only_xr is set according to whether we only have -(*R)-> left.
1728 */
1729static enum bfs_result __bfs(struct lock_list *source_entry,
1730			     void *data,
1731			     bool (*match)(struct lock_list *entry, void *data),
1732			     bool (*skip)(struct lock_list *entry, void *data),
1733			     struct lock_list **target_entry,
1734			     int offset)
1735{
1736	struct circular_queue *cq = &lock_cq;
1737	struct lock_list *lock = NULL;
1738	struct lock_list *entry;
1739	struct list_head *head;
1740	unsigned int cq_depth;
1741	bool first;
1742
1743	lockdep_assert_locked();
1744
1745	__cq_init(cq);
1746	__cq_enqueue(cq, source_entry);
1747
1748	while ((lock = __bfs_next(lock, offset)) || (lock = __cq_dequeue(cq))) {
1749		if (!lock->class)
1750			return BFS_EINVALIDNODE;
1751
1752		/*
1753		 * Step 1: check whether we already finish on this one.
1754		 *
1755		 * If we have visited all the dependencies from this @lock to
1756		 * others (iow, if we have visited all lock_list entries in
1757		 * @lock->class->locks_{after,before}) we skip, otherwise go
1758		 * and visit all the dependencies in the list and mark this
1759		 * list accessed.
1760		 */
1761		if (lock_accessed(lock))
1762			continue;
1763		else
1764			mark_lock_accessed(lock);
1765
1766		/*
1767		 * Step 2: check whether prev dependency and this form a strong
1768		 *         dependency path.
1769		 */
1770		if (lock->parent) { /* Parent exists, check prev dependency */
1771			u8 dep = lock->dep;
1772			bool prev_only_xr = lock->parent->only_xr;
1773
1774			/*
1775			 * Mask out all -(S*)-> if we only have *R in previous
1776			 * step, because -(*R)-> -(S*)-> don't make up a strong
1777			 * dependency.
1778			 */
1779			if (prev_only_xr)
1780				dep &= ~(DEP_SR_MASK | DEP_SN_MASK);
1781
1782			/* If nothing left, we skip */
1783			if (!dep)
1784				continue;
1785
1786			/* If there are only -(*R)-> left, set that for the next step */
1787			lock->only_xr = !(dep & (DEP_SN_MASK | DEP_EN_MASK));
1788		}
1789
1790		/*
1791		 * Step 3: we haven't visited this and there is a strong
1792		 *         dependency path to this, so check with @match.
1793		 *         If @skip is provide and returns true, we skip this
1794		 *         lock (and any path this lock is in).
1795		 */
1796		if (skip && skip(lock, data))
1797			continue;
1798
1799		if (match(lock, data)) {
1800			*target_entry = lock;
1801			return BFS_RMATCH;
1802		}
1803
1804		/*
1805		 * Step 4: if not match, expand the path by adding the
1806		 *         forward or backwards dependencies in the search
1807		 *
1808		 */
1809		first = true;
1810		head = get_dep_list(lock, offset);
1811		list_for_each_entry_rcu(entry, head, entry) {
1812			visit_lock_entry(entry, lock);
1813
1814			/*
1815			 * Note we only enqueue the first of the list into the
1816			 * queue, because we can always find a sibling
1817			 * dependency from one (see __bfs_next()), as a result
1818			 * the space of queue is saved.
1819			 */
1820			if (!first)
1821				continue;
1822
1823			first = false;
1824
1825			if (__cq_enqueue(cq, entry))
1826				return BFS_EQUEUEFULL;
1827
1828			cq_depth = __cq_get_elem_count(cq);
1829			if (max_bfs_queue_depth < cq_depth)
1830				max_bfs_queue_depth = cq_depth;
1831		}
1832	}
1833
1834	return BFS_RNOMATCH;
1835}
1836
1837static inline enum bfs_result
1838__bfs_forwards(struct lock_list *src_entry,
1839	       void *data,
1840	       bool (*match)(struct lock_list *entry, void *data),
1841	       bool (*skip)(struct lock_list *entry, void *data),
1842	       struct lock_list **target_entry)
1843{
1844	return __bfs(src_entry, data, match, skip, target_entry,
1845		     offsetof(struct lock_class, locks_after));
1846
1847}
1848
1849static inline enum bfs_result
1850__bfs_backwards(struct lock_list *src_entry,
1851		void *data,
1852		bool (*match)(struct lock_list *entry, void *data),
1853	       bool (*skip)(struct lock_list *entry, void *data),
1854		struct lock_list **target_entry)
1855{
1856	return __bfs(src_entry, data, match, skip, target_entry,
1857		     offsetof(struct lock_class, locks_before));
1858
1859}
1860
1861static void print_lock_trace(const struct lock_trace *trace,
1862			     unsigned int spaces)
1863{
1864	stack_trace_print(trace->entries, trace->nr_entries, spaces);
1865}
1866
1867/*
1868 * Print a dependency chain entry (this is only done when a deadlock
1869 * has been detected):
1870 */
1871static noinline void
1872print_circular_bug_entry(struct lock_list *target, int depth)
1873{
1874	if (debug_locks_silent)
1875		return;
1876	printk("\n-> #%u", depth);
1877	print_lock_name(NULL, target->class);
1878	printk(KERN_CONT ":\n");
1879	print_lock_trace(target->trace, 6);
1880}
1881
1882static void
1883print_circular_lock_scenario(struct held_lock *src,
1884			     struct held_lock *tgt,
1885			     struct lock_list *prt)
1886{
1887	struct lock_class *source = hlock_class(src);
1888	struct lock_class *target = hlock_class(tgt);
1889	struct lock_class *parent = prt->class;
1890	int src_read = src->read;
1891	int tgt_read = tgt->read;
1892
1893	/*
1894	 * A direct locking problem where unsafe_class lock is taken
1895	 * directly by safe_class lock, then all we need to show
1896	 * is the deadlock scenario, as it is obvious that the
1897	 * unsafe lock is taken under the safe lock.
1898	 *
1899	 * But if there is a chain instead, where the safe lock takes
1900	 * an intermediate lock (middle_class) where this lock is
1901	 * not the same as the safe lock, then the lock chain is
1902	 * used to describe the problem. Otherwise we would need
1903	 * to show a different CPU case for each link in the chain
1904	 * from the safe_class lock to the unsafe_class lock.
1905	 */
1906	if (parent != source) {
1907		printk("Chain exists of:\n  ");
1908		__print_lock_name(src, source);
1909		printk(KERN_CONT " --> ");
1910		__print_lock_name(NULL, parent);
1911		printk(KERN_CONT " --> ");
1912		__print_lock_name(tgt, target);
1913		printk(KERN_CONT "\n\n");
1914	}
1915
1916	printk(" Possible unsafe locking scenario:\n\n");
1917	printk("       CPU0                    CPU1\n");
1918	printk("       ----                    ----\n");
1919	if (tgt_read != 0)
1920		printk("  rlock(");
1921	else
1922		printk("  lock(");
1923	__print_lock_name(tgt, target);
1924	printk(KERN_CONT ");\n");
1925	printk("                               lock(");
1926	__print_lock_name(NULL, parent);
1927	printk(KERN_CONT ");\n");
1928	printk("                               lock(");
1929	__print_lock_name(tgt, target);
1930	printk(KERN_CONT ");\n");
1931	if (src_read != 0)
1932		printk("  rlock(");
1933	else if (src->sync)
1934		printk("  sync(");
1935	else
1936		printk("  lock(");
1937	__print_lock_name(src, source);
1938	printk(KERN_CONT ");\n");
1939	printk("\n *** DEADLOCK ***\n\n");
1940}
1941
1942/*
1943 * When a circular dependency is detected, print the
1944 * header first:
1945 */
1946static noinline void
1947print_circular_bug_header(struct lock_list *entry, unsigned int depth,
1948			struct held_lock *check_src,
1949			struct held_lock *check_tgt)
1950{
1951	struct task_struct *curr = current;
1952
1953	if (debug_locks_silent)
1954		return;
1955
1956	pr_warn("\n");
1957	pr_warn("======================================================\n");
1958	pr_warn("WARNING: possible circular locking dependency detected\n");
1959	print_kernel_ident();
1960	pr_warn("------------------------------------------------------\n");
1961	pr_warn("%s/%d is trying to acquire lock:\n",
1962		curr->comm, task_pid_nr(curr));
1963	print_lock(check_src);
1964
1965	pr_warn("\nbut task is already holding lock:\n");
1966
1967	print_lock(check_tgt);
1968	pr_warn("\nwhich lock already depends on the new lock.\n\n");
1969	pr_warn("\nthe existing dependency chain (in reverse order) is:\n");
1970
1971	print_circular_bug_entry(entry, depth);
1972}
1973
1974/*
1975 * We are about to add A -> B into the dependency graph, and in __bfs() a
1976 * strong dependency path A -> .. -> B is found: hlock_class equals
1977 * entry->class.
1978 *
1979 * If A -> .. -> B can replace A -> B in any __bfs() search (means the former
1980 * is _stronger_ than or equal to the latter), we consider A -> B as redundant.
1981 * For example if A -> .. -> B is -(EN)-> (i.e. A -(E*)-> .. -(*N)-> B), and A
1982 * -> B is -(ER)-> or -(EN)->, then we don't need to add A -> B into the
1983 * dependency graph, as any strong path ..-> A -> B ->.. we can get with
1984 * having dependency A -> B, we could already get a equivalent path ..-> A ->
1985 * .. -> B -> .. with A -> .. -> B. Therefore A -> B is redundant.
1986 *
1987 * We need to make sure both the start and the end of A -> .. -> B is not
1988 * weaker than A -> B. For the start part, please see the comment in
1989 * check_redundant(). For the end part, we need:
1990 *
1991 * Either
1992 *
1993 *     a) A -> B is -(*R)-> (everything is not weaker than that)
1994 *
1995 * or
1996 *
1997 *     b) A -> .. -> B is -(*N)-> (nothing is stronger than this)
1998 *
1999 */
2000static inline bool hlock_equal(struct lock_list *entry, void *data)
2001{
2002	struct held_lock *hlock = (struct held_lock *)data;
2003
2004	return hlock_class(hlock) == entry->class && /* Found A -> .. -> B */
2005	       (hlock->read == 2 ||  /* A -> B is -(*R)-> */
2006		!entry->only_xr); /* A -> .. -> B is -(*N)-> */
2007}
2008
2009/*
2010 * We are about to add B -> A into the dependency graph, and in __bfs() a
2011 * strong dependency path A -> .. -> B is found: hlock_class equals
2012 * entry->class.
2013 *
2014 * We will have a deadlock case (conflict) if A -> .. -> B -> A is a strong
2015 * dependency cycle, that means:
2016 *
2017 * Either
2018 *
2019 *     a) B -> A is -(E*)->
2020 *
2021 * or
2022 *
2023 *     b) A -> .. -> B is -(*N)-> (i.e. A -> .. -(*N)-> B)
2024 *
2025 * as then we don't have -(*R)-> -(S*)-> in the cycle.
2026 */
2027static inline bool hlock_conflict(struct lock_list *entry, void *data)
2028{
2029	struct held_lock *hlock = (struct held_lock *)data;
2030
2031	return hlock_class(hlock) == entry->class && /* Found A -> .. -> B */
2032	       (hlock->read == 0 || /* B -> A is -(E*)-> */
2033		!entry->only_xr); /* A -> .. -> B is -(*N)-> */
2034}
2035
2036static noinline void print_circular_bug(struct lock_list *this,
2037				struct lock_list *target,
2038				struct held_lock *check_src,
2039				struct held_lock *check_tgt)
2040{
2041	struct task_struct *curr = current;
2042	struct lock_list *parent;
2043	struct lock_list *first_parent;
2044	int depth;
2045
2046	if (!debug_locks_off_graph_unlock() || debug_locks_silent)
2047		return;
2048
2049	this->trace = save_trace();
2050	if (!this->trace)
2051		return;
2052
2053	depth = get_lock_depth(target);
2054
2055	nbcon_cpu_emergency_enter();
2056
2057	print_circular_bug_header(target, depth, check_src, check_tgt);
2058
2059	parent = get_lock_parent(target);
2060	first_parent = parent;
2061
2062	while (parent) {
2063		print_circular_bug_entry(parent, --depth);
2064		parent = get_lock_parent(parent);
2065	}
2066
2067	printk("\nother info that might help us debug this:\n\n");
2068	print_circular_lock_scenario(check_src, check_tgt,
2069				     first_parent);
2070
2071	lockdep_print_held_locks(curr);
2072
2073	printk("\nstack backtrace:\n");
2074	dump_stack();
2075
2076	nbcon_cpu_emergency_exit();
2077}
2078
2079static noinline void print_bfs_bug(int ret)
2080{
2081	if (!debug_locks_off_graph_unlock())
2082		return;
2083
2084	/*
2085	 * Breadth-first-search failed, graph got corrupted?
2086	 */
2087	if (ret == BFS_EQUEUEFULL)
2088		pr_warn("Increase LOCKDEP_CIRCULAR_QUEUE_BITS to avoid this warning:\n");
2089
2090	WARN(1, "lockdep bfs error:%d\n", ret);
2091}
2092
2093static bool noop_count(struct lock_list *entry, void *data)
2094{
2095	(*(unsigned long *)data)++;
2096	return false;
2097}
2098
2099static unsigned long __lockdep_count_forward_deps(struct lock_list *this)
2100{
2101	unsigned long  count = 0;
2102	struct lock_list *target_entry;
2103
2104	__bfs_forwards(this, (void *)&count, noop_count, NULL, &target_entry);
2105
2106	return count;
2107}
2108unsigned long lockdep_count_forward_deps(struct lock_class *class)
2109{
2110	unsigned long ret, flags;
2111	struct lock_list this;
2112
2113	__bfs_init_root(&this, class);
2114
2115	raw_local_irq_save(flags);
2116	lockdep_lock();
2117	ret = __lockdep_count_forward_deps(&this);
2118	lockdep_unlock();
2119	raw_local_irq_restore(flags);
2120
2121	return ret;
2122}
2123
2124static unsigned long __lockdep_count_backward_deps(struct lock_list *this)
2125{
2126	unsigned long  count = 0;
2127	struct lock_list *target_entry;
2128
2129	__bfs_backwards(this, (void *)&count, noop_count, NULL, &target_entry);
2130
2131	return count;
2132}
2133
2134unsigned long lockdep_count_backward_deps(struct lock_class *class)
2135{
2136	unsigned long ret, flags;
2137	struct lock_list this;
2138
2139	__bfs_init_root(&this, class);
2140
2141	raw_local_irq_save(flags);
2142	lockdep_lock();
2143	ret = __lockdep_count_backward_deps(&this);
2144	lockdep_unlock();
2145	raw_local_irq_restore(flags);
2146
2147	return ret;
2148}
2149
2150/*
2151 * Check that the dependency graph starting at <src> can lead to
2152 * <target> or not.
2153 */
2154static noinline enum bfs_result
2155check_path(struct held_lock *target, struct lock_list *src_entry,
2156	   bool (*match)(struct lock_list *entry, void *data),
2157	   bool (*skip)(struct lock_list *entry, void *data),
2158	   struct lock_list **target_entry)
2159{
2160	enum bfs_result ret;
2161
2162	ret = __bfs_forwards(src_entry, target, match, skip, target_entry);
2163
2164	if (unlikely(bfs_error(ret)))
2165		print_bfs_bug(ret);
2166
2167	return ret;
2168}
2169
2170static void print_deadlock_bug(struct task_struct *, struct held_lock *, struct held_lock *);
2171
2172/*
2173 * Prove that the dependency graph starting at <src> can not
2174 * lead to <target>. If it can, there is a circle when adding
2175 * <target> -> <src> dependency.
2176 *
2177 * Print an error and return BFS_RMATCH if it does.
2178 */
2179static noinline enum bfs_result
2180check_noncircular(struct held_lock *src, struct held_lock *target,
2181		  struct lock_trace **const trace)
2182{
2183	enum bfs_result ret;
2184	struct lock_list *target_entry;
2185	struct lock_list src_entry;
2186
2187	bfs_init_root(&src_entry, src);
2188
2189	debug_atomic_inc(nr_cyclic_checks);
2190
2191	ret = check_path(target, &src_entry, hlock_conflict, NULL, &target_entry);
2192
2193	if (unlikely(ret == BFS_RMATCH)) {
2194		if (!*trace) {
2195			/*
2196			 * If save_trace fails here, the printing might
2197			 * trigger a WARN but because of the !nr_entries it
2198			 * should not do bad things.
2199			 */
2200			*trace = save_trace();
2201		}
2202
2203		if (src->class_idx == target->class_idx)
2204			print_deadlock_bug(current, src, target);
2205		else
2206			print_circular_bug(&src_entry, target_entry, src, target);
2207	}
2208
2209	return ret;
2210}
2211
2212#ifdef CONFIG_TRACE_IRQFLAGS
2213
2214/*
2215 * Forwards and backwards subgraph searching, for the purposes of
2216 * proving that two subgraphs can be connected by a new dependency
2217 * without creating any illegal irq-safe -> irq-unsafe lock dependency.
2218 *
2219 * A irq safe->unsafe deadlock happens with the following conditions:
2220 *
2221 * 1) We have a strong dependency path A -> ... -> B
2222 *
2223 * 2) and we have ENABLED_IRQ usage of B and USED_IN_IRQ usage of A, therefore
2224 *    irq can create a new dependency B -> A (consider the case that a holder
2225 *    of B gets interrupted by an irq whose handler will try to acquire A).
2226 *
2227 * 3) the dependency circle A -> ... -> B -> A we get from 1) and 2) is a
2228 *    strong circle:
2229 *
2230 *      For the usage bits of B:
2231 *        a) if A -> B is -(*N)->, then B -> A could be any type, so any
2232 *           ENABLED_IRQ usage suffices.
2233 *        b) if A -> B is -(*R)->, then B -> A must be -(E*)->, so only
2234 *           ENABLED_IRQ_*_READ usage suffices.
2235 *
2236 *      For the usage bits of A:
2237 *        c) if A -> B is -(E*)->, then B -> A could be any type, so any
2238 *           USED_IN_IRQ usage suffices.
2239 *        d) if A -> B is -(S*)->, then B -> A must be -(*N)->, so only
2240 *           USED_IN_IRQ_*_READ usage suffices.
2241 */
2242
2243/*
2244 * There is a strong dependency path in the dependency graph: A -> B, and now
2245 * we need to decide which usage bit of A should be accumulated to detect
2246 * safe->unsafe bugs.
2247 *
2248 * Note that usage_accumulate() is used in backwards search, so ->only_xr
2249 * stands for whether A -> B only has -(S*)-> (in this case ->only_xr is true).
2250 *
2251 * As above, if only_xr is false, which means A -> B has -(E*)-> dependency
2252 * path, any usage of A should be considered. Otherwise, we should only
2253 * consider _READ usage.
2254 */
2255static inline bool usage_accumulate(struct lock_list *entry, void *mask)
2256{
2257	if (!entry->only_xr)
2258		*(unsigned long *)mask |= entry->class->usage_mask;
2259	else /* Mask out _READ usage bits */
2260		*(unsigned long *)mask |= (entry->class->usage_mask & LOCKF_IRQ);
2261
2262	return false;
2263}
2264
2265/*
2266 * There is a strong dependency path in the dependency graph: A -> B, and now
2267 * we need to decide which usage bit of B conflicts with the usage bits of A,
2268 * i.e. which usage bit of B may introduce safe->unsafe deadlocks.
2269 *
2270 * As above, if only_xr is false, which means A -> B has -(*N)-> dependency
2271 * path, any usage of B should be considered. Otherwise, we should only
2272 * consider _READ usage.
2273 */
2274static inline bool usage_match(struct lock_list *entry, void *mask)
2275{
2276	if (!entry->only_xr)
2277		return !!(entry->class->usage_mask & *(unsigned long *)mask);
2278	else /* Mask out _READ usage bits */
2279		return !!((entry->class->usage_mask & LOCKF_IRQ) & *(unsigned long *)mask);
2280}
2281
2282static inline bool usage_skip(struct lock_list *entry, void *mask)
2283{
2284	if (entry->class->lock_type == LD_LOCK_NORMAL)
2285		return false;
2286
2287	/*
2288	 * Skip local_lock() for irq inversion detection.
2289	 *
2290	 * For !RT, local_lock() is not a real lock, so it won't carry any
2291	 * dependency.
2292	 *
2293	 * For RT, an irq inversion happens when we have lock A and B, and on
2294	 * some CPU we can have:
2295	 *
2296	 *	lock(A);
2297	 *	<interrupted>
2298	 *	  lock(B);
2299	 *
2300	 * where lock(B) cannot sleep, and we have a dependency B -> ... -> A.
2301	 *
2302	 * Now we prove local_lock() cannot exist in that dependency. First we
2303	 * have the observation for any lock chain L1 -> ... -> Ln, for any
2304	 * 1 <= i <= n, Li.inner_wait_type <= L1.inner_wait_type, otherwise
2305	 * wait context check will complain. And since B is not a sleep lock,
2306	 * therefore B.inner_wait_type >= 2, and since the inner_wait_type of
2307	 * local_lock() is 3, which is greater than 2, therefore there is no
2308	 * way the local_lock() exists in the dependency B -> ... -> A.
2309	 *
2310	 * As a result, we will skip local_lock(), when we search for irq
2311	 * inversion bugs.
2312	 */
2313	if (entry->class->lock_type == LD_LOCK_PERCPU &&
2314	    DEBUG_LOCKS_WARN_ON(entry->class->wait_type_inner < LD_WAIT_CONFIG))
2315		return false;
2316
2317	/*
2318	 * Skip WAIT_OVERRIDE for irq inversion detection -- it's not actually
2319	 * a lock and only used to override the wait_type.
2320	 */
2321
2322	return true;
2323}
2324
2325/*
2326 * Find a node in the forwards-direction dependency sub-graph starting
2327 * at @root->class that matches @bit.
2328 *
2329 * Return BFS_MATCH if such a node exists in the subgraph, and put that node
2330 * into *@target_entry.
2331 */
2332static enum bfs_result
2333find_usage_forwards(struct lock_list *root, unsigned long usage_mask,
2334			struct lock_list **target_entry)
2335{
2336	enum bfs_result result;
2337
2338	debug_atomic_inc(nr_find_usage_forwards_checks);
2339
2340	result = __bfs_forwards(root, &usage_mask, usage_match, usage_skip, target_entry);
2341
2342	return result;
2343}
2344
2345/*
2346 * Find a node in the backwards-direction dependency sub-graph starting
2347 * at @root->class that matches @bit.
2348 */
2349static enum bfs_result
2350find_usage_backwards(struct lock_list *root, unsigned long usage_mask,
2351			struct lock_list **target_entry)
2352{
2353	enum bfs_result result;
2354
2355	debug_atomic_inc(nr_find_usage_backwards_checks);
2356
2357	result = __bfs_backwards(root, &usage_mask, usage_match, usage_skip, target_entry);
2358
2359	return result;
2360}
2361
2362static void print_lock_class_header(struct lock_class *class, int depth)
2363{
2364	int bit;
2365
2366	printk("%*s->", depth, "");
2367	print_lock_name(NULL, class);
2368#ifdef CONFIG_DEBUG_LOCKDEP
2369	printk(KERN_CONT " ops: %lu", debug_class_ops_read(class));
2370#endif
2371	printk(KERN_CONT " {\n");
2372
2373	for (bit = 0; bit < LOCK_TRACE_STATES; bit++) {
2374		if (class->usage_mask & (1 << bit)) {
2375			int len = depth;
2376
2377			len += printk("%*s   %s", depth, "", usage_str[bit]);
2378			len += printk(KERN_CONT " at:\n");
2379			print_lock_trace(class->usage_traces[bit], len);
2380		}
2381	}
2382	printk("%*s }\n", depth, "");
2383
2384	printk("%*s ... key      at: [<%px>] %pS\n",
2385		depth, "", class->key, class->key);
2386}
2387
2388/*
2389 * Dependency path printing:
2390 *
2391 * After BFS we get a lock dependency path (linked via ->parent of lock_list),
2392 * printing out each lock in the dependency path will help on understanding how
2393 * the deadlock could happen. Here are some details about dependency path
2394 * printing:
2395 *
2396 * 1)	A lock_list can be either forwards or backwards for a lock dependency,
2397 * 	for a lock dependency A -> B, there are two lock_lists:
2398 *
2399 * 	a)	lock_list in the ->locks_after list of A, whose ->class is B and
2400 * 		->links_to is A. In this case, we can say the lock_list is
2401 * 		"A -> B" (forwards case).
2402 *
2403 * 	b)	lock_list in the ->locks_before list of B, whose ->class is A
2404 * 		and ->links_to is B. In this case, we can say the lock_list is
2405 * 		"B <- A" (bacwards case).
2406 *
2407 * 	The ->trace of both a) and b) point to the call trace where B was
2408 * 	acquired with A held.
2409 *
2410 * 2)	A "helper" lock_list is introduced during BFS, this lock_list doesn't
2411 * 	represent a certain lock dependency, it only provides an initial entry
2412 * 	for BFS. For example, BFS may introduce a "helper" lock_list whose
2413 * 	->class is A, as a result BFS will search all dependencies starting with
2414 * 	A, e.g. A -> B or A -> C.
2415 *
2416 * 	The notation of a forwards helper lock_list is like "-> A", which means
2417 * 	we should search the forwards dependencies starting with "A", e.g A -> B
2418 * 	or A -> C.
2419 *
2420 * 	The notation of a bacwards helper lock_list is like "<- B", which means
2421 * 	we should search the backwards dependencies ending with "B", e.g.
2422 * 	B <- A or B <- C.
2423 */
2424
2425/*
2426 * printk the shortest lock dependencies from @root to @leaf in reverse order.
2427 *
2428 * We have a lock dependency path as follow:
2429 *
2430 *    @root                                                                 @leaf
2431 *      |                                                                     |
2432 *      V                                                                     V
2433 *	          ->parent                                   ->parent
2434 * | lock_list | <--------- | lock_list | ... | lock_list  | <--------- | lock_list |
2435 * |    -> L1  |            | L1 -> L2  | ... |Ln-2 -> Ln-1|            | Ln-1 -> Ln|
2436 *
2437 * , so it's natural that we start from @leaf and print every ->class and
2438 * ->trace until we reach the @root.
2439 */
2440static void __used
2441print_shortest_lock_dependencies(struct lock_list *leaf,
2442				 struct lock_list *root)
2443{
2444	struct lock_list *entry = leaf;
2445	int depth;
2446
2447	/*compute depth from generated tree by BFS*/
2448	depth = get_lock_depth(leaf);
2449
2450	do {
2451		print_lock_class_header(entry->class, depth);
2452		printk("%*s ... acquired at:\n", depth, "");
2453		print_lock_trace(entry->trace, 2);
2454		printk("\n");
2455
2456		if (depth == 0 && (entry != root)) {
2457			printk("lockdep:%s bad path found in chain graph\n", __func__);
2458			break;
2459		}
2460
2461		entry = get_lock_parent(entry);
2462		depth--;
2463	} while (entry && (depth >= 0));
2464}
2465
2466/*
2467 * printk the shortest lock dependencies from @leaf to @root.
2468 *
2469 * We have a lock dependency path (from a backwards search) as follow:
2470 *
2471 *    @leaf                                                                 @root
2472 *      |                                                                     |
2473 *      V                                                                     V
2474 *	          ->parent                                   ->parent
2475 * | lock_list | ---------> | lock_list | ... | lock_list  | ---------> | lock_list |
2476 * | L2 <- L1  |            | L3 <- L2  | ... | Ln <- Ln-1 |            |    <- Ln  |
2477 *
2478 * , so when we iterate from @leaf to @root, we actually print the lock
2479 * dependency path L1 -> L2 -> .. -> Ln in the non-reverse order.
2480 *
2481 * Another thing to notice here is that ->class of L2 <- L1 is L1, while the
2482 * ->trace of L2 <- L1 is the call trace of L2, in fact we don't have the call
2483 * trace of L1 in the dependency path, which is alright, because most of the
2484 * time we can figure out where L1 is held from the call trace of L2.
2485 */
2486static void __used
2487print_shortest_lock_dependencies_backwards(struct lock_list *leaf,
2488					   struct lock_list *root)
2489{
2490	struct lock_list *entry = leaf;
2491	const struct lock_trace *trace = NULL;
2492	int depth;
2493
2494	/*compute depth from generated tree by BFS*/
2495	depth = get_lock_depth(leaf);
2496
2497	do {
2498		print_lock_class_header(entry->class, depth);
2499		if (trace) {
2500			printk("%*s ... acquired at:\n", depth, "");
2501			print_lock_trace(trace, 2);
2502			printk("\n");
2503		}
2504
2505		/*
2506		 * Record the pointer to the trace for the next lock_list
2507		 * entry, see the comments for the function.
2508		 */
2509		trace = entry->trace;
2510
2511		if (depth == 0 && (entry != root)) {
2512			printk("lockdep:%s bad path found in chain graph\n", __func__);
2513			break;
2514		}
2515
2516		entry = get_lock_parent(entry);
2517		depth--;
2518	} while (entry && (depth >= 0));
2519}
2520
2521static void
2522print_irq_lock_scenario(struct lock_list *safe_entry,
2523			struct lock_list *unsafe_entry,
2524			struct lock_class *prev_class,
2525			struct lock_class *next_class)
2526{
2527	struct lock_class *safe_class = safe_entry->class;
2528	struct lock_class *unsafe_class = unsafe_entry->class;
2529	struct lock_class *middle_class = prev_class;
2530
2531	if (middle_class == safe_class)
2532		middle_class = next_class;
2533
2534	/*
2535	 * A direct locking problem where unsafe_class lock is taken
2536	 * directly by safe_class lock, then all we need to show
2537	 * is the deadlock scenario, as it is obvious that the
2538	 * unsafe lock is taken under the safe lock.
2539	 *
2540	 * But if there is a chain instead, where the safe lock takes
2541	 * an intermediate lock (middle_class) where this lock is
2542	 * not the same as the safe lock, then the lock chain is
2543	 * used to describe the problem. Otherwise we would need
2544	 * to show a different CPU case for each link in the chain
2545	 * from the safe_class lock to the unsafe_class lock.
2546	 */
2547	if (middle_class != unsafe_class) {
2548		printk("Chain exists of:\n  ");
2549		__print_lock_name(NULL, safe_class);
2550		printk(KERN_CONT " --> ");
2551		__print_lock_name(NULL, middle_class);
2552		printk(KERN_CONT " --> ");
2553		__print_lock_name(NULL, unsafe_class);
2554		printk(KERN_CONT "\n\n");
2555	}
2556
2557	printk(" Possible interrupt unsafe locking scenario:\n\n");
2558	printk("       CPU0                    CPU1\n");
2559	printk("       ----                    ----\n");
2560	printk("  lock(");
2561	__print_lock_name(NULL, unsafe_class);
2562	printk(KERN_CONT ");\n");
2563	printk("                               local_irq_disable();\n");
2564	printk("                               lock(");
2565	__print_lock_name(NULL, safe_class);
2566	printk(KERN_CONT ");\n");
2567	printk("                               lock(");
2568	__print_lock_name(NULL, middle_class);
2569	printk(KERN_CONT ");\n");
2570	printk("  <Interrupt>\n");
2571	printk("    lock(");
2572	__print_lock_name(NULL, safe_class);
2573	printk(KERN_CONT ");\n");
2574	printk("\n *** DEADLOCK ***\n\n");
2575}
2576
2577static void
2578print_bad_irq_dependency(struct task_struct *curr,
2579			 struct lock_list *prev_root,
2580			 struct lock_list *next_root,
2581			 struct lock_list *backwards_entry,
2582			 struct lock_list *forwards_entry,
2583			 struct held_lock *prev,
2584			 struct held_lock *next,
2585			 enum lock_usage_bit bit1,
2586			 enum lock_usage_bit bit2,
2587			 const char *irqclass)
2588{
2589	if (!debug_locks_off_graph_unlock() || debug_locks_silent)
2590		return;
2591
2592	nbcon_cpu_emergency_enter();
2593
2594	pr_warn("\n");
2595	pr_warn("=====================================================\n");
2596	pr_warn("WARNING: %s-safe -> %s-unsafe lock order detected\n",
2597		irqclass, irqclass);
2598	print_kernel_ident();
2599	pr_warn("-----------------------------------------------------\n");
2600	pr_warn("%s/%d [HC%u[%lu]:SC%u[%lu]:HE%u:SE%u] is trying to acquire:\n",
2601		curr->comm, task_pid_nr(curr),
2602		lockdep_hardirq_context(), hardirq_count() >> HARDIRQ_SHIFT,
2603		curr->softirq_context, softirq_count() >> SOFTIRQ_SHIFT,
2604		lockdep_hardirqs_enabled(),
2605		curr->softirqs_enabled);
2606	print_lock(next);
2607
2608	pr_warn("\nand this task is already holding:\n");
2609	print_lock(prev);
2610	pr_warn("which would create a new lock dependency:\n");
2611	print_lock_name(prev, hlock_class(prev));
2612	pr_cont(" ->");
2613	print_lock_name(next, hlock_class(next));
2614	pr_cont("\n");
2615
2616	pr_warn("\nbut this new dependency connects a %s-irq-safe lock:\n",
2617		irqclass);
2618	print_lock_name(NULL, backwards_entry->class);
2619	pr_warn("\n... which became %s-irq-safe at:\n", irqclass);
2620
2621	print_lock_trace(backwards_entry->class->usage_traces[bit1], 1);
2622
2623	pr_warn("\nto a %s-irq-unsafe lock:\n", irqclass);
2624	print_lock_name(NULL, forwards_entry->class);
2625	pr_warn("\n... which became %s-irq-unsafe at:\n", irqclass);
2626	pr_warn("...");
2627
2628	print_lock_trace(forwards_entry->class->usage_traces[bit2], 1);
2629
2630	pr_warn("\nother info that might help us debug this:\n\n");
2631	print_irq_lock_scenario(backwards_entry, forwards_entry,
2632				hlock_class(prev), hlock_class(next));
2633
2634	lockdep_print_held_locks(curr);
2635
2636	pr_warn("\nthe dependencies between %s-irq-safe lock and the holding lock:\n", irqclass);
2637	print_shortest_lock_dependencies_backwards(backwards_entry, prev_root);
2638
2639	pr_warn("\nthe dependencies between the lock to be acquired");
2640	pr_warn(" and %s-irq-unsafe lock:\n", irqclass);
2641	next_root->trace = save_trace();
2642	if (!next_root->trace)
2643		goto out;
2644	print_shortest_lock_dependencies(forwards_entry, next_root);
2645
2646	pr_warn("\nstack backtrace:\n");
2647	dump_stack();
2648out:
2649	nbcon_cpu_emergency_exit();
2650}
2651
2652static const char *state_names[] = {
2653#define LOCKDEP_STATE(__STATE) \
2654	__stringify(__STATE),
2655#include "lockdep_states.h"
2656#undef LOCKDEP_STATE
2657};
2658
2659static const char *state_rnames[] = {
2660#define LOCKDEP_STATE(__STATE) \
2661	__stringify(__STATE)"-READ",
2662#include "lockdep_states.h"
2663#undef LOCKDEP_STATE
2664};
2665
2666static inline const char *state_name(enum lock_usage_bit bit)
2667{
2668	if (bit & LOCK_USAGE_READ_MASK)
2669		return state_rnames[bit >> LOCK_USAGE_DIR_MASK];
2670	else
2671		return state_names[bit >> LOCK_USAGE_DIR_MASK];
2672}
2673
2674/*
2675 * The bit number is encoded like:
2676 *
2677 *  bit0: 0 exclusive, 1 read lock
2678 *  bit1: 0 used in irq, 1 irq enabled
2679 *  bit2-n: state
2680 */
2681static int exclusive_bit(int new_bit)
2682{
2683	int state = new_bit & LOCK_USAGE_STATE_MASK;
2684	int dir = new_bit & LOCK_USAGE_DIR_MASK;
2685
2686	/*
2687	 * keep state, bit flip the direction and strip read.
2688	 */
2689	return state | (dir ^ LOCK_USAGE_DIR_MASK);
2690}
2691
2692/*
2693 * Observe that when given a bitmask where each bitnr is encoded as above, a
2694 * right shift of the mask transforms the individual bitnrs as -1 and
2695 * conversely, a left shift transforms into +1 for the individual bitnrs.
2696 *
2697 * So for all bits whose number have LOCK_ENABLED_* set (bitnr1 == 1), we can
2698 * create the mask with those bit numbers using LOCK_USED_IN_* (bitnr1 == 0)
2699 * instead by subtracting the bit number by 2, or shifting the mask right by 2.
2700 *
2701 * Similarly, bitnr1 == 0 becomes bitnr1 == 1 by adding 2, or shifting left 2.
2702 *
2703 * So split the mask (note that LOCKF_ENABLED_IRQ_ALL|LOCKF_USED_IN_IRQ_ALL is
2704 * all bits set) and recompose with bitnr1 flipped.
2705 */
2706static unsigned long invert_dir_mask(unsigned long mask)
2707{
2708	unsigned long excl = 0;
2709
2710	/* Invert dir */
2711	excl |= (mask & LOCKF_ENABLED_IRQ_ALL) >> LOCK_USAGE_DIR_MASK;
2712	excl |= (mask & LOCKF_USED_IN_IRQ_ALL) << LOCK_USAGE_DIR_MASK;
2713
2714	return excl;
2715}
2716
2717/*
2718 * Note that a LOCK_ENABLED_IRQ_*_READ usage and a LOCK_USED_IN_IRQ_*_READ
2719 * usage may cause deadlock too, for example:
2720 *
2721 * P1				P2
2722 * <irq disabled>
2723 * write_lock(l1);		<irq enabled>
2724 *				read_lock(l2);
2725 * write_lock(l2);
2726 * 				<in irq>
2727 * 				read_lock(l1);
2728 *
2729 * , in above case, l1 will be marked as LOCK_USED_IN_IRQ_HARDIRQ_READ and l2
2730 * will marked as LOCK_ENABLE_IRQ_HARDIRQ_READ, and this is a possible
2731 * deadlock.
2732 *
2733 * In fact, all of the following cases may cause deadlocks:
2734 *
2735 * 	 LOCK_USED_IN_IRQ_* -> LOCK_ENABLED_IRQ_*
2736 * 	 LOCK_USED_IN_IRQ_*_READ -> LOCK_ENABLED_IRQ_*
2737 * 	 LOCK_USED_IN_IRQ_* -> LOCK_ENABLED_IRQ_*_READ
2738 * 	 LOCK_USED_IN_IRQ_*_READ -> LOCK_ENABLED_IRQ_*_READ
2739 *
2740 * As a result, to calculate the "exclusive mask", first we invert the
2741 * direction (USED_IN/ENABLED) of the original mask, and 1) for all bits with
2742 * bitnr0 set (LOCK_*_READ), add those with bitnr0 cleared (LOCK_*). 2) for all
2743 * bits with bitnr0 cleared (LOCK_*_READ), add those with bitnr0 set (LOCK_*).
2744 */
2745static unsigned long exclusive_mask(unsigned long mask)
2746{
2747	unsigned long excl = invert_dir_mask(mask);
2748
2749	excl |= (excl & LOCKF_IRQ_READ) >> LOCK_USAGE_READ_MASK;
2750	excl |= (excl & LOCKF_IRQ) << LOCK_USAGE_READ_MASK;
2751
2752	return excl;
2753}
2754
2755/*
2756 * Retrieve the _possible_ original mask to which @mask is
2757 * exclusive. Ie: this is the opposite of exclusive_mask().
2758 * Note that 2 possible original bits can match an exclusive
2759 * bit: one has LOCK_USAGE_READ_MASK set, the other has it
2760 * cleared. So both are returned for each exclusive bit.
2761 */
2762static unsigned long original_mask(unsigned long mask)
2763{
2764	unsigned long excl = invert_dir_mask(mask);
2765
2766	/* Include read in existing usages */
2767	excl |= (excl & LOCKF_IRQ_READ) >> LOCK_USAGE_READ_MASK;
2768	excl |= (excl & LOCKF_IRQ) << LOCK_USAGE_READ_MASK;
2769
2770	return excl;
2771}
2772
2773/*
2774 * Find the first pair of bit match between an original
2775 * usage mask and an exclusive usage mask.
2776 */
2777static int find_exclusive_match(unsigned long mask,
2778				unsigned long excl_mask,
2779				enum lock_usage_bit *bitp,
2780				enum lock_usage_bit *excl_bitp)
2781{
2782	int bit, excl, excl_read;
2783
2784	for_each_set_bit(bit, &mask, LOCK_USED) {
2785		/*
2786		 * exclusive_bit() strips the read bit, however,
2787		 * LOCK_ENABLED_IRQ_*_READ may cause deadlocks too, so we need
2788		 * to search excl | LOCK_USAGE_READ_MASK as well.
2789		 */
2790		excl = exclusive_bit(bit);
2791		excl_read = excl | LOCK_USAGE_READ_MASK;
2792		if (excl_mask & lock_flag(excl)) {
2793			*bitp = bit;
2794			*excl_bitp = excl;
2795			return 0;
2796		} else if (excl_mask & lock_flag(excl_read)) {
2797			*bitp = bit;
2798			*excl_bitp = excl_read;
2799			return 0;
2800		}
2801	}
2802	return -1;
2803}
2804
2805/*
2806 * Prove that the new dependency does not connect a hardirq-safe(-read)
2807 * lock with a hardirq-unsafe lock - to achieve this we search
2808 * the backwards-subgraph starting at <prev>, and the
2809 * forwards-subgraph starting at <next>:
2810 */
2811static int check_irq_usage(struct task_struct *curr, struct held_lock *prev,
2812			   struct held_lock *next)
2813{
2814	unsigned long usage_mask = 0, forward_mask, backward_mask;
2815	enum lock_usage_bit forward_bit = 0, backward_bit = 0;
2816	struct lock_list *target_entry1;
2817	struct lock_list *target_entry;
2818	struct lock_list this, that;
2819	enum bfs_result ret;
2820
2821	/*
2822	 * Step 1: gather all hard/soft IRQs usages backward in an
2823	 * accumulated usage mask.
2824	 */
2825	bfs_init_rootb(&this, prev);
2826
2827	ret = __bfs_backwards(&this, &usage_mask, usage_accumulate, usage_skip, NULL);
2828	if (bfs_error(ret)) {
2829		print_bfs_bug(ret);
2830		return 0;
2831	}
2832
2833	usage_mask &= LOCKF_USED_IN_IRQ_ALL;
2834	if (!usage_mask)
2835		return 1;
2836
2837	/*
2838	 * Step 2: find exclusive uses forward that match the previous
2839	 * backward accumulated mask.
2840	 */
2841	forward_mask = exclusive_mask(usage_mask);
2842
2843	bfs_init_root(&that, next);
2844
2845	ret = find_usage_forwards(&that, forward_mask, &target_entry1);
2846	if (bfs_error(ret)) {
2847		print_bfs_bug(ret);
2848		return 0;
2849	}
2850	if (ret == BFS_RNOMATCH)
2851		return 1;
2852
2853	/*
2854	 * Step 3: we found a bad match! Now retrieve a lock from the backward
2855	 * list whose usage mask matches the exclusive usage mask from the
2856	 * lock found on the forward list.
2857	 *
2858	 * Note, we should only keep the LOCKF_ENABLED_IRQ_ALL bits, considering
2859	 * the follow case:
2860	 *
2861	 * When trying to add A -> B to the graph, we find that there is a
2862	 * hardirq-safe L, that L -> ... -> A, and another hardirq-unsafe M,
2863	 * that B -> ... -> M. However M is **softirq-safe**, if we use exact
2864	 * invert bits of M's usage_mask, we will find another lock N that is
2865	 * **softirq-unsafe** and N -> ... -> A, however N -> .. -> M will not
2866	 * cause a inversion deadlock.
2867	 */
2868	backward_mask = original_mask(target_entry1->class->usage_mask & LOCKF_ENABLED_IRQ_ALL);
2869
2870	ret = find_usage_backwards(&this, backward_mask, &target_entry);
2871	if (bfs_error(ret)) {
2872		print_bfs_bug(ret);
2873		return 0;
2874	}
2875	if (DEBUG_LOCKS_WARN_ON(ret == BFS_RNOMATCH))
2876		return 1;
2877
2878	/*
2879	 * Step 4: narrow down to a pair of incompatible usage bits
2880	 * and report it.
2881	 */
2882	ret = find_exclusive_match(target_entry->class->usage_mask,
2883				   target_entry1->class->usage_mask,
2884				   &backward_bit, &forward_bit);
2885	if (DEBUG_LOCKS_WARN_ON(ret == -1))
2886		return 1;
2887
2888	print_bad_irq_dependency(curr, &this, &that,
2889				 target_entry, target_entry1,
2890				 prev, next,
2891				 backward_bit, forward_bit,
2892				 state_name(backward_bit));
2893
2894	return 0;
2895}
2896
2897#else
2898
2899static inline int check_irq_usage(struct task_struct *curr,
2900				  struct held_lock *prev, struct held_lock *next)
2901{
2902	return 1;
2903}
2904
2905static inline bool usage_skip(struct lock_list *entry, void *mask)
2906{
2907	return false;
2908}
2909
2910#endif /* CONFIG_TRACE_IRQFLAGS */
2911
2912#ifdef CONFIG_LOCKDEP_SMALL
2913/*
2914 * Check that the dependency graph starting at <src> can lead to
2915 * <target> or not. If it can, <src> -> <target> dependency is already
2916 * in the graph.
2917 *
2918 * Return BFS_RMATCH if it does, or BFS_RNOMATCH if it does not, return BFS_E* if
2919 * any error appears in the bfs search.
2920 */
2921static noinline enum bfs_result
2922check_redundant(struct held_lock *src, struct held_lock *target)
2923{
2924	enum bfs_result ret;
2925	struct lock_list *target_entry;
2926	struct lock_list src_entry;
2927
2928	bfs_init_root(&src_entry, src);
2929	/*
2930	 * Special setup for check_redundant().
2931	 *
2932	 * To report redundant, we need to find a strong dependency path that
2933	 * is equal to or stronger than <src> -> <target>. So if <src> is E,
2934	 * we need to let __bfs() only search for a path starting at a -(E*)->,
2935	 * we achieve this by setting the initial node's ->only_xr to true in
2936	 * that case. And if <prev> is S, we set initial ->only_xr to false
2937	 * because both -(S*)-> (equal) and -(E*)-> (stronger) are redundant.
2938	 */
2939	src_entry.only_xr = src->read == 0;
2940
2941	debug_atomic_inc(nr_redundant_checks);
2942
2943	/*
2944	 * Note: we skip local_lock() for redundant check, because as the
2945	 * comment in usage_skip(), A -> local_lock() -> B and A -> B are not
2946	 * the same.
2947	 */
2948	ret = check_path(target, &src_entry, hlock_equal, usage_skip, &target_entry);
2949
2950	if (ret == BFS_RMATCH)
2951		debug_atomic_inc(nr_redundant);
2952
2953	return ret;
2954}
2955
2956#else
2957
2958static inline enum bfs_result
2959check_redundant(struct held_lock *src, struct held_lock *target)
2960{
2961	return BFS_RNOMATCH;
2962}
2963
2964#endif
2965
2966static void inc_chains(int irq_context)
2967{
2968	if (irq_context & LOCK_CHAIN_HARDIRQ_CONTEXT)
2969		nr_hardirq_chains++;
2970	else if (irq_context & LOCK_CHAIN_SOFTIRQ_CONTEXT)
2971		nr_softirq_chains++;
2972	else
2973		nr_process_chains++;
2974}
2975
2976static void dec_chains(int irq_context)
2977{
2978	if (irq_context & LOCK_CHAIN_HARDIRQ_CONTEXT)
2979		nr_hardirq_chains--;
2980	else if (irq_context & LOCK_CHAIN_SOFTIRQ_CONTEXT)
2981		nr_softirq_chains--;
2982	else
2983		nr_process_chains--;
2984}
2985
2986static void
2987print_deadlock_scenario(struct held_lock *nxt, struct held_lock *prv)
2988{
2989	struct lock_class *next = hlock_class(nxt);
2990	struct lock_class *prev = hlock_class(prv);
2991
2992	printk(" Possible unsafe locking scenario:\n\n");
2993	printk("       CPU0\n");
2994	printk("       ----\n");
2995	printk("  lock(");
2996	__print_lock_name(prv, prev);
2997	printk(KERN_CONT ");\n");
2998	printk("  lock(");
2999	__print_lock_name(nxt, next);
3000	printk(KERN_CONT ");\n");
3001	printk("\n *** DEADLOCK ***\n\n");
3002	printk(" May be due to missing lock nesting notation\n\n");
3003}
3004
3005static void
3006print_deadlock_bug(struct task_struct *curr, struct held_lock *prev,
3007		   struct held_lock *next)
3008{
3009	struct lock_class *class = hlock_class(prev);
3010
3011	if (!debug_locks_off_graph_unlock() || debug_locks_silent)
3012		return;
3013
3014	nbcon_cpu_emergency_enter();
3015
3016	pr_warn("\n");
3017	pr_warn("============================================\n");
3018	pr_warn("WARNING: possible recursive locking detected\n");
3019	print_kernel_ident();
3020	pr_warn("--------------------------------------------\n");
3021	pr_warn("%s/%d is trying to acquire lock:\n",
3022		curr->comm, task_pid_nr(curr));
3023	print_lock(next);
3024	pr_warn("\nbut task is already holding lock:\n");
3025	print_lock(prev);
3026
3027	if (class->cmp_fn) {
3028		pr_warn("and the lock comparison function returns %i:\n",
3029			class->cmp_fn(prev->instance, next->instance));
3030	}
3031
3032	pr_warn("\nother info that might help us debug this:\n");
3033	print_deadlock_scenario(next, prev);
3034	lockdep_print_held_locks(curr);
3035
3036	pr_warn("\nstack backtrace:\n");
3037	dump_stack();
3038
3039	nbcon_cpu_emergency_exit();
3040}
3041
3042/*
3043 * Check whether we are holding such a class already.
3044 *
3045 * (Note that this has to be done separately, because the graph cannot
3046 * detect such classes of deadlocks.)
3047 *
3048 * Returns: 0 on deadlock detected, 1 on OK, 2 if another lock with the same
3049 * lock class is held but nest_lock is also held, i.e. we rely on the
3050 * nest_lock to avoid the deadlock.
3051 */
3052static int
3053check_deadlock(struct task_struct *curr, struct held_lock *next)
3054{
3055	struct lock_class *class;
3056	struct held_lock *prev;
3057	struct held_lock *nest = NULL;
3058	int i;
3059
3060	for (i = 0; i < curr->lockdep_depth; i++) {
3061		prev = curr->held_locks + i;
3062
3063		if (prev->instance == next->nest_lock)
3064			nest = prev;
3065
3066		if (hlock_class(prev) != hlock_class(next))
3067			continue;
3068
3069		/*
3070		 * Allow read-after-read recursion of the same
3071		 * lock class (i.e. read_lock(lock)+read_lock(lock)):
3072		 */
3073		if ((next->read == 2) && prev->read)
3074			continue;
3075
3076		class = hlock_class(prev);
3077
3078		if (class->cmp_fn &&
3079		    class->cmp_fn(prev->instance, next->instance) < 0)
3080			continue;
3081
3082		/*
3083		 * We're holding the nest_lock, which serializes this lock's
3084		 * nesting behaviour.
3085		 */
3086		if (nest)
3087			return 2;
3088
3089		print_deadlock_bug(curr, prev, next);
3090		return 0;
3091	}
3092	return 1;
3093}
3094
3095/*
3096 * There was a chain-cache miss, and we are about to add a new dependency
3097 * to a previous lock. We validate the following rules:
3098 *
3099 *  - would the adding of the <prev> -> <next> dependency create a
3100 *    circular dependency in the graph? [== circular deadlock]
3101 *
3102 *  - does the new prev->next dependency connect any hardirq-safe lock
3103 *    (in the full backwards-subgraph starting at <prev>) with any
3104 *    hardirq-unsafe lock (in the full forwards-subgraph starting at
3105 *    <next>)? [== illegal lock inversion with hardirq contexts]
3106 *
3107 *  - does the new prev->next dependency connect any softirq-safe lock
3108 *    (in the full backwards-subgraph starting at <prev>) with any
3109 *    softirq-unsafe lock (in the full forwards-subgraph starting at
3110 *    <next>)? [== illegal lock inversion with softirq contexts]
3111 *
3112 * any of these scenarios could lead to a deadlock.
3113 *
3114 * Then if all the validations pass, we add the forwards and backwards
3115 * dependency.
3116 */
3117static int
3118check_prev_add(struct task_struct *curr, struct held_lock *prev,
3119	       struct held_lock *next, u16 distance,
3120	       struct lock_trace **const trace)
3121{
3122	struct lock_list *entry;
3123	enum bfs_result ret;
3124
3125	if (!hlock_class(prev)->key || !hlock_class(next)->key) {
3126		/*
3127		 * The warning statements below may trigger a use-after-free
3128		 * of the class name. It is better to trigger a use-after free
3129		 * and to have the class name most of the time instead of not
3130		 * having the class name available.
3131		 */
3132		WARN_ONCE(!debug_locks_silent && !hlock_class(prev)->key,
3133			  "Detected use-after-free of lock class %px/%s\n",
3134			  hlock_class(prev),
3135			  hlock_class(prev)->name);
3136		WARN_ONCE(!debug_locks_silent && !hlock_class(next)->key,
3137			  "Detected use-after-free of lock class %px/%s\n",
3138			  hlock_class(next),
3139			  hlock_class(next)->name);
3140		return 2;
3141	}
3142
3143	if (prev->class_idx == next->class_idx) {
3144		struct lock_class *class = hlock_class(prev);
3145
3146		if (class->cmp_fn &&
3147		    class->cmp_fn(prev->instance, next->instance) < 0)
3148			return 2;
3149	}
3150
3151	/*
3152	 * Prove that the new <prev> -> <next> dependency would not
3153	 * create a circular dependency in the graph. (We do this by
3154	 * a breadth-first search into the graph starting at <next>,
3155	 * and check whether we can reach <prev>.)
3156	 *
3157	 * The search is limited by the size of the circular queue (i.e.,
3158	 * MAX_CIRCULAR_QUEUE_SIZE) which keeps track of a breadth of nodes
3159	 * in the graph whose neighbours are to be checked.
3160	 */
3161	ret = check_noncircular(next, prev, trace);
3162	if (unlikely(bfs_error(ret) || ret == BFS_RMATCH))
3163		return 0;
3164
3165	if (!check_irq_usage(curr, prev, next))
3166		return 0;
3167
3168	/*
3169	 * Is the <prev> -> <next> dependency already present?
3170	 *
3171	 * (this may occur even though this is a new chain: consider
3172	 *  e.g. the L1 -> L2 -> L3 -> L4 and the L5 -> L1 -> L2 -> L3
3173	 *  chains - the second one will be new, but L1 already has
3174	 *  L2 added to its dependency list, due to the first chain.)
3175	 */
3176	list_for_each_entry(entry, &hlock_class(prev)->locks_after, entry) {
3177		if (entry->class == hlock_class(next)) {
3178			if (distance == 1)
3179				entry->distance = 1;
3180			entry->dep |= calc_dep(prev, next);
3181
3182			/*
3183			 * Also, update the reverse dependency in @next's
3184			 * ->locks_before list.
3185			 *
3186			 *  Here we reuse @entry as the cursor, which is fine
3187			 *  because we won't go to the next iteration of the
3188			 *  outer loop:
3189			 *
3190			 *  For normal cases, we return in the inner loop.
3191			 *
3192			 *  If we fail to return, we have inconsistency, i.e.
3193			 *  <prev>::locks_after contains <next> while
3194			 *  <next>::locks_before doesn't contain <prev>. In
3195			 *  that case, we return after the inner and indicate
3196			 *  something is wrong.
3197			 */
3198			list_for_each_entry(entry, &hlock_class(next)->locks_before, entry) {
3199				if (entry->class == hlock_class(prev)) {
3200					if (distance == 1)
3201						entry->distance = 1;
3202					entry->dep |= calc_depb(prev, next);
3203					return 1;
3204				}
3205			}
3206
3207			/* <prev> is not found in <next>::locks_before */
3208			return 0;
3209		}
3210	}
3211
3212	/*
3213	 * Is the <prev> -> <next> link redundant?
3214	 */
3215	ret = check_redundant(prev, next);
3216	if (bfs_error(ret))
3217		return 0;
3218	else if (ret == BFS_RMATCH)
3219		return 2;
3220
3221	if (!*trace) {
3222		*trace = save_trace();
3223		if (!*trace)
3224			return 0;
3225	}
3226
3227	/*
3228	 * Ok, all validations passed, add the new lock
3229	 * to the previous lock's dependency list:
3230	 */
3231	ret = add_lock_to_list(hlock_class(next), hlock_class(prev),
3232			       &hlock_class(prev)->locks_after, distance,
3233			       calc_dep(prev, next), *trace);
3234
3235	if (!ret)
3236		return 0;
3237
3238	ret = add_lock_to_list(hlock_class(prev), hlock_class(next),
3239			       &hlock_class(next)->locks_before, distance,
3240			       calc_depb(prev, next), *trace);
3241	if (!ret)
3242		return 0;
3243
3244	return 2;
3245}
3246
3247/*
3248 * Add the dependency to all directly-previous locks that are 'relevant'.
3249 * The ones that are relevant are (in increasing distance from curr):
3250 * all consecutive trylock entries and the final non-trylock entry - or
3251 * the end of this context's lock-chain - whichever comes first.
3252 */
3253static int
3254check_prevs_add(struct task_struct *curr, struct held_lock *next)
3255{
3256	struct lock_trace *trace = NULL;
3257	int depth = curr->lockdep_depth;
3258	struct held_lock *hlock;
3259
3260	/*
3261	 * Debugging checks.
3262	 *
3263	 * Depth must not be zero for a non-head lock:
3264	 */
3265	if (!depth)
3266		goto out_bug;
3267	/*
3268	 * At least two relevant locks must exist for this
3269	 * to be a head:
3270	 */
3271	if (curr->held_locks[depth].irq_context !=
3272			curr->held_locks[depth-1].irq_context)
3273		goto out_bug;
3274
3275	for (;;) {
3276		u16 distance = curr->lockdep_depth - depth + 1;
3277		hlock = curr->held_locks + depth - 1;
3278
3279		if (hlock->check) {
3280			int ret = check_prev_add(curr, hlock, next, distance, &trace);
3281			if (!ret)
3282				return 0;
3283
3284			/*
3285			 * Stop after the first non-trylock entry,
3286			 * as non-trylock entries have added their
3287			 * own direct dependencies already, so this
3288			 * lock is connected to them indirectly:
3289			 */
3290			if (!hlock->trylock)
3291				break;
3292		}
3293
3294		depth--;
3295		/*
3296		 * End of lock-stack?
3297		 */
3298		if (!depth)
3299			break;
3300		/*
3301		 * Stop the search if we cross into another context:
3302		 */
3303		if (curr->held_locks[depth].irq_context !=
3304				curr->held_locks[depth-1].irq_context)
3305			break;
3306	}
3307	return 1;
3308out_bug:
3309	if (!debug_locks_off_graph_unlock())
3310		return 0;
3311
3312	/*
3313	 * Clearly we all shouldn't be here, but since we made it we
3314	 * can reliable say we messed up our state. See the above two
3315	 * gotos for reasons why we could possibly end up here.
3316	 */
3317	WARN_ON(1);
3318
3319	return 0;
3320}
3321
3322struct lock_chain lock_chains[MAX_LOCKDEP_CHAINS];
3323static DECLARE_BITMAP(lock_chains_in_use, MAX_LOCKDEP_CHAINS);
3324static u16 chain_hlocks[MAX_LOCKDEP_CHAIN_HLOCKS];
3325unsigned long nr_zapped_lock_chains;
3326unsigned int nr_free_chain_hlocks;	/* Free chain_hlocks in buckets */
3327unsigned int nr_lost_chain_hlocks;	/* Lost chain_hlocks */
3328unsigned int nr_large_chain_blocks;	/* size > MAX_CHAIN_BUCKETS */
3329
3330/*
3331 * The first 2 chain_hlocks entries in the chain block in the bucket
3332 * list contains the following meta data:
3333 *
3334 *   entry[0]:
3335 *     Bit    15 - always set to 1 (it is not a class index)
3336 *     Bits 0-14 - upper 15 bits of the next block index
3337 *   entry[1]    - lower 16 bits of next block index
3338 *
3339 * A next block index of all 1 bits means it is the end of the list.
3340 *
3341 * On the unsized bucket (bucket-0), the 3rd and 4th entries contain
3342 * the chain block size:
3343 *
3344 *   entry[2] - upper 16 bits of the chain block size
3345 *   entry[3] - lower 16 bits of the chain block size
3346 */
3347#define MAX_CHAIN_BUCKETS	16
3348#define CHAIN_BLK_FLAG		(1U << 15)
3349#define CHAIN_BLK_LIST_END	0xFFFFU
3350
3351static int chain_block_buckets[MAX_CHAIN_BUCKETS];
3352
3353static inline int size_to_bucket(int size)
3354{
3355	if (size > MAX_CHAIN_BUCKETS)
3356		return 0;
3357
3358	return size - 1;
3359}
3360
3361/*
3362 * Iterate all the chain blocks in a bucket.
3363 */
3364#define for_each_chain_block(bucket, prev, curr)		\
3365	for ((prev) = -1, (curr) = chain_block_buckets[bucket];	\
3366	     (curr) >= 0;					\
3367	     (prev) = (curr), (curr) = chain_block_next(curr))
3368
3369/*
3370 * next block or -1
3371 */
3372static inline int chain_block_next(int offset)
3373{
3374	int next = chain_hlocks[offset];
3375
3376	WARN_ON_ONCE(!(next & CHAIN_BLK_FLAG));
3377
3378	if (next == CHAIN_BLK_LIST_END)
3379		return -1;
3380
3381	next &= ~CHAIN_BLK_FLAG;
3382	next <<= 16;
3383	next |= chain_hlocks[offset + 1];
3384
3385	return next;
3386}
3387
3388/*
3389 * bucket-0 only
3390 */
3391static inline int chain_block_size(int offset)
3392{
3393	return (chain_hlocks[offset + 2] << 16) | chain_hlocks[offset + 3];
3394}
3395
3396static inline void init_chain_block(int offset, int next, int bucket, int size)
3397{
3398	chain_hlocks[offset] = (next >> 16) | CHAIN_BLK_FLAG;
3399	chain_hlocks[offset + 1] = (u16)next;
3400
3401	if (size && !bucket) {
3402		chain_hlocks[offset + 2] = size >> 16;
3403		chain_hlocks[offset + 3] = (u16)size;
3404	}
3405}
3406
3407static inline void add_chain_block(int offset, int size)
3408{
3409	int bucket = size_to_bucket(size);
3410	int next = chain_block_buckets[bucket];
3411	int prev, curr;
3412
3413	if (unlikely(size < 2)) {
3414		/*
3415		 * We can't store single entries on the freelist. Leak them.
3416		 *
3417		 * One possible way out would be to uniquely mark them, other
3418		 * than with CHAIN_BLK_FLAG, such that we can recover them when
3419		 * the block before it is re-added.
3420		 */
3421		if (size)
3422			nr_lost_chain_hlocks++;
3423		return;
3424	}
3425
3426	nr_free_chain_hlocks += size;
3427	if (!bucket) {
3428		nr_large_chain_blocks++;
3429
3430		/*
3431		 * Variable sized, sort large to small.
3432		 */
3433		for_each_chain_block(0, prev, curr) {
3434			if (size >= chain_block_size(curr))
3435				break;
3436		}
3437		init_chain_block(offset, curr, 0, size);
3438		if (prev < 0)
3439			chain_block_buckets[0] = offset;
3440		else
3441			init_chain_block(prev, offset, 0, 0);
3442		return;
3443	}
3444	/*
3445	 * Fixed size, add to head.
3446	 */
3447	init_chain_block(offset, next, bucket, size);
3448	chain_block_buckets[bucket] = offset;
3449}
3450
3451/*
3452 * Only the first block in the list can be deleted.
3453 *
3454 * For the variable size bucket[0], the first block (the largest one) is
3455 * returned, broken up and put back into the pool. So if a chain block of
3456 * length > MAX_CHAIN_BUCKETS is ever used and zapped, it will just be
3457 * queued up after the primordial chain block and never be used until the
3458 * hlock entries in the primordial chain block is almost used up. That
3459 * causes fragmentation and reduce allocation efficiency. That can be
3460 * monitored by looking at the "large chain blocks" number in lockdep_stats.
3461 */
3462static inline void del_chain_block(int bucket, int size, int next)
3463{
3464	nr_free_chain_hlocks -= size;
3465	chain_block_buckets[bucket] = next;
3466
3467	if (!bucket)
3468		nr_large_chain_blocks--;
3469}
3470
3471static void init_chain_block_buckets(void)
3472{
3473	int i;
3474
3475	for (i = 0; i < MAX_CHAIN_BUCKETS; i++)
3476		chain_block_buckets[i] = -1;
3477
3478	add_chain_block(0, ARRAY_SIZE(chain_hlocks));
3479}
3480
3481/*
3482 * Return offset of a chain block of the right size or -1 if not found.
3483 *
3484 * Fairly simple worst-fit allocator with the addition of a number of size
3485 * specific free lists.
3486 */
3487static int alloc_chain_hlocks(int req)
3488{
3489	int bucket, curr, size;
3490
3491	/*
3492	 * We rely on the MSB to act as an escape bit to denote freelist
3493	 * pointers. Make sure this bit isn't set in 'normal' class_idx usage.
3494	 */
3495	BUILD_BUG_ON((MAX_LOCKDEP_KEYS-1) & CHAIN_BLK_FLAG);
3496
3497	init_data_structures_once();
3498
3499	if (nr_free_chain_hlocks < req)
3500		return -1;
3501
3502	/*
3503	 * We require a minimum of 2 (u16) entries to encode a freelist
3504	 * 'pointer'.
3505	 */
3506	req = max(req, 2);
3507	bucket = size_to_bucket(req);
3508	curr = chain_block_buckets[bucket];
3509
3510	if (bucket) {
3511		if (curr >= 0) {
3512			del_chain_block(bucket, req, chain_block_next(curr));
3513			return curr;
3514		}
3515		/* Try bucket 0 */
3516		curr = chain_block_buckets[0];
3517	}
3518
3519	/*
3520	 * The variable sized freelist is sorted by size; the first entry is
3521	 * the largest. Use it if it fits.
3522	 */
3523	if (curr >= 0) {
3524		size = chain_block_size(curr);
3525		if (likely(size >= req)) {
3526			del_chain_block(0, size, chain_block_next(curr));
3527			if (size > req)
3528				add_chain_block(curr + req, size - req);
3529			return curr;
3530		}
3531	}
3532
3533	/*
3534	 * Last resort, split a block in a larger sized bucket.
3535	 */
3536	for (size = MAX_CHAIN_BUCKETS; size > req; size--) {
3537		bucket = size_to_bucket(size);
3538		curr = chain_block_buckets[bucket];
3539		if (curr < 0)
3540			continue;
3541
3542		del_chain_block(bucket, size, chain_block_next(curr));
3543		add_chain_block(curr + req, size - req);
3544		return curr;
3545	}
3546
3547	return -1;
3548}
3549
3550static inline void free_chain_hlocks(int base, int size)
3551{
3552	add_chain_block(base, max(size, 2));
3553}
3554
3555struct lock_class *lock_chain_get_class(struct lock_chain *chain, int i)
3556{
3557	u16 chain_hlock = chain_hlocks[chain->base + i];
3558	unsigned int class_idx = chain_hlock_class_idx(chain_hlock);
3559
3560	return lock_classes + class_idx;
3561}
3562
3563/*
3564 * Returns the index of the first held_lock of the current chain
3565 */
3566static inline int get_first_held_lock(struct task_struct *curr,
3567					struct held_lock *hlock)
3568{
3569	int i;
3570	struct held_lock *hlock_curr;
3571
3572	for (i = curr->lockdep_depth - 1; i >= 0; i--) {
3573		hlock_curr = curr->held_locks + i;
3574		if (hlock_curr->irq_context != hlock->irq_context)
3575			break;
3576
3577	}
3578
3579	return ++i;
3580}
3581
3582#ifdef CONFIG_DEBUG_LOCKDEP
3583/*
3584 * Returns the next chain_key iteration
3585 */
3586static u64 print_chain_key_iteration(u16 hlock_id, u64 chain_key)
3587{
3588	u64 new_chain_key = iterate_chain_key(chain_key, hlock_id);
3589
3590	printk(" hlock_id:%d -> chain_key:%016Lx",
3591		(unsigned int)hlock_id,
3592		(unsigned long long)new_chain_key);
3593	return new_chain_key;
3594}
3595
3596static void
3597print_chain_keys_held_locks(struct task_struct *curr, struct held_lock *hlock_next)
3598{
3599	struct held_lock *hlock;
3600	u64 chain_key = INITIAL_CHAIN_KEY;
3601	int depth = curr->lockdep_depth;
3602	int i = get_first_held_lock(curr, hlock_next);
3603
3604	printk("depth: %u (irq_context %u)\n", depth - i + 1,
3605		hlock_next->irq_context);
3606	for (; i < depth; i++) {
3607		hlock = curr->held_locks + i;
3608		chain_key = print_chain_key_iteration(hlock_id(hlock), chain_key);
3609
3610		print_lock(hlock);
3611	}
3612
3613	print_chain_key_iteration(hlock_id(hlock_next), chain_key);
3614	print_lock(hlock_next);
3615}
3616
3617static void print_chain_keys_chain(struct lock_chain *chain)
3618{
3619	int i;
3620	u64 chain_key = INITIAL_CHAIN_KEY;
3621	u16 hlock_id;
3622
3623	printk("depth: %u\n", chain->depth);
3624	for (i = 0; i < chain->depth; i++) {
3625		hlock_id = chain_hlocks[chain->base + i];
3626		chain_key = print_chain_key_iteration(hlock_id, chain_key);
3627
3628		print_lock_name(NULL, lock_classes + chain_hlock_class_idx(hlock_id));
3629		printk("\n");
3630	}
3631}
3632
3633static void print_collision(struct task_struct *curr,
3634			struct held_lock *hlock_next,
3635			struct lock_chain *chain)
3636{
3637	nbcon_cpu_emergency_enter();
3638
3639	pr_warn("\n");
3640	pr_warn("============================\n");
3641	pr_warn("WARNING: chain_key collision\n");
3642	print_kernel_ident();
3643	pr_warn("----------------------------\n");
3644	pr_warn("%s/%d: ", current->comm, task_pid_nr(current));
3645	pr_warn("Hash chain already cached but the contents don't match!\n");
3646
3647	pr_warn("Held locks:");
3648	print_chain_keys_held_locks(curr, hlock_next);
3649
3650	pr_warn("Locks in cached chain:");
3651	print_chain_keys_chain(chain);
3652
3653	pr_warn("\nstack backtrace:\n");
3654	dump_stack();
3655
3656	nbcon_cpu_emergency_exit();
3657}
3658#endif
3659
3660/*
3661 * Checks whether the chain and the current held locks are consistent
3662 * in depth and also in content. If they are not it most likely means
3663 * that there was a collision during the calculation of the chain_key.
3664 * Returns: 0 not passed, 1 passed
3665 */
3666static int check_no_collision(struct task_struct *curr,
3667			struct held_lock *hlock,
3668			struct lock_chain *chain)
3669{
3670#ifdef CONFIG_DEBUG_LOCKDEP
3671	int i, j, id;
3672
3673	i = get_first_held_lock(curr, hlock);
3674
3675	if (DEBUG_LOCKS_WARN_ON(chain->depth != curr->lockdep_depth - (i - 1))) {
3676		print_collision(curr, hlock, chain);
3677		return 0;
3678	}
3679
3680	for (j = 0; j < chain->depth - 1; j++, i++) {
3681		id = hlock_id(&curr->held_locks[i]);
3682
3683		if (DEBUG_LOCKS_WARN_ON(chain_hlocks[chain->base + j] != id)) {
3684			print_collision(curr, hlock, chain);
3685			return 0;
3686		}
3687	}
3688#endif
3689	return 1;
3690}
3691
3692/*
3693 * Given an index that is >= -1, return the index of the next lock chain.
3694 * Return -2 if there is no next lock chain.
3695 */
3696long lockdep_next_lockchain(long i)
3697{
3698	i = find_next_bit(lock_chains_in_use, ARRAY_SIZE(lock_chains), i + 1);
3699	return i < ARRAY_SIZE(lock_chains) ? i : -2;
3700}
3701
3702unsigned long lock_chain_count(void)
3703{
3704	return bitmap_weight(lock_chains_in_use, ARRAY_SIZE(lock_chains));
3705}
3706
3707/* Must be called with the graph lock held. */
3708static struct lock_chain *alloc_lock_chain(void)
3709{
3710	int idx = find_first_zero_bit(lock_chains_in_use,
3711				      ARRAY_SIZE(lock_chains));
3712
3713	if (unlikely(idx >= ARRAY_SIZE(lock_chains)))
3714		return NULL;
3715	__set_bit(idx, lock_chains_in_use);
3716	return lock_chains + idx;
3717}
3718
3719/*
3720 * Adds a dependency chain into chain hashtable. And must be called with
3721 * graph_lock held.
3722 *
3723 * Return 0 if fail, and graph_lock is released.
3724 * Return 1 if succeed, with graph_lock held.
3725 */
3726static inline int add_chain_cache(struct task_struct *curr,
3727				  struct held_lock *hlock,
3728				  u64 chain_key)
3729{
3730	struct hlist_head *hash_head = chainhashentry(chain_key);
3731	struct lock_chain *chain;
3732	int i, j;
3733
3734	/*
3735	 * The caller must hold the graph lock, ensure we've got IRQs
3736	 * disabled to make this an IRQ-safe lock.. for recursion reasons
3737	 * lockdep won't complain about its own locking errors.
3738	 */
3739	if (lockdep_assert_locked())
3740		return 0;
3741
3742	chain = alloc_lock_chain();
3743	if (!chain) {
3744		if (!debug_locks_off_graph_unlock())
3745			return 0;
3746
3747		nbcon_cpu_emergency_enter();
3748		print_lockdep_off("BUG: MAX_LOCKDEP_CHAINS too low!");
3749		dump_stack();
3750		nbcon_cpu_emergency_exit();
3751		return 0;
3752	}
3753	chain->chain_key = chain_key;
3754	chain->irq_context = hlock->irq_context;
3755	i = get_first_held_lock(curr, hlock);
3756	chain->depth = curr->lockdep_depth + 1 - i;
3757
3758	BUILD_BUG_ON((1UL << 24) <= ARRAY_SIZE(chain_hlocks));
3759	BUILD_BUG_ON((1UL << 6)  <= ARRAY_SIZE(curr->held_locks));
3760	BUILD_BUG_ON((1UL << 8*sizeof(chain_hlocks[0])) <= ARRAY_SIZE(lock_classes));
3761
3762	j = alloc_chain_hlocks(chain->depth);
3763	if (j < 0) {
3764		if (!debug_locks_off_graph_unlock())
3765			return 0;
3766
3767		nbcon_cpu_emergency_enter();
3768		print_lockdep_off("BUG: MAX_LOCKDEP_CHAIN_HLOCKS too low!");
3769		dump_stack();
3770		nbcon_cpu_emergency_exit();
3771		return 0;
3772	}
3773
3774	chain->base = j;
3775	for (j = 0; j < chain->depth - 1; j++, i++) {
3776		int lock_id = hlock_id(curr->held_locks + i);
3777
3778		chain_hlocks[chain->base + j] = lock_id;
3779	}
3780	chain_hlocks[chain->base + j] = hlock_id(hlock);
3781	hlist_add_head_rcu(&chain->entry, hash_head);
3782	debug_atomic_inc(chain_lookup_misses);
3783	inc_chains(chain->irq_context);
3784
3785	return 1;
3786}
3787
3788/*
3789 * Look up a dependency chain. Must be called with either the graph lock or
3790 * the RCU read lock held.
3791 */
3792static inline struct lock_chain *lookup_chain_cache(u64 chain_key)
3793{
3794	struct hlist_head *hash_head = chainhashentry(chain_key);
3795	struct lock_chain *chain;
3796
3797	hlist_for_each_entry_rcu(chain, hash_head, entry) {
3798		if (READ_ONCE(chain->chain_key) == chain_key) {
3799			debug_atomic_inc(chain_lookup_hits);
3800			return chain;
3801		}
3802	}
3803	return NULL;
3804}
3805
3806/*
3807 * If the key is not present yet in dependency chain cache then
3808 * add it and return 1 - in this case the new dependency chain is
3809 * validated. If the key is already hashed, return 0.
3810 * (On return with 1 graph_lock is held.)
3811 */
3812static inline int lookup_chain_cache_add(struct task_struct *curr,
3813					 struct held_lock *hlock,
3814					 u64 chain_key)
3815{
3816	struct lock_class *class = hlock_class(hlock);
3817	struct lock_chain *chain = lookup_chain_cache(chain_key);
3818
3819	if (chain) {
3820cache_hit:
3821		if (!check_no_collision(curr, hlock, chain))
3822			return 0;
3823
3824		if (very_verbose(class)) {
3825			printk("\nhash chain already cached, key: "
3826					"%016Lx tail class: [%px] %s\n",
3827					(unsigned long long)chain_key,
3828					class->key, class->name);
3829		}
3830
3831		return 0;
3832	}
3833
3834	if (very_verbose(class)) {
3835		printk("\nnew hash chain, key: %016Lx tail class: [%px] %s\n",
3836			(unsigned long long)chain_key, class->key, class->name);
3837	}
3838
3839	if (!graph_lock())
3840		return 0;
3841
3842	/*
3843	 * We have to walk the chain again locked - to avoid duplicates:
3844	 */
3845	chain = lookup_chain_cache(chain_key);
3846	if (chain) {
3847		graph_unlock();
3848		goto cache_hit;
3849	}
3850
3851	if (!add_chain_cache(curr, hlock, chain_key))
3852		return 0;
3853
3854	return 1;
3855}
3856
3857static int validate_chain(struct task_struct *curr,
3858			  struct held_lock *hlock,
3859			  int chain_head, u64 chain_key)
3860{
3861	/*
3862	 * Trylock needs to maintain the stack of held locks, but it
3863	 * does not add new dependencies, because trylock can be done
3864	 * in any order.
3865	 *
3866	 * We look up the chain_key and do the O(N^2) check and update of
3867	 * the dependencies only if this is a new dependency chain.
3868	 * (If lookup_chain_cache_add() return with 1 it acquires
3869	 * graph_lock for us)
3870	 */
3871	if (!hlock->trylock && hlock->check &&
3872	    lookup_chain_cache_add(curr, hlock, chain_key)) {
3873		/*
3874		 * Check whether last held lock:
3875		 *
3876		 * - is irq-safe, if this lock is irq-unsafe
3877		 * - is softirq-safe, if this lock is hardirq-unsafe
3878		 *
3879		 * And check whether the new lock's dependency graph
3880		 * could lead back to the previous lock:
3881		 *
3882		 * - within the current held-lock stack
3883		 * - across our accumulated lock dependency records
3884		 *
3885		 * any of these scenarios could lead to a deadlock.
3886		 */
3887		/*
3888		 * The simple case: does the current hold the same lock
3889		 * already?
3890		 */
3891		int ret = check_deadlock(curr, hlock);
3892
3893		if (!ret)
3894			return 0;
3895		/*
3896		 * Add dependency only if this lock is not the head
3897		 * of the chain, and if the new lock introduces no more
3898		 * lock dependency (because we already hold a lock with the
3899		 * same lock class) nor deadlock (because the nest_lock
3900		 * serializes nesting locks), see the comments for
3901		 * check_deadlock().
3902		 */
3903		if (!chain_head && ret != 2) {
3904			if (!check_prevs_add(curr, hlock))
3905				return 0;
3906		}
3907
3908		graph_unlock();
3909	} else {
3910		/* after lookup_chain_cache_add(): */
3911		if (unlikely(!debug_locks))
3912			return 0;
3913	}
3914
3915	return 1;
3916}
3917#else
3918static inline int validate_chain(struct task_struct *curr,
3919				 struct held_lock *hlock,
3920				 int chain_head, u64 chain_key)
3921{
3922	return 1;
3923}
3924
3925static void init_chain_block_buckets(void)	{ }
3926#endif /* CONFIG_PROVE_LOCKING */
3927
3928/*
3929 * We are building curr_chain_key incrementally, so double-check
3930 * it from scratch, to make sure that it's done correctly:
3931 */
3932static void check_chain_key(struct task_struct *curr)
3933{
3934#ifdef CONFIG_DEBUG_LOCKDEP
3935	struct held_lock *hlock, *prev_hlock = NULL;
3936	unsigned int i;
3937	u64 chain_key = INITIAL_CHAIN_KEY;
3938
3939	for (i = 0; i < curr->lockdep_depth; i++) {
3940		hlock = curr->held_locks + i;
3941		if (chain_key != hlock->prev_chain_key) {
3942			debug_locks_off();
3943			/*
3944			 * We got mighty confused, our chain keys don't match
3945			 * with what we expect, someone trample on our task state?
3946			 */
3947			WARN(1, "hm#1, depth: %u [%u], %016Lx != %016Lx\n",
3948				curr->lockdep_depth, i,
3949				(unsigned long long)chain_key,
3950				(unsigned long long)hlock->prev_chain_key);
3951			return;
3952		}
3953
3954		/*
3955		 * hlock->class_idx can't go beyond MAX_LOCKDEP_KEYS, but is
3956		 * it registered lock class index?
3957		 */
3958		if (DEBUG_LOCKS_WARN_ON(!test_bit(hlock->class_idx, lock_classes_in_use)))
3959			return;
3960
3961		if (prev_hlock && (prev_hlock->irq_context !=
3962							hlock->irq_context))
3963			chain_key = INITIAL_CHAIN_KEY;
3964		chain_key = iterate_chain_key(chain_key, hlock_id(hlock));
3965		prev_hlock = hlock;
3966	}
3967	if (chain_key != curr->curr_chain_key) {
3968		debug_locks_off();
3969		/*
3970		 * More smoking hash instead of calculating it, damn see these
3971		 * numbers float.. I bet that a pink elephant stepped on my memory.
3972		 */
3973		WARN(1, "hm#2, depth: %u [%u], %016Lx != %016Lx\n",
3974			curr->lockdep_depth, i,
3975			(unsigned long long)chain_key,
3976			(unsigned long long)curr->curr_chain_key);
3977	}
3978#endif
3979}
3980
3981#ifdef CONFIG_PROVE_LOCKING
3982static int mark_lock(struct task_struct *curr, struct held_lock *this,
3983		     enum lock_usage_bit new_bit);
3984
3985static void print_usage_bug_scenario(struct held_lock *lock)
3986{
3987	struct lock_class *class = hlock_class(lock);
3988
3989	printk(" Possible unsafe locking scenario:\n\n");
3990	printk("       CPU0\n");
3991	printk("       ----\n");
3992	printk("  lock(");
3993	__print_lock_name(lock, class);
3994	printk(KERN_CONT ");\n");
3995	printk("  <Interrupt>\n");
3996	printk("    lock(");
3997	__print_lock_name(lock, class);
3998	printk(KERN_CONT ");\n");
3999	printk("\n *** DEADLOCK ***\n\n");
4000}
4001
4002static void
4003print_usage_bug(struct task_struct *curr, struct held_lock *this,
4004		enum lock_usage_bit prev_bit, enum lock_usage_bit new_bit)
4005{
4006	if (!debug_locks_off() || debug_locks_silent)
4007		return;
4008
4009	nbcon_cpu_emergency_enter();
4010
4011	pr_warn("\n");
4012	pr_warn("================================\n");
4013	pr_warn("WARNING: inconsistent lock state\n");
4014	print_kernel_ident();
4015	pr_warn("--------------------------------\n");
4016
4017	pr_warn("inconsistent {%s} -> {%s} usage.\n",
4018		usage_str[prev_bit], usage_str[new_bit]);
4019
4020	pr_warn("%s/%d [HC%u[%lu]:SC%u[%lu]:HE%u:SE%u] takes:\n",
4021		curr->comm, task_pid_nr(curr),
4022		lockdep_hardirq_context(), hardirq_count() >> HARDIRQ_SHIFT,
4023		lockdep_softirq_context(curr), softirq_count() >> SOFTIRQ_SHIFT,
4024		lockdep_hardirqs_enabled(),
4025		lockdep_softirqs_enabled(curr));
4026	print_lock(this);
4027
4028	pr_warn("{%s} state was registered at:\n", usage_str[prev_bit]);
4029	print_lock_trace(hlock_class(this)->usage_traces[prev_bit], 1);
4030
4031	print_irqtrace_events(curr);
4032	pr_warn("\nother info that might help us debug this:\n");
4033	print_usage_bug_scenario(this);
4034
4035	lockdep_print_held_locks(curr);
4036
4037	pr_warn("\nstack backtrace:\n");
4038	dump_stack();
4039
4040	nbcon_cpu_emergency_exit();
4041}
4042
4043/*
4044 * Print out an error if an invalid bit is set:
4045 */
4046static inline int
4047valid_state(struct task_struct *curr, struct held_lock *this,
4048	    enum lock_usage_bit new_bit, enum lock_usage_bit bad_bit)
4049{
4050	if (unlikely(hlock_class(this)->usage_mask & (1 << bad_bit))) {
4051		graph_unlock();
4052		print_usage_bug(curr, this, bad_bit, new_bit);
4053		return 0;
4054	}
4055	return 1;
4056}
4057
4058
4059/*
4060 * print irq inversion bug:
4061 */
4062static void
4063print_irq_inversion_bug(struct task_struct *curr,
4064			struct lock_list *root, struct lock_list *other,
4065			struct held_lock *this, int forwards,
4066			const char *irqclass)
4067{
4068	struct lock_list *entry = other;
4069	struct lock_list *middle = NULL;
4070	int depth;
4071
4072	if (!debug_locks_off_graph_unlock() || debug_locks_silent)
4073		return;
4074
4075	nbcon_cpu_emergency_enter();
4076
4077	pr_warn("\n");
4078	pr_warn("========================================================\n");
4079	pr_warn("WARNING: possible irq lock inversion dependency detected\n");
4080	print_kernel_ident();
4081	pr_warn("--------------------------------------------------------\n");
4082	pr_warn("%s/%d just changed the state of lock:\n",
4083		curr->comm, task_pid_nr(curr));
4084	print_lock(this);
4085	if (forwards)
4086		pr_warn("but this lock took another, %s-unsafe lock in the past:\n", irqclass);
4087	else
4088		pr_warn("but this lock was taken by another, %s-safe lock in the past:\n", irqclass);
4089	print_lock_name(NULL, other->class);
4090	pr_warn("\n\nand interrupts could create inverse lock ordering between them.\n\n");
4091
4092	pr_warn("\nother info that might help us debug this:\n");
4093
4094	/* Find a middle lock (if one exists) */
4095	depth = get_lock_depth(other);
4096	do {
4097		if (depth == 0 && (entry != root)) {
4098			pr_warn("lockdep:%s bad path found in chain graph\n", __func__);
4099			break;
4100		}
4101		middle = entry;
4102		entry = get_lock_parent(entry);
4103		depth--;
4104	} while (entry && entry != root && (depth >= 0));
4105	if (forwards)
4106		print_irq_lock_scenario(root, other,
4107			middle ? middle->class : root->class, other->class);
4108	else
4109		print_irq_lock_scenario(other, root,
4110			middle ? middle->class : other->class, root->class);
4111
4112	lockdep_print_held_locks(curr);
4113
4114	pr_warn("\nthe shortest dependencies between 2nd lock and 1st lock:\n");
4115	root->trace = save_trace();
4116	if (!root->trace)
4117		goto out;
4118	print_shortest_lock_dependencies(other, root);
4119
4120	pr_warn("\nstack backtrace:\n");
4121	dump_stack();
4122out:
4123	nbcon_cpu_emergency_exit();
4124}
4125
4126/*
4127 * Prove that in the forwards-direction subgraph starting at <this>
4128 * there is no lock matching <mask>:
4129 */
4130static int
4131check_usage_forwards(struct task_struct *curr, struct held_lock *this,
4132		     enum lock_usage_bit bit)
4133{
4134	enum bfs_result ret;
4135	struct lock_list root;
4136	struct lock_list *target_entry;
4137	enum lock_usage_bit read_bit = bit + LOCK_USAGE_READ_MASK;
4138	unsigned usage_mask = lock_flag(bit) | lock_flag(read_bit);
4139
4140	bfs_init_root(&root, this);
4141	ret = find_usage_forwards(&root, usage_mask, &target_entry);
4142	if (bfs_error(ret)) {
4143		print_bfs_bug(ret);
4144		return 0;
4145	}
4146	if (ret == BFS_RNOMATCH)
4147		return 1;
4148
4149	/* Check whether write or read usage is the match */
4150	if (target_entry->class->usage_mask & lock_flag(bit)) {
4151		print_irq_inversion_bug(curr, &root, target_entry,
4152					this, 1, state_name(bit));
4153	} else {
4154		print_irq_inversion_bug(curr, &root, target_entry,
4155					this, 1, state_name(read_bit));
4156	}
4157
4158	return 0;
4159}
4160
4161/*
4162 * Prove that in the backwards-direction subgraph starting at <this>
4163 * there is no lock matching <mask>:
4164 */
4165static int
4166check_usage_backwards(struct task_struct *curr, struct held_lock *this,
4167		      enum lock_usage_bit bit)
4168{
4169	enum bfs_result ret;
4170	struct lock_list root;
4171	struct lock_list *target_entry;
4172	enum lock_usage_bit read_bit = bit + LOCK_USAGE_READ_MASK;
4173	unsigned usage_mask = lock_flag(bit) | lock_flag(read_bit);
4174
4175	bfs_init_rootb(&root, this);
4176	ret = find_usage_backwards(&root, usage_mask, &target_entry);
4177	if (bfs_error(ret)) {
4178		print_bfs_bug(ret);
4179		return 0;
4180	}
4181	if (ret == BFS_RNOMATCH)
4182		return 1;
4183
4184	/* Check whether write or read usage is the match */
4185	if (target_entry->class->usage_mask & lock_flag(bit)) {
4186		print_irq_inversion_bug(curr, &root, target_entry,
4187					this, 0, state_name(bit));
4188	} else {
4189		print_irq_inversion_bug(curr, &root, target_entry,
4190					this, 0, state_name(read_bit));
4191	}
4192
4193	return 0;
4194}
4195
4196void print_irqtrace_events(struct task_struct *curr)
4197{
4198	const struct irqtrace_events *trace = &curr->irqtrace;
4199
4200	nbcon_cpu_emergency_enter();
4201
4202	printk("irq event stamp: %u\n", trace->irq_events);
4203	printk("hardirqs last  enabled at (%u): [<%px>] %pS\n",
4204		trace->hardirq_enable_event, (void *)trace->hardirq_enable_ip,
4205		(void *)trace->hardirq_enable_ip);
4206	printk("hardirqs last disabled at (%u): [<%px>] %pS\n",
4207		trace->hardirq_disable_event, (void *)trace->hardirq_disable_ip,
4208		(void *)trace->hardirq_disable_ip);
4209	printk("softirqs last  enabled at (%u): [<%px>] %pS\n",
4210		trace->softirq_enable_event, (void *)trace->softirq_enable_ip,
4211		(void *)trace->softirq_enable_ip);
4212	printk("softirqs last disabled at (%u): [<%px>] %pS\n",
4213		trace->softirq_disable_event, (void *)trace->softirq_disable_ip,
4214		(void *)trace->softirq_disable_ip);
4215
4216	nbcon_cpu_emergency_exit();
4217}
4218
4219static int HARDIRQ_verbose(struct lock_class *class)
4220{
4221#if HARDIRQ_VERBOSE
4222	return class_filter(class);
4223#endif
4224	return 0;
4225}
4226
4227static int SOFTIRQ_verbose(struct lock_class *class)
4228{
4229#if SOFTIRQ_VERBOSE
4230	return class_filter(class);
4231#endif
4232	return 0;
4233}
4234
4235static int (*state_verbose_f[])(struct lock_class *class) = {
4236#define LOCKDEP_STATE(__STATE) \
4237	__STATE##_verbose,
4238#include "lockdep_states.h"
4239#undef LOCKDEP_STATE
4240};
4241
4242static inline int state_verbose(enum lock_usage_bit bit,
4243				struct lock_class *class)
4244{
4245	return state_verbose_f[bit >> LOCK_USAGE_DIR_MASK](class);
4246}
4247
4248typedef int (*check_usage_f)(struct task_struct *, struct held_lock *,
4249			     enum lock_usage_bit bit, const char *name);
4250
4251static int
4252mark_lock_irq(struct task_struct *curr, struct held_lock *this,
4253		enum lock_usage_bit new_bit)
4254{
4255	int excl_bit = exclusive_bit(new_bit);
4256	int read = new_bit & LOCK_USAGE_READ_MASK;
4257	int dir = new_bit & LOCK_USAGE_DIR_MASK;
4258
4259	/*
4260	 * Validate that this particular lock does not have conflicting
4261	 * usage states.
4262	 */
4263	if (!valid_state(curr, this, new_bit, excl_bit))
4264		return 0;
4265
4266	/*
4267	 * Check for read in write conflicts
4268	 */
4269	if (!read && !valid_state(curr, this, new_bit,
4270				  excl_bit + LOCK_USAGE_READ_MASK))
4271		return 0;
4272
4273
4274	/*
4275	 * Validate that the lock dependencies don't have conflicting usage
4276	 * states.
4277	 */
4278	if (dir) {
4279		/*
4280		 * mark ENABLED has to look backwards -- to ensure no dependee
4281		 * has USED_IN state, which, again, would allow  recursion deadlocks.
4282		 */
4283		if (!check_usage_backwards(curr, this, excl_bit))
4284			return 0;
4285	} else {
4286		/*
4287		 * mark USED_IN has to look forwards -- to ensure no dependency
4288		 * has ENABLED state, which would allow recursion deadlocks.
4289		 */
4290		if (!check_usage_forwards(curr, this, excl_bit))
4291			return 0;
4292	}
4293
4294	if (state_verbose(new_bit, hlock_class(this)))
4295		return 2;
4296
4297	return 1;
4298}
4299
4300/*
4301 * Mark all held locks with a usage bit:
4302 */
4303static int
4304mark_held_locks(struct task_struct *curr, enum lock_usage_bit base_bit)
4305{
4306	struct held_lock *hlock;
4307	int i;
4308
4309	for (i = 0; i < curr->lockdep_depth; i++) {
4310		enum lock_usage_bit hlock_bit = base_bit;
4311		hlock = curr->held_locks + i;
4312
4313		if (hlock->read)
4314			hlock_bit += LOCK_USAGE_READ_MASK;
4315
4316		BUG_ON(hlock_bit >= LOCK_USAGE_STATES);
4317
4318		if (!hlock->check)
4319			continue;
4320
4321		if (!mark_lock(curr, hlock, hlock_bit))
4322			return 0;
4323	}
4324
4325	return 1;
4326}
4327
4328/*
4329 * Hardirqs will be enabled:
4330 */
4331static void __trace_hardirqs_on_caller(void)
4332{
4333	struct task_struct *curr = current;
4334
4335	/*
4336	 * We are going to turn hardirqs on, so set the
4337	 * usage bit for all held locks:
4338	 */
4339	if (!mark_held_locks(curr, LOCK_ENABLED_HARDIRQ))
4340		return;
4341	/*
4342	 * If we have softirqs enabled, then set the usage
4343	 * bit for all held locks. (disabled hardirqs prevented
4344	 * this bit from being set before)
4345	 */
4346	if (curr->softirqs_enabled)
4347		mark_held_locks(curr, LOCK_ENABLED_SOFTIRQ);
4348}
4349
4350/**
4351 * lockdep_hardirqs_on_prepare - Prepare for enabling interrupts
4352 *
4353 * Invoked before a possible transition to RCU idle from exit to user or
4354 * guest mode. This ensures that all RCU operations are done before RCU
4355 * stops watching. After the RCU transition lockdep_hardirqs_on() has to be
4356 * invoked to set the final state.
4357 */
4358void lockdep_hardirqs_on_prepare(void)
4359{
4360	if (unlikely(!debug_locks))
4361		return;
4362
4363	/*
4364	 * NMIs do not (and cannot) track lock dependencies, nothing to do.
4365	 */
4366	if (unlikely(in_nmi()))
4367		return;
4368
4369	if (unlikely(this_cpu_read(lockdep_recursion)))
4370		return;
4371
4372	if (unlikely(lockdep_hardirqs_enabled())) {
4373		/*
4374		 * Neither irq nor preemption are disabled here
4375		 * so this is racy by nature but losing one hit
4376		 * in a stat is not a big deal.
4377		 */
4378		__debug_atomic_inc(redundant_hardirqs_on);
4379		return;
4380	}
4381
4382	/*
4383	 * We're enabling irqs and according to our state above irqs weren't
4384	 * already enabled, yet we find the hardware thinks they are in fact
4385	 * enabled.. someone messed up their IRQ state tracing.
4386	 */
4387	if (DEBUG_LOCKS_WARN_ON(!irqs_disabled()))
4388		return;
4389
4390	/*
4391	 * See the fine text that goes along with this variable definition.
4392	 */
4393	if (DEBUG_LOCKS_WARN_ON(early_boot_irqs_disabled))
4394		return;
4395
4396	/*
4397	 * Can't allow enabling interrupts while in an interrupt handler,
4398	 * that's general bad form and such. Recursion, limited stack etc..
4399	 */
4400	if (DEBUG_LOCKS_WARN_ON(lockdep_hardirq_context()))
4401		return;
4402
4403	current->hardirq_chain_key = current->curr_chain_key;
4404
4405	lockdep_recursion_inc();
4406	__trace_hardirqs_on_caller();
4407	lockdep_recursion_finish();
4408}
4409EXPORT_SYMBOL_GPL(lockdep_hardirqs_on_prepare);
4410
4411void noinstr lockdep_hardirqs_on(unsigned long ip)
4412{
4413	struct irqtrace_events *trace = &current->irqtrace;
4414
4415	if (unlikely(!debug_locks))
4416		return;
4417
4418	/*
4419	 * NMIs can happen in the middle of local_irq_{en,dis}able() where the
4420	 * tracking state and hardware state are out of sync.
4421	 *
4422	 * NMIs must save lockdep_hardirqs_enabled() to restore IRQ state from,
4423	 * and not rely on hardware state like normal interrupts.
4424	 */
4425	if (unlikely(in_nmi())) {
4426		if (!IS_ENABLED(CONFIG_TRACE_IRQFLAGS_NMI))
4427			return;
4428
4429		/*
4430		 * Skip:
4431		 *  - recursion check, because NMI can hit lockdep;
4432		 *  - hardware state check, because above;
4433		 *  - chain_key check, see lockdep_hardirqs_on_prepare().
4434		 */
4435		goto skip_checks;
4436	}
4437
4438	if (unlikely(this_cpu_read(lockdep_recursion)))
4439		return;
4440
4441	if (lockdep_hardirqs_enabled()) {
4442		/*
4443		 * Neither irq nor preemption are disabled here
4444		 * so this is racy by nature but losing one hit
4445		 * in a stat is not a big deal.
4446		 */
4447		__debug_atomic_inc(redundant_hardirqs_on);
4448		return;
4449	}
4450
4451	/*
4452	 * We're enabling irqs and according to our state above irqs weren't
4453	 * already enabled, yet we find the hardware thinks they are in fact
4454	 * enabled.. someone messed up their IRQ state tracing.
4455	 */
4456	if (DEBUG_LOCKS_WARN_ON(!irqs_disabled()))
4457		return;
4458
4459	/*
4460	 * Ensure the lock stack remained unchanged between
4461	 * lockdep_hardirqs_on_prepare() and lockdep_hardirqs_on().
4462	 */
4463	DEBUG_LOCKS_WARN_ON(current->hardirq_chain_key !=
4464			    current->curr_chain_key);
4465
4466skip_checks:
4467	/* we'll do an OFF -> ON transition: */
4468	__this_cpu_write(hardirqs_enabled, 1);
4469	trace->hardirq_enable_ip = ip;
4470	trace->hardirq_enable_event = ++trace->irq_events;
4471	debug_atomic_inc(hardirqs_on_events);
4472}
4473EXPORT_SYMBOL_GPL(lockdep_hardirqs_on);
4474
4475/*
4476 * Hardirqs were disabled:
4477 */
4478void noinstr lockdep_hardirqs_off(unsigned long ip)
4479{
4480	if (unlikely(!debug_locks))
4481		return;
4482
4483	/*
4484	 * Matching lockdep_hardirqs_on(), allow NMIs in the middle of lockdep;
4485	 * they will restore the software state. This ensures the software
4486	 * state is consistent inside NMIs as well.
4487	 */
4488	if (in_nmi()) {
4489		if (!IS_ENABLED(CONFIG_TRACE_IRQFLAGS_NMI))
4490			return;
4491	} else if (__this_cpu_read(lockdep_recursion))
4492		return;
4493
4494	/*
4495	 * So we're supposed to get called after you mask local IRQs, but for
4496	 * some reason the hardware doesn't quite think you did a proper job.
4497	 */
4498	if (DEBUG_LOCKS_WARN_ON(!irqs_disabled()))
4499		return;
4500
4501	if (lockdep_hardirqs_enabled()) {
4502		struct irqtrace_events *trace = &current->irqtrace;
4503
4504		/*
4505		 * We have done an ON -> OFF transition:
4506		 */
4507		__this_cpu_write(hardirqs_enabled, 0);
4508		trace->hardirq_disable_ip = ip;
4509		trace->hardirq_disable_event = ++trace->irq_events;
4510		debug_atomic_inc(hardirqs_off_events);
4511	} else {
4512		debug_atomic_inc(redundant_hardirqs_off);
4513	}
4514}
4515EXPORT_SYMBOL_GPL(lockdep_hardirqs_off);
4516
4517/*
4518 * Softirqs will be enabled:
4519 */
4520void lockdep_softirqs_on(unsigned long ip)
4521{
4522	struct irqtrace_events *trace = &current->irqtrace;
4523
4524	if (unlikely(!lockdep_enabled()))
4525		return;
4526
4527	/*
4528	 * We fancy IRQs being disabled here, see softirq.c, avoids
4529	 * funny state and nesting things.
4530	 */
4531	if (DEBUG_LOCKS_WARN_ON(!irqs_disabled()))
4532		return;
4533
4534	if (current->softirqs_enabled) {
4535		debug_atomic_inc(redundant_softirqs_on);
4536		return;
4537	}
4538
4539	lockdep_recursion_inc();
4540	/*
4541	 * We'll do an OFF -> ON transition:
4542	 */
4543	current->softirqs_enabled = 1;
4544	trace->softirq_enable_ip = ip;
4545	trace->softirq_enable_event = ++trace->irq_events;
4546	debug_atomic_inc(softirqs_on_events);
4547	/*
4548	 * We are going to turn softirqs on, so set the
4549	 * usage bit for all held locks, if hardirqs are
4550	 * enabled too:
4551	 */
4552	if (lockdep_hardirqs_enabled())
4553		mark_held_locks(current, LOCK_ENABLED_SOFTIRQ);
4554	lockdep_recursion_finish();
4555}
4556
4557/*
4558 * Softirqs were disabled:
4559 */
4560void lockdep_softirqs_off(unsigned long ip)
4561{
4562	if (unlikely(!lockdep_enabled()))
4563		return;
4564
4565	/*
4566	 * We fancy IRQs being disabled here, see softirq.c
4567	 */
4568	if (DEBUG_LOCKS_WARN_ON(!irqs_disabled()))
4569		return;
4570
4571	if (current->softirqs_enabled) {
4572		struct irqtrace_events *trace = &current->irqtrace;
4573
4574		/*
4575		 * We have done an ON -> OFF transition:
4576		 */
4577		current->softirqs_enabled = 0;
4578		trace->softirq_disable_ip = ip;
4579		trace->softirq_disable_event = ++trace->irq_events;
4580		debug_atomic_inc(softirqs_off_events);
4581		/*
4582		 * Whoops, we wanted softirqs off, so why aren't they?
4583		 */
4584		DEBUG_LOCKS_WARN_ON(!softirq_count());
4585	} else
4586		debug_atomic_inc(redundant_softirqs_off);
4587}
4588
4589/**
4590 * lockdep_cleanup_dead_cpu - Ensure CPU lockdep state is cleanly stopped
4591 *
4592 * @cpu: index of offlined CPU
4593 * @idle: task pointer for offlined CPU's idle thread
4594 *
4595 * Invoked after the CPU is dead. Ensures that the tracing infrastructure
4596 * is left in a suitable state for the CPU to be subsequently brought
4597 * online again.
4598 */
4599void lockdep_cleanup_dead_cpu(unsigned int cpu, struct task_struct *idle)
4600{
4601	if (unlikely(!debug_locks))
4602		return;
4603
4604	if (unlikely(per_cpu(hardirqs_enabled, cpu))) {
4605		pr_warn("CPU %u left hardirqs enabled!", cpu);
4606		if (idle)
4607			print_irqtrace_events(idle);
4608		/* Clean it up for when the CPU comes online again. */
4609		per_cpu(hardirqs_enabled, cpu) = 0;
4610	}
4611}
4612
4613static int
4614mark_usage(struct task_struct *curr, struct held_lock *hlock, int check)
4615{
4616	if (!check)
4617		goto lock_used;
4618
4619	/*
4620	 * If non-trylock use in a hardirq or softirq context, then
4621	 * mark the lock as used in these contexts:
4622	 */
4623	if (!hlock->trylock) {
4624		if (hlock->read) {
4625			if (lockdep_hardirq_context())
4626				if (!mark_lock(curr, hlock,
4627						LOCK_USED_IN_HARDIRQ_READ))
4628					return 0;
4629			if (curr->softirq_context)
4630				if (!mark_lock(curr, hlock,
4631						LOCK_USED_IN_SOFTIRQ_READ))
4632					return 0;
4633		} else {
4634			if (lockdep_hardirq_context())
4635				if (!mark_lock(curr, hlock, LOCK_USED_IN_HARDIRQ))
4636					return 0;
4637			if (curr->softirq_context)
4638				if (!mark_lock(curr, hlock, LOCK_USED_IN_SOFTIRQ))
4639					return 0;
4640		}
4641	}
4642
4643	/*
4644	 * For lock_sync(), don't mark the ENABLED usage, since lock_sync()
4645	 * creates no critical section and no extra dependency can be introduced
4646	 * by interrupts
4647	 */
4648	if (!hlock->hardirqs_off && !hlock->sync) {
4649		if (hlock->read) {
4650			if (!mark_lock(curr, hlock,
4651					LOCK_ENABLED_HARDIRQ_READ))
4652				return 0;
4653			if (curr->softirqs_enabled)
4654				if (!mark_lock(curr, hlock,
4655						LOCK_ENABLED_SOFTIRQ_READ))
4656					return 0;
4657		} else {
4658			if (!mark_lock(curr, hlock,
4659					LOCK_ENABLED_HARDIRQ))
4660				return 0;
4661			if (curr->softirqs_enabled)
4662				if (!mark_lock(curr, hlock,
4663						LOCK_ENABLED_SOFTIRQ))
4664					return 0;
4665		}
4666	}
4667
4668lock_used:
4669	/* mark it as used: */
4670	if (!mark_lock(curr, hlock, LOCK_USED))
4671		return 0;
4672
4673	return 1;
4674}
4675
4676static inline unsigned int task_irq_context(struct task_struct *task)
4677{
4678	return LOCK_CHAIN_HARDIRQ_CONTEXT * !!lockdep_hardirq_context() +
4679	       LOCK_CHAIN_SOFTIRQ_CONTEXT * !!task->softirq_context;
4680}
4681
4682static int separate_irq_context(struct task_struct *curr,
4683		struct held_lock *hlock)
4684{
4685	unsigned int depth = curr->lockdep_depth;
4686
4687	/*
4688	 * Keep track of points where we cross into an interrupt context:
4689	 */
4690	if (depth) {
4691		struct held_lock *prev_hlock;
4692
4693		prev_hlock = curr->held_locks + depth-1;
4694		/*
4695		 * If we cross into another context, reset the
4696		 * hash key (this also prevents the checking and the
4697		 * adding of the dependency to 'prev'):
4698		 */
4699		if (prev_hlock->irq_context != hlock->irq_context)
4700			return 1;
4701	}
4702	return 0;
4703}
4704
4705/*
4706 * Mark a lock with a usage bit, and validate the state transition:
4707 */
4708static int mark_lock(struct task_struct *curr, struct held_lock *this,
4709			     enum lock_usage_bit new_bit)
4710{
4711	unsigned int new_mask, ret = 1;
4712
4713	if (new_bit >= LOCK_USAGE_STATES) {
4714		DEBUG_LOCKS_WARN_ON(1);
4715		return 0;
4716	}
4717
4718	if (new_bit == LOCK_USED && this->read)
4719		new_bit = LOCK_USED_READ;
4720
4721	new_mask = 1 << new_bit;
4722
4723	/*
4724	 * If already set then do not dirty the cacheline,
4725	 * nor do any checks:
4726	 */
4727	if (likely(hlock_class(this)->usage_mask & new_mask))
4728		return 1;
4729
4730	if (!graph_lock())
4731		return 0;
4732	/*
4733	 * Make sure we didn't race:
4734	 */
4735	if (unlikely(hlock_class(this)->usage_mask & new_mask))
4736		goto unlock;
4737
4738	if (!hlock_class(this)->usage_mask)
4739		debug_atomic_dec(nr_unused_locks);
4740
4741	hlock_class(this)->usage_mask |= new_mask;
4742
4743	if (new_bit < LOCK_TRACE_STATES) {
4744		if (!(hlock_class(this)->usage_traces[new_bit] = save_trace()))
4745			return 0;
4746	}
4747
4748	if (new_bit < LOCK_USED) {
4749		ret = mark_lock_irq(curr, this, new_bit);
4750		if (!ret)
4751			return 0;
4752	}
4753
4754unlock:
4755	graph_unlock();
4756
4757	/*
4758	 * We must printk outside of the graph_lock:
4759	 */
4760	if (ret == 2) {
4761		nbcon_cpu_emergency_enter();
4762		printk("\nmarked lock as {%s}:\n", usage_str[new_bit]);
4763		print_lock(this);
4764		print_irqtrace_events(curr);
4765		dump_stack();
4766		nbcon_cpu_emergency_exit();
4767	}
4768
4769	return ret;
4770}
4771
4772static inline short task_wait_context(struct task_struct *curr)
4773{
4774	/*
4775	 * Set appropriate wait type for the context; for IRQs we have to take
4776	 * into account force_irqthread as that is implied by PREEMPT_RT.
4777	 */
4778	if (lockdep_hardirq_context()) {
4779		/*
4780		 * Check if force_irqthreads will run us threaded.
4781		 */
4782		if (curr->hardirq_threaded || curr->irq_config)
4783			return LD_WAIT_CONFIG;
4784
4785		return LD_WAIT_SPIN;
4786	} else if (curr->softirq_context) {
4787		/*
4788		 * Softirqs are always threaded.
4789		 */
4790		return LD_WAIT_CONFIG;
4791	}
4792
4793	return LD_WAIT_MAX;
4794}
4795
4796static int
4797print_lock_invalid_wait_context(struct task_struct *curr,
4798				struct held_lock *hlock)
4799{
4800	short curr_inner;
4801
4802	if (!debug_locks_off())
4803		return 0;
4804	if (debug_locks_silent)
4805		return 0;
4806
4807	nbcon_cpu_emergency_enter();
4808
4809	pr_warn("\n");
4810	pr_warn("=============================\n");
4811	pr_warn("[ BUG: Invalid wait context ]\n");
4812	print_kernel_ident();
4813	pr_warn("-----------------------------\n");
4814
4815	pr_warn("%s/%d is trying to lock:\n", curr->comm, task_pid_nr(curr));
4816	print_lock(hlock);
4817
4818	pr_warn("other info that might help us debug this:\n");
4819
4820	curr_inner = task_wait_context(curr);
4821	pr_warn("context-{%d:%d}\n", curr_inner, curr_inner);
4822
4823	lockdep_print_held_locks(curr);
4824
4825	pr_warn("stack backtrace:\n");
4826	dump_stack();
4827
4828	nbcon_cpu_emergency_exit();
4829
4830	return 0;
4831}
4832
4833/*
4834 * Verify the wait_type context.
4835 *
4836 * This check validates we take locks in the right wait-type order; that is it
4837 * ensures that we do not take mutexes inside spinlocks and do not attempt to
4838 * acquire spinlocks inside raw_spinlocks and the sort.
4839 *
4840 * The entire thing is slightly more complex because of RCU, RCU is a lock that
4841 * can be taken from (pretty much) any context but also has constraints.
4842 * However when taken in a stricter environment the RCU lock does not loosen
4843 * the constraints.
4844 *
4845 * Therefore we must look for the strictest environment in the lock stack and
4846 * compare that to the lock we're trying to acquire.
4847 */
4848static int check_wait_context(struct task_struct *curr, struct held_lock *next)
4849{
4850	u8 next_inner = hlock_class(next)->wait_type_inner;
4851	u8 next_outer = hlock_class(next)->wait_type_outer;
4852	u8 curr_inner;
4853	int depth;
4854
4855	if (!next_inner || next->trylock)
4856		return 0;
4857
4858	if (!next_outer)
4859		next_outer = next_inner;
4860
4861	/*
4862	 * Find start of current irq_context..
4863	 */
4864	for (depth = curr->lockdep_depth - 1; depth >= 0; depth--) {
4865		struct held_lock *prev = curr->held_locks + depth;
4866		if (prev->irq_context != next->irq_context)
4867			break;
4868	}
4869	depth++;
4870
4871	curr_inner = task_wait_context(curr);
4872
4873	for (; depth < curr->lockdep_depth; depth++) {
4874		struct held_lock *prev = curr->held_locks + depth;
4875		struct lock_class *class = hlock_class(prev);
4876		u8 prev_inner = class->wait_type_inner;
4877
4878		if (prev_inner) {
4879			/*
4880			 * We can have a bigger inner than a previous one
4881			 * when outer is smaller than inner, as with RCU.
4882			 *
4883			 * Also due to trylocks.
4884			 */
4885			curr_inner = min(curr_inner, prev_inner);
4886
4887			/*
4888			 * Allow override for annotations -- this is typically
4889			 * only valid/needed for code that only exists when
4890			 * CONFIG_PREEMPT_RT=n.
4891			 */
4892			if (unlikely(class->lock_type == LD_LOCK_WAIT_OVERRIDE))
4893				curr_inner = prev_inner;
4894		}
4895	}
4896
4897	if (next_outer > curr_inner)
4898		return print_lock_invalid_wait_context(curr, next);
4899
4900	return 0;
4901}
4902
4903#else /* CONFIG_PROVE_LOCKING */
4904
4905static inline int
4906mark_usage(struct task_struct *curr, struct held_lock *hlock, int check)
4907{
4908	return 1;
4909}
4910
4911static inline unsigned int task_irq_context(struct task_struct *task)
4912{
4913	return 0;
4914}
4915
4916static inline int separate_irq_context(struct task_struct *curr,
4917		struct held_lock *hlock)
4918{
4919	return 0;
4920}
4921
4922static inline int check_wait_context(struct task_struct *curr,
4923				     struct held_lock *next)
4924{
4925	return 0;
4926}
4927
4928#endif /* CONFIG_PROVE_LOCKING */
4929
4930/*
4931 * Initialize a lock instance's lock-class mapping info:
4932 */
4933void lockdep_init_map_type(struct lockdep_map *lock, const char *name,
4934			    struct lock_class_key *key, int subclass,
4935			    u8 inner, u8 outer, u8 lock_type)
4936{
4937	int i;
4938
4939	for (i = 0; i < NR_LOCKDEP_CACHING_CLASSES; i++)
4940		lock->class_cache[i] = NULL;
4941
4942#ifdef CONFIG_LOCK_STAT
4943	lock->cpu = raw_smp_processor_id();
4944#endif
4945
4946	/*
4947	 * Can't be having no nameless bastards around this place!
4948	 */
4949	if (DEBUG_LOCKS_WARN_ON(!name)) {
4950		lock->name = "NULL";
4951		return;
4952	}
4953
4954	lock->name = name;
4955
4956	lock->wait_type_outer = outer;
4957	lock->wait_type_inner = inner;
4958	lock->lock_type = lock_type;
4959
4960	/*
4961	 * No key, no joy, we need to hash something.
4962	 */
4963	if (DEBUG_LOCKS_WARN_ON(!key))
4964		return;
4965	/*
4966	 * Sanity check, the lock-class key must either have been allocated
4967	 * statically or must have been registered as a dynamic key.
4968	 */
4969	if (!static_obj(key) && !is_dynamic_key(key)) {
4970		if (debug_locks)
4971			printk(KERN_ERR "BUG: key %px has not been registered!\n", key);
4972		DEBUG_LOCKS_WARN_ON(1);
4973		return;
4974	}
4975	lock->key = key;
4976
4977	if (unlikely(!debug_locks))
4978		return;
4979
4980	if (subclass) {
4981		unsigned long flags;
4982
4983		if (DEBUG_LOCKS_WARN_ON(!lockdep_enabled()))
4984			return;
4985
4986		raw_local_irq_save(flags);
4987		lockdep_recursion_inc();
4988		register_lock_class(lock, subclass, 1);
4989		lockdep_recursion_finish();
4990		raw_local_irq_restore(flags);
4991	}
4992}
4993EXPORT_SYMBOL_GPL(lockdep_init_map_type);
4994
4995struct lock_class_key __lockdep_no_validate__;
4996EXPORT_SYMBOL_GPL(__lockdep_no_validate__);
4997
4998struct lock_class_key __lockdep_no_track__;
4999EXPORT_SYMBOL_GPL(__lockdep_no_track__);
5000
5001#ifdef CONFIG_PROVE_LOCKING
5002void lockdep_set_lock_cmp_fn(struct lockdep_map *lock, lock_cmp_fn cmp_fn,
5003			     lock_print_fn print_fn)
5004{
5005	struct lock_class *class = lock->class_cache[0];
5006	unsigned long flags;
5007
5008	raw_local_irq_save(flags);
5009	lockdep_recursion_inc();
5010
5011	if (!class)
5012		class = register_lock_class(lock, 0, 0);
5013
5014	if (class) {
5015		WARN_ON(class->cmp_fn	&& class->cmp_fn != cmp_fn);
5016		WARN_ON(class->print_fn && class->print_fn != print_fn);
5017
5018		class->cmp_fn	= cmp_fn;
5019		class->print_fn = print_fn;
5020	}
5021
5022	lockdep_recursion_finish();
5023	raw_local_irq_restore(flags);
5024}
5025EXPORT_SYMBOL_GPL(lockdep_set_lock_cmp_fn);
5026#endif
5027
5028static void
5029print_lock_nested_lock_not_held(struct task_struct *curr,
5030				struct held_lock *hlock)
5031{
5032	if (!debug_locks_off())
5033		return;
5034	if (debug_locks_silent)
5035		return;
5036
5037	nbcon_cpu_emergency_enter();
5038
5039	pr_warn("\n");
5040	pr_warn("==================================\n");
5041	pr_warn("WARNING: Nested lock was not taken\n");
5042	print_kernel_ident();
5043	pr_warn("----------------------------------\n");
5044
5045	pr_warn("%s/%d is trying to lock:\n", curr->comm, task_pid_nr(curr));
5046	print_lock(hlock);
5047
5048	pr_warn("\nbut this task is not holding:\n");
5049	pr_warn("%s\n", hlock->nest_lock->name);
5050
5051	pr_warn("\nstack backtrace:\n");
5052	dump_stack();
5053
5054	pr_warn("\nother info that might help us debug this:\n");
5055	lockdep_print_held_locks(curr);
5056
5057	pr_warn("\nstack backtrace:\n");
5058	dump_stack();
5059
5060	nbcon_cpu_emergency_exit();
5061}
5062
5063static int __lock_is_held(const struct lockdep_map *lock, int read);
5064
5065/*
5066 * This gets called for every mutex_lock*()/spin_lock*() operation.
5067 * We maintain the dependency maps and validate the locking attempt:
5068 *
5069 * The callers must make sure that IRQs are disabled before calling it,
5070 * otherwise we could get an interrupt which would want to take locks,
5071 * which would end up in lockdep again.
5072 */
5073static int __lock_acquire(struct lockdep_map *lock, unsigned int subclass,
5074			  int trylock, int read, int check, int hardirqs_off,
5075			  struct lockdep_map *nest_lock, unsigned long ip,
5076			  int references, int pin_count, int sync)
5077{
5078	struct task_struct *curr = current;
5079	struct lock_class *class = NULL;
5080	struct held_lock *hlock;
5081	unsigned int depth;
5082	int chain_head = 0;
5083	int class_idx;
5084	u64 chain_key;
5085
5086	if (unlikely(!debug_locks))
5087		return 0;
5088
5089	if (unlikely(lock->key == &__lockdep_no_track__))
5090		return 0;
5091
5092	if (!prove_locking || lock->key == &__lockdep_no_validate__)
5093		check = 0;
5094
5095	if (subclass < NR_LOCKDEP_CACHING_CLASSES)
5096		class = lock->class_cache[subclass];
5097	/*
5098	 * Not cached?
5099	 */
5100	if (unlikely(!class)) {
5101		class = register_lock_class(lock, subclass, 0);
5102		if (!class)
5103			return 0;
5104	}
5105
5106	debug_class_ops_inc(class);
5107
5108	if (very_verbose(class)) {
5109		nbcon_cpu_emergency_enter();
5110		printk("\nacquire class [%px] %s", class->key, class->name);
5111		if (class->name_version > 1)
5112			printk(KERN_CONT "#%d", class->name_version);
5113		printk(KERN_CONT "\n");
5114		dump_stack();
5115		nbcon_cpu_emergency_exit();
5116	}
5117
5118	/*
5119	 * Add the lock to the list of currently held locks.
5120	 * (we dont increase the depth just yet, up until the
5121	 * dependency checks are done)
5122	 */
5123	depth = curr->lockdep_depth;
5124	/*
5125	 * Ran out of static storage for our per-task lock stack again have we?
5126	 */
5127	if (DEBUG_LOCKS_WARN_ON(depth >= MAX_LOCK_DEPTH))
5128		return 0;
5129
5130	class_idx = class - lock_classes;
5131
5132	if (depth && !sync) {
5133		/* we're holding locks and the new held lock is not a sync */
5134		hlock = curr->held_locks + depth - 1;
5135		if (hlock->class_idx == class_idx && nest_lock) {
5136			if (!references)
5137				references++;
5138
5139			if (!hlock->references)
5140				hlock->references++;
5141
5142			hlock->references += references;
5143
5144			/* Overflow */
5145			if (DEBUG_LOCKS_WARN_ON(hlock->references < references))
5146				return 0;
5147
5148			return 2;
5149		}
5150	}
5151
5152	hlock = curr->held_locks + depth;
5153	/*
5154	 * Plain impossible, we just registered it and checked it weren't no
5155	 * NULL like.. I bet this mushroom I ate was good!
5156	 */
5157	if (DEBUG_LOCKS_WARN_ON(!class))
5158		return 0;
5159	hlock->class_idx = class_idx;
5160	hlock->acquire_ip = ip;
5161	hlock->instance = lock;
5162	hlock->nest_lock = nest_lock;
5163	hlock->irq_context = task_irq_context(curr);
5164	hlock->trylock = trylock;
5165	hlock->read = read;
5166	hlock->check = check;
5167	hlock->sync = !!sync;
5168	hlock->hardirqs_off = !!hardirqs_off;
5169	hlock->references = references;
5170#ifdef CONFIG_LOCK_STAT
5171	hlock->waittime_stamp = 0;
5172	hlock->holdtime_stamp = lockstat_clock();
5173#endif
5174	hlock->pin_count = pin_count;
5175
5176	if (check_wait_context(curr, hlock))
5177		return 0;
5178
5179	/* Initialize the lock usage bit */
5180	if (!mark_usage(curr, hlock, check))
5181		return 0;
5182
5183	/*
5184	 * Calculate the chain hash: it's the combined hash of all the
5185	 * lock keys along the dependency chain. We save the hash value
5186	 * at every step so that we can get the current hash easily
5187	 * after unlock. The chain hash is then used to cache dependency
5188	 * results.
5189	 *
5190	 * The 'key ID' is what is the most compact key value to drive
5191	 * the hash, not class->key.
5192	 */
5193	/*
5194	 * Whoops, we did it again.. class_idx is invalid.
5195	 */
5196	if (DEBUG_LOCKS_WARN_ON(!test_bit(class_idx, lock_classes_in_use)))
5197		return 0;
5198
5199	chain_key = curr->curr_chain_key;
5200	if (!depth) {
5201		/*
5202		 * How can we have a chain hash when we ain't got no keys?!
5203		 */
5204		if (DEBUG_LOCKS_WARN_ON(chain_key != INITIAL_CHAIN_KEY))
5205			return 0;
5206		chain_head = 1;
5207	}
5208
5209	hlock->prev_chain_key = chain_key;
5210	if (separate_irq_context(curr, hlock)) {
5211		chain_key = INITIAL_CHAIN_KEY;
5212		chain_head = 1;
5213	}
5214	chain_key = iterate_chain_key(chain_key, hlock_id(hlock));
5215
5216	if (nest_lock && !__lock_is_held(nest_lock, -1)) {
5217		print_lock_nested_lock_not_held(curr, hlock);
5218		return 0;
5219	}
5220
5221	if (!debug_locks_silent) {
5222		WARN_ON_ONCE(depth && !hlock_class(hlock - 1)->key);
5223		WARN_ON_ONCE(!hlock_class(hlock)->key);
5224	}
5225
5226	if (!validate_chain(curr, hlock, chain_head, chain_key))
5227		return 0;
5228
5229	/* For lock_sync(), we are done here since no actual critical section */
5230	if (hlock->sync)
5231		return 1;
5232
5233	curr->curr_chain_key = chain_key;
5234	curr->lockdep_depth++;
5235	check_chain_key(curr);
5236#ifdef CONFIG_DEBUG_LOCKDEP
5237	if (unlikely(!debug_locks))
5238		return 0;
5239#endif
5240	if (unlikely(curr->lockdep_depth >= MAX_LOCK_DEPTH)) {
5241		debug_locks_off();
5242		nbcon_cpu_emergency_enter();
5243		print_lockdep_off("BUG: MAX_LOCK_DEPTH too low!");
5244		printk(KERN_DEBUG "depth: %i  max: %lu!\n",
5245		       curr->lockdep_depth, MAX_LOCK_DEPTH);
5246
5247		lockdep_print_held_locks(current);
5248		debug_show_all_locks();
5249		dump_stack();
5250		nbcon_cpu_emergency_exit();
5251
5252		return 0;
5253	}
5254
5255	if (unlikely(curr->lockdep_depth > max_lockdep_depth))
5256		max_lockdep_depth = curr->lockdep_depth;
5257
5258	return 1;
5259}
5260
5261static void print_unlock_imbalance_bug(struct task_struct *curr,
5262				       struct lockdep_map *lock,
5263				       unsigned long ip)
5264{
5265	if (!debug_locks_off())
5266		return;
5267	if (debug_locks_silent)
5268		return;
5269
5270	nbcon_cpu_emergency_enter();
5271
5272	pr_warn("\n");
5273	pr_warn("=====================================\n");
5274	pr_warn("WARNING: bad unlock balance detected!\n");
5275	print_kernel_ident();
5276	pr_warn("-------------------------------------\n");
5277	pr_warn("%s/%d is trying to release lock (",
5278		curr->comm, task_pid_nr(curr));
5279	print_lockdep_cache(lock);
5280	pr_cont(") at:\n");
5281	print_ip_sym(KERN_WARNING, ip);
5282	pr_warn("but there are no more locks to release!\n");
5283	pr_warn("\nother info that might help us debug this:\n");
5284	lockdep_print_held_locks(curr);
5285
5286	pr_warn("\nstack backtrace:\n");
5287	dump_stack();
5288
5289	nbcon_cpu_emergency_exit();
5290}
5291
5292static noinstr int match_held_lock(const struct held_lock *hlock,
5293				   const struct lockdep_map *lock)
5294{
5295	if (hlock->instance == lock)
5296		return 1;
5297
5298	if (hlock->references) {
5299		const struct lock_class *class = lock->class_cache[0];
5300
5301		if (!class)
5302			class = look_up_lock_class(lock, 0);
5303
5304		/*
5305		 * If look_up_lock_class() failed to find a class, we're trying
5306		 * to test if we hold a lock that has never yet been acquired.
5307		 * Clearly if the lock hasn't been acquired _ever_, we're not
5308		 * holding it either, so report failure.
5309		 */
5310		if (!class)
5311			return 0;
5312
5313		/*
5314		 * References, but not a lock we're actually ref-counting?
5315		 * State got messed up, follow the sites that change ->references
5316		 * and try to make sense of it.
5317		 */
5318		if (DEBUG_LOCKS_WARN_ON(!hlock->nest_lock))
5319			return 0;
5320
5321		if (hlock->class_idx == class - lock_classes)
5322			return 1;
5323	}
5324
5325	return 0;
5326}
5327
5328/* @depth must not be zero */
5329static struct held_lock *find_held_lock(struct task_struct *curr,
5330					struct lockdep_map *lock,
5331					unsigned int depth, int *idx)
5332{
5333	struct held_lock *ret, *hlock, *prev_hlock;
5334	int i;
5335
5336	i = depth - 1;
5337	hlock = curr->held_locks + i;
5338	ret = hlock;
5339	if (match_held_lock(hlock, lock))
5340		goto out;
5341
5342	ret = NULL;
5343	for (i--, prev_hlock = hlock--;
5344	     i >= 0;
5345	     i--, prev_hlock = hlock--) {
5346		/*
5347		 * We must not cross into another context:
5348		 */
5349		if (prev_hlock->irq_context != hlock->irq_context) {
5350			ret = NULL;
5351			break;
5352		}
5353		if (match_held_lock(hlock, lock)) {
5354			ret = hlock;
5355			break;
5356		}
5357	}
5358
5359out:
5360	*idx = i;
5361	return ret;
5362}
5363
5364static int reacquire_held_locks(struct task_struct *curr, unsigned int depth,
5365				int idx, unsigned int *merged)
5366{
5367	struct held_lock *hlock;
5368	int first_idx = idx;
5369
5370	if (DEBUG_LOCKS_WARN_ON(!irqs_disabled()))
5371		return 0;
5372
5373	for (hlock = curr->held_locks + idx; idx < depth; idx++, hlock++) {
5374		switch (__lock_acquire(hlock->instance,
5375				    hlock_class(hlock)->subclass,
5376				    hlock->trylock,
5377				    hlock->read, hlock->check,
5378				    hlock->hardirqs_off,
5379				    hlock->nest_lock, hlock->acquire_ip,
5380				    hlock->references, hlock->pin_count, 0)) {
5381		case 0:
5382			return 1;
5383		case 1:
5384			break;
5385		case 2:
5386			*merged += (idx == first_idx);
5387			break;
5388		default:
5389			WARN_ON(1);
5390			return 0;
5391		}
5392	}
5393	return 0;
5394}
5395
5396static int
5397__lock_set_class(struct lockdep_map *lock, const char *name,
5398		 struct lock_class_key *key, unsigned int subclass,
5399		 unsigned long ip)
5400{
5401	struct task_struct *curr = current;
5402	unsigned int depth, merged = 0;
5403	struct held_lock *hlock;
5404	struct lock_class *class;
5405	int i;
5406
5407	if (unlikely(!debug_locks))
5408		return 0;
5409
5410	depth = curr->lockdep_depth;
5411	/*
5412	 * This function is about (re)setting the class of a held lock,
5413	 * yet we're not actually holding any locks. Naughty user!
5414	 */
5415	if (DEBUG_LOCKS_WARN_ON(!depth))
5416		return 0;
5417
5418	hlock = find_held_lock(curr, lock, depth, &i);
5419	if (!hlock) {
5420		print_unlock_imbalance_bug(curr, lock, ip);
5421		return 0;
5422	}
5423
5424	lockdep_init_map_type(lock, name, key, 0,
5425			      lock->wait_type_inner,
5426			      lock->wait_type_outer,
5427			      lock->lock_type);
5428	class = register_lock_class(lock, subclass, 0);
5429	hlock->class_idx = class - lock_classes;
5430
5431	curr->lockdep_depth = i;
5432	curr->curr_chain_key = hlock->prev_chain_key;
5433
5434	if (reacquire_held_locks(curr, depth, i, &merged))
5435		return 0;
5436
5437	/*
5438	 * I took it apart and put it back together again, except now I have
5439	 * these 'spare' parts.. where shall I put them.
5440	 */
5441	if (DEBUG_LOCKS_WARN_ON(curr->lockdep_depth != depth - merged))
5442		return 0;
5443	return 1;
5444}
5445
5446static int __lock_downgrade(struct lockdep_map *lock, unsigned long ip)
5447{
5448	struct task_struct *curr = current;
5449	unsigned int depth, merged = 0;
5450	struct held_lock *hlock;
5451	int i;
5452
5453	if (unlikely(!debug_locks))
5454		return 0;
5455
5456	depth = curr->lockdep_depth;
5457	/*
5458	 * This function is about (re)setting the class of a held lock,
5459	 * yet we're not actually holding any locks. Naughty user!
5460	 */
5461	if (DEBUG_LOCKS_WARN_ON(!depth))
5462		return 0;
5463
5464	hlock = find_held_lock(curr, lock, depth, &i);
5465	if (!hlock) {
5466		print_unlock_imbalance_bug(curr, lock, ip);
5467		return 0;
5468	}
5469
5470	curr->lockdep_depth = i;
5471	curr->curr_chain_key = hlock->prev_chain_key;
5472
5473	WARN(hlock->read, "downgrading a read lock");
5474	hlock->read = 1;
5475	hlock->acquire_ip = ip;
5476
5477	if (reacquire_held_locks(curr, depth, i, &merged))
5478		return 0;
5479
5480	/* Merging can't happen with unchanged classes.. */
5481	if (DEBUG_LOCKS_WARN_ON(merged))
5482		return 0;
5483
5484	/*
5485	 * I took it apart and put it back together again, except now I have
5486	 * these 'spare' parts.. where shall I put them.
5487	 */
5488	if (DEBUG_LOCKS_WARN_ON(curr->lockdep_depth != depth))
5489		return 0;
5490
5491	return 1;
5492}
5493
5494/*
5495 * Remove the lock from the list of currently held locks - this gets
5496 * called on mutex_unlock()/spin_unlock*() (or on a failed
5497 * mutex_lock_interruptible()).
5498 */
5499static int
5500__lock_release(struct lockdep_map *lock, unsigned long ip)
5501{
5502	struct task_struct *curr = current;
5503	unsigned int depth, merged = 1;
5504	struct held_lock *hlock;
5505	int i;
5506
5507	if (unlikely(!debug_locks))
5508		return 0;
5509
5510	depth = curr->lockdep_depth;
5511	/*
5512	 * So we're all set to release this lock.. wait what lock? We don't
5513	 * own any locks, you've been drinking again?
5514	 */
5515	if (depth <= 0) {
5516		print_unlock_imbalance_bug(curr, lock, ip);
5517		return 0;
5518	}
5519
5520	/*
5521	 * Check whether the lock exists in the current stack
5522	 * of held locks:
5523	 */
5524	hlock = find_held_lock(curr, lock, depth, &i);
5525	if (!hlock) {
5526		print_unlock_imbalance_bug(curr, lock, ip);
5527		return 0;
5528	}
5529
5530	if (hlock->instance == lock)
5531		lock_release_holdtime(hlock);
5532
5533	WARN(hlock->pin_count, "releasing a pinned lock\n");
5534
5535	if (hlock->references) {
5536		hlock->references--;
5537		if (hlock->references) {
5538			/*
5539			 * We had, and after removing one, still have
5540			 * references, the current lock stack is still
5541			 * valid. We're done!
5542			 */
5543			return 1;
5544		}
5545	}
5546
5547	/*
5548	 * We have the right lock to unlock, 'hlock' points to it.
5549	 * Now we remove it from the stack, and add back the other
5550	 * entries (if any), recalculating the hash along the way:
5551	 */
5552
5553	curr->lockdep_depth = i;
5554	curr->curr_chain_key = hlock->prev_chain_key;
5555
5556	/*
5557	 * The most likely case is when the unlock is on the innermost
5558	 * lock. In this case, we are done!
5559	 */
5560	if (i == depth-1)
5561		return 1;
5562
5563	if (reacquire_held_locks(curr, depth, i + 1, &merged))
5564		return 0;
5565
5566	/*
5567	 * We had N bottles of beer on the wall, we drank one, but now
5568	 * there's not N-1 bottles of beer left on the wall...
5569	 * Pouring two of the bottles together is acceptable.
5570	 */
5571	DEBUG_LOCKS_WARN_ON(curr->lockdep_depth != depth - merged);
5572
5573	/*
5574	 * Since reacquire_held_locks() would have called check_chain_key()
5575	 * indirectly via __lock_acquire(), we don't need to do it again
5576	 * on return.
5577	 */
5578	return 0;
5579}
5580
5581static __always_inline
5582int __lock_is_held(const struct lockdep_map *lock, int read)
5583{
5584	struct task_struct *curr = current;
5585	int i;
5586
5587	for (i = 0; i < curr->lockdep_depth; i++) {
5588		struct held_lock *hlock = curr->held_locks + i;
5589
5590		if (match_held_lock(hlock, lock)) {
5591			if (read == -1 || !!hlock->read == read)
5592				return LOCK_STATE_HELD;
5593
5594			return LOCK_STATE_NOT_HELD;
5595		}
5596	}
5597
5598	return LOCK_STATE_NOT_HELD;
5599}
5600
5601static struct pin_cookie __lock_pin_lock(struct lockdep_map *lock)
5602{
5603	struct pin_cookie cookie = NIL_COOKIE;
5604	struct task_struct *curr = current;
5605	int i;
5606
5607	if (unlikely(!debug_locks))
5608		return cookie;
5609
5610	for (i = 0; i < curr->lockdep_depth; i++) {
5611		struct held_lock *hlock = curr->held_locks + i;
5612
5613		if (match_held_lock(hlock, lock)) {
5614			/*
5615			 * Grab 16bits of randomness; this is sufficient to not
5616			 * be guessable and still allows some pin nesting in
5617			 * our u32 pin_count.
5618			 */
5619			cookie.val = 1 + (sched_clock() & 0xffff);
5620			hlock->pin_count += cookie.val;
5621			return cookie;
5622		}
5623	}
5624
5625	WARN(1, "pinning an unheld lock\n");
5626	return cookie;
5627}
5628
5629static void __lock_repin_lock(struct lockdep_map *lock, struct pin_cookie cookie)
5630{
5631	struct task_struct *curr = current;
5632	int i;
5633
5634	if (unlikely(!debug_locks))
5635		return;
5636
5637	for (i = 0; i < curr->lockdep_depth; i++) {
5638		struct held_lock *hlock = curr->held_locks + i;
5639
5640		if (match_held_lock(hlock, lock)) {
5641			hlock->pin_count += cookie.val;
5642			return;
5643		}
5644	}
5645
5646	WARN(1, "pinning an unheld lock\n");
5647}
5648
5649static void __lock_unpin_lock(struct lockdep_map *lock, struct pin_cookie cookie)
5650{
5651	struct task_struct *curr = current;
5652	int i;
5653
5654	if (unlikely(!debug_locks))
5655		return;
5656
5657	for (i = 0; i < curr->lockdep_depth; i++) {
5658		struct held_lock *hlock = curr->held_locks + i;
5659
5660		if (match_held_lock(hlock, lock)) {
5661			if (WARN(!hlock->pin_count, "unpinning an unpinned lock\n"))
5662				return;
5663
5664			hlock->pin_count -= cookie.val;
5665
5666			if (WARN((int)hlock->pin_count < 0, "pin count corrupted\n"))
5667				hlock->pin_count = 0;
5668
5669			return;
5670		}
5671	}
5672
5673	WARN(1, "unpinning an unheld lock\n");
5674}
5675
5676/*
5677 * Check whether we follow the irq-flags state precisely:
5678 */
5679static noinstr void check_flags(unsigned long flags)
5680{
5681#if defined(CONFIG_PROVE_LOCKING) && defined(CONFIG_DEBUG_LOCKDEP)
5682	if (!debug_locks)
5683		return;
5684
5685	/* Get the warning out..  */
5686	instrumentation_begin();
5687
5688	if (irqs_disabled_flags(flags)) {
5689		if (DEBUG_LOCKS_WARN_ON(lockdep_hardirqs_enabled())) {
5690			printk("possible reason: unannotated irqs-off.\n");
5691		}
5692	} else {
5693		if (DEBUG_LOCKS_WARN_ON(!lockdep_hardirqs_enabled())) {
5694			printk("possible reason: unannotated irqs-on.\n");
5695		}
5696	}
5697
5698#ifndef CONFIG_PREEMPT_RT
5699	/*
5700	 * We dont accurately track softirq state in e.g.
5701	 * hardirq contexts (such as on 4KSTACKS), so only
5702	 * check if not in hardirq contexts:
5703	 */
5704	if (!hardirq_count()) {
5705		if (softirq_count()) {
5706			/* like the above, but with softirqs */
5707			DEBUG_LOCKS_WARN_ON(current->softirqs_enabled);
5708		} else {
5709			/* lick the above, does it taste good? */
5710			DEBUG_LOCKS_WARN_ON(!current->softirqs_enabled);
5711		}
5712	}
5713#endif
5714
5715	if (!debug_locks)
5716		print_irqtrace_events(current);
5717
5718	instrumentation_end();
5719#endif
5720}
5721
5722void lock_set_class(struct lockdep_map *lock, const char *name,
5723		    struct lock_class_key *key, unsigned int subclass,
5724		    unsigned long ip)
5725{
5726	unsigned long flags;
5727
5728	if (unlikely(!lockdep_enabled()))
5729		return;
5730
5731	raw_local_irq_save(flags);
5732	lockdep_recursion_inc();
5733	check_flags(flags);
5734	if (__lock_set_class(lock, name, key, subclass, ip))
5735		check_chain_key(current);
5736	lockdep_recursion_finish();
5737	raw_local_irq_restore(flags);
5738}
5739EXPORT_SYMBOL_GPL(lock_set_class);
5740
5741void lock_downgrade(struct lockdep_map *lock, unsigned long ip)
5742{
5743	unsigned long flags;
5744
5745	if (unlikely(!lockdep_enabled()))
5746		return;
5747
5748	raw_local_irq_save(flags);
5749	lockdep_recursion_inc();
5750	check_flags(flags);
5751	if (__lock_downgrade(lock, ip))
5752		check_chain_key(current);
5753	lockdep_recursion_finish();
5754	raw_local_irq_restore(flags);
5755}
5756EXPORT_SYMBOL_GPL(lock_downgrade);
5757
5758/* NMI context !!! */
5759static void verify_lock_unused(struct lockdep_map *lock, struct held_lock *hlock, int subclass)
5760{
5761#ifdef CONFIG_PROVE_LOCKING
5762	struct lock_class *class = look_up_lock_class(lock, subclass);
5763	unsigned long mask = LOCKF_USED;
5764
5765	/* if it doesn't have a class (yet), it certainly hasn't been used yet */
5766	if (!class)
5767		return;
5768
5769	/*
5770	 * READ locks only conflict with USED, such that if we only ever use
5771	 * READ locks, there is no deadlock possible -- RCU.
5772	 */
5773	if (!hlock->read)
5774		mask |= LOCKF_USED_READ;
5775
5776	if (!(class->usage_mask & mask))
5777		return;
5778
5779	hlock->class_idx = class - lock_classes;
5780
5781	print_usage_bug(current, hlock, LOCK_USED, LOCK_USAGE_STATES);
5782#endif
5783}
5784
5785static bool lockdep_nmi(void)
5786{
5787	if (raw_cpu_read(lockdep_recursion))
5788		return false;
5789
5790	if (!in_nmi())
5791		return false;
5792
5793	return true;
5794}
5795
5796/*
5797 * read_lock() is recursive if:
5798 * 1. We force lockdep think this way in selftests or
5799 * 2. The implementation is not queued read/write lock or
5800 * 3. The locker is at an in_interrupt() context.
5801 */
5802bool read_lock_is_recursive(void)
5803{
5804	return force_read_lock_recursive ||
5805	       !IS_ENABLED(CONFIG_QUEUED_RWLOCKS) ||
5806	       in_interrupt();
5807}
5808EXPORT_SYMBOL_GPL(read_lock_is_recursive);
5809
5810/*
5811 * We are not always called with irqs disabled - do that here,
5812 * and also avoid lockdep recursion:
5813 */
5814void lock_acquire(struct lockdep_map *lock, unsigned int subclass,
5815			  int trylock, int read, int check,
5816			  struct lockdep_map *nest_lock, unsigned long ip)
5817{
5818	unsigned long flags;
5819
5820	trace_lock_acquire(lock, subclass, trylock, read, check, nest_lock, ip);
5821
5822	if (!debug_locks)
5823		return;
5824
5825	if (unlikely(!lockdep_enabled())) {
5826		/* XXX allow trylock from NMI ?!? */
5827		if (lockdep_nmi() && !trylock) {
5828			struct held_lock hlock;
5829
5830			hlock.acquire_ip = ip;
5831			hlock.instance = lock;
5832			hlock.nest_lock = nest_lock;
5833			hlock.irq_context = 2; // XXX
5834			hlock.trylock = trylock;
5835			hlock.read = read;
5836			hlock.check = check;
5837			hlock.hardirqs_off = true;
5838			hlock.references = 0;
5839
5840			verify_lock_unused(lock, &hlock, subclass);
5841		}
5842		return;
5843	}
5844
5845	raw_local_irq_save(flags);
5846	check_flags(flags);
5847
5848	lockdep_recursion_inc();
5849	__lock_acquire(lock, subclass, trylock, read, check,
5850		       irqs_disabled_flags(flags), nest_lock, ip, 0, 0, 0);
5851	lockdep_recursion_finish();
5852	raw_local_irq_restore(flags);
5853}
5854EXPORT_SYMBOL_GPL(lock_acquire);
5855
5856void lock_release(struct lockdep_map *lock, unsigned long ip)
5857{
5858	unsigned long flags;
5859
5860	trace_lock_release(lock, ip);
5861
5862	if (unlikely(!lockdep_enabled() ||
5863		     lock->key == &__lockdep_no_track__))
5864		return;
5865
5866	raw_local_irq_save(flags);
5867	check_flags(flags);
5868
5869	lockdep_recursion_inc();
5870	if (__lock_release(lock, ip))
5871		check_chain_key(current);
5872	lockdep_recursion_finish();
5873	raw_local_irq_restore(flags);
5874}
5875EXPORT_SYMBOL_GPL(lock_release);
5876
5877/*
5878 * lock_sync() - A special annotation for synchronize_{s,}rcu()-like API.
5879 *
5880 * No actual critical section is created by the APIs annotated with this: these
5881 * APIs are used to wait for one or multiple critical sections (on other CPUs
5882 * or threads), and it means that calling these APIs inside these critical
5883 * sections is potential deadlock.
5884 */
5885void lock_sync(struct lockdep_map *lock, unsigned subclass, int read,
5886	       int check, struct lockdep_map *nest_lock, unsigned long ip)
5887{
5888	unsigned long flags;
5889
5890	if (unlikely(!lockdep_enabled()))
5891		return;
5892
5893	raw_local_irq_save(flags);
5894	check_flags(flags);
5895
5896	lockdep_recursion_inc();
5897	__lock_acquire(lock, subclass, 0, read, check,
5898		       irqs_disabled_flags(flags), nest_lock, ip, 0, 0, 1);
5899	check_chain_key(current);
5900	lockdep_recursion_finish();
5901	raw_local_irq_restore(flags);
5902}
5903EXPORT_SYMBOL_GPL(lock_sync);
5904
5905noinstr int lock_is_held_type(const struct lockdep_map *lock, int read)
5906{
5907	unsigned long flags;
5908	int ret = LOCK_STATE_NOT_HELD;
5909
5910	/*
5911	 * Avoid false negative lockdep_assert_held() and
5912	 * lockdep_assert_not_held().
5913	 */
5914	if (unlikely(!lockdep_enabled()))
5915		return LOCK_STATE_UNKNOWN;
5916
5917	raw_local_irq_save(flags);
5918	check_flags(flags);
5919
5920	lockdep_recursion_inc();
5921	ret = __lock_is_held(lock, read);
5922	lockdep_recursion_finish();
5923	raw_local_irq_restore(flags);
5924
5925	return ret;
5926}
5927EXPORT_SYMBOL_GPL(lock_is_held_type);
5928NOKPROBE_SYMBOL(lock_is_held_type);
5929
5930struct pin_cookie lock_pin_lock(struct lockdep_map *lock)
5931{
5932	struct pin_cookie cookie = NIL_COOKIE;
5933	unsigned long flags;
5934
5935	if (unlikely(!lockdep_enabled()))
5936		return cookie;
5937
5938	raw_local_irq_save(flags);
5939	check_flags(flags);
5940
5941	lockdep_recursion_inc();
5942	cookie = __lock_pin_lock(lock);
5943	lockdep_recursion_finish();
5944	raw_local_irq_restore(flags);
5945
5946	return cookie;
5947}
5948EXPORT_SYMBOL_GPL(lock_pin_lock);
5949
5950void lock_repin_lock(struct lockdep_map *lock, struct pin_cookie cookie)
5951{
5952	unsigned long flags;
5953
5954	if (unlikely(!lockdep_enabled()))
5955		return;
5956
5957	raw_local_irq_save(flags);
5958	check_flags(flags);
5959
5960	lockdep_recursion_inc();
5961	__lock_repin_lock(lock, cookie);
5962	lockdep_recursion_finish();
5963	raw_local_irq_restore(flags);
5964}
5965EXPORT_SYMBOL_GPL(lock_repin_lock);
5966
5967void lock_unpin_lock(struct lockdep_map *lock, struct pin_cookie cookie)
5968{
5969	unsigned long flags;
5970
5971	if (unlikely(!lockdep_enabled()))
5972		return;
5973
5974	raw_local_irq_save(flags);
5975	check_flags(flags);
5976
5977	lockdep_recursion_inc();
5978	__lock_unpin_lock(lock, cookie);
5979	lockdep_recursion_finish();
5980	raw_local_irq_restore(flags);
5981}
5982EXPORT_SYMBOL_GPL(lock_unpin_lock);
5983
5984#ifdef CONFIG_LOCK_STAT
5985static void print_lock_contention_bug(struct task_struct *curr,
5986				      struct lockdep_map *lock,
5987				      unsigned long ip)
5988{
5989	if (!debug_locks_off())
5990		return;
5991	if (debug_locks_silent)
5992		return;
5993
5994	nbcon_cpu_emergency_enter();
5995
5996	pr_warn("\n");
5997	pr_warn("=================================\n");
5998	pr_warn("WARNING: bad contention detected!\n");
5999	print_kernel_ident();
6000	pr_warn("---------------------------------\n");
6001	pr_warn("%s/%d is trying to contend lock (",
6002		curr->comm, task_pid_nr(curr));
6003	print_lockdep_cache(lock);
6004	pr_cont(") at:\n");
6005	print_ip_sym(KERN_WARNING, ip);
6006	pr_warn("but there are no locks held!\n");
6007	pr_warn("\nother info that might help us debug this:\n");
6008	lockdep_print_held_locks(curr);
6009
6010	pr_warn("\nstack backtrace:\n");
6011	dump_stack();
6012
6013	nbcon_cpu_emergency_exit();
6014}
6015
6016static void
6017__lock_contended(struct lockdep_map *lock, unsigned long ip)
6018{
6019	struct task_struct *curr = current;
6020	struct held_lock *hlock;
6021	struct lock_class_stats *stats;
6022	unsigned int depth;
6023	int i, contention_point, contending_point;
6024
6025	depth = curr->lockdep_depth;
6026	/*
6027	 * Whee, we contended on this lock, except it seems we're not
6028	 * actually trying to acquire anything much at all..
6029	 */
6030	if (DEBUG_LOCKS_WARN_ON(!depth))
6031		return;
6032
6033	if (unlikely(lock->key == &__lockdep_no_track__))
6034		return;
6035
6036	hlock = find_held_lock(curr, lock, depth, &i);
6037	if (!hlock) {
6038		print_lock_contention_bug(curr, lock, ip);
6039		return;
6040	}
6041
6042	if (hlock->instance != lock)
6043		return;
6044
6045	hlock->waittime_stamp = lockstat_clock();
6046
6047	contention_point = lock_point(hlock_class(hlock)->contention_point, ip);
6048	contending_point = lock_point(hlock_class(hlock)->contending_point,
6049				      lock->ip);
6050
6051	stats = get_lock_stats(hlock_class(hlock));
6052	if (contention_point < LOCKSTAT_POINTS)
6053		stats->contention_point[contention_point]++;
6054	if (contending_point < LOCKSTAT_POINTS)
6055		stats->contending_point[contending_point]++;
6056	if (lock->cpu != smp_processor_id())
6057		stats->bounces[bounce_contended + !!hlock->read]++;
6058}
6059
6060static void
6061__lock_acquired(struct lockdep_map *lock, unsigned long ip)
6062{
6063	struct task_struct *curr = current;
6064	struct held_lock *hlock;
6065	struct lock_class_stats *stats;
6066	unsigned int depth;
6067	u64 now, waittime = 0;
6068	int i, cpu;
6069
6070	depth = curr->lockdep_depth;
6071	/*
6072	 * Yay, we acquired ownership of this lock we didn't try to
6073	 * acquire, how the heck did that happen?
6074	 */
6075	if (DEBUG_LOCKS_WARN_ON(!depth))
6076		return;
6077
6078	if (unlikely(lock->key == &__lockdep_no_track__))
6079		return;
6080
6081	hlock = find_held_lock(curr, lock, depth, &i);
6082	if (!hlock) {
6083		print_lock_contention_bug(curr, lock, _RET_IP_);
6084		return;
6085	}
6086
6087	if (hlock->instance != lock)
6088		return;
6089
6090	cpu = smp_processor_id();
6091	if (hlock->waittime_stamp) {
6092		now = lockstat_clock();
6093		waittime = now - hlock->waittime_stamp;
6094		hlock->holdtime_stamp = now;
6095	}
6096
6097	stats = get_lock_stats(hlock_class(hlock));
6098	if (waittime) {
6099		if (hlock->read)
6100			lock_time_inc(&stats->read_waittime, waittime);
6101		else
6102			lock_time_inc(&stats->write_waittime, waittime);
6103	}
6104	if (lock->cpu != cpu)
6105		stats->bounces[bounce_acquired + !!hlock->read]++;
6106
6107	lock->cpu = cpu;
6108	lock->ip = ip;
6109}
6110
6111void lock_contended(struct lockdep_map *lock, unsigned long ip)
6112{
6113	unsigned long flags;
6114
6115	trace_lock_contended(lock, ip);
6116
6117	if (unlikely(!lock_stat || !lockdep_enabled()))
6118		return;
6119
6120	raw_local_irq_save(flags);
6121	check_flags(flags);
6122	lockdep_recursion_inc();
6123	__lock_contended(lock, ip);
6124	lockdep_recursion_finish();
6125	raw_local_irq_restore(flags);
6126}
6127EXPORT_SYMBOL_GPL(lock_contended);
6128
6129void lock_acquired(struct lockdep_map *lock, unsigned long ip)
6130{
6131	unsigned long flags;
6132
6133	trace_lock_acquired(lock, ip);
6134
6135	if (unlikely(!lock_stat || !lockdep_enabled()))
6136		return;
6137
6138	raw_local_irq_save(flags);
6139	check_flags(flags);
6140	lockdep_recursion_inc();
6141	__lock_acquired(lock, ip);
6142	lockdep_recursion_finish();
6143	raw_local_irq_restore(flags);
6144}
6145EXPORT_SYMBOL_GPL(lock_acquired);
6146#endif
6147
6148/*
6149 * Used by the testsuite, sanitize the validator state
6150 * after a simulated failure:
6151 */
6152
6153void lockdep_reset(void)
6154{
6155	unsigned long flags;
6156	int i;
6157
6158	raw_local_irq_save(flags);
6159	lockdep_init_task(current);
6160	memset(current->held_locks, 0, MAX_LOCK_DEPTH*sizeof(struct held_lock));
6161	nr_hardirq_chains = 0;
6162	nr_softirq_chains = 0;
6163	nr_process_chains = 0;
6164	debug_locks = 1;
6165	for (i = 0; i < CHAINHASH_SIZE; i++)
6166		INIT_HLIST_HEAD(chainhash_table + i);
6167	raw_local_irq_restore(flags);
6168}
6169
6170/* Remove a class from a lock chain. Must be called with the graph lock held. */
6171static void remove_class_from_lock_chain(struct pending_free *pf,
6172					 struct lock_chain *chain,
6173					 struct lock_class *class)
6174{
6175#ifdef CONFIG_PROVE_LOCKING
6176	int i;
6177
6178	for (i = chain->base; i < chain->base + chain->depth; i++) {
6179		if (chain_hlock_class_idx(chain_hlocks[i]) != class - lock_classes)
6180			continue;
6181		/*
6182		 * Each lock class occurs at most once in a lock chain so once
6183		 * we found a match we can break out of this loop.
6184		 */
6185		goto free_lock_chain;
6186	}
6187	/* Since the chain has not been modified, return. */
6188	return;
6189
6190free_lock_chain:
6191	free_chain_hlocks(chain->base, chain->depth);
6192	/* Overwrite the chain key for concurrent RCU readers. */
6193	WRITE_ONCE(chain->chain_key, INITIAL_CHAIN_KEY);
6194	dec_chains(chain->irq_context);
6195
6196	/*
6197	 * Note: calling hlist_del_rcu() from inside a
6198	 * hlist_for_each_entry_rcu() loop is safe.
6199	 */
6200	hlist_del_rcu(&chain->entry);
6201	__set_bit(chain - lock_chains, pf->lock_chains_being_freed);
6202	nr_zapped_lock_chains++;
6203#endif
6204}
6205
6206/* Must be called with the graph lock held. */
6207static void remove_class_from_lock_chains(struct pending_free *pf,
6208					  struct lock_class *class)
6209{
6210	struct lock_chain *chain;
6211	struct hlist_head *head;
6212	int i;
6213
6214	for (i = 0; i < ARRAY_SIZE(chainhash_table); i++) {
6215		head = chainhash_table + i;
6216		hlist_for_each_entry_rcu(chain, head, entry) {
6217			remove_class_from_lock_chain(pf, chain, class);
6218		}
6219	}
6220}
6221
6222/*
6223 * Remove all references to a lock class. The caller must hold the graph lock.
6224 */
6225static void zap_class(struct pending_free *pf, struct lock_class *class)
6226{
6227	struct lock_list *entry;
6228	int i;
6229
6230	WARN_ON_ONCE(!class->key);
6231
6232	/*
6233	 * Remove all dependencies this lock is
6234	 * involved in:
6235	 */
6236	for_each_set_bit(i, list_entries_in_use, ARRAY_SIZE(list_entries)) {
6237		entry = list_entries + i;
6238		if (entry->class != class && entry->links_to != class)
6239			continue;
6240		__clear_bit(i, list_entries_in_use);
6241		nr_list_entries--;
6242		list_del_rcu(&entry->entry);
6243	}
6244	if (list_empty(&class->locks_after) &&
6245	    list_empty(&class->locks_before)) {
6246		list_move_tail(&class->lock_entry, &pf->zapped);
6247		hlist_del_rcu(&class->hash_entry);
6248		WRITE_ONCE(class->key, NULL);
6249		WRITE_ONCE(class->name, NULL);
6250		nr_lock_classes--;
6251		__clear_bit(class - lock_classes, lock_classes_in_use);
6252		if (class - lock_classes == max_lock_class_idx)
6253			max_lock_class_idx--;
6254	} else {
6255		WARN_ONCE(true, "%s() failed for class %s\n", __func__,
6256			  class->name);
6257	}
6258
6259	remove_class_from_lock_chains(pf, class);
6260	nr_zapped_classes++;
6261}
6262
6263static void reinit_class(struct lock_class *class)
6264{
6265	WARN_ON_ONCE(!class->lock_entry.next);
6266	WARN_ON_ONCE(!list_empty(&class->locks_after));
6267	WARN_ON_ONCE(!list_empty(&class->locks_before));
6268	memset_startat(class, 0, key);
6269	WARN_ON_ONCE(!class->lock_entry.next);
6270	WARN_ON_ONCE(!list_empty(&class->locks_after));
6271	WARN_ON_ONCE(!list_empty(&class->locks_before));
6272}
6273
6274static inline int within(const void *addr, void *start, unsigned long size)
6275{
6276	return addr >= start && addr < start + size;
6277}
6278
6279static bool inside_selftest(void)
6280{
6281	return current == lockdep_selftest_task_struct;
6282}
6283
6284/* The caller must hold the graph lock. */
6285static struct pending_free *get_pending_free(void)
6286{
6287	return delayed_free.pf + delayed_free.index;
6288}
6289
6290static void free_zapped_rcu(struct rcu_head *cb);
6291
6292/*
6293* See if we need to queue an RCU callback, must called with
6294* the lockdep lock held, returns false if either we don't have
6295* any pending free or the callback is already scheduled.
6296* Otherwise, a call_rcu() must follow this function call.
6297*/
6298static bool prepare_call_rcu_zapped(struct pending_free *pf)
6299{
6300	WARN_ON_ONCE(inside_selftest());
6301
6302	if (list_empty(&pf->zapped))
6303		return false;
6304
6305	if (delayed_free.scheduled)
6306		return false;
6307
6308	delayed_free.scheduled = true;
6309
6310	WARN_ON_ONCE(delayed_free.pf + delayed_free.index != pf);
6311	delayed_free.index ^= 1;
6312
6313	return true;
6314}
6315
6316/* The caller must hold the graph lock. May be called from RCU context. */
6317static void __free_zapped_classes(struct pending_free *pf)
6318{
6319	struct lock_class *class;
6320
6321	check_data_structures();
6322
6323	list_for_each_entry(class, &pf->zapped, lock_entry)
6324		reinit_class(class);
6325
6326	list_splice_init(&pf->zapped, &free_lock_classes);
6327
6328#ifdef CONFIG_PROVE_LOCKING
6329	bitmap_andnot(lock_chains_in_use, lock_chains_in_use,
6330		      pf->lock_chains_being_freed, ARRAY_SIZE(lock_chains));
6331	bitmap_clear(pf->lock_chains_being_freed, 0, ARRAY_SIZE(lock_chains));
6332#endif
6333}
6334
6335static void free_zapped_rcu(struct rcu_head *ch)
6336{
6337	struct pending_free *pf;
6338	unsigned long flags;
6339	bool need_callback;
6340
6341	if (WARN_ON_ONCE(ch != &delayed_free.rcu_head))
6342		return;
6343
6344	raw_local_irq_save(flags);
6345	lockdep_lock();
6346
6347	/* closed head */
6348	pf = delayed_free.pf + (delayed_free.index ^ 1);
6349	__free_zapped_classes(pf);
6350	delayed_free.scheduled = false;
6351	need_callback =
6352		prepare_call_rcu_zapped(delayed_free.pf + delayed_free.index);
6353	lockdep_unlock();
6354	raw_local_irq_restore(flags);
6355
6356	/*
6357	* If there's pending free and its callback has not been scheduled,
6358	* queue an RCU callback.
6359	*/
6360	if (need_callback)
6361		call_rcu(&delayed_free.rcu_head, free_zapped_rcu);
6362
6363}
6364
6365/*
6366 * Remove all lock classes from the class hash table and from the
6367 * all_lock_classes list whose key or name is in the address range [start,
6368 * start + size). Move these lock classes to the zapped_classes list. Must
6369 * be called with the graph lock held.
6370 */
6371static void __lockdep_free_key_range(struct pending_free *pf, void *start,
6372				     unsigned long size)
6373{
6374	struct lock_class *class;
6375	struct hlist_head *head;
6376	int i;
6377
6378	/* Unhash all classes that were created by a module. */
6379	for (i = 0; i < CLASSHASH_SIZE; i++) {
6380		head = classhash_table + i;
6381		hlist_for_each_entry_rcu(class, head, hash_entry) {
6382			if (!within(class->key, start, size) &&
6383			    !within(class->name, start, size))
6384				continue;
6385			zap_class(pf, class);
6386		}
6387	}
6388}
6389
6390/*
6391 * Used in module.c to remove lock classes from memory that is going to be
6392 * freed; and possibly re-used by other modules.
6393 *
6394 * We will have had one synchronize_rcu() before getting here, so we're
6395 * guaranteed nobody will look up these exact classes -- they're properly dead
6396 * but still allocated.
6397 */
6398static void lockdep_free_key_range_reg(void *start, unsigned long size)
6399{
6400	struct pending_free *pf;
6401	unsigned long flags;
6402	bool need_callback;
6403
6404	init_data_structures_once();
6405
6406	raw_local_irq_save(flags);
6407	lockdep_lock();
6408	pf = get_pending_free();
6409	__lockdep_free_key_range(pf, start, size);
6410	need_callback = prepare_call_rcu_zapped(pf);
6411	lockdep_unlock();
6412	raw_local_irq_restore(flags);
6413	if (need_callback)
6414		call_rcu(&delayed_free.rcu_head, free_zapped_rcu);
6415	/*
6416	 * Wait for any possible iterators from look_up_lock_class() to pass
6417	 * before continuing to free the memory they refer to.
6418	 */
6419	synchronize_rcu();
6420}
6421
6422/*
6423 * Free all lockdep keys in the range [start, start+size). Does not sleep.
6424 * Ignores debug_locks. Must only be used by the lockdep selftests.
6425 */
6426static void lockdep_free_key_range_imm(void *start, unsigned long size)
6427{
6428	struct pending_free *pf = delayed_free.pf;
6429	unsigned long flags;
6430
6431	init_data_structures_once();
6432
6433	raw_local_irq_save(flags);
6434	lockdep_lock();
6435	__lockdep_free_key_range(pf, start, size);
6436	__free_zapped_classes(pf);
6437	lockdep_unlock();
6438	raw_local_irq_restore(flags);
6439}
6440
6441void lockdep_free_key_range(void *start, unsigned long size)
6442{
6443	init_data_structures_once();
6444
6445	if (inside_selftest())
6446		lockdep_free_key_range_imm(start, size);
6447	else
6448		lockdep_free_key_range_reg(start, size);
6449}
6450
6451/*
6452 * Check whether any element of the @lock->class_cache[] array refers to a
6453 * registered lock class. The caller must hold either the graph lock or the
6454 * RCU read lock.
6455 */
6456static bool lock_class_cache_is_registered(struct lockdep_map *lock)
6457{
6458	struct lock_class *class;
6459	struct hlist_head *head;
6460	int i, j;
6461
6462	for (i = 0; i < CLASSHASH_SIZE; i++) {
6463		head = classhash_table + i;
6464		hlist_for_each_entry_rcu(class, head, hash_entry) {
6465			for (j = 0; j < NR_LOCKDEP_CACHING_CLASSES; j++)
6466				if (lock->class_cache[j] == class)
6467					return true;
6468		}
6469	}
6470	return false;
6471}
6472
6473/* The caller must hold the graph lock. Does not sleep. */
6474static void __lockdep_reset_lock(struct pending_free *pf,
6475				 struct lockdep_map *lock)
6476{
6477	struct lock_class *class;
6478	int j;
6479
6480	/*
6481	 * Remove all classes this lock might have:
6482	 */
6483	for (j = 0; j < MAX_LOCKDEP_SUBCLASSES; j++) {
6484		/*
6485		 * If the class exists we look it up and zap it:
6486		 */
6487		class = look_up_lock_class(lock, j);
6488		if (class)
6489			zap_class(pf, class);
6490	}
6491	/*
6492	 * Debug check: in the end all mapped classes should
6493	 * be gone.
6494	 */
6495	if (WARN_ON_ONCE(lock_class_cache_is_registered(lock)))
6496		debug_locks_off();
6497}
6498
6499/*
6500 * Remove all information lockdep has about a lock if debug_locks == 1. Free
6501 * released data structures from RCU context.
6502 */
6503static void lockdep_reset_lock_reg(struct lockdep_map *lock)
6504{
6505	struct pending_free *pf;
6506	unsigned long flags;
6507	int locked;
6508	bool need_callback = false;
6509
6510	raw_local_irq_save(flags);
6511	locked = graph_lock();
6512	if (!locked)
6513		goto out_irq;
6514
6515	pf = get_pending_free();
6516	__lockdep_reset_lock(pf, lock);
6517	need_callback = prepare_call_rcu_zapped(pf);
6518
6519	graph_unlock();
6520out_irq:
6521	raw_local_irq_restore(flags);
6522	if (need_callback)
6523		call_rcu(&delayed_free.rcu_head, free_zapped_rcu);
6524}
6525
6526/*
6527 * Reset a lock. Does not sleep. Ignores debug_locks. Must only be used by the
6528 * lockdep selftests.
6529 */
6530static void lockdep_reset_lock_imm(struct lockdep_map *lock)
6531{
6532	struct pending_free *pf = delayed_free.pf;
6533	unsigned long flags;
6534
6535	raw_local_irq_save(flags);
6536	lockdep_lock();
6537	__lockdep_reset_lock(pf, lock);
6538	__free_zapped_classes(pf);
6539	lockdep_unlock();
6540	raw_local_irq_restore(flags);
6541}
6542
6543void lockdep_reset_lock(struct lockdep_map *lock)
6544{
6545	init_data_structures_once();
6546
6547	if (inside_selftest())
6548		lockdep_reset_lock_imm(lock);
6549	else
6550		lockdep_reset_lock_reg(lock);
6551}
6552
6553/*
6554 * Unregister a dynamically allocated key.
6555 *
6556 * Unlike lockdep_register_key(), a search is always done to find a matching
6557 * key irrespective of debug_locks to avoid potential invalid access to freed
6558 * memory in lock_class entry.
6559 */
6560void lockdep_unregister_key(struct lock_class_key *key)
6561{
6562	struct hlist_head *hash_head = keyhashentry(key);
6563	struct lock_class_key *k;
6564	struct pending_free *pf;
6565	unsigned long flags;
6566	bool found = false;
6567	bool need_callback = false;
6568
6569	might_sleep();
6570
6571	if (WARN_ON_ONCE(static_obj(key)))
6572		return;
6573
6574	raw_local_irq_save(flags);
6575	lockdep_lock();
6576
6577	hlist_for_each_entry_rcu(k, hash_head, hash_entry) {
6578		if (k == key) {
6579			hlist_del_rcu(&k->hash_entry);
6580			found = true;
6581			break;
6582		}
6583	}
6584	WARN_ON_ONCE(!found && debug_locks);
6585	if (found) {
6586		pf = get_pending_free();
6587		__lockdep_free_key_range(pf, key, 1);
6588		need_callback = prepare_call_rcu_zapped(pf);
6589	}
6590	lockdep_unlock();
6591	raw_local_irq_restore(flags);
6592
6593	if (need_callback)
6594		call_rcu(&delayed_free.rcu_head, free_zapped_rcu);
6595
6596	/* Wait until is_dynamic_key() has finished accessing k->hash_entry. */
6597	synchronize_rcu();
6598}
6599EXPORT_SYMBOL_GPL(lockdep_unregister_key);
6600
6601void __init lockdep_init(void)
6602{
6603	pr_info("Lock dependency validator: Copyright (c) 2006 Red Hat, Inc., Ingo Molnar\n");
6604
6605	pr_info("... MAX_LOCKDEP_SUBCLASSES:  %lu\n", MAX_LOCKDEP_SUBCLASSES);
6606	pr_info("... MAX_LOCK_DEPTH:          %lu\n", MAX_LOCK_DEPTH);
6607	pr_info("... MAX_LOCKDEP_KEYS:        %lu\n", MAX_LOCKDEP_KEYS);
6608	pr_info("... CLASSHASH_SIZE:          %lu\n", CLASSHASH_SIZE);
6609	pr_info("... MAX_LOCKDEP_ENTRIES:     %lu\n", MAX_LOCKDEP_ENTRIES);
6610	pr_info("... MAX_LOCKDEP_CHAINS:      %lu\n", MAX_LOCKDEP_CHAINS);
6611	pr_info("... CHAINHASH_SIZE:          %lu\n", CHAINHASH_SIZE);
6612
6613	pr_info(" memory used by lock dependency info: %zu kB\n",
6614	       (sizeof(lock_classes) +
6615		sizeof(lock_classes_in_use) +
6616		sizeof(classhash_table) +
6617		sizeof(list_entries) +
6618		sizeof(list_entries_in_use) +
6619		sizeof(chainhash_table) +
6620		sizeof(delayed_free)
6621#ifdef CONFIG_PROVE_LOCKING
6622		+ sizeof(lock_cq)
6623		+ sizeof(lock_chains)
6624		+ sizeof(lock_chains_in_use)
6625		+ sizeof(chain_hlocks)
6626#endif
6627		) / 1024
6628		);
6629
6630#if defined(CONFIG_TRACE_IRQFLAGS) && defined(CONFIG_PROVE_LOCKING)
6631	pr_info(" memory used for stack traces: %zu kB\n",
6632	       (sizeof(stack_trace) + sizeof(stack_trace_hash)) / 1024
6633	       );
6634#endif
6635
6636	pr_info(" per task-struct memory footprint: %zu bytes\n",
6637	       sizeof(((struct task_struct *)NULL)->held_locks));
6638}
6639
6640static void
6641print_freed_lock_bug(struct task_struct *curr, const void *mem_from,
6642		     const void *mem_to, struct held_lock *hlock)
6643{
6644	if (!debug_locks_off())
6645		return;
6646	if (debug_locks_silent)
6647		return;
6648
6649	nbcon_cpu_emergency_enter();
6650
6651	pr_warn("\n");
6652	pr_warn("=========================\n");
6653	pr_warn("WARNING: held lock freed!\n");
6654	print_kernel_ident();
6655	pr_warn("-------------------------\n");
6656	pr_warn("%s/%d is freeing memory %px-%px, with a lock still held there!\n",
6657		curr->comm, task_pid_nr(curr), mem_from, mem_to-1);
6658	print_lock(hlock);
6659	lockdep_print_held_locks(curr);
6660
6661	pr_warn("\nstack backtrace:\n");
6662	dump_stack();
6663
6664	nbcon_cpu_emergency_exit();
6665}
6666
6667static inline int not_in_range(const void* mem_from, unsigned long mem_len,
6668				const void* lock_from, unsigned long lock_len)
6669{
6670	return lock_from + lock_len <= mem_from ||
6671		mem_from + mem_len <= lock_from;
6672}
6673
6674/*
6675 * Called when kernel memory is freed (or unmapped), or if a lock
6676 * is destroyed or reinitialized - this code checks whether there is
6677 * any held lock in the memory range of <from> to <to>:
6678 */
6679void debug_check_no_locks_freed(const void *mem_from, unsigned long mem_len)
6680{
6681	struct task_struct *curr = current;
6682	struct held_lock *hlock;
6683	unsigned long flags;
6684	int i;
6685
6686	if (unlikely(!debug_locks))
6687		return;
6688
6689	raw_local_irq_save(flags);
6690	for (i = 0; i < curr->lockdep_depth; i++) {
6691		hlock = curr->held_locks + i;
6692
6693		if (not_in_range(mem_from, mem_len, hlock->instance,
6694					sizeof(*hlock->instance)))
6695			continue;
6696
6697		print_freed_lock_bug(curr, mem_from, mem_from + mem_len, hlock);
6698		break;
6699	}
6700	raw_local_irq_restore(flags);
6701}
6702EXPORT_SYMBOL_GPL(debug_check_no_locks_freed);
6703
6704static void print_held_locks_bug(void)
6705{
6706	if (!debug_locks_off())
6707		return;
6708	if (debug_locks_silent)
6709		return;
6710
6711	nbcon_cpu_emergency_enter();
6712
6713	pr_warn("\n");
6714	pr_warn("====================================\n");
6715	pr_warn("WARNING: %s/%d still has locks held!\n",
6716	       current->comm, task_pid_nr(current));
6717	print_kernel_ident();
6718	pr_warn("------------------------------------\n");
6719	lockdep_print_held_locks(current);
6720	pr_warn("\nstack backtrace:\n");
6721	dump_stack();
6722
6723	nbcon_cpu_emergency_exit();
6724}
6725
6726void debug_check_no_locks_held(void)
6727{
6728	if (unlikely(current->lockdep_depth > 0))
6729		print_held_locks_bug();
6730}
6731EXPORT_SYMBOL_GPL(debug_check_no_locks_held);
6732
6733#ifdef __KERNEL__
6734void debug_show_all_locks(void)
6735{
6736	struct task_struct *g, *p;
6737
6738	if (unlikely(!debug_locks)) {
6739		pr_warn("INFO: lockdep is turned off.\n");
6740		return;
6741	}
6742	pr_warn("\nShowing all locks held in the system:\n");
6743
6744	rcu_read_lock();
6745	for_each_process_thread(g, p) {
6746		if (!p->lockdep_depth)
6747			continue;
6748		lockdep_print_held_locks(p);
6749		touch_nmi_watchdog();
6750		touch_all_softlockup_watchdogs();
6751	}
6752	rcu_read_unlock();
6753
6754	pr_warn("\n");
6755	pr_warn("=============================================\n\n");
6756}
6757EXPORT_SYMBOL_GPL(debug_show_all_locks);
6758#endif
6759
6760/*
6761 * Careful: only use this function if you are sure that
6762 * the task cannot run in parallel!
6763 */
6764void debug_show_held_locks(struct task_struct *task)
6765{
6766	if (unlikely(!debug_locks)) {
6767		printk("INFO: lockdep is turned off.\n");
6768		return;
6769	}
6770	lockdep_print_held_locks(task);
6771}
6772EXPORT_SYMBOL_GPL(debug_show_held_locks);
6773
6774asmlinkage __visible void lockdep_sys_exit(void)
6775{
6776	struct task_struct *curr = current;
6777
6778	if (unlikely(curr->lockdep_depth)) {
6779		if (!debug_locks_off())
6780			return;
6781		nbcon_cpu_emergency_enter();
6782		pr_warn("\n");
6783		pr_warn("================================================\n");
6784		pr_warn("WARNING: lock held when returning to user space!\n");
6785		print_kernel_ident();
6786		pr_warn("------------------------------------------------\n");
6787		pr_warn("%s/%d is leaving the kernel with locks still held!\n",
6788				curr->comm, curr->pid);
6789		lockdep_print_held_locks(curr);
6790		nbcon_cpu_emergency_exit();
6791	}
6792
6793	/*
6794	 * The lock history for each syscall should be independent. So wipe the
6795	 * slate clean on return to userspace.
6796	 */
6797	lockdep_invariant_state(false);
6798}
6799
6800void lockdep_rcu_suspicious(const char *file, const int line, const char *s)
6801{
6802	struct task_struct *curr = current;
6803	int dl = READ_ONCE(debug_locks);
6804	bool rcu = warn_rcu_enter();
6805
6806	/* Note: the following can be executed concurrently, so be careful. */
6807	nbcon_cpu_emergency_enter();
6808	pr_warn("\n");
6809	pr_warn("=============================\n");
6810	pr_warn("WARNING: suspicious RCU usage\n");
6811	print_kernel_ident();
6812	pr_warn("-----------------------------\n");
6813	pr_warn("%s:%d %s!\n", file, line, s);
6814	pr_warn("\nother info that might help us debug this:\n\n");
6815	pr_warn("\n%srcu_scheduler_active = %d, debug_locks = %d\n%s",
6816	       !rcu_lockdep_current_cpu_online()
6817			? "RCU used illegally from offline CPU!\n"
6818			: "",
6819	       rcu_scheduler_active, dl,
6820	       dl ? "" : "Possible false positive due to lockdep disabling via debug_locks = 0\n");
6821
6822	/*
6823	 * If a CPU is in the RCU-free window in idle (ie: in the section
6824	 * between ct_idle_enter() and ct_idle_exit(), then RCU
6825	 * considers that CPU to be in an "extended quiescent state",
6826	 * which means that RCU will be completely ignoring that CPU.
6827	 * Therefore, rcu_read_lock() and friends have absolutely no
6828	 * effect on a CPU running in that state. In other words, even if
6829	 * such an RCU-idle CPU has called rcu_read_lock(), RCU might well
6830	 * delete data structures out from under it.  RCU really has no
6831	 * choice here: we need to keep an RCU-free window in idle where
6832	 * the CPU may possibly enter into low power mode. This way we can
6833	 * notice an extended quiescent state to other CPUs that started a grace
6834	 * period. Otherwise we would delay any grace period as long as we run
6835	 * in the idle task.
6836	 *
6837	 * So complain bitterly if someone does call rcu_read_lock(),
6838	 * rcu_read_lock_bh() and so on from extended quiescent states.
6839	 */
6840	if (!rcu_is_watching())
6841		pr_warn("RCU used illegally from extended quiescent state!\n");
6842
6843	lockdep_print_held_locks(curr);
6844	pr_warn("\nstack backtrace:\n");
6845	dump_stack();
6846	nbcon_cpu_emergency_exit();
6847	warn_rcu_exit(rcu);
6848}
6849EXPORT_SYMBOL_GPL(lockdep_rcu_suspicious);