| 1 | |
| 2 | * kernel/lockdep.c
|
| 3 | *
|
| 4 | * Runtime locking correctness validator
|
| 5 | *
|
| 6 | * Started by Ingo Molnar:
|
| 7 | *
|
| 8 | * Copyright (C) 2006,2007 Red Hat, Inc., Ingo Molnar <[email protected]>
|
| 9 | * Copyright (C) 2007 Red Hat, Inc., Peter Zijlstra
|
| 10 | *
|
| 11 | * this code maps all the lock dependencies as they occur in a live kernel
|
| 12 | * and will warn about the following classes of locking bugs:
|
| 13 | *
|
| 14 | * - lock inversion scenarios
|
| 15 | * - circular lock dependencies
|
| 16 | * - hardirq/softirq safe/unsafe locking bugs
|
| 17 | *
|
| 18 | * Bugs are reported even if the current locking scenario does not cause
|
| 19 | * any deadlock at this point.
|
| 20 | *
|
| 21 | * I.e. if anytime in the past two locks were taken in a different order,
|
| 22 | * even if it happened for another task, even if those were different
|
| 23 | * locks (but of the same class as this lock), this code will detect it.
|
| 24 | *
|
| 25 | * Thanks to Arjan van de Ven for coming up with the initial idea of
|
| 26 | * mapping lock dependencies runtime.
|
| 27 |
|
| 28 | #define DISABLE_BRANCH_PROFILING
|
| 29 | #include <linux/mutex.h>
|
| 30 | #include <linux/sched.h>
|
| 31 | #include <linux/sched/clock.h>
|
| 32 | #include <linux/sched/task.h>
|
| 33 | #include <linux/sched/mm.h>
|
| 34 | #include <linux/delay.h>
|
| 35 | #include <linux/module.h>
|
| 36 | #include <linux/proc_fs.h>
|
| 37 | #include <linux/seq_file.h>
|
| 38 | #include <linux/spinlock.h>
|
| 39 | #include <linux/kallsyms.h>
|
| 40 | #include <linux/interrupt.h>
|
| 41 | #include <linux/stacktrace.h>
|
| 42 | #include <linux/debug_locks.h>
|
| 43 | #include <linux/irqflags.h>
|
| 44 | #include <linux/utsname.h>
|
| 45 | #include <linux/hash.h>
|
| 46 | #include <linux/ftrace.h>
|
| 47 | #include <linux/stringify.h>
|
| 48 | #include <linux/bitops.h>
|
| 49 | #include <linux/gfp.h>
|
| 50 | #include <linux/kmemcheck.h>
|
| 51 | #include <linux/random.h>
|
| 52 | #include <linux/jhash.h>
|
| 53 |
|
| 54 | #include <asm/sections.h>
|
| 55 |
|
| 56 | #include "lockdep_internals.h"
|
| 57 |
|
| 58 | #define CREATE_TRACE_POINTS
|
| 59 | #include <trace/events/lock.h>
|
| 60 |
|
| 61 | #ifdef CONFIG_PROVE_LOCKING
|
| 62 | int prove_locking = 1;
|
| 63 | module_param(prove_locking, int, 0644);
|
| 64 | #else
|
| 65 | #define prove_locking 0
|
| 66 | #endif
|
| 67 |
|
| 68 | #ifdef CONFIG_LOCK_STAT
|
| 69 | int lock_stat = 1;
|
| 70 | module_param(lock_stat, int, 0644);
|
| 71 | #else
|
| 72 | #define lock_stat 0
|
| 73 | #endif
|
| 74 |
|
| 75 | |
| 76 | * lockdep_lock: protects the lockdep graph, the hashes and the
|
| 77 | * class/list/hash allocators.
|
| 78 | *
|
| 79 | * This is one of the rare exceptions where it's justified
|
| 80 | * to use a raw spinlock - we really dont want the spinlock
|
| 81 | * code to recurse back into the lockdep code...
|
| 82 |
|
| 83 | static arch_spinlock_t lockdep_lock = (arch_spinlock_t)__ARCH_SPIN_LOCK_UNLOCKED;
|
| 84 |
|
| 85 | static int graph_lock(void)
|
| 86 | {
|
| 87 | arch_spin_lock(&lockdep_lock);
|
| 88 | |
| 89 | * Make sure that if another CPU detected a bug while
|
| 90 | * walking the graph we dont change it (while the other
|
| 91 | * CPU is busy printing out stuff with the graph lock
|
| 92 | * dropped already)
|
| 93 |
|
| 94 | if (!debug_locks) {
|
| 95 | arch_spin_unlock(&lockdep_lock);
|
| 96 | return 0;
|
| 97 | }
|
| 98 |
|
| 99 | current->lockdep_recursion++;
|
| 100 | return 1;
|
| 101 | }
|
| 102 |
|
| 103 | static inline int graph_unlock(void)
|
| 104 | {
|
| 105 | if (debug_locks && !arch_spin_is_locked(&lockdep_lock)) {
|
| 106 | |
| 107 | * The lockdep graph lock isn't locked while we expect it to
|
| 108 | * be, we're confused now, bye!
|
| 109 |
|
| 110 | return DEBUG_LOCKS_WARN_ON(1);
|
| 111 | }
|
| 112 |
|
| 113 | current->lockdep_recursion--;
|
| 114 | arch_spin_unlock(&lockdep_lock);
|
| 115 | return 0;
|
| 116 | }
|
| 117 |
|
| 118 | |
| 119 | * Turn lock debugging off and return with 0 if it was off already,
|
| 120 | * and also release the graph lock:
|
| 121 |
|
| 122 | static inline int debug_locks_off_graph_unlock(void)
|
| 123 | {
|
| 124 | int ret = debug_locks_off();
|
| 125 |
|
| 126 | arch_spin_unlock(&lockdep_lock);
|
| 127 |
|
| 128 | return ret;
|
| 129 | }
|
| 130 |
|
| 131 | unsigned long nr_list_entries;
|
| 132 | static struct lock_list list_entries[MAX_LOCKDEP_ENTRIES];
|
| 133 |
|
| 134 | |
| 135 | * All data structures here are protected by the global debug_lock.
|
| 136 | *
|
| 137 | * Mutex key structs only get allocated, once during bootup, and never
|
| 138 | * get freed - this significantly simplifies the debugging code.
|
| 139 |
|
| 140 | unsigned long nr_lock_classes;
|
| 141 | static struct lock_class lock_classes[MAX_LOCKDEP_KEYS];
|
| 142 |
|
| 143 | static inline struct lock_class *hlock_class(struct held_lock *hlock)
|
| 144 | {
|
| 145 | if (!hlock->class_idx) {
|
| 146 | |
| 147 | * Someone passed in garbage, we give up.
|
| 148 |
|
| 149 | DEBUG_LOCKS_WARN_ON(1);
|
| 150 | return NULL;
|
| 151 | }
|
| 152 | return lock_classes + hlock->class_idx - 1;
|
| 153 | }
|
| 154 |
|
| 155 | #ifdef CONFIG_LOCK_STAT
|
| 156 | static DEFINE_PER_CPU(struct lock_class_stats[MAX_LOCKDEP_KEYS], cpu_lock_stats);
|
| 157 |
|
| 158 | static inline u64 lockstat_clock(void)
|
| 159 | {
|
| 160 | return local_clock();
|
| 161 | }
|
| 162 |
|
| 163 | static int lock_point(unsigned long points[], unsigned long ip)
|
| 164 | {
|
| 165 | int i;
|
| 166 |
|
| 167 | for (i = 0; i < LOCKSTAT_POINTS; i++) {
|
| 168 | if (points[i] == 0) {
|
| 169 | points[i] = ip;
|
| 170 | break;
|
| 171 | }
|
| 172 | if (points[i] == ip)
|
| 173 | break;
|
| 174 | }
|
| 175 |
|
| 176 | return i;
|
| 177 | }
|
| 178 |
|
| 179 | static void lock_time_inc(struct lock_time *lt, u64 time)
|
| 180 | {
|
| 181 | if (time > lt->max)
|
| 182 | lt->max = time;
|
| 183 |
|
| 184 | if (time < lt->min || !lt->nr)
|
| 185 | lt->min = time;
|
| 186 |
|
| 187 | lt->total += time;
|
| 188 | lt->nr++;
|
| 189 | }
|
| 190 |
|
| 191 | static inline void lock_time_add(struct lock_time *src, struct lock_time *dst)
|
| 192 | {
|
| 193 | if (!src->nr)
|
| 194 | return;
|
| 195 |
|
| 196 | if (src->max > dst->max)
|
| 197 | dst->max = src->max;
|
| 198 |
|
| 199 | if (src->min < dst->min || !dst->nr)
|
| 200 | dst->min = src->min;
|
| 201 |
|
| 202 | dst->total += src->total;
|
| 203 | dst->nr += src->nr;
|
| 204 | }
|
| 205 |
|
| 206 | struct lock_class_stats lock_stats(struct lock_class *class)
|
| 207 | {
|
| 208 | struct lock_class_stats stats;
|
| 209 | int cpu, i;
|
| 210 |
|
| 211 | memset(&stats, 0, sizeof(struct lock_class_stats));
|
| 212 | for_each_possible_cpu(cpu) {
|
| 213 | struct lock_class_stats *pcs =
|
| 214 | &per_cpu(cpu_lock_stats, cpu)[class - lock_classes];
|
| 215 |
|
| 216 | for (i = 0; i < ARRAY_SIZE(stats.contention_point); i++)
|
| 217 | stats.contention_point[i] += pcs->contention_point[i];
|
| 218 |
|
| 219 | for (i = 0; i < ARRAY_SIZE(stats.contending_point); i++)
|
| 220 | stats.contending_point[i] += pcs->contending_point[i];
|
| 221 |
|
| 222 | lock_time_add(&pcs->read_waittime, &stats.read_waittime);
|
| 223 | lock_time_add(&pcs->write_waittime, &stats.write_waittime);
|
| 224 |
|
| 225 | lock_time_add(&pcs->read_holdtime, &stats.read_holdtime);
|
| 226 | lock_time_add(&pcs->write_holdtime, &stats.write_holdtime);
|
| 227 |
|
| 228 | for (i = 0; i < ARRAY_SIZE(stats.bounces); i++)
|
| 229 | stats.bounces[i] += pcs->bounces[i];
|
| 230 | }
|
| 231 |
|
| 232 | return stats;
|
| 233 | }
|
| 234 |
|
| 235 | void clear_lock_stats(struct lock_class *class)
|
| 236 | {
|
| 237 | int cpu;
|
| 238 |
|
| 239 | for_each_possible_cpu(cpu) {
|
| 240 | struct lock_class_stats *cpu_stats =
|
| 241 | &per_cpu(cpu_lock_stats, cpu)[class - lock_classes];
|
| 242 |
|
| 243 | memset(cpu_stats, 0, sizeof(struct lock_class_stats));
|
| 244 | }
|
| 245 | memset(class->contention_point, 0, sizeof(class->contention_point));
|
| 246 | memset(class->contending_point, 0, sizeof(class->contending_point));
|
| 247 | }
|
| 248 |
|
| 249 | static struct lock_class_stats *get_lock_stats(struct lock_class *class)
|
| 250 | {
|
| 251 | return &get_cpu_var(cpu_lock_stats)[class - lock_classes];
|
| 252 | }
|
| 253 |
|
| 254 | static void put_lock_stats(struct lock_class_stats *stats)
|
| 255 | {
|
| 256 | put_cpu_var(cpu_lock_stats);
|
| 257 | }
|
| 258 |
|
| 259 | static void lock_release_holdtime(struct held_lock *hlock)
|
| 260 | {
|
| 261 | struct lock_class_stats *stats;
|
| 262 | u64 holdtime;
|
| 263 |
|
| 264 | if (!lock_stat)
|
| 265 | return;
|
| 266 |
|
| 267 | holdtime = lockstat_clock() - hlock->holdtime_stamp;
|
| 268 |
|
| 269 | stats = get_lock_stats(hlock_class(hlock));
|
| 270 | if (hlock->read)
|
| 271 | lock_time_inc(&stats->read_holdtime, holdtime);
|
| 272 | else
|
| 273 | lock_time_inc(&stats->write_holdtime, holdtime);
|
| 274 | put_lock_stats(stats);
|
| 275 | }
|
| 276 | #else
|
| 277 | static inline void lock_release_holdtime(struct held_lock *hlock)
|
| 278 | {
|
| 279 | }
|
| 280 | #endif
|
| 281 |
|
| 282 | |
| 283 | * We keep a global list of all lock classes. The list only grows,
|
| 284 | * never shrinks. The list is only accessed with the lockdep
|
| 285 | * spinlock lock held.
|
| 286 |
|
| 287 | LIST_HEAD(all_lock_classes);
|
| 288 |
|
| 289 | |
| 290 | * The lockdep classes are in a hash-table as well, for fast lookup:
|
| 291 |
|
| 292 | #define CLASSHASH_BITS (MAX_LOCKDEP_KEYS_BITS - 1)
|
| 293 | #define CLASSHASH_SIZE (1UL << CLASSHASH_BITS)
|
| 294 | #define __classhashfn(key) hash_long((unsigned long)key, CLASSHASH_BITS)
|
| 295 | #define classhashentry(key) (classhash_table + __classhashfn((key)))
|
| 296 |
|
| 297 | static struct hlist_head classhash_table[CLASSHASH_SIZE];
|
| 298 |
|
| 299 | |
| 300 | * We put the lock dependency chains into a hash-table as well, to cache
|
| 301 | * their existence:
|
| 302 |
|
| 303 | #define CHAINHASH_BITS (MAX_LOCKDEP_CHAINS_BITS-1)
|
| 304 | #define CHAINHASH_SIZE (1UL << CHAINHASH_BITS)
|
| 305 | #define __chainhashfn(chain) hash_long(chain, CHAINHASH_BITS)
|
| 306 | #define chainhashentry(chain) (chainhash_table + __chainhashfn((chain)))
|
| 307 |
|
| 308 | static struct hlist_head chainhash_table[CHAINHASH_SIZE];
|
| 309 |
|
| 310 | |
| 311 | * The hash key of the lock dependency chains is a hash itself too:
|
| 312 | * it's a hash of all locks taken up to that lock, including that lock.
|
| 313 | * It's a 64-bit hash, because it's important for the keys to be
|
| 314 | * unique.
|
| 315 |
|
| 316 | static inline u64 iterate_chain_key(u64 key, u32 idx)
|
| 317 | {
|
| 318 | u32 k0 = key, k1 = key >> 32;
|
| 319 |
|
| 320 | __jhash_mix(idx, k0, k1);
|
| 321 |
|
| 322 | return k0 | (u64)k1 << 32;
|
| 323 | }
|
| 324 |
|
| 325 | void lockdep_off(void)
|
| 326 | {
|
| 327 | current->lockdep_recursion++;
|
| 328 | }
|
| 329 | EXPORT_SYMBOL(lockdep_off);
|
| 330 |
|
| 331 | void lockdep_on(void)
|
| 332 | {
|
| 333 | current->lockdep_recursion--;
|
| 334 | }
|
| 335 | EXPORT_SYMBOL(lockdep_on);
|
| 336 |
|
| 337 | |
| 338 | * Debugging switches:
|
| 339 |
|
| 340 |
|
| 341 | #define VERBOSE 0
|
| 342 | #define VERY_VERBOSE 0
|
| 343 |
|
| 344 | #if VERBOSE
|
| 345 | # define HARDIRQ_VERBOSE 1
|
| 346 | # define SOFTIRQ_VERBOSE 1
|
| 347 | # define RECLAIM_VERBOSE 1
|
| 348 | #else
|
| 349 | # define HARDIRQ_VERBOSE 0
|
| 350 | # define SOFTIRQ_VERBOSE 0
|
| 351 | # define RECLAIM_VERBOSE 0
|
| 352 | #endif
|
| 353 |
|
| 354 | #if VERBOSE || HARDIRQ_VERBOSE || SOFTIRQ_VERBOSE || RECLAIM_VERBOSE
|
| 355 | |
| 356 | * Quick filtering for interesting events:
|
| 357 |
|
| 358 | static int class_filter(struct lock_class *class)
|
| 359 | {
|
| 360 | #if 0 |
| 361 | /* Example */
|
| 362 | if (class->name_version == 1 &&
|
| 363 | !strcmp(class->name, "lockname"))
|
| 364 | return 1;
|
| 365 | if (class->name_version == 1 &&
|
| 366 | !strcmp(class->name, "&struct->lockfield"))
|
| 367 | return 1;
|
| 368 | #endif
|
| 369 |
|
| 370 | return 0;
|
| 371 | }
|
| 372 | #endif
|
| 373 |
|
| 374 | static int verbose(struct lock_class *class)
|
| 375 | {
|
| 376 | #if VERBOSE
|
| 377 | return class_filter(class);
|
| 378 | #endif
|
| 379 | return 0;
|
| 380 | }
|
| 381 |
|
| 382 | |
| 383 | * Stack-trace: tightly packed array of stack backtrace
|
| 384 | * addresses. Protected by the graph_lock.
|
| 385 |
|
| 386 | unsigned long nr_stack_trace_entries;
|
| 387 | static unsigned long stack_trace[MAX_STACK_TRACE_ENTRIES];
|
| 388 |
|
| 389 | static void print_lockdep_off(const char *bug_msg)
|
| 390 | {
|
| 391 | printk(KERN_DEBUG "%s\n", bug_msg);
|
| 392 | printk(KERN_DEBUG "turning off the locking correctness validator.\n");
|
| 393 | #ifdef CONFIG_LOCK_STAT
|
| 394 | printk(KERN_DEBUG "Please attach the output of /proc/lock_stat to the bug report\n");
|
| 395 | #endif
|
| 396 | }
|
| 397 |
|
| 398 | static int save_trace(struct stack_trace *trace)
|
| 399 | {
|
| 400 | trace->nr_entries = 0;
|
| 401 | trace->max_entries = MAX_STACK_TRACE_ENTRIES - nr_stack_trace_entries;
|
| 402 | trace->entries = stack_trace + nr_stack_trace_entries;
|
| 403 |
|
| 404 | trace->skip = 3;
|
| 405 |
|
| 406 | save_stack_trace(trace);
|
| 407 |
|
| 408 | |
| 409 | * Some daft arches put -1 at the end to indicate its a full trace.
|
| 410 | *
|
| 411 | * <rant> this is buggy anyway, since it takes a whole extra entry so a
|
| 412 | * complete trace that maxes out the entries provided will be reported
|
| 413 | * as incomplete, friggin useless </rant>
|
| 414 |
|
| 415 | if (trace->nr_entries != 0 &&
|
| 416 | trace->entries[trace->nr_entries-1] == ULONG_MAX)
|
| 417 | trace->nr_entries--;
|
| 418 |
|
| 419 | trace->max_entries = trace->nr_entries;
|
| 420 |
|
| 421 | nr_stack_trace_entries += trace->nr_entries;
|
| 422 |
|
| 423 | if (nr_stack_trace_entries >= MAX_STACK_TRACE_ENTRIES-1) {
|
| 424 | if (!debug_locks_off_graph_unlock())
|
| 425 | return 0;
|
| 426 |
|
| 427 | print_lockdep_off("BUG: MAX_STACK_TRACE_ENTRIES too low!");
|
| 428 | dump_stack();
|
| 429 |
|
| 430 | return 0;
|
| 431 | }
|
| 432 |
|
| 433 | return 1;
|
| 434 | }
|
| 435 |
|
| 436 | unsigned int nr_hardirq_chains;
|
| 437 | unsigned int nr_softirq_chains;
|
| 438 | unsigned int nr_process_chains;
|
| 439 | unsigned int max_lockdep_depth;
|
| 440 |
|
| 441 | #ifdef CONFIG_DEBUG_LOCKDEP
|
| 442 | |
| 443 | * Various lockdep statistics:
|
| 444 |
|
| 445 | DEFINE_PER_CPU(struct lockdep_stats, lockdep_stats);
|
| 446 | #endif
|
| 447 |
|
| 448 | |
| 449 | * Locking printouts:
|
| 450 |
|
| 451 |
|
| 452 | #define __USAGE(__STATE) \
|
| 453 | [LOCK_USED_IN_##__STATE] = "IN-"__stringify(__STATE)"-W", \
|
| 454 | [LOCK_ENABLED_##__STATE] = __stringify(__STATE)"-ON-W", \
|
| 455 | [LOCK_USED_IN_##__STATE##_READ] = "IN-"__stringify(__STATE)"-R",\
|
| 456 | [LOCK_ENABLED_##__STATE##_READ] = __stringify(__STATE)"-ON-R",
|
| 457 |
|
| 458 | static const char *usage_str[] =
|
| 459 | {
|
| 460 | #define LOCKDEP_STATE(__STATE) __USAGE(__STATE)
|
| 461 | #include "lockdep_states.h"
|
| 462 | #undef LOCKDEP_STATE
|
| 463 | [LOCK_USED] = "INITIAL USE",
|
| 464 | };
|
| 465 |
|
| 466 | const char * __get_key_name(struct lockdep_subclass_key *key, char *str)
|
| 467 | {
|
| 468 | return kallsyms_lookup((unsigned long)key, NULL, NULL, NULL, str);
|
| 469 | }
|
| 470 |
|
| 471 | static inline unsigned long lock_flag(enum lock_usage_bit bit)
|
| 472 | {
|
| 473 | return 1UL << bit;
|
| 474 | }
|
| 475 |
|
| 476 | static char get_usage_char(struct lock_class *class, enum lock_usage_bit bit)
|
| 477 | {
|
| 478 | char c = '.';
|
| 479 |
|
| 480 | if (class->usage_mask & lock_flag(bit + 2))
|
| 481 | c = '+';
|
| 482 | if (class->usage_mask & lock_flag(bit)) {
|
| 483 | c = '-';
|
| 484 | if (class->usage_mask & lock_flag(bit + 2))
|
| 485 | c = '?';
|
| 486 | }
|
| 487 |
|
| 488 | return c;
|
| 489 | }
|
| 490 |
|
| 491 | void get_usage_chars(struct lock_class *class, char usage[LOCK_USAGE_CHARS])
|
| 492 | {
|
| 493 | int i = 0;
|
| 494 |
|
| 495 | #define LOCKDEP_STATE(__STATE) \
|
| 496 | usage[i++] = get_usage_char(class, LOCK_USED_IN_##__STATE); \
|
| 497 | usage[i++] = get_usage_char(class, LOCK_USED_IN_##__STATE##_READ);
|
| 498 | #include "lockdep_states.h"
|
| 499 | #undef LOCKDEP_STATE
|
| 500 |
|
| 501 | usage[i] = '\0';
|
| 502 | }
|
| 503 |
|
| 504 | static void __print_lock_name(struct lock_class *class)
|
| 505 | {
|
| 506 | char str[KSYM_NAME_LEN];
|
| 507 | const char *name;
|
| 508 |
|
| 509 | name = class->name;
|
| 510 | if (!name) {
|
| 511 | name = __get_key_name(class->key, str);
|
| 512 | printk(KERN_CONT "%s", name);
|
| 513 | } else {
|
| 514 | printk(KERN_CONT "%s", name);
|
| 515 | if (class->name_version > 1)
|
| 516 | printk(KERN_CONT "#%d", class->name_version);
|
| 517 | if (class->subclass)
|
| 518 | printk(KERN_CONT "/%d", class->subclass);
|
| 519 | }
|
| 520 | }
|
| 521 |
|
| 522 | static void print_lock_name(struct lock_class *class)
|
| 523 | {
|
| 524 | char usage[LOCK_USAGE_CHARS];
|
| 525 |
|
| 526 | get_usage_chars(class, usage);
|
| 527 |
|
| 528 | printk(KERN_CONT " (");
|
| 529 | __print_lock_name(class);
|
| 530 | printk(KERN_CONT "){%s}", usage);
|
| 531 | }
|
| 532 |
|
| 533 | static void print_lockdep_cache(struct lockdep_map *lock)
|
| 534 | {
|
| 535 | const char *name;
|
| 536 | char str[KSYM_NAME_LEN];
|
| 537 |
|
| 538 | name = lock->name;
|
| 539 | if (!name)
|
| 540 | name = __get_key_name(lock->key->subkeys, str);
|
| 541 |
|
| 542 | printk(KERN_CONT "%s", name);
|
| 543 | }
|
| 544 |
|
| 545 | static void print_lock(struct held_lock *hlock)
|
| 546 | {
|
| 547 | |
| 548 | * We can be called locklessly through debug_show_all_locks() so be
|
| 549 | * extra careful, the hlock might have been released and cleared.
|
| 550 |
|
| 551 | unsigned int class_idx = hlock->class_idx;
|
| 552 |
|
| 553 |
|
| 554 | barrier();
|
| 555 |
|
| 556 | if (!class_idx || (class_idx - 1) >= MAX_LOCKDEP_KEYS) {
|
| 557 | printk(KERN_CONT "<RELEASED>\n");
|
| 558 | return;
|
| 559 | }
|
| 560 |
|
| 561 | print_lock_name(lock_classes + class_idx - 1);
|
| 562 | printk(KERN_CONT ", at: [<%p>] %pS\n",
|
| 563 | (void *)hlock->acquire_ip, (void *)hlock->acquire_ip);
|
| 564 | }
|
| 565 |
|
| 566 | static void lockdep_print_held_locks(struct task_struct *curr)
|
| 567 | {
|
| 568 | int i, depth = curr->lockdep_depth;
|
| 569 |
|
| 570 | if (!depth) {
|
| 571 | printk("no locks held by %s/%d.\n", curr->comm, task_pid_nr(curr));
|
| 572 | return;
|
| 573 | }
|
| 574 | printk("%d lock%s held by %s/%d:\n",
|
| 575 | depth, depth > 1 ? "s" : "", curr->comm, task_pid_nr(curr));
|
| 576 |
|
| 577 | for (i = 0; i < depth; i++) {
|
| 578 | printk(" #%d: ", i);
|
| 579 | print_lock(curr->held_locks + i);
|
| 580 | }
|
| 581 | }
|
| 582 |
|
| 583 | static void print_kernel_ident(void)
|
| 584 | {
|
| 585 | printk("%s %.*s %s\n", init_utsname()->release,
|
| 586 | (int)strcspn(init_utsname()->version, " "),
|
| 587 | init_utsname()->version,
|
| 588 | print_tainted());
|
| 589 | }
|
| 590 |
|
| 591 | static int very_verbose(struct lock_class *class)
|
| 592 | {
|
| 593 | #if VERY_VERBOSE
|
| 594 | return class_filter(class);
|
| 595 | #endif
|
| 596 | return 0;
|
| 597 | }
|
| 598 |
|
| 599 | |
| 600 | * Is this the address of a static object:
|
| 601 |
|
| 602 | #ifdef __KERNEL__
|
| 603 | static int static_obj(void *obj)
|
| 604 | {
|
| 605 | unsigned long start = (unsigned long) &_stext,
|
| 606 | end = (unsigned long) &_end,
|
| 607 | addr = (unsigned long) obj;
|
| 608 |
|
| 609 | |
| 610 | * static variable?
|
| 611 |
|
| 612 | if ((addr >= start) && (addr < end))
|
| 613 | return 1;
|
| 614 |
|
| 615 | if (arch_is_kernel_data(addr))
|
| 616 | return 1;
|
| 617 |
|
| 618 | |
| 619 | * in-kernel percpu var?
|
| 620 |
|
| 621 | if (is_kernel_percpu_address(addr))
|
| 622 | return 1;
|
| 623 |
|
| 624 | |
| 625 | * module static or percpu var?
|
| 626 |
|
| 627 | return is_module_address(addr) || is_module_percpu_address(addr);
|
| 628 | }
|
| 629 | #endif
|
| 630 |
|
| 631 | |
| 632 | * To make lock name printouts unique, we calculate a unique
|
| 633 | * class->name_version generation counter:
|
| 634 |
|
| 635 | static int count_matching_names(struct lock_class *new_class)
|
| 636 | {
|
| 637 | struct lock_class *class;
|
| 638 | int count = 0;
|
| 639 |
|
| 640 | if (!new_class->name)
|
| 641 | return 0;
|
| 642 |
|
| 643 | list_for_each_entry_rcu(class, &all_lock_classes, lock_entry) {
|
| 644 | if (new_class->key - new_class->subclass == class->key)
|
| 645 | return class->name_version;
|
| 646 | if (class->name && !strcmp(class->name, new_class->name))
|
| 647 | count = max(count, class->name_version);
|
| 648 | }
|
| 649 |
|
| 650 | return count + 1;
|
| 651 | }
|
| 652 |
|
| 653 | |
| 654 | * Register a lock's class in the hash-table, if the class is not present
|
| 655 | * yet. Otherwise we look it up. We cache the result in the lock object
|
| 656 | * itself, so actual lookup of the hash should be once per lock object.
|
| 657 |
|
| 658 | static inline struct lock_class *
|
| 659 | look_up_lock_class(struct lockdep_map *lock, unsigned int subclass)
|
| 660 | {
|
| 661 | struct lockdep_subclass_key *key;
|
| 662 | struct hlist_head *hash_head;
|
| 663 | struct lock_class *class;
|
| 664 | bool is_static = false;
|
| 665 |
|
| 666 | if (unlikely(subclass >= MAX_LOCKDEP_SUBCLASSES)) {
|
| 667 | debug_locks_off();
|
| 668 | printk(KERN_ERR
|
| 669 | "BUG: looking up invalid subclass: %u\n", subclass);
|
| 670 | printk(KERN_ERR
|
| 671 | "turning off the locking correctness validator.\n");
|
| 672 | dump_stack();
|
| 673 | return NULL;
|
| 674 | }
|
| 675 |
|
| 676 | |
| 677 | * Static locks do not have their class-keys yet - for them the key
|
| 678 | * is the lock object itself. If the lock is in the per cpu area,
|
| 679 | * the canonical address of the lock (per cpu offset removed) is
|
| 680 | * used.
|
| 681 |
|
| 682 | if (unlikely(!lock->key)) {
|
| 683 | unsigned long can_addr, addr = (unsigned long)lock;
|
| 684 |
|
| 685 | if (__is_kernel_percpu_address(addr, &can_addr))
|
| 686 | lock->key = (void *)can_addr;
|
| 687 | else if (__is_module_percpu_address(addr, &can_addr))
|
| 688 | lock->key = (void *)can_addr;
|
| 689 | else if (static_obj(lock))
|
| 690 | lock->key = (void *)lock;
|
| 691 | else
|
| 692 | return ERR_PTR(-EINVAL);
|
| 693 | is_static = true;
|
| 694 | }
|
| 695 |
|
| 696 | |
| 697 | * NOTE: the class-key must be unique. For dynamic locks, a static
|
| 698 | * lock_class_key variable is passed in through the mutex_init()
|
| 699 | * (or spin_lock_init()) call - which acts as the key. For static
|
| 700 | * locks we use the lock object itself as the key.
|
| 701 |
|
| 702 | BUILD_BUG_ON(sizeof(struct lock_class_key) >
|
| 703 | sizeof(struct lockdep_map));
|
| 704 |
|
| 705 | key = lock->key->subkeys + subclass;
|
| 706 |
|
| 707 | hash_head = classhashentry(key);
|
| 708 |
|
| 709 | |
| 710 | * We do an RCU walk of the hash, see lockdep_free_key_range().
|
| 711 |
|
| 712 | if (DEBUG_LOCKS_WARN_ON(!irqs_disabled()))
|
| 713 | return NULL;
|
| 714 |
|
| 715 | hlist_for_each_entry_rcu(class, hash_head, hash_entry) {
|
| 716 | if (class->key == key) {
|
| 717 | |
| 718 | * Huh! same key, different name? Did someone trample
|
| 719 | * on some memory? We're most confused.
|
| 720 |
|
| 721 | WARN_ON_ONCE(class->name != lock->name);
|
| 722 | return class;
|
| 723 | }
|
| 724 | }
|
| 725 |
|
| 726 | return is_static || static_obj(lock->key) ? NULL : ERR_PTR(-EINVAL);
|
| 727 | }
|
| 728 |
|
| 729 | |
| 730 | * Register a lock's class in the hash-table, if the class is not present
|
| 731 | * yet. Otherwise we look it up. We cache the result in the lock object
|
| 732 | * itself, so actual lookup of the hash should be once per lock object.
|
| 733 |
|
| 734 | static struct lock_class *
|
| 735 | register_lock_class(struct lockdep_map *lock, unsigned int subclass, int force)
|
| 736 | {
|
| 737 | struct lockdep_subclass_key *key;
|
| 738 | struct hlist_head *hash_head;
|
| 739 | struct lock_class *class;
|
| 740 |
|
| 741 | DEBUG_LOCKS_WARN_ON(!irqs_disabled());
|
| 742 |
|
| 743 | class = look_up_lock_class(lock, subclass);
|
| 744 | if (likely(!IS_ERR_OR_NULL(class)))
|
| 745 | goto out_set_class_cache;
|
| 746 |
|
| 747 | |
| 748 | * Debug-check: all keys must be persistent!
|
| 749 |
|
| 750 | if (IS_ERR(class)) {
|
| 751 | debug_locks_off();
|
| 752 | printk("INFO: trying to register non-static key.\n");
|
| 753 | printk("the code is fine but needs lockdep annotation.\n");
|
| 754 | printk("turning off the locking correctness validator.\n");
|
| 755 | dump_stack();
|
| 756 | return NULL;
|
| 757 | }
|
| 758 |
|
| 759 | key = lock->key->subkeys + subclass;
|
| 760 | hash_head = classhashentry(key);
|
| 761 |
|
| 762 | if (!graph_lock()) {
|
| 763 | return NULL;
|
| 764 | }
|
| 765 | |
| 766 | * We have to do the hash-walk again, to avoid races
|
| 767 | * with another CPU:
|
| 768 |
|
| 769 | hlist_for_each_entry_rcu(class, hash_head, hash_entry) {
|
| 770 | if (class->key == key)
|
| 771 | goto out_unlock_set;
|
| 772 | }
|
| 773 |
|
| 774 | |
| 775 | * Allocate a new key from the static array, and add it to
|
| 776 | * the hash:
|
| 777 |
|
| 778 | if (nr_lock_classes >= MAX_LOCKDEP_KEYS) {
|
| 779 | if (!debug_locks_off_graph_unlock()) {
|
| 780 | return NULL;
|
| 781 | }
|
| 782 |
|
| 783 | print_lockdep_off("BUG: MAX_LOCKDEP_KEYS too low!");
|
| 784 | dump_stack();
|
| 785 | return NULL;
|
| 786 | }
|
| 787 | class = lock_classes + nr_lock_classes++;
|
| 788 | debug_atomic_inc(nr_unused_locks);
|
| 789 | class->key = key;
|
| 790 | class->name = lock->name;
|
| 791 | class->subclass = subclass;
|
| 792 | INIT_LIST_HEAD(&class->lock_entry);
|
| 793 | INIT_LIST_HEAD(&class->locks_before);
|
| 794 | INIT_LIST_HEAD(&class->locks_after);
|
| 795 | class->name_version = count_matching_names(class);
|
| 796 | |
| 797 | * We use RCU's safe list-add method to make
|
| 798 | * parallel walking of the hash-list safe:
|
| 799 |
|
| 800 | hlist_add_head_rcu(&class->hash_entry, hash_head);
|
| 801 | |
| 802 | * Add it to the global list of classes:
|
| 803 |
|
| 804 | list_add_tail_rcu(&class->lock_entry, &all_lock_classes);
|
| 805 |
|
| 806 | if (verbose(class)) {
|
| 807 | graph_unlock();
|
| 808 |
|
| 809 | printk("\nnew class %p: %s", class->key, class->name);
|
| 810 | if (class->name_version > 1)
|
| 811 | printk(KERN_CONT "#%d", class->name_version);
|
| 812 | printk(KERN_CONT "\n");
|
| 813 | dump_stack();
|
| 814 |
|
| 815 | if (!graph_lock()) {
|
| 816 | return NULL;
|
| 817 | }
|
| 818 | }
|
| 819 | out_unlock_set:
|
| 820 | graph_unlock();
|
| 821 |
|
| 822 | out_set_class_cache:
|
| 823 | if (!subclass || force)
|
| 824 | lock->class_cache[0] = class;
|
| 825 | else if (subclass < NR_LOCKDEP_CACHING_CLASSES)
|
| 826 | lock->class_cache[subclass] = class;
|
| 827 |
|
| 828 | |
| 829 | * Hash collision, did we smoke some? We found a class with a matching
|
| 830 | * hash but the subclass -- which is hashed in -- didn't match.
|
| 831 |
|
| 832 | if (DEBUG_LOCKS_WARN_ON(class->subclass != subclass))
|
| 833 | return NULL;
|
| 834 |
|
| 835 | return class;
|
| 836 | }
|
| 837 |
|
| 838 | #ifdef CONFIG_PROVE_LOCKING
|
| 839 | |
| 840 | * Allocate a lockdep entry. (assumes the graph_lock held, returns
|
| 841 | * with NULL on failure)
|
| 842 |
|
| 843 | static struct lock_list *alloc_list_entry(void)
|
| 844 | {
|
| 845 | if (nr_list_entries >= MAX_LOCKDEP_ENTRIES) {
|
| 846 | if (!debug_locks_off_graph_unlock())
|
| 847 | return NULL;
|
| 848 |
|
| 849 | print_lockdep_off("BUG: MAX_LOCKDEP_ENTRIES too low!");
|
| 850 | dump_stack();
|
| 851 | return NULL;
|
| 852 | }
|
| 853 | return list_entries + nr_list_entries++;
|
| 854 | }
|
| 855 |
|
| 856 | |
| 857 | * Add a new dependency to the head of the list:
|
| 858 |
|
| 859 | static int add_lock_to_list(struct lock_class *this, struct list_head *head,
|
| 860 | unsigned long ip, int distance,
|
| 861 | struct stack_trace *trace)
|
| 862 | {
|
| 863 | struct lock_list *entry;
|
| 864 | |
| 865 | * Lock not present yet - get a new dependency struct and
|
| 866 | * add it to the list:
|
| 867 |
|
| 868 | entry = alloc_list_entry();
|
| 869 | if (!entry)
|
| 870 | return 0;
|
| 871 |
|
| 872 | entry->class = this;
|
| 873 | entry->distance = distance;
|
| 874 | entry->trace = *trace;
|
| 875 | |
| 876 | * Both allocation and removal are done under the graph lock; but
|
| 877 | * iteration is under RCU-sched; see look_up_lock_class() and
|
| 878 | * lockdep_free_key_range().
|
| 879 |
|
| 880 | list_add_tail_rcu(&entry->entry, head);
|
| 881 |
|
| 882 | return 1;
|
| 883 | }
|
| 884 |
|
| 885 | |
| 886 | * For good efficiency of modular, we use power of 2
|
| 887 |
|
| 888 | #define MAX_CIRCULAR_QUEUE_SIZE 4096UL
|
| 889 | #define CQ_MASK (MAX_CIRCULAR_QUEUE_SIZE-1)
|
| 890 |
|
| 891 | |
| 892 | * The circular_queue and helpers is used to implement the
|
| 893 | * breadth-first search(BFS)algorithem, by which we can build
|
| 894 | * the shortest path from the next lock to be acquired to the
|
| 895 | * previous held lock if there is a circular between them.
|
| 896 |
|
| 897 | struct circular_queue {
|
| 898 | unsigned long element[MAX_CIRCULAR_QUEUE_SIZE];
|
| 899 | unsigned int front, rear;
|
| 900 | };
|
| 901 |
|
| 902 | static struct circular_queue lock_cq;
|
| 903 |
|
| 904 | unsigned int max_bfs_queue_depth;
|
| 905 |
|
| 906 | static unsigned int lockdep_dependency_gen_id;
|
| 907 |
|
| 908 | static inline void __cq_init(struct circular_queue *cq)
|
| 909 | {
|
| 910 | cq->front = cq->rear = 0;
|
| 911 | lockdep_dependency_gen_id++;
|
| 912 | }
|
| 913 |
|
| 914 | static inline int __cq_empty(struct circular_queue *cq)
|
| 915 | {
|
| 916 | return (cq->front == cq->rear);
|
| 917 | }
|
| 918 |
|
| 919 | static inline int __cq_full(struct circular_queue *cq)
|
| 920 | {
|
| 921 | return ((cq->rear + 1) & CQ_MASK) == cq->front;
|
| 922 | }
|
| 923 |
|
| 924 | static inline int __cq_enqueue(struct circular_queue *cq, unsigned long elem)
|
| 925 | {
|
| 926 | if (__cq_full(cq))
|
| 927 | return -1;
|
| 928 |
|
| 929 | cq->element[cq->rear] = elem;
|
| 930 | cq->rear = (cq->rear + 1) & CQ_MASK;
|
| 931 | return 0;
|
| 932 | }
|
| 933 |
|
| 934 | static inline int __cq_dequeue(struct circular_queue *cq, unsigned long *elem)
|
| 935 | {
|
| 936 | if (__cq_empty(cq))
|
| 937 | return -1;
|
| 938 |
|
| 939 | *elem = cq->element[cq->front];
|
| 940 | cq->front = (cq->front + 1) & CQ_MASK;
|
| 941 | return 0;
|
| 942 | }
|
| 943 |
|
| 944 | static inline unsigned int __cq_get_elem_count(struct circular_queue *cq)
|
| 945 | {
|
| 946 | return (cq->rear - cq->front) & CQ_MASK;
|
| 947 | }
|
| 948 |
|
| 949 | static inline void mark_lock_accessed(struct lock_list *lock,
|
| 950 | struct lock_list *parent)
|
| 951 | {
|
| 952 | unsigned long nr;
|
| 953 |
|
| 954 | nr = lock - list_entries;
|
| 955 | WARN_ON(nr >= nr_list_entries);
|
| 956 | lock->parent = parent;
|
| 957 | lock->class->dep_gen_id = lockdep_dependency_gen_id;
|
| 958 | }
|
| 959 |
|
| 960 | static inline unsigned long lock_accessed(struct lock_list *lock)
|
| 961 | {
|
| 962 | unsigned long nr;
|
| 963 |
|
| 964 | nr = lock - list_entries;
|
| 965 | WARN_ON(nr >= nr_list_entries);
|
| 966 | return lock->class->dep_gen_id == lockdep_dependency_gen_id;
|
| 967 | }
|
| 968 |
|
| 969 | static inline struct lock_list *get_lock_parent(struct lock_list *child)
|
| 970 | {
|
| 971 | return child->parent;
|
| 972 | }
|
| 973 |
|
| 974 | static inline int get_lock_depth(struct lock_list *child)
|
| 975 | {
|
| 976 | int depth = 0;
|
| 977 | struct lock_list *parent;
|
| 978 |
|
| 979 | while ((parent = get_lock_parent(child))) {
|
| 980 | child = parent;
|
| 981 | depth++;
|
| 982 | }
|
| 983 | return depth;
|
| 984 | }
|
| 985 |
|
| 986 | static int __bfs(struct lock_list *source_entry,
|
| 987 | void *data,
|
| 988 | int (*match)(struct lock_list *entry, void *data),
|
| 989 | struct lock_list **target_entry,
|
| 990 | int forward)
|
| 991 | {
|
| 992 | struct lock_list *entry;
|
| 993 | struct list_head *head;
|
| 994 | struct circular_queue *cq = &lock_cq;
|
| 995 | int ret = 1;
|
| 996 |
|
| 997 | if (match(source_entry, data)) {
|
| 998 | *target_entry = source_entry;
|
| 999 | ret = 0;
|
| 1000 | goto exit;
|
| 1001 | }
|
| 1002 |
|
| 1003 | if (forward)
|
| 1004 | head = &source_entry->class->locks_after;
|
| 1005 | else
|
| 1006 | head = &source_entry->class->locks_before;
|
| 1007 |
|
| 1008 | if (list_empty(head))
|
| 1009 | goto exit;
|
| 1010 |
|
| 1011 | __cq_init(cq);
|
| 1012 | __cq_enqueue(cq, (unsigned long)source_entry);
|
| 1013 |
|
| 1014 | while (!__cq_empty(cq)) {
|
| 1015 | struct lock_list *lock;
|
| 1016 |
|
| 1017 | __cq_dequeue(cq, (unsigned long *)&lock);
|
| 1018 |
|
| 1019 | if (!lock->class) {
|
| 1020 | ret = -2;
|
| 1021 | goto exit;
|
| 1022 | }
|
| 1023 |
|
| 1024 | if (forward)
|
| 1025 | head = &lock->class->locks_after;
|
| 1026 | else
|
| 1027 | head = &lock->class->locks_before;
|
| 1028 |
|
| 1029 | DEBUG_LOCKS_WARN_ON(!irqs_disabled());
|
| 1030 |
|
| 1031 | list_for_each_entry_rcu(entry, head, entry) {
|
| 1032 | if (!lock_accessed(entry)) {
|
| 1033 | unsigned int cq_depth;
|
| 1034 | mark_lock_accessed(entry, lock);
|
| 1035 | if (match(entry, data)) {
|
| 1036 | *target_entry = entry;
|
| 1037 | ret = 0;
|
| 1038 | goto exit;
|
| 1039 | }
|
| 1040 |
|
| 1041 | if (__cq_enqueue(cq, (unsigned long)entry)) {
|
| 1042 | ret = -1;
|
| 1043 | goto exit;
|
| 1044 | }
|
| 1045 | cq_depth = __cq_get_elem_count(cq);
|
| 1046 | if (max_bfs_queue_depth < cq_depth)
|
| 1047 | max_bfs_queue_depth = cq_depth;
|
| 1048 | }
|
| 1049 | }
|
| 1050 | }
|
| 1051 | exit:
|
| 1052 | return ret;
|
| 1053 | }
|
| 1054 |
|
| 1055 | static inline int __bfs_forwards(struct lock_list *src_entry,
|
| 1056 | void *data,
|
| 1057 | int (*match)(struct lock_list *entry, void *data),
|
| 1058 | struct lock_list **target_entry)
|
| 1059 | {
|
| 1060 | return __bfs(src_entry, data, match, target_entry, 1);
|
| 1061 |
|
| 1062 | }
|
| 1063 |
|
| 1064 | static inline int __bfs_backwards(struct lock_list *src_entry,
|
| 1065 | void *data,
|
| 1066 | int (*match)(struct lock_list *entry, void *data),
|
| 1067 | struct lock_list **target_entry)
|
| 1068 | {
|
| 1069 | return __bfs(src_entry, data, match, target_entry, 0);
|
| 1070 |
|
| 1071 | }
|
| 1072 |
|
| 1073 | |
| 1074 | * Recursive, forwards-direction lock-dependency checking, used for
|
| 1075 | * both noncyclic checking and for hardirq-unsafe/softirq-unsafe
|
| 1076 | * checking.
|
| 1077 |
|
| 1078 |
|
| 1079 | |
| 1080 | * Print a dependency chain entry (this is only done when a deadlock
|
| 1081 | * has been detected):
|
| 1082 |
|
| 1083 | static noinline int
|
| 1084 | print_circular_bug_entry(struct lock_list *target, int depth)
|
| 1085 | {
|
| 1086 | if (debug_locks_silent)
|
| 1087 | return 0;
|
| 1088 | printk("\n-> #%u", depth);
|
| 1089 | print_lock_name(target->class);
|
| 1090 | printk(KERN_CONT ":\n");
|
| 1091 | print_stack_trace(&target->trace, 6);
|
| 1092 |
|
| 1093 | return 0;
|
| 1094 | }
|
| 1095 |
|
| 1096 | static void
|
| 1097 | print_circular_lock_scenario(struct held_lock *src,
|
| 1098 | struct held_lock *tgt,
|
| 1099 | struct lock_list *prt)
|
| 1100 | {
|
| 1101 | struct lock_class *source = hlock_class(src);
|
| 1102 | struct lock_class *target = hlock_class(tgt);
|
| 1103 | struct lock_class *parent = prt->class;
|
| 1104 |
|
| 1105 | |
| 1106 | * A direct locking problem where unsafe_class lock is taken
|
| 1107 | * directly by safe_class lock, then all we need to show
|
| 1108 | * is the deadlock scenario, as it is obvious that the
|
| 1109 | * unsafe lock is taken under the safe lock.
|
| 1110 | *
|
| 1111 | * But if there is a chain instead, where the safe lock takes
|
| 1112 | * an intermediate lock (middle_class) where this lock is
|
| 1113 | * not the same as the safe lock, then the lock chain is
|
| 1114 | * used to describe the problem. Otherwise we would need
|
| 1115 | * to show a different CPU case for each link in the chain
|
| 1116 | * from the safe_class lock to the unsafe_class lock.
|
| 1117 |
|
| 1118 | if (parent != source) {
|
| 1119 | printk("Chain exists of:\n ");
|
| 1120 | __print_lock_name(source);
|
| 1121 | printk(KERN_CONT " --> ");
|
| 1122 | __print_lock_name(parent);
|
| 1123 | printk(KERN_CONT " --> ");
|
| 1124 | __print_lock_name(target);
|
| 1125 | printk(KERN_CONT "\n\n");
|
| 1126 | }
|
| 1127 |
|
| 1128 | printk(" Possible unsafe locking scenario:\n\n");
|
| 1129 | printk(" CPU0 CPU1\n");
|
| 1130 | printk(" ---- ----\n");
|
| 1131 | printk(" lock(");
|
| 1132 | __print_lock_name(target);
|
| 1133 | printk(KERN_CONT ");\n");
|
| 1134 | printk(" lock(");
|
| 1135 | __print_lock_name(parent);
|
| 1136 | printk(KERN_CONT ");\n");
|
| 1137 | printk(" lock(");
|
| 1138 | __print_lock_name(target);
|
| 1139 | printk(KERN_CONT ");\n");
|
| 1140 | printk(" lock(");
|
| 1141 | __print_lock_name(source);
|
| 1142 | printk(KERN_CONT ");\n");
|
| 1143 | printk("\n *** DEADLOCK ***\n\n");
|
| 1144 | }
|
| 1145 |
|
| 1146 | |
| 1147 | * When a circular dependency is detected, print the
|
| 1148 | * header first:
|
| 1149 |
|
| 1150 | static noinline int
|
| 1151 | print_circular_bug_header(struct lock_list *entry, unsigned int depth,
|
| 1152 | struct held_lock *check_src,
|
| 1153 | struct held_lock *check_tgt)
|
| 1154 | {
|
| 1155 | struct task_struct *curr = current;
|
| 1156 |
|
| 1157 | if (debug_locks_silent)
|
| 1158 | return 0;
|
| 1159 |
|
| 1160 | pr_warn("\n");
|
| 1161 | pr_warn("======================================================\n");
|
| 1162 | pr_warn("WARNING: possible circular locking dependency detected\n");
|
| 1163 | print_kernel_ident();
|
| 1164 | pr_warn("------------------------------------------------------\n");
|
| 1165 | pr_warn("%s/%d is trying to acquire lock:\n",
|
| 1166 | curr->comm, task_pid_nr(curr));
|
| 1167 | print_lock(check_src);
|
| 1168 | pr_warn("\nbut task is already holding lock:\n");
|
| 1169 | print_lock(check_tgt);
|
| 1170 | pr_warn("\nwhich lock already depends on the new lock.\n\n");
|
| 1171 | pr_warn("\nthe existing dependency chain (in reverse order) is:\n");
|
| 1172 |
|
| 1173 | print_circular_bug_entry(entry, depth);
|
| 1174 |
|
| 1175 | return 0;
|
| 1176 | }
|
| 1177 |
|
| 1178 | static inline int class_equal(struct lock_list *entry, void *data)
|
| 1179 | {
|
| 1180 | return entry->class == data;
|
| 1181 | }
|
| 1182 |
|
| 1183 | static noinline int print_circular_bug(struct lock_list *this,
|
| 1184 | struct lock_list *target,
|
| 1185 | struct held_lock *check_src,
|
| 1186 | struct held_lock *check_tgt)
|
| 1187 | {
|
| 1188 | struct task_struct *curr = current;
|
| 1189 | struct lock_list *parent;
|
| 1190 | struct lock_list *first_parent;
|
| 1191 | int depth;
|
| 1192 |
|
| 1193 | if (!debug_locks_off_graph_unlock() || debug_locks_silent)
|
| 1194 | return 0;
|
| 1195 |
|
| 1196 | if (!save_trace(&this->trace))
|
| 1197 | return 0;
|
| 1198 |
|
| 1199 | depth = get_lock_depth(target);
|
| 1200 |
|
| 1201 | print_circular_bug_header(target, depth, check_src, check_tgt);
|
| 1202 |
|
| 1203 | parent = get_lock_parent(target);
|
| 1204 | first_parent = parent;
|
| 1205 |
|
| 1206 | while (parent) {
|
| 1207 | print_circular_bug_entry(parent, --depth);
|
| 1208 | parent = get_lock_parent(parent);
|
| 1209 | }
|
| 1210 |
|
| 1211 | printk("\nother info that might help us debug this:\n\n");
|
| 1212 | print_circular_lock_scenario(check_src, check_tgt,
|
| 1213 | first_parent);
|
| 1214 |
|
| 1215 | lockdep_print_held_locks(curr);
|
| 1216 |
|
| 1217 | printk("\nstack backtrace:\n");
|
| 1218 | dump_stack();
|
| 1219 |
|
| 1220 | return 0;
|
| 1221 | }
|
| 1222 |
|
| 1223 | static noinline int print_bfs_bug(int ret)
|
| 1224 | {
|
| 1225 | if (!debug_locks_off_graph_unlock())
|
| 1226 | return 0;
|
| 1227 |
|
| 1228 | |
| 1229 | * Breadth-first-search failed, graph got corrupted?
|
| 1230 |
|
| 1231 | WARN(1, "lockdep bfs error:%d\n", ret);
|
| 1232 |
|
| 1233 | return 0;
|
| 1234 | }
|
| 1235 |
|
| 1236 | static int noop_count(struct lock_list *entry, void *data)
|
| 1237 | {
|
| 1238 | (*(unsigned long *)data)++;
|
| 1239 | return 0;
|
| 1240 | }
|
| 1241 |
|
| 1242 | static unsigned long __lockdep_count_forward_deps(struct lock_list *this)
|
| 1243 | {
|
| 1244 | unsigned long count = 0;
|
| 1245 | struct lock_list *uninitialized_var(target_entry);
|
| 1246 |
|
| 1247 | __bfs_forwards(this, (void *)&count, noop_count, &target_entry);
|
| 1248 |
|
| 1249 | return count;
|
| 1250 | }
|
| 1251 | unsigned long lockdep_count_forward_deps(struct lock_class *class)
|
| 1252 | {
|
| 1253 | unsigned long ret, flags;
|
| 1254 | struct lock_list this;
|
| 1255 |
|
| 1256 | this.parent = NULL;
|
| 1257 | this.class = class;
|
| 1258 |
|
| 1259 | local_irq_save(flags);
|
| 1260 | arch_spin_lock(&lockdep_lock);
|
| 1261 | ret = __lockdep_count_forward_deps(&this);
|
| 1262 | arch_spin_unlock(&lockdep_lock);
|
| 1263 | local_irq_restore(flags);
|
| 1264 |
|
| 1265 | return ret;
|
| 1266 | }
|
| 1267 |
|
| 1268 | static unsigned long __lockdep_count_backward_deps(struct lock_list *this)
|
| 1269 | {
|
| 1270 | unsigned long count = 0;
|
| 1271 | struct lock_list *uninitialized_var(target_entry);
|
| 1272 |
|
| 1273 | __bfs_backwards(this, (void *)&count, noop_count, &target_entry);
|
| 1274 |
|
| 1275 | return count;
|
| 1276 | }
|
| 1277 |
|
| 1278 | unsigned long lockdep_count_backward_deps(struct lock_class *class)
|
| 1279 | {
|
| 1280 | unsigned long ret, flags;
|
| 1281 | struct lock_list this;
|
| 1282 |
|
| 1283 | this.parent = NULL;
|
| 1284 | this.class = class;
|
| 1285 |
|
| 1286 | local_irq_save(flags);
|
| 1287 | arch_spin_lock(&lockdep_lock);
|
| 1288 | ret = __lockdep_count_backward_deps(&this);
|
| 1289 | arch_spin_unlock(&lockdep_lock);
|
| 1290 | local_irq_restore(flags);
|
| 1291 |
|
| 1292 | return ret;
|
| 1293 | }
|
| 1294 |
|
| 1295 | |
| 1296 | * Prove that the dependency graph starting at <entry> can not
|
| 1297 | * lead to <target>. Print an error and return 0 if it does.
|
| 1298 |
|
| 1299 | static noinline int
|
| 1300 | check_noncircular(struct lock_list *root, struct lock_class *target,
|
| 1301 | struct lock_list **target_entry)
|
| 1302 | {
|
| 1303 | int result;
|
| 1304 |
|
| 1305 | debug_atomic_inc(nr_cyclic_checks);
|
| 1306 |
|
| 1307 | result = __bfs_forwards(root, target, class_equal, target_entry);
|
| 1308 |
|
| 1309 | return result;
|
| 1310 | }
|
| 1311 |
|
| 1312 | #if defined(CONFIG_TRACE_IRQFLAGS) && defined(CONFIG_PROVE_LOCKING)
|
| 1313 | |
| 1314 | * Forwards and backwards subgraph searching, for the purposes of
|
| 1315 | * proving that two subgraphs can be connected by a new dependency
|
| 1316 | * without creating any illegal irq-safe -> irq-unsafe lock dependency.
|
| 1317 |
|
| 1318 |
|
| 1319 | static inline int usage_match(struct lock_list *entry, void *bit)
|
| 1320 | {
|
| 1321 | return entry->class->usage_mask & (1 << (enum lock_usage_bit)bit);
|
| 1322 | }
|
| 1323 |
|
| 1324 |
|
| 1325 |
|
| 1326 | |
| 1327 | * Find a node in the forwards-direction dependency sub-graph starting
|
| 1328 | * at @root->class that matches @bit.
|
| 1329 | *
|
| 1330 | * Return 0 if such a node exists in the subgraph, and put that node
|
| 1331 | * into *@target_entry.
|
| 1332 | *
|
| 1333 | * Return 1 otherwise and keep *@target_entry unchanged.
|
| 1334 | * Return <0 on error.
|
| 1335 |
|
| 1336 | static int
|
| 1337 | find_usage_forwards(struct lock_list *root, enum lock_usage_bit bit,
|
| 1338 | struct lock_list **target_entry)
|
| 1339 | {
|
| 1340 | int result;
|
| 1341 |
|
| 1342 | debug_atomic_inc(nr_find_usage_forwards_checks);
|
| 1343 |
|
| 1344 | result = __bfs_forwards(root, (void *)bit, usage_match, target_entry);
|
| 1345 |
|
| 1346 | return result;
|
| 1347 | }
|
| 1348 |
|
| 1349 | |
| 1350 | * Find a node in the backwards-direction dependency sub-graph starting
|
| 1351 | * at @root->class that matches @bit.
|
| 1352 | *
|
| 1353 | * Return 0 if such a node exists in the subgraph, and put that node
|
| 1354 | * into *@target_entry.
|
| 1355 | *
|
| 1356 | * Return 1 otherwise and keep *@target_entry unchanged.
|
| 1357 | * Return <0 on error.
|
| 1358 |
|
| 1359 | static int
|
| 1360 | find_usage_backwards(struct lock_list *root, enum lock_usage_bit bit,
|
| 1361 | struct lock_list **target_entry)
|
| 1362 | {
|
| 1363 | int result;
|
| 1364 |
|
| 1365 | debug_atomic_inc(nr_find_usage_backwards_checks);
|
| 1366 |
|
| 1367 | result = __bfs_backwards(root, (void *)bit, usage_match, target_entry);
|
| 1368 |
|
| 1369 | return result;
|
| 1370 | }
|
| 1371 |
|
| 1372 | static void print_lock_class_header(struct lock_class *class, int depth)
|
| 1373 | {
|
| 1374 | int bit;
|
| 1375 |
|
| 1376 | printk("%*s->", depth, "");
|
| 1377 | print_lock_name(class);
|
| 1378 | printk(KERN_CONT " ops: %lu", class->ops);
|
| 1379 | printk(KERN_CONT " {\n");
|
| 1380 |
|
| 1381 | for (bit = 0; bit < LOCK_USAGE_STATES; bit++) {
|
| 1382 | if (class->usage_mask & (1 << bit)) {
|
| 1383 | int len = depth;
|
| 1384 |
|
| 1385 | len += printk("%*s %s", depth, "", usage_str[bit]);
|
| 1386 | len += printk(KERN_CONT " at:\n");
|
| 1387 | print_stack_trace(class->usage_traces + bit, len);
|
| 1388 | }
|
| 1389 | }
|
| 1390 | printk("%*s }\n", depth, "");
|
| 1391 |
|
| 1392 | printk("%*s ... key at: [<%p>] %pS\n",
|
| 1393 | depth, "", class->key, class->key);
|
| 1394 | }
|
| 1395 |
|
| 1396 | |
| 1397 | * printk the shortest lock dependencies from @start to @end in reverse order:
|
| 1398 |
|
| 1399 | static void __used
|
| 1400 | print_shortest_lock_dependencies(struct lock_list *leaf,
|
| 1401 | struct lock_list *root)
|
| 1402 | {
|
| 1403 | struct lock_list *entry = leaf;
|
| 1404 | int depth;
|
| 1405 |
|
| 1406 |
|
| 1407 | depth = get_lock_depth(leaf);
|
| 1408 |
|
| 1409 | do {
|
| 1410 | print_lock_class_header(entry->class, depth);
|
| 1411 | printk("%*s ... acquired at:\n", depth, "");
|
| 1412 | print_stack_trace(&entry->trace, 2);
|
| 1413 | printk("\n");
|
| 1414 |
|
| 1415 | if (depth == 0 && (entry != root)) {
|
| 1416 | printk("lockdep:%s bad path found in chain graph\n", __func__);
|
| 1417 | break;
|
| 1418 | }
|
| 1419 |
|
| 1420 | entry = get_lock_parent(entry);
|
| 1421 | depth--;
|
| 1422 | } while (entry && (depth >= 0));
|
| 1423 |
|
| 1424 | return;
|
| 1425 | }
|
| 1426 |
|
| 1427 | static void
|
| 1428 | print_irq_lock_scenario(struct lock_list *safe_entry,
|
| 1429 | struct lock_list *unsafe_entry,
|
| 1430 | struct lock_class *prev_class,
|
| 1431 | struct lock_class *next_class)
|
| 1432 | {
|
| 1433 | struct lock_class *safe_class = safe_entry->class;
|
| 1434 | struct lock_class *unsafe_class = unsafe_entry->class;
|
| 1435 | struct lock_class *middle_class = prev_class;
|
| 1436 |
|
| 1437 | if (middle_class == safe_class)
|
| 1438 | middle_class = next_class;
|
| 1439 |
|
| 1440 | |
| 1441 | * A direct locking problem where unsafe_class lock is taken
|
| 1442 | * directly by safe_class lock, then all we need to show
|
| 1443 | * is the deadlock scenario, as it is obvious that the
|
| 1444 | * unsafe lock is taken under the safe lock.
|
| 1445 | *
|
| 1446 | * But if there is a chain instead, where the safe lock takes
|
| 1447 | * an intermediate lock (middle_class) where this lock is
|
| 1448 | * not the same as the safe lock, then the lock chain is
|
| 1449 | * used to describe the problem. Otherwise we would need
|
| 1450 | * to show a different CPU case for each link in the chain
|
| 1451 | * from the safe_class lock to the unsafe_class lock.
|
| 1452 |
|
| 1453 | if (middle_class != unsafe_class) {
|
| 1454 | printk("Chain exists of:\n ");
|
| 1455 | __print_lock_name(safe_class);
|
| 1456 | printk(KERN_CONT " --> ");
|
| 1457 | __print_lock_name(middle_class);
|
| 1458 | printk(KERN_CONT " --> ");
|
| 1459 | __print_lock_name(unsafe_class);
|
| 1460 | printk(KERN_CONT "\n\n");
|
| 1461 | }
|
| 1462 |
|
| 1463 | printk(" Possible interrupt unsafe locking scenario:\n\n");
|
| 1464 | printk(" CPU0 CPU1\n");
|
| 1465 | printk(" ---- ----\n");
|
| 1466 | printk(" lock(");
|
| 1467 | __print_lock_name(unsafe_class);
|
| 1468 | printk(KERN_CONT ");\n");
|
| 1469 | printk(" local_irq_disable();\n");
|
| 1470 | printk(" lock(");
|
| 1471 | __print_lock_name(safe_class);
|
| 1472 | printk(KERN_CONT ");\n");
|
| 1473 | printk(" lock(");
|
| 1474 | __print_lock_name(middle_class);
|
| 1475 | printk(KERN_CONT ");\n");
|
| 1476 | printk(" <Interrupt>\n");
|
| 1477 | printk(" lock(");
|
| 1478 | __print_lock_name(safe_class);
|
| 1479 | printk(KERN_CONT ");\n");
|
| 1480 | printk("\n *** DEADLOCK ***\n\n");
|
| 1481 | }
|
| 1482 |
|
| 1483 | static int
|
| 1484 | print_bad_irq_dependency(struct task_struct *curr,
|
| 1485 | struct lock_list *prev_root,
|
| 1486 | struct lock_list *next_root,
|
| 1487 | struct lock_list *backwards_entry,
|
| 1488 | struct lock_list *forwards_entry,
|
| 1489 | struct held_lock *prev,
|
| 1490 | struct held_lock *next,
|
| 1491 | enum lock_usage_bit bit1,
|
| 1492 | enum lock_usage_bit bit2,
|
| 1493 | const char *irqclass)
|
| 1494 | {
|
| 1495 | if (!debug_locks_off_graph_unlock() || debug_locks_silent)
|
| 1496 | return 0;
|
| 1497 |
|
| 1498 | pr_warn("\n");
|
| 1499 | pr_warn("=====================================================\n");
|
| 1500 | pr_warn("WARNING: %s-safe -> %s-unsafe lock order detected\n",
|
| 1501 | irqclass, irqclass);
|
| 1502 | print_kernel_ident();
|
| 1503 | pr_warn("-----------------------------------------------------\n");
|
| 1504 | pr_warn("%s/%d [HC%u[%lu]:SC%u[%lu]:HE%u:SE%u] is trying to acquire:\n",
|
| 1505 | curr->comm, task_pid_nr(curr),
|
| 1506 | curr->hardirq_context, hardirq_count() >> HARDIRQ_SHIFT,
|
| 1507 | curr->softirq_context, softirq_count() >> SOFTIRQ_SHIFT,
|
| 1508 | curr->hardirqs_enabled,
|
| 1509 | curr->softirqs_enabled);
|
| 1510 | print_lock(next);
|
| 1511 |
|
| 1512 | pr_warn("\nand this task is already holding:\n");
|
| 1513 | print_lock(prev);
|
| 1514 | pr_warn("which would create a new lock dependency:\n");
|
| 1515 | print_lock_name(hlock_class(prev));
|
| 1516 | pr_cont(" ->");
|
| 1517 | print_lock_name(hlock_class(next));
|
| 1518 | pr_cont("\n");
|
| 1519 |
|
| 1520 | pr_warn("\nbut this new dependency connects a %s-irq-safe lock:\n",
|
| 1521 | irqclass);
|
| 1522 | print_lock_name(backwards_entry->class);
|
| 1523 | pr_warn("\n... which became %s-irq-safe at:\n", irqclass);
|
| 1524 |
|
| 1525 | print_stack_trace(backwards_entry->class->usage_traces + bit1, 1);
|
| 1526 |
|
| 1527 | pr_warn("\nto a %s-irq-unsafe lock:\n", irqclass);
|
| 1528 | print_lock_name(forwards_entry->class);
|
| 1529 | pr_warn("\n... which became %s-irq-unsafe at:\n", irqclass);
|
| 1530 | pr_warn("...");
|
| 1531 |
|
| 1532 | print_stack_trace(forwards_entry->class->usage_traces + bit2, 1);
|
| 1533 |
|
| 1534 | pr_warn("\nother info that might help us debug this:\n\n");
|
| 1535 | print_irq_lock_scenario(backwards_entry, forwards_entry,
|
| 1536 | hlock_class(prev), hlock_class(next));
|
| 1537 |
|
| 1538 | lockdep_print_held_locks(curr);
|
| 1539 |
|
| 1540 | pr_warn("\nthe dependencies between %s-irq-safe lock and the holding lock:\n", irqclass);
|
| 1541 | if (!save_trace(&prev_root->trace))
|
| 1542 | return 0;
|
| 1543 | print_shortest_lock_dependencies(backwards_entry, prev_root);
|
| 1544 |
|
| 1545 | pr_warn("\nthe dependencies between the lock to be acquired");
|
| 1546 | pr_warn(" and %s-irq-unsafe lock:\n", irqclass);
|
| 1547 | if (!save_trace(&next_root->trace))
|
| 1548 | return 0;
|
| 1549 | print_shortest_lock_dependencies(forwards_entry, next_root);
|
| 1550 |
|
| 1551 | pr_warn("\nstack backtrace:\n");
|
| 1552 | dump_stack();
|
| 1553 |
|
| 1554 | return 0;
|
| 1555 | }
|
| 1556 |
|
| 1557 | static int
|
| 1558 | check_usage(struct task_struct *curr, struct held_lock *prev,
|
| 1559 | struct held_lock *next, enum lock_usage_bit bit_backwards,
|
| 1560 | enum lock_usage_bit bit_forwards, const char *irqclass)
|
| 1561 | {
|
| 1562 | int ret;
|
| 1563 | struct lock_list this, that;
|
| 1564 | struct lock_list *uninitialized_var(target_entry);
|
| 1565 | struct lock_list *uninitialized_var(target_entry1);
|
| 1566 |
|
| 1567 | this.parent = NULL;
|
| 1568 |
|
| 1569 | this.class = hlock_class(prev);
|
| 1570 | ret = find_usage_backwards(&this, bit_backwards, &target_entry);
|
| 1571 | if (ret < 0)
|
| 1572 | return print_bfs_bug(ret);
|
| 1573 | if (ret == 1)
|
| 1574 | return ret;
|
| 1575 |
|
| 1576 | that.parent = NULL;
|
| 1577 | that.class = hlock_class(next);
|
| 1578 | ret = find_usage_forwards(&that, bit_forwards, &target_entry1);
|
| 1579 | if (ret < 0)
|
| 1580 | return print_bfs_bug(ret);
|
| 1581 | if (ret == 1)
|
| 1582 | return ret;
|
| 1583 |
|
| 1584 | return print_bad_irq_dependency(curr, &this, &that,
|
| 1585 | target_entry, target_entry1,
|
| 1586 | prev, next,
|
| 1587 | bit_backwards, bit_forwards, irqclass);
|
| 1588 | }
|
| 1589 |
|
| 1590 | static const char *state_names[] = {
|
| 1591 | #define LOCKDEP_STATE(__STATE) \
|
| 1592 | __stringify(__STATE),
|
| 1593 | #include "lockdep_states.h"
|
| 1594 | #undef LOCKDEP_STATE
|
| 1595 | };
|
| 1596 |
|
| 1597 | static const char *state_rnames[] = {
|
| 1598 | #define LOCKDEP_STATE(__STATE) \
|
| 1599 | __stringify(__STATE)"-READ",
|
| 1600 | #include "lockdep_states.h"
|
| 1601 | #undef LOCKDEP_STATE
|
| 1602 | };
|
| 1603 |
|
| 1604 | static inline const char *state_name(enum lock_usage_bit bit)
|
| 1605 | {
|
| 1606 | return (bit & 1) ? state_rnames[bit >> 2] : state_names[bit >> 2];
|
| 1607 | }
|
| 1608 |
|
| 1609 | static int exclusive_bit(int new_bit)
|
| 1610 | {
|
| 1611 | |
| 1612 | * USED_IN
|
| 1613 | * USED_IN_READ
|
| 1614 | * ENABLED
|
| 1615 | * ENABLED_READ
|
| 1616 | *
|
| 1617 | * bit 0 - write/read
|
| 1618 | * bit 1 - used_in/enabled
|
| 1619 | * bit 2+ state
|
| 1620 |
|
| 1621 |
|
| 1622 | int state = new_bit & ~3;
|
| 1623 | int dir = new_bit & 2;
|
| 1624 |
|
| 1625 | |
| 1626 | * keep state, bit flip the direction and strip read.
|
| 1627 |
|
| 1628 | return state | (dir ^ 2);
|
| 1629 | }
|
| 1630 |
|
| 1631 | static int check_irq_usage(struct task_struct *curr, struct held_lock *prev,
|
| 1632 | struct held_lock *next, enum lock_usage_bit bit)
|
| 1633 | {
|
| 1634 | |
| 1635 | * Prove that the new dependency does not connect a hardirq-safe
|
| 1636 | * lock with a hardirq-unsafe lock - to achieve this we search
|
| 1637 | * the backwards-subgraph starting at <prev>, and the
|
| 1638 | * forwards-subgraph starting at <next>:
|
| 1639 |
|
| 1640 | if (!check_usage(curr, prev, next, bit,
|
| 1641 | exclusive_bit(bit), state_name(bit)))
|
| 1642 | return 0;
|
| 1643 |
|
| 1644 | bit++;
|
| 1645 |
|
| 1646 | |
| 1647 | * Prove that the new dependency does not connect a hardirq-safe-read
|
| 1648 | * lock with a hardirq-unsafe lock - to achieve this we search
|
| 1649 | * the backwards-subgraph starting at <prev>, and the
|
| 1650 | * forwards-subgraph starting at <next>:
|
| 1651 |
|
| 1652 | if (!check_usage(curr, prev, next, bit,
|
| 1653 | exclusive_bit(bit), state_name(bit)))
|
| 1654 | return 0;
|
| 1655 |
|
| 1656 | return 1;
|
| 1657 | }
|
| 1658 |
|
| 1659 | static int
|
| 1660 | check_prev_add_irq(struct task_struct *curr, struct held_lock *prev,
|
| 1661 | struct held_lock *next)
|
| 1662 | {
|
| 1663 | #define LOCKDEP_STATE(__STATE) \
|
| 1664 | if (!check_irq_usage(curr, prev, next, LOCK_USED_IN_##__STATE)) \
|
| 1665 | return 0;
|
| 1666 | #include "lockdep_states.h"
|
| 1667 | #undef LOCKDEP_STATE
|
| 1668 |
|
| 1669 | return 1;
|
| 1670 | }
|
| 1671 |
|
| 1672 | static void inc_chains(void)
|
| 1673 | {
|
| 1674 | if (current->hardirq_context)
|
| 1675 | nr_hardirq_chains++;
|
| 1676 | else {
|
| 1677 | if (current->softirq_context)
|
| 1678 | nr_softirq_chains++;
|
| 1679 | else
|
| 1680 | nr_process_chains++;
|
| 1681 | }
|
| 1682 | }
|
| 1683 |
|
| 1684 | #else
|
| 1685 |
|
| 1686 | static inline int
|
| 1687 | check_prev_add_irq(struct task_struct *curr, struct held_lock *prev,
|
| 1688 | struct held_lock *next)
|
| 1689 | {
|
| 1690 | return 1;
|
| 1691 | }
|
| 1692 |
|
| 1693 | static inline void inc_chains(void)
|
| 1694 | {
|
| 1695 | nr_process_chains++;
|
| 1696 | }
|
| 1697 |
|
| 1698 | #endif
|
| 1699 |
|
| 1700 | static void
|
| 1701 | print_deadlock_scenario(struct held_lock *nxt,
|
| 1702 | struct held_lock *prv)
|
| 1703 | {
|
| 1704 | struct lock_class *next = hlock_class(nxt);
|
| 1705 | struct lock_class *prev = hlock_class(prv);
|
| 1706 |
|
| 1707 | printk(" Possible unsafe locking scenario:\n\n");
|
| 1708 | printk(" CPU0\n");
|
| 1709 | printk(" ----\n");
|
| 1710 | printk(" lock(");
|
| 1711 | __print_lock_name(prev);
|
| 1712 | printk(KERN_CONT ");\n");
|
| 1713 | printk(" lock(");
|
| 1714 | __print_lock_name(next);
|
| 1715 | printk(KERN_CONT ");\n");
|
| 1716 | printk("\n *** DEADLOCK ***\n\n");
|
| 1717 | printk(" May be due to missing lock nesting notation\n\n");
|
| 1718 | }
|
| 1719 |
|
| 1720 | static int
|
| 1721 | print_deadlock_bug(struct task_struct *curr, struct held_lock *prev,
|
| 1722 | struct held_lock *next)
|
| 1723 | {
|
| 1724 | if (!debug_locks_off_graph_unlock() || debug_locks_silent)
|
| 1725 | return 0;
|
| 1726 |
|
| 1727 | pr_warn("\n");
|
| 1728 | pr_warn("============================================\n");
|
| 1729 | pr_warn("WARNING: possible recursive locking detected\n");
|
| 1730 | print_kernel_ident();
|
| 1731 | pr_warn("--------------------------------------------\n");
|
| 1732 | pr_warn("%s/%d is trying to acquire lock:\n",
|
| 1733 | curr->comm, task_pid_nr(curr));
|
| 1734 | print_lock(next);
|
| 1735 | pr_warn("\nbut task is already holding lock:\n");
|
| 1736 | print_lock(prev);
|
| 1737 |
|
| 1738 | pr_warn("\nother info that might help us debug this:\n");
|
| 1739 | print_deadlock_scenario(next, prev);
|
| 1740 | lockdep_print_held_locks(curr);
|
| 1741 |
|
| 1742 | pr_warn("\nstack backtrace:\n");
|
| 1743 | dump_stack();
|
| 1744 |
|
| 1745 | return 0;
|
| 1746 | }
|
| 1747 |
|
| 1748 | |
| 1749 | * Check whether we are holding such a class already.
|
| 1750 | *
|
| 1751 | * (Note that this has to be done separately, because the graph cannot
|
| 1752 | * detect such classes of deadlocks.)
|
| 1753 | *
|
| 1754 | * Returns: 0 on deadlock detected, 1 on OK, 2 on recursive read
|
| 1755 |
|
| 1756 | static int
|
| 1757 | check_deadlock(struct task_struct *curr, struct held_lock *next,
|
| 1758 | struct lockdep_map *next_instance, int read)
|
| 1759 | {
|
| 1760 | struct held_lock *prev;
|
| 1761 | struct held_lock *nest = NULL;
|
| 1762 | int i;
|
| 1763 |
|
| 1764 | for (i = 0; i < curr->lockdep_depth; i++) {
|
| 1765 | prev = curr->held_locks + i;
|
| 1766 |
|
| 1767 | if (prev->instance == next->nest_lock)
|
| 1768 | nest = prev;
|
| 1769 |
|
| 1770 | if (hlock_class(prev) != hlock_class(next))
|
| 1771 | continue;
|
| 1772 |
|
| 1773 | |
| 1774 | * Allow read-after-read recursion of the same
|
| 1775 | * lock class (i.e. read_lock(lock)+read_lock(lock)):
|
| 1776 |
|
| 1777 | if ((read == 2) && prev->read)
|
| 1778 | return 2;
|
| 1779 |
|
| 1780 | |
| 1781 | * We're holding the nest_lock, which serializes this lock's
|
| 1782 | * nesting behaviour.
|
| 1783 |
|
| 1784 | if (nest)
|
| 1785 | return 2;
|
| 1786 |
|
| 1787 | return print_deadlock_bug(curr, prev, next);
|
| 1788 | }
|
| 1789 | return 1;
|
| 1790 | }
|
| 1791 |
|
| 1792 | |
| 1793 | * There was a chain-cache miss, and we are about to add a new dependency
|
| 1794 | * to a previous lock. We recursively validate the following rules:
|
| 1795 | *
|
| 1796 | * - would the adding of the <prev> -> <next> dependency create a
|
| 1797 | * circular dependency in the graph? [== circular deadlock]
|
| 1798 | *
|
| 1799 | * - does the new prev->next dependency connect any hardirq-safe lock
|
| 1800 | * (in the full backwards-subgraph starting at <prev>) with any
|
| 1801 | * hardirq-unsafe lock (in the full forwards-subgraph starting at
|
| 1802 | * <next>)? [== illegal lock inversion with hardirq contexts]
|
| 1803 | *
|
| 1804 | * - does the new prev->next dependency connect any softirq-safe lock
|
| 1805 | * (in the full backwards-subgraph starting at <prev>) with any
|
| 1806 | * softirq-unsafe lock (in the full forwards-subgraph starting at
|
| 1807 | * <next>)? [== illegal lock inversion with softirq contexts]
|
| 1808 | *
|
| 1809 | * any of these scenarios could lead to a deadlock.
|
| 1810 | *
|
| 1811 | * Then if all the validations pass, we add the forwards and backwards
|
| 1812 | * dependency.
|
| 1813 |
|
| 1814 | static int
|
| 1815 | check_prev_add(struct task_struct *curr, struct held_lock *prev,
|
| 1816 | struct held_lock *next, int distance, int *stack_saved)
|
| 1817 | {
|
| 1818 | struct lock_list *entry;
|
| 1819 | int ret;
|
| 1820 | struct lock_list this;
|
| 1821 | struct lock_list *uninitialized_var(target_entry);
|
| 1822 | |
| 1823 | * Static variable, serialized by the graph_lock().
|
| 1824 | *
|
| 1825 | * We use this static variable to save the stack trace in case
|
| 1826 | * we call into this function multiple times due to encountering
|
| 1827 | * trylocks in the held lock stack.
|
| 1828 |
|
| 1829 | static struct stack_trace trace;
|
| 1830 |
|
| 1831 | |
| 1832 | * Prove that the new <prev> -> <next> dependency would not
|
| 1833 | * create a circular dependency in the graph. (We do this by
|
| 1834 | * forward-recursing into the graph starting at <next>, and
|
| 1835 | * checking whether we can reach <prev>.)
|
| 1836 | *
|
| 1837 | * We are using global variables to control the recursion, to
|
| 1838 | * keep the stackframe size of the recursive functions low:
|
| 1839 |
|
| 1840 | this.class = hlock_class(next);
|
| 1841 | this.parent = NULL;
|
| 1842 | ret = check_noncircular(&this, hlock_class(prev), &target_entry);
|
| 1843 | if (unlikely(!ret))
|
| 1844 | return print_circular_bug(&this, target_entry, next, prev);
|
| 1845 | else if (unlikely(ret < 0))
|
| 1846 | return print_bfs_bug(ret);
|
| 1847 |
|
| 1848 | if (!check_prev_add_irq(curr, prev, next))
|
| 1849 | return 0;
|
| 1850 |
|
| 1851 | |
| 1852 | * For recursive read-locks we do all the dependency checks,
|
| 1853 | * but we dont store read-triggered dependencies (only
|
| 1854 | * write-triggered dependencies). This ensures that only the
|
| 1855 | * write-side dependencies matter, and that if for example a
|
| 1856 | * write-lock never takes any other locks, then the reads are
|
| 1857 | * equivalent to a NOP.
|
| 1858 |
|
| 1859 | if (next->read == 2 || prev->read == 2)
|
| 1860 | return 1;
|
| 1861 | |
| 1862 | * Is the <prev> -> <next> dependency already present?
|
| 1863 | *
|
| 1864 | * (this may occur even though this is a new chain: consider
|
| 1865 | * e.g. the L1 -> L2 -> L3 -> L4 and the L5 -> L1 -> L2 -> L3
|
| 1866 | * chains - the second one will be new, but L1 already has
|
| 1867 | * L2 added to its dependency list, due to the first chain.)
|
| 1868 |
|
| 1869 | list_for_each_entry(entry, &hlock_class(prev)->locks_after, entry) {
|
| 1870 | if (entry->class == hlock_class(next)) {
|
| 1871 | if (distance == 1)
|
| 1872 | entry->distance = 1;
|
| 1873 | return 2;
|
| 1874 | }
|
| 1875 | }
|
| 1876 |
|
| 1877 | if (!*stack_saved) {
|
| 1878 | if (!save_trace(&trace))
|
| 1879 | return 0;
|
| 1880 | *stack_saved = 1;
|
| 1881 | }
|
| 1882 |
|
| 1883 | |
| 1884 | * Ok, all validations passed, add the new lock
|
| 1885 | * to the previous lock's dependency list:
|
| 1886 |
|
| 1887 | ret = add_lock_to_list(hlock_class(next),
|
| 1888 | &hlock_class(prev)->locks_after,
|
| 1889 | next->acquire_ip, distance, &trace);
|
| 1890 |
|
| 1891 | if (!ret)
|
| 1892 | return 0;
|
| 1893 |
|
| 1894 | ret = add_lock_to_list(hlock_class(prev),
|
| 1895 | &hlock_class(next)->locks_before,
|
| 1896 | next->acquire_ip, distance, &trace);
|
| 1897 | if (!ret)
|
| 1898 | return 0;
|
| 1899 |
|
| 1900 | |
| 1901 | * Debugging printouts:
|
| 1902 |
|
| 1903 | if (verbose(hlock_class(prev)) || verbose(hlock_class(next))) {
|
| 1904 |
|
| 1905 | *stack_saved = 0;
|
| 1906 | graph_unlock();
|
| 1907 | printk("\n new dependency: ");
|
| 1908 | print_lock_name(hlock_class(prev));
|
| 1909 | printk(KERN_CONT " => ");
|
| 1910 | print_lock_name(hlock_class(next));
|
| 1911 | printk(KERN_CONT "\n");
|
| 1912 | dump_stack();
|
| 1913 | return graph_lock();
|
| 1914 | }
|
| 1915 | return 1;
|
| 1916 | }
|
| 1917 |
|
| 1918 | |
| 1919 | * Add the dependency to all directly-previous locks that are 'relevant'.
|
| 1920 | * The ones that are relevant are (in increasing distance from curr):
|
| 1921 | * all consecutive trylock entries and the final non-trylock entry - or
|
| 1922 | * the end of this context's lock-chain - whichever comes first.
|
| 1923 |
|
| 1924 | static int
|
| 1925 | check_prevs_add(struct task_struct *curr, struct held_lock *next)
|
| 1926 | {
|
| 1927 | int depth = curr->lockdep_depth;
|
| 1928 | int stack_saved = 0;
|
| 1929 | struct held_lock *hlock;
|
| 1930 |
|
| 1931 | |
| 1932 | * Debugging checks.
|
| 1933 | *
|
| 1934 | * Depth must not be zero for a non-head lock:
|
| 1935 |
|
| 1936 | if (!depth)
|
| 1937 | goto out_bug;
|
| 1938 | |
| 1939 | * At least two relevant locks must exist for this
|
| 1940 | * to be a head:
|
| 1941 |
|
| 1942 | if (curr->held_locks[depth].irq_context !=
|
| 1943 | curr->held_locks[depth-1].irq_context)
|
| 1944 | goto out_bug;
|
| 1945 |
|
| 1946 | for (;;) {
|
| 1947 | int distance = curr->lockdep_depth - depth + 1;
|
| 1948 | hlock = curr->held_locks + depth - 1;
|
| 1949 | |
| 1950 | * Only non-recursive-read entries get new dependencies
|
| 1951 | * added:
|
| 1952 |
|
| 1953 | if (hlock->read != 2 && hlock->check) {
|
| 1954 | if (!check_prev_add(curr, hlock, next,
|
| 1955 | distance, &stack_saved))
|
| 1956 | return 0;
|
| 1957 | |
| 1958 | * Stop after the first non-trylock entry,
|
| 1959 | * as non-trylock entries have added their
|
| 1960 | * own direct dependencies already, so this
|
| 1961 | * lock is connected to them indirectly:
|
| 1962 |
|
| 1963 | if (!hlock->trylock)
|
| 1964 | break;
|
| 1965 | }
|
| 1966 | depth--;
|
| 1967 | |
| 1968 | * End of lock-stack?
|
| 1969 |
|
| 1970 | if (!depth)
|
| 1971 | break;
|
| 1972 | |
| 1973 | * Stop the search if we cross into another context:
|
| 1974 |
|
| 1975 | if (curr->held_locks[depth].irq_context !=
|
| 1976 | curr->held_locks[depth-1].irq_context)
|
| 1977 | break;
|
| 1978 | }
|
| 1979 | return 1;
|
| 1980 | out_bug:
|
| 1981 | if (!debug_locks_off_graph_unlock())
|
| 1982 | return 0;
|
| 1983 |
|
| 1984 | |
| 1985 | * Clearly we all shouldn't be here, but since we made it we
|
| 1986 | * can reliable say we messed up our state. See the above two
|
| 1987 | * gotos for reasons why we could possibly end up here.
|
| 1988 |
|
| 1989 | WARN_ON(1);
|
| 1990 |
|
| 1991 | return 0;
|
| 1992 | }
|
| 1993 |
|
| 1994 | unsigned long nr_lock_chains;
|
| 1995 | struct lock_chain lock_chains[MAX_LOCKDEP_CHAINS];
|
| 1996 | int nr_chain_hlocks;
|
| 1997 | static u16 chain_hlocks[MAX_LOCKDEP_CHAIN_HLOCKS];
|
| 1998 |
|
| 1999 | struct lock_class *lock_chain_get_class(struct lock_chain *chain, int i)
|
| 2000 | {
|
| 2001 | return lock_classes + chain_hlocks[chain->base + i];
|
| 2002 | }
|
| 2003 |
|
| 2004 | |
| 2005 | * Returns the index of the first held_lock of the current chain
|
| 2006 |
|
| 2007 | static inline int get_first_held_lock(struct task_struct *curr,
|
| 2008 | struct held_lock *hlock)
|
| 2009 | {
|
| 2010 | int i;
|
| 2011 | struct held_lock *hlock_curr;
|
| 2012 |
|
| 2013 | for (i = curr->lockdep_depth - 1; i >= 0; i--) {
|
| 2014 | hlock_curr = curr->held_locks + i;
|
| 2015 | if (hlock_curr->irq_context != hlock->irq_context)
|
| 2016 | break;
|
| 2017 |
|
| 2018 | }
|
| 2019 |
|
| 2020 | return ++i;
|
| 2021 | }
|
| 2022 |
|
| 2023 | #ifdef CONFIG_DEBUG_LOCKDEP
|
| 2024 | |
| 2025 | * Returns the next chain_key iteration
|
| 2026 |
|
| 2027 | static u64 print_chain_key_iteration(int class_idx, u64 chain_key)
|
| 2028 | {
|
| 2029 | u64 new_chain_key = iterate_chain_key(chain_key, class_idx);
|
| 2030 |
|
| 2031 | printk(" class_idx:%d -> chain_key:%016Lx",
|
| 2032 | class_idx,
|
| 2033 | (unsigned long long)new_chain_key);
|
| 2034 | return new_chain_key;
|
| 2035 | }
|
| 2036 |
|
| 2037 | static void
|
| 2038 | print_chain_keys_held_locks(struct task_struct *curr, struct held_lock *hlock_next)
|
| 2039 | {
|
| 2040 | struct held_lock *hlock;
|
| 2041 | u64 chain_key = 0;
|
| 2042 | int depth = curr->lockdep_depth;
|
| 2043 | int i;
|
| 2044 |
|
| 2045 | printk("depth: %u\n", depth + 1);
|
| 2046 | for (i = get_first_held_lock(curr, hlock_next); i < depth; i++) {
|
| 2047 | hlock = curr->held_locks + i;
|
| 2048 | chain_key = print_chain_key_iteration(hlock->class_idx, chain_key);
|
| 2049 |
|
| 2050 | print_lock(hlock);
|
| 2051 | }
|
| 2052 |
|
| 2053 | print_chain_key_iteration(hlock_next->class_idx, chain_key);
|
| 2054 | print_lock(hlock_next);
|
| 2055 | }
|
| 2056 |
|
| 2057 | static void print_chain_keys_chain(struct lock_chain *chain)
|
| 2058 | {
|
| 2059 | int i;
|
| 2060 | u64 chain_key = 0;
|
| 2061 | int class_id;
|
| 2062 |
|
| 2063 | printk("depth: %u\n", chain->depth);
|
| 2064 | for (i = 0; i < chain->depth; i++) {
|
| 2065 | class_id = chain_hlocks[chain->base + i];
|
| 2066 | chain_key = print_chain_key_iteration(class_id + 1, chain_key);
|
| 2067 |
|
| 2068 | print_lock_name(lock_classes + class_id);
|
| 2069 | printk("\n");
|
| 2070 | }
|
| 2071 | }
|
| 2072 |
|
| 2073 | static void print_collision(struct task_struct *curr,
|
| 2074 | struct held_lock *hlock_next,
|
| 2075 | struct lock_chain *chain)
|
| 2076 | {
|
| 2077 | pr_warn("\n");
|
| 2078 | pr_warn("============================\n");
|
| 2079 | pr_warn("WARNING: chain_key collision\n");
|
| 2080 | print_kernel_ident();
|
| 2081 | pr_warn("----------------------------\n");
|
| 2082 | pr_warn("%s/%d: ", current->comm, task_pid_nr(current));
|
| 2083 | pr_warn("Hash chain already cached but the contents don't match!\n");
|
| 2084 |
|
| 2085 | pr_warn("Held locks:");
|
| 2086 | print_chain_keys_held_locks(curr, hlock_next);
|
| 2087 |
|
| 2088 | pr_warn("Locks in cached chain:");
|
| 2089 | print_chain_keys_chain(chain);
|
| 2090 |
|
| 2091 | pr_warn("\nstack backtrace:\n");
|
| 2092 | dump_stack();
|
| 2093 | }
|
| 2094 | #endif
|
| 2095 |
|
| 2096 | |
| 2097 | * Checks whether the chain and the current held locks are consistent
|
| 2098 | * in depth and also in content. If they are not it most likely means
|
| 2099 | * that there was a collision during the calculation of the chain_key.
|
| 2100 | * Returns: 0 not passed, 1 passed
|
| 2101 |
|
| 2102 | static int check_no_collision(struct task_struct *curr,
|
| 2103 | struct held_lock *hlock,
|
| 2104 | struct lock_chain *chain)
|
| 2105 | {
|
| 2106 | #ifdef CONFIG_DEBUG_LOCKDEP
|
| 2107 | int i, j, id;
|
| 2108 |
|
| 2109 | i = get_first_held_lock(curr, hlock);
|
| 2110 |
|
| 2111 | if (DEBUG_LOCKS_WARN_ON(chain->depth != curr->lockdep_depth - (i - 1))) {
|
| 2112 | print_collision(curr, hlock, chain);
|
| 2113 | return 0;
|
| 2114 | }
|
| 2115 |
|
| 2116 | for (j = 0; j < chain->depth - 1; j++, i++) {
|
| 2117 | id = curr->held_locks[i].class_idx - 1;
|
| 2118 |
|
| 2119 | if (DEBUG_LOCKS_WARN_ON(chain_hlocks[chain->base + j] != id)) {
|
| 2120 | print_collision(curr, hlock, chain);
|
| 2121 | return 0;
|
| 2122 | }
|
| 2123 | }
|
| 2124 | #endif
|
| 2125 | return 1;
|
| 2126 | }
|
| 2127 |
|
| 2128 | |
| 2129 | * Look up a dependency chain. If the key is not present yet then
|
| 2130 | * add it and return 1 - in this case the new dependency chain is
|
| 2131 | * validated. If the key is already hashed, return 0.
|
| 2132 | * (On return with 1 graph_lock is held.)
|
| 2133 |
|
| 2134 | static inline int lookup_chain_cache(struct task_struct *curr,
|
| 2135 | struct held_lock *hlock,
|
| 2136 | u64 chain_key)
|
| 2137 | {
|
| 2138 | struct lock_class *class = hlock_class(hlock);
|
| 2139 | struct hlist_head *hash_head = chainhashentry(chain_key);
|
| 2140 | struct lock_chain *chain;
|
| 2141 | int i, j;
|
| 2142 |
|
| 2143 | |
| 2144 | * We might need to take the graph lock, ensure we've got IRQs
|
| 2145 | * disabled to make this an IRQ-safe lock.. for recursion reasons
|
| 2146 | * lockdep won't complain about its own locking errors.
|
| 2147 |
|
| 2148 | if (DEBUG_LOCKS_WARN_ON(!irqs_disabled()))
|
| 2149 | return 0;
|
| 2150 | |
| 2151 | * We can walk it lock-free, because entries only get added
|
| 2152 | * to the hash:
|
| 2153 |
|
| 2154 | hlist_for_each_entry_rcu(chain, hash_head, entry) {
|
| 2155 | if (chain->chain_key == chain_key) {
|
| 2156 | cache_hit:
|
| 2157 | debug_atomic_inc(chain_lookup_hits);
|
| 2158 | if (!check_no_collision(curr, hlock, chain))
|
| 2159 | return 0;
|
| 2160 |
|
| 2161 | if (very_verbose(class))
|
| 2162 | printk("\nhash chain already cached, key: "
|
| 2163 | "%016Lx tail class: [%p] %s\n",
|
| 2164 | (unsigned long long)chain_key,
|
| 2165 | class->key, class->name);
|
| 2166 | return 0;
|
| 2167 | }
|
| 2168 | }
|
| 2169 | if (very_verbose(class))
|
| 2170 | printk("\nnew hash chain, key: %016Lx tail class: [%p] %s\n",
|
| 2171 | (unsigned long long)chain_key, class->key, class->name);
|
| 2172 | |
| 2173 | * Allocate a new chain entry from the static array, and add
|
| 2174 | * it to the hash:
|
| 2175 |
|
| 2176 | if (!graph_lock())
|
| 2177 | return 0;
|
| 2178 | |
| 2179 | * We have to walk the chain again locked - to avoid duplicates:
|
| 2180 |
|
| 2181 | hlist_for_each_entry(chain, hash_head, entry) {
|
| 2182 | if (chain->chain_key == chain_key) {
|
| 2183 | graph_unlock();
|
| 2184 | goto cache_hit;
|
| 2185 | }
|
| 2186 | }
|
| 2187 | if (unlikely(nr_lock_chains >= MAX_LOCKDEP_CHAINS)) {
|
| 2188 | if (!debug_locks_off_graph_unlock())
|
| 2189 | return 0;
|
| 2190 |
|
| 2191 | print_lockdep_off("BUG: MAX_LOCKDEP_CHAINS too low!");
|
| 2192 | dump_stack();
|
| 2193 | return 0;
|
| 2194 | }
|
| 2195 | chain = lock_chains + nr_lock_chains++;
|
| 2196 | chain->chain_key = chain_key;
|
| 2197 | chain->irq_context = hlock->irq_context;
|
| 2198 | i = get_first_held_lock(curr, hlock);
|
| 2199 | chain->depth = curr->lockdep_depth + 1 - i;
|
| 2200 |
|
| 2201 | BUILD_BUG_ON((1UL << 24) <= ARRAY_SIZE(chain_hlocks));
|
| 2202 | BUILD_BUG_ON((1UL << 6) <= ARRAY_SIZE(curr->held_locks));
|
| 2203 | BUILD_BUG_ON((1UL << 8*sizeof(chain_hlocks[0])) <= ARRAY_SIZE(lock_classes));
|
| 2204 |
|
| 2205 | if (likely(nr_chain_hlocks + chain->depth <= MAX_LOCKDEP_CHAIN_HLOCKS)) {
|
| 2206 | chain->base = nr_chain_hlocks;
|
| 2207 | for (j = 0; j < chain->depth - 1; j++, i++) {
|
| 2208 | int lock_id = curr->held_locks[i].class_idx - 1;
|
| 2209 | chain_hlocks[chain->base + j] = lock_id;
|
| 2210 | }
|
| 2211 | chain_hlocks[chain->base + j] = class - lock_classes;
|
| 2212 | }
|
| 2213 |
|
| 2214 | if (nr_chain_hlocks < MAX_LOCKDEP_CHAIN_HLOCKS)
|
| 2215 | nr_chain_hlocks += chain->depth;
|
| 2216 |
|
| 2217 | #ifdef CONFIG_DEBUG_LOCKDEP
|
| 2218 | |
| 2219 | * Important for check_no_collision().
|
| 2220 |
|
| 2221 | if (unlikely(nr_chain_hlocks > MAX_LOCKDEP_CHAIN_HLOCKS)) {
|
| 2222 | if (!debug_locks_off_graph_unlock())
|
| 2223 | return 0;
|
| 2224 |
|
| 2225 | print_lockdep_off("BUG: MAX_LOCKDEP_CHAIN_HLOCKS too low!");
|
| 2226 | dump_stack();
|
| 2227 | return 0;
|
| 2228 | }
|
| 2229 | #endif
|
| 2230 |
|
| 2231 | hlist_add_head_rcu(&chain->entry, hash_head);
|
| 2232 | debug_atomic_inc(chain_lookup_misses);
|
| 2233 | inc_chains();
|
| 2234 |
|
| 2235 | return 1;
|
| 2236 | }
|
| 2237 |
|
| 2238 | static int validate_chain(struct task_struct *curr, struct lockdep_map *lock,
|
| 2239 | struct held_lock *hlock, int chain_head, u64 chain_key)
|
| 2240 | {
|
| 2241 | |
| 2242 | * Trylock needs to maintain the stack of held locks, but it
|
| 2243 | * does not add new dependencies, because trylock can be done
|
| 2244 | * in any order.
|
| 2245 | *
|
| 2246 | * We look up the chain_key and do the O(N^2) check and update of
|
| 2247 | * the dependencies only if this is a new dependency chain.
|
| 2248 | * (If lookup_chain_cache() returns with 1 it acquires
|
| 2249 | * graph_lock for us)
|
| 2250 |
|
| 2251 | if (!hlock->trylock && hlock->check &&
|
| 2252 | lookup_chain_cache(curr, hlock, chain_key)) {
|
| 2253 | |
| 2254 | * Check whether last held lock:
|
| 2255 | *
|
| 2256 | * - is irq-safe, if this lock is irq-unsafe
|
| 2257 | * - is softirq-safe, if this lock is hardirq-unsafe
|
| 2258 | *
|
| 2259 | * And check whether the new lock's dependency graph
|
| 2260 | * could lead back to the previous lock.
|
| 2261 | *
|
| 2262 | * any of these scenarios could lead to a deadlock. If
|
| 2263 | * All validations
|
| 2264 |
|
| 2265 | int ret = check_deadlock(curr, hlock, lock, hlock->read);
|
| 2266 |
|
| 2267 | if (!ret)
|
| 2268 | return 0;
|
| 2269 | |
| 2270 | * Mark recursive read, as we jump over it when
|
| 2271 | * building dependencies (just like we jump over
|
| 2272 | * trylock entries):
|
| 2273 |
|
| 2274 | if (ret == 2)
|
| 2275 | hlock->read = 2;
|
| 2276 | |
| 2277 | * Add dependency only if this lock is not the head
|
| 2278 | * of the chain, and if it's not a secondary read-lock:
|
| 2279 |
|
| 2280 | if (!chain_head && ret != 2)
|
| 2281 | if (!check_prevs_add(curr, hlock))
|
| 2282 | return 0;
|
| 2283 | graph_unlock();
|
| 2284 | } else
|
| 2285 |
|
| 2286 | if (unlikely(!debug_locks))
|
| 2287 | return 0;
|
| 2288 |
|
| 2289 | return 1;
|
| 2290 | }
|
| 2291 | #else
|
| 2292 | static inline int validate_chain(struct task_struct *curr,
|
| 2293 | struct lockdep_map *lock, struct held_lock *hlock,
|
| 2294 | int chain_head, u64 chain_key)
|
| 2295 | {
|
| 2296 | return 1;
|
| 2297 | }
|
| 2298 | #endif
|
| 2299 |
|
| 2300 | |
| 2301 | * We are building curr_chain_key incrementally, so double-check
|
| 2302 | * it from scratch, to make sure that it's done correctly:
|
| 2303 |
|
| 2304 | static void check_chain_key(struct task_struct *curr)
|
| 2305 | {
|
| 2306 | #ifdef CONFIG_DEBUG_LOCKDEP
|
| 2307 | struct held_lock *hlock, *prev_hlock = NULL;
|
| 2308 | unsigned int i;
|
| 2309 | u64 chain_key = 0;
|
| 2310 |
|
| 2311 | for (i = 0; i < curr->lockdep_depth; i++) {
|
| 2312 | hlock = curr->held_locks + i;
|
| 2313 | if (chain_key != hlock->prev_chain_key) {
|
| 2314 | debug_locks_off();
|
| 2315 | |
| 2316 | * We got mighty confused, our chain keys don't match
|
| 2317 | * with what we expect, someone trample on our task state?
|
| 2318 |
|
| 2319 | WARN(1, "hm#1, depth: %u [%u], %016Lx != %016Lx\n",
|
| 2320 | curr->lockdep_depth, i,
|
| 2321 | (unsigned long long)chain_key,
|
| 2322 | (unsigned long long)hlock->prev_chain_key);
|
| 2323 | return;
|
| 2324 | }
|
| 2325 | |
| 2326 | * Whoops ran out of static storage again?
|
| 2327 |
|
| 2328 | if (DEBUG_LOCKS_WARN_ON(hlock->class_idx > MAX_LOCKDEP_KEYS))
|
| 2329 | return;
|
| 2330 |
|
| 2331 | if (prev_hlock && (prev_hlock->irq_context !=
|
| 2332 | hlock->irq_context))
|
| 2333 | chain_key = 0;
|
| 2334 | chain_key = iterate_chain_key(chain_key, hlock->class_idx);
|
| 2335 | prev_hlock = hlock;
|
| 2336 | }
|
| 2337 | if (chain_key != curr->curr_chain_key) {
|
| 2338 | debug_locks_off();
|
| 2339 | |
| 2340 | * More smoking hash instead of calculating it, damn see these
|
| 2341 | * numbers float.. I bet that a pink elephant stepped on my memory.
|
| 2342 |
|
| 2343 | WARN(1, "hm#2, depth: %u [%u], %016Lx != %016Lx\n",
|
| 2344 | curr->lockdep_depth, i,
|
| 2345 | (unsigned long long)chain_key,
|
| 2346 | (unsigned long long)curr->curr_chain_key);
|
| 2347 | }
|
| 2348 | #endif
|
| 2349 | }
|
| 2350 |
|
| 2351 | static void
|
| 2352 | print_usage_bug_scenario(struct held_lock *lock)
|
| 2353 | {
|
| 2354 | struct lock_class *class = hlock_class(lock);
|
| 2355 |
|
| 2356 | printk(" Possible unsafe locking scenario:\n\n");
|
| 2357 | printk(" CPU0\n");
|
| 2358 | printk(" ----\n");
|
| 2359 | printk(" lock(");
|
| 2360 | __print_lock_name(class);
|
| 2361 | printk(KERN_CONT ");\n");
|
| 2362 | printk(" <Interrupt>\n");
|
| 2363 | printk(" lock(");
|
| 2364 | __print_lock_name(class);
|
| 2365 | printk(KERN_CONT ");\n");
|
| 2366 | printk("\n *** DEADLOCK ***\n\n");
|
| 2367 | }
|
| 2368 |
|
| 2369 | static int
|
| 2370 | print_usage_bug(struct task_struct *curr, struct held_lock *this,
|
| 2371 | enum lock_usage_bit prev_bit, enum lock_usage_bit new_bit)
|
| 2372 | {
|
| 2373 | if (!debug_locks_off_graph_unlock() || debug_locks_silent)
|
| 2374 | return 0;
|
| 2375 |
|
| 2376 | pr_warn("\n");
|
| 2377 | pr_warn("================================\n");
|
| 2378 | pr_warn("WARNING: inconsistent lock state\n");
|
| 2379 | print_kernel_ident();
|
| 2380 | pr_warn("--------------------------------\n");
|
| 2381 |
|
| 2382 | pr_warn("inconsistent {%s} -> {%s} usage.\n",
|
| 2383 | usage_str[prev_bit], usage_str[new_bit]);
|
| 2384 |
|
| 2385 | pr_warn("%s/%d [HC%u[%lu]:SC%u[%lu]:HE%u:SE%u] takes:\n",
|
| 2386 | curr->comm, task_pid_nr(curr),
|
| 2387 | trace_hardirq_context(curr), hardirq_count() >> HARDIRQ_SHIFT,
|
| 2388 | trace_softirq_context(curr), softirq_count() >> SOFTIRQ_SHIFT,
|
| 2389 | trace_hardirqs_enabled(curr),
|
| 2390 | trace_softirqs_enabled(curr));
|
| 2391 | print_lock(this);
|
| 2392 |
|
| 2393 | pr_warn("{%s} state was registered at:\n", usage_str[prev_bit]);
|
| 2394 | print_stack_trace(hlock_class(this)->usage_traces + prev_bit, 1);
|
| 2395 |
|
| 2396 | print_irqtrace_events(curr);
|
| 2397 | pr_warn("\nother info that might help us debug this:\n");
|
| 2398 | print_usage_bug_scenario(this);
|
| 2399 |
|
| 2400 | lockdep_print_held_locks(curr);
|
| 2401 |
|
| 2402 | pr_warn("\nstack backtrace:\n");
|
| 2403 | dump_stack();
|
| 2404 |
|
| 2405 | return 0;
|
| 2406 | }
|
| 2407 |
|
| 2408 | |
| 2409 | * Print out an error if an invalid bit is set:
|
| 2410 |
|
| 2411 | static inline int
|
| 2412 | valid_state(struct task_struct *curr, struct held_lock *this,
|
| 2413 | enum lock_usage_bit new_bit, enum lock_usage_bit bad_bit)
|
| 2414 | {
|
| 2415 | if (unlikely(hlock_class(this)->usage_mask & (1 << bad_bit)))
|
| 2416 | return print_usage_bug(curr, this, bad_bit, new_bit);
|
| 2417 | return 1;
|
| 2418 | }
|
| 2419 |
|
| 2420 | static int mark_lock(struct task_struct *curr, struct held_lock *this,
|
| 2421 | enum lock_usage_bit new_bit);
|
| 2422 |
|
| 2423 | #if defined(CONFIG_TRACE_IRQFLAGS) && defined(CONFIG_PROVE_LOCKING)
|
| 2424 |
|
| 2425 | |
| 2426 | * print irq inversion bug:
|
| 2427 |
|
| 2428 | static int
|
| 2429 | print_irq_inversion_bug(struct task_struct *curr,
|
| 2430 | struct lock_list *root, struct lock_list *other,
|
| 2431 | struct held_lock *this, int forwards,
|
| 2432 | const char *irqclass)
|
| 2433 | {
|
| 2434 | struct lock_list *entry = other;
|
| 2435 | struct lock_list *middle = NULL;
|
| 2436 | int depth;
|
| 2437 |
|
| 2438 | if (!debug_locks_off_graph_unlock() || debug_locks_silent)
|
| 2439 | return 0;
|
| 2440 |
|
| 2441 | pr_warn("\n");
|
| 2442 | pr_warn("========================================================\n");
|
| 2443 | pr_warn("WARNING: possible irq lock inversion dependency detected\n");
|
| 2444 | print_kernel_ident();
|
| 2445 | pr_warn("--------------------------------------------------------\n");
|
| 2446 | pr_warn("%s/%d just changed the state of lock:\n",
|
| 2447 | curr->comm, task_pid_nr(curr));
|
| 2448 | print_lock(this);
|
| 2449 | if (forwards)
|
| 2450 | pr_warn("but this lock took another, %s-unsafe lock in the past:\n", irqclass);
|
| 2451 | else
|
| 2452 | pr_warn("but this lock was taken by another, %s-safe lock in the past:\n", irqclass);
|
| 2453 | print_lock_name(other->class);
|
| 2454 | pr_warn("\n\nand interrupts could create inverse lock ordering between them.\n\n");
|
| 2455 |
|
| 2456 | pr_warn("\nother info that might help us debug this:\n");
|
| 2457 |
|
| 2458 |
|
| 2459 | depth = get_lock_depth(other);
|
| 2460 | do {
|
| 2461 | if (depth == 0 && (entry != root)) {
|
| 2462 | pr_warn("lockdep:%s bad path found in chain graph\n", __func__);
|
| 2463 | break;
|
| 2464 | }
|
| 2465 | middle = entry;
|
| 2466 | entry = get_lock_parent(entry);
|
| 2467 | depth--;
|
| 2468 | } while (entry && entry != root && (depth >= 0));
|
| 2469 | if (forwards)
|
| 2470 | print_irq_lock_scenario(root, other,
|
| 2471 | middle ? middle->class : root->class, other->class);
|
| 2472 | else
|
| 2473 | print_irq_lock_scenario(other, root,
|
| 2474 | middle ? middle->class : other->class, root->class);
|
| 2475 |
|
| 2476 | lockdep_print_held_locks(curr);
|
| 2477 |
|
| 2478 | pr_warn("\nthe shortest dependencies between 2nd lock and 1st lock:\n");
|
| 2479 | if (!save_trace(&root->trace))
|
| 2480 | return 0;
|
| 2481 | print_shortest_lock_dependencies(other, root);
|
| 2482 |
|
| 2483 | pr_warn("\nstack backtrace:\n");
|
| 2484 | dump_stack();
|
| 2485 |
|
| 2486 | return 0;
|
| 2487 | }
|
| 2488 |
|
| 2489 | |
| 2490 | * Prove that in the forwards-direction subgraph starting at <this>
|
| 2491 | * there is no lock matching <mask>:
|
| 2492 |
|
| 2493 | static int
|
| 2494 | check_usage_forwards(struct task_struct *curr, struct held_lock *this,
|
| 2495 | enum lock_usage_bit bit, const char *irqclass)
|
| 2496 | {
|
| 2497 | int ret;
|
| 2498 | struct lock_list root;
|
| 2499 | struct lock_list *uninitialized_var(target_entry);
|
| 2500 |
|
| 2501 | root.parent = NULL;
|
| 2502 | root.class = hlock_class(this);
|
| 2503 | ret = find_usage_forwards(&root, bit, &target_entry);
|
| 2504 | if (ret < 0)
|
| 2505 | return print_bfs_bug(ret);
|
| 2506 | if (ret == 1)
|
| 2507 | return ret;
|
| 2508 |
|
| 2509 | return print_irq_inversion_bug(curr, &root, target_entry,
|
| 2510 | this, 1, irqclass);
|
| 2511 | }
|
| 2512 |
|
| 2513 | |
| 2514 | * Prove that in the backwards-direction subgraph starting at <this>
|
| 2515 | * there is no lock matching <mask>:
|
| 2516 |
|
| 2517 | static int
|
| 2518 | check_usage_backwards(struct task_struct *curr, struct held_lock *this,
|
| 2519 | enum lock_usage_bit bit, const char *irqclass)
|
| 2520 | {
|
| 2521 | int ret;
|
| 2522 | struct lock_list root;
|
| 2523 | struct lock_list *uninitialized_var(target_entry);
|
| 2524 |
|
| 2525 | root.parent = NULL;
|
| 2526 | root.class = hlock_class(this);
|
| 2527 | ret = find_usage_backwards(&root, bit, &target_entry);
|
| 2528 | if (ret < 0)
|
| 2529 | return print_bfs_bug(ret);
|
| 2530 | if (ret == 1)
|
| 2531 | return ret;
|
| 2532 |
|
| 2533 | return print_irq_inversion_bug(curr, &root, target_entry,
|
| 2534 | this, 0, irqclass);
|
| 2535 | }
|
| 2536 |
|
| 2537 | void print_irqtrace_events(struct task_struct *curr)
|
| 2538 | {
|
| 2539 | printk("irq event stamp: %u\n", curr->irq_events);
|
| 2540 | printk("hardirqs last enabled at (%u): [<%p>] %pS\n",
|
| 2541 | curr->hardirq_enable_event, (void *)curr->hardirq_enable_ip,
|
| 2542 | (void *)curr->hardirq_enable_ip);
|
| 2543 | printk("hardirqs last disabled at (%u): [<%p>] %pS\n",
|
| 2544 | curr->hardirq_disable_event, (void *)curr->hardirq_disable_ip,
|
| 2545 | (void *)curr->hardirq_disable_ip);
|
| 2546 | printk("softirqs last enabled at (%u): [<%p>] %pS\n",
|
| 2547 | curr->softirq_enable_event, (void *)curr->softirq_enable_ip,
|
| 2548 | (void *)curr->softirq_enable_ip);
|
| 2549 | printk("softirqs last disabled at (%u): [<%p>] %pS\n",
|
| 2550 | curr->softirq_disable_event, (void *)curr->softirq_disable_ip,
|
| 2551 | (void *)curr->softirq_disable_ip);
|
| 2552 | }
|
| 2553 |
|
| 2554 | static int HARDIRQ_verbose(struct lock_class *class)
|
| 2555 | {
|
| 2556 | #if HARDIRQ_VERBOSE
|
| 2557 | return class_filter(class);
|
| 2558 | #endif
|
| 2559 | return 0;
|
| 2560 | }
|
| 2561 |
|
| 2562 | static int SOFTIRQ_verbose(struct lock_class *class)
|
| 2563 | {
|
| 2564 | #if SOFTIRQ_VERBOSE
|
| 2565 | return class_filter(class);
|
| 2566 | #endif
|
| 2567 | return 0;
|
| 2568 | }
|
| 2569 |
|
| 2570 | static int RECLAIM_FS_verbose(struct lock_class *class)
|
| 2571 | {
|
| 2572 | #if RECLAIM_VERBOSE
|
| 2573 | return class_filter(class);
|
| 2574 | #endif
|
| 2575 | return 0;
|
| 2576 | }
|
| 2577 |
|
| 2578 | #define STRICT_READ_CHECKS 1
|
| 2579 |
|
| 2580 | static int (*state_verbose_f[])(struct lock_class *class) = {
|
| 2581 | #define LOCKDEP_STATE(__STATE) \
|
| 2582 | __STATE##_verbose,
|
| 2583 | #include "lockdep_states.h"
|
| 2584 | #undef LOCKDEP_STATE
|
| 2585 | };
|
| 2586 |
|
| 2587 | static inline int state_verbose(enum lock_usage_bit bit,
|
| 2588 | struct lock_class *class)
|
| 2589 | {
|
| 2590 | return state_verbose_f[bit >> 2](class);
|
| 2591 | }
|
| 2592 |
|
| 2593 | typedef int (*check_usage_f)(struct task_struct *, struct held_lock *,
|
| 2594 | enum lock_usage_bit bit, const char *name);
|
| 2595 |
|
| 2596 | static int
|
| 2597 | mark_lock_irq(struct task_struct *curr, struct held_lock *this,
|
| 2598 | enum lock_usage_bit new_bit)
|
| 2599 | {
|
| 2600 | int excl_bit = exclusive_bit(new_bit);
|
| 2601 | int read = new_bit & 1;
|
| 2602 | int dir = new_bit & 2;
|
| 2603 |
|
| 2604 | |
| 2605 | * mark USED_IN has to look forwards -- to ensure no dependency
|
| 2606 | * has ENABLED state, which would allow recursion deadlocks.
|
| 2607 | *
|
| 2608 | * mark ENABLED has to look backwards -- to ensure no dependee
|
| 2609 | * has USED_IN state, which, again, would allow recursion deadlocks.
|
| 2610 |
|
| 2611 | check_usage_f usage = dir ?
|
| 2612 | check_usage_backwards : check_usage_forwards;
|
| 2613 |
|
| 2614 | |
| 2615 | * Validate that this particular lock does not have conflicting
|
| 2616 | * usage states.
|
| 2617 |
|
| 2618 | if (!valid_state(curr, this, new_bit, excl_bit))
|
| 2619 | return 0;
|
| 2620 |
|
| 2621 | |
| 2622 | * Validate that the lock dependencies don't have conflicting usage
|
| 2623 | * states.
|
| 2624 |
|
| 2625 | if ((!read || !dir || STRICT_READ_CHECKS) &&
|
| 2626 | !usage(curr, this, excl_bit, state_name(new_bit & ~1)))
|
| 2627 | return 0;
|
| 2628 |
|
| 2629 | |
| 2630 | * Check for read in write conflicts
|
| 2631 |
|
| 2632 | if (!read) {
|
| 2633 | if (!valid_state(curr, this, new_bit, excl_bit + 1))
|
| 2634 | return 0;
|
| 2635 |
|
| 2636 | if (STRICT_READ_CHECKS &&
|
| 2637 | !usage(curr, this, excl_bit + 1,
|
| 2638 | state_name(new_bit + 1)))
|
| 2639 | return 0;
|
| 2640 | }
|
| 2641 |
|
| 2642 | if (state_verbose(new_bit, hlock_class(this)))
|
| 2643 | return 2;
|
| 2644 |
|
| 2645 | return 1;
|
| 2646 | }
|
| 2647 |
|
| 2648 | enum mark_type {
|
| 2649 | #define LOCKDEP_STATE(__STATE) __STATE,
|
| 2650 | #include "lockdep_states.h"
|
| 2651 | #undef LOCKDEP_STATE
|
| 2652 | };
|
| 2653 |
|
| 2654 | |
| 2655 | * Mark all held locks with a usage bit:
|
| 2656 |
|
| 2657 | static int
|
| 2658 | mark_held_locks(struct task_struct *curr, enum mark_type mark)
|
| 2659 | {
|
| 2660 | enum lock_usage_bit usage_bit;
|
| 2661 | struct held_lock *hlock;
|
| 2662 | int i;
|
| 2663 |
|
| 2664 | for (i = 0; i < curr->lockdep_depth; i++) {
|
| 2665 | hlock = curr->held_locks + i;
|
| 2666 |
|
| 2667 | usage_bit = 2 + (mark << 2);
|
| 2668 | if (hlock->read)
|
| 2669 | usage_bit += 1;
|
| 2670 |
|
| 2671 | BUG_ON(usage_bit >= LOCK_USAGE_STATES);
|
| 2672 |
|
| 2673 | if (!hlock->check)
|
| 2674 | continue;
|
| 2675 |
|
| 2676 | if (!mark_lock(curr, hlock, usage_bit))
|
| 2677 | return 0;
|
| 2678 | }
|
| 2679 |
|
| 2680 | return 1;
|
| 2681 | }
|
| 2682 |
|
| 2683 | |
| 2684 | * Hardirqs will be enabled:
|
| 2685 |
|
| 2686 | static void __trace_hardirqs_on_caller(unsigned long ip)
|
| 2687 | {
|
| 2688 | struct task_struct *curr = current;
|
| 2689 |
|
| 2690 |
|
| 2691 | curr->hardirqs_enabled = 1;
|
| 2692 |
|
| 2693 | |
| 2694 | * We are going to turn hardirqs on, so set the
|
| 2695 | * usage bit for all held locks:
|
| 2696 |
|
| 2697 | if (!mark_held_locks(curr, HARDIRQ))
|
| 2698 | return;
|
| 2699 | |
| 2700 | * If we have softirqs enabled, then set the usage
|
| 2701 | * bit for all held locks. (disabled hardirqs prevented
|
| 2702 | * this bit from being set before)
|
| 2703 |
|
| 2704 | if (curr->softirqs_enabled)
|
| 2705 | if (!mark_held_locks(curr, SOFTIRQ))
|
| 2706 | return;
|
| 2707 |
|
| 2708 | curr->hardirq_enable_ip = ip;
|
| 2709 | curr->hardirq_enable_event = ++curr->irq_events;
|
| 2710 | debug_atomic_inc(hardirqs_on_events);
|
| 2711 | }
|
| 2712 |
|
| 2713 | __visible void trace_hardirqs_on_caller(unsigned long ip)
|
| 2714 | {
|
| 2715 | time_hardirqs_on(CALLER_ADDR0, ip);
|
| 2716 |
|
| 2717 | if (unlikely(!debug_locks || current->lockdep_recursion))
|
| 2718 | return;
|
| 2719 |
|
| 2720 | if (unlikely(current->hardirqs_enabled)) {
|
| 2721 | |
| 2722 | * Neither irq nor preemption are disabled here
|
| 2723 | * so this is racy by nature but losing one hit
|
| 2724 | * in a stat is not a big deal.
|
| 2725 |
|
| 2726 | __debug_atomic_inc(redundant_hardirqs_on);
|
| 2727 | return;
|
| 2728 | }
|
| 2729 |
|
| 2730 | |
| 2731 | * We're enabling irqs and according to our state above irqs weren't
|
| 2732 | * already enabled, yet we find the hardware thinks they are in fact
|
| 2733 | * enabled.. someone messed up their IRQ state tracing.
|
| 2734 |
|
| 2735 | if (DEBUG_LOCKS_WARN_ON(!irqs_disabled()))
|
| 2736 | return;
|
| 2737 |
|
| 2738 | |
| 2739 | * See the fine text that goes along with this variable definition.
|
| 2740 |
|
| 2741 | if (DEBUG_LOCKS_WARN_ON(unlikely(early_boot_irqs_disabled)))
|
| 2742 | return;
|
| 2743 |
|
| 2744 | |
| 2745 | * Can't allow enabling interrupts while in an interrupt handler,
|
| 2746 | * that's general bad form and such. Recursion, limited stack etc..
|
| 2747 |
|
| 2748 | if (DEBUG_LOCKS_WARN_ON(current->hardirq_context))
|
| 2749 | return;
|
| 2750 |
|
| 2751 | current->lockdep_recursion = 1;
|
| 2752 | __trace_hardirqs_on_caller(ip);
|
| 2753 | current->lockdep_recursion = 0;
|
| 2754 | }
|
| 2755 | EXPORT_SYMBOL(trace_hardirqs_on_caller);
|
| 2756 |
|
| 2757 | void trace_hardirqs_on(void)
|
| 2758 | {
|
| 2759 | trace_hardirqs_on_caller(CALLER_ADDR0);
|
| 2760 | }
|
| 2761 | EXPORT_SYMBOL(trace_hardirqs_on);
|
| 2762 |
|
| 2763 | |
| 2764 | * Hardirqs were disabled:
|
| 2765 |
|
| 2766 | __visible void trace_hardirqs_off_caller(unsigned long ip)
|
| 2767 | {
|
| 2768 | struct task_struct *curr = current;
|
| 2769 |
|
| 2770 | time_hardirqs_off(CALLER_ADDR0, ip);
|
| 2771 |
|
| 2772 | if (unlikely(!debug_locks || current->lockdep_recursion))
|
| 2773 | return;
|
| 2774 |
|
| 2775 | |
| 2776 | * So we're supposed to get called after you mask local IRQs, but for
|
| 2777 | * some reason the hardware doesn't quite think you did a proper job.
|
| 2778 |
|
| 2779 | if (DEBUG_LOCKS_WARN_ON(!irqs_disabled()))
|
| 2780 | return;
|
| 2781 |
|
| 2782 | if (curr->hardirqs_enabled) {
|
| 2783 | |
| 2784 | * We have done an ON -> OFF transition:
|
| 2785 |
|
| 2786 | curr->hardirqs_enabled = 0;
|
| 2787 | curr->hardirq_disable_ip = ip;
|
| 2788 | curr->hardirq_disable_event = ++curr->irq_events;
|
| 2789 | debug_atomic_inc(hardirqs_off_events);
|
| 2790 | } else
|
| 2791 | debug_atomic_inc(redundant_hardirqs_off);
|
| 2792 | }
|
| 2793 | EXPORT_SYMBOL(trace_hardirqs_off_caller);
|
| 2794 |
|
| 2795 | void trace_hardirqs_off(void)
|
| 2796 | {
|
| 2797 | trace_hardirqs_off_caller(CALLER_ADDR0);
|
| 2798 | }
|
| 2799 | EXPORT_SYMBOL(trace_hardirqs_off);
|
| 2800 |
|
| 2801 | |
| 2802 | * Softirqs will be enabled:
|
| 2803 |
|
| 2804 | void trace_softirqs_on(unsigned long ip)
|
| 2805 | {
|
| 2806 | struct task_struct *curr = current;
|
| 2807 |
|
| 2808 | if (unlikely(!debug_locks || current->lockdep_recursion))
|
| 2809 | return;
|
| 2810 |
|
| 2811 | |
| 2812 | * We fancy IRQs being disabled here, see softirq.c, avoids
|
| 2813 | * funny state and nesting things.
|
| 2814 |
|
| 2815 | if (DEBUG_LOCKS_WARN_ON(!irqs_disabled()))
|
| 2816 | return;
|
| 2817 |
|
| 2818 | if (curr->softirqs_enabled) {
|
| 2819 | debug_atomic_inc(redundant_softirqs_on);
|
| 2820 | return;
|
| 2821 | }
|
| 2822 |
|
| 2823 | current->lockdep_recursion = 1;
|
| 2824 | |
| 2825 | * We'll do an OFF -> ON transition:
|
| 2826 |
|
| 2827 | curr->softirqs_enabled = 1;
|
| 2828 | curr->softirq_enable_ip = ip;
|
| 2829 | curr->softirq_enable_event = ++curr->irq_events;
|
| 2830 | debug_atomic_inc(softirqs_on_events);
|
| 2831 | |
| 2832 | * We are going to turn softirqs on, so set the
|
| 2833 | * usage bit for all held locks, if hardirqs are
|
| 2834 | * enabled too:
|
| 2835 |
|
| 2836 | if (curr->hardirqs_enabled)
|
| 2837 | mark_held_locks(curr, SOFTIRQ);
|
| 2838 | current->lockdep_recursion = 0;
|
| 2839 | }
|
| 2840 |
|
| 2841 | |
| 2842 | * Softirqs were disabled:
|
| 2843 |
|
| 2844 | void trace_softirqs_off(unsigned long ip)
|
| 2845 | {
|
| 2846 | struct task_struct *curr = current;
|
| 2847 |
|
| 2848 | if (unlikely(!debug_locks || current->lockdep_recursion))
|
| 2849 | return;
|
| 2850 |
|
| 2851 | |
| 2852 | * We fancy IRQs being disabled here, see softirq.c
|
| 2853 |
|
| 2854 | if (DEBUG_LOCKS_WARN_ON(!irqs_disabled()))
|
| 2855 | return;
|
| 2856 |
|
| 2857 | if (curr->softirqs_enabled) {
|
| 2858 | |
| 2859 | * We have done an ON -> OFF transition:
|
| 2860 |
|
| 2861 | curr->softirqs_enabled = 0;
|
| 2862 | curr->softirq_disable_ip = ip;
|
| 2863 | curr->softirq_disable_event = ++curr->irq_events;
|
| 2864 | debug_atomic_inc(softirqs_off_events);
|
| 2865 | |
| 2866 | * Whoops, we wanted softirqs off, so why aren't they?
|
| 2867 |
|
| 2868 | DEBUG_LOCKS_WARN_ON(!softirq_count());
|
| 2869 | } else
|
| 2870 | debug_atomic_inc(redundant_softirqs_off);
|
| 2871 | }
|
| 2872 |
|
| 2873 | static void __lockdep_trace_alloc(gfp_t gfp_mask, unsigned long flags)
|
| 2874 | {
|
| 2875 | struct task_struct *curr = current;
|
| 2876 |
|
| 2877 | if (unlikely(!debug_locks))
|
| 2878 | return;
|
| 2879 |
|
| 2880 | gfp_mask = current_gfp_context(gfp_mask);
|
| 2881 |
|
| 2882 |
|
| 2883 | if (!(gfp_mask & __GFP_DIRECT_RECLAIM))
|
| 2884 | return;
|
| 2885 |
|
| 2886 |
|
| 2887 | if ((curr->flags & PF_MEMALLOC) && !(gfp_mask & __GFP_NOMEMALLOC))
|
| 2888 | return;
|
| 2889 |
|
| 2890 |
|
| 2891 | if (!(gfp_mask & __GFP_FS) || (curr->flags & PF_MEMALLOC_NOFS))
|
| 2892 | return;
|
| 2893 |
|
| 2894 | |
| 2895 | * Oi! Can't be having __GFP_FS allocations with IRQs disabled.
|
| 2896 |
|
| 2897 | if (DEBUG_LOCKS_WARN_ON(irqs_disabled_flags(flags)))
|
| 2898 | return;
|
| 2899 |
|
| 2900 |
|
| 2901 | if (gfp_mask & __GFP_NOLOCKDEP)
|
| 2902 | return;
|
| 2903 |
|
| 2904 | mark_held_locks(curr, RECLAIM_FS);
|
| 2905 | }
|
| 2906 |
|
| 2907 | static void check_flags(unsigned long flags);
|
| 2908 |
|
| 2909 | void lockdep_trace_alloc(gfp_t gfp_mask)
|
| 2910 | {
|
| 2911 | unsigned long flags;
|
| 2912 |
|
| 2913 | if (unlikely(current->lockdep_recursion))
|
| 2914 | return;
|
| 2915 |
|
| 2916 | raw_local_irq_save(flags);
|
| 2917 | check_flags(flags);
|
| 2918 | current->lockdep_recursion = 1;
|
| 2919 | __lockdep_trace_alloc(gfp_mask, flags);
|
| 2920 | current->lockdep_recursion = 0;
|
| 2921 | raw_local_irq_restore(flags);
|
| 2922 | }
|
| 2923 |
|
| 2924 | static int mark_irqflags(struct task_struct *curr, struct held_lock *hlock)
|
| 2925 | {
|
| 2926 | |
| 2927 | * If non-trylock use in a hardirq or softirq context, then
|
| 2928 | * mark the lock as used in these contexts:
|
| 2929 |
|
| 2930 | if (!hlock->trylock) {
|
| 2931 | if (hlock->read) {
|
| 2932 | if (curr->hardirq_context)
|
| 2933 | if (!mark_lock(curr, hlock,
|
| 2934 | LOCK_USED_IN_HARDIRQ_READ))
|
| 2935 | return 0;
|
| 2936 | if (curr->softirq_context)
|
| 2937 | if (!mark_lock(curr, hlock,
|
| 2938 | LOCK_USED_IN_SOFTIRQ_READ))
|
| 2939 | return 0;
|
| 2940 | } else {
|
| 2941 | if (curr->hardirq_context)
|
| 2942 | if (!mark_lock(curr, hlock, LOCK_USED_IN_HARDIRQ))
|
| 2943 | return 0;
|
| 2944 | if (curr->softirq_context)
|
| 2945 | if (!mark_lock(curr, hlock, LOCK_USED_IN_SOFTIRQ))
|
| 2946 | return 0;
|
| 2947 | }
|
| 2948 | }
|
| 2949 | if (!hlock->hardirqs_off) {
|
| 2950 | if (hlock->read) {
|
| 2951 | if (!mark_lock(curr, hlock,
|
| 2952 | LOCK_ENABLED_HARDIRQ_READ))
|
| 2953 | return 0;
|
| 2954 | if (curr->softirqs_enabled)
|
| 2955 | if (!mark_lock(curr, hlock,
|
| 2956 | LOCK_ENABLED_SOFTIRQ_READ))
|
| 2957 | return 0;
|
| 2958 | } else {
|
| 2959 | if (!mark_lock(curr, hlock,
|
| 2960 | LOCK_ENABLED_HARDIRQ))
|
| 2961 | return 0;
|
| 2962 | if (curr->softirqs_enabled)
|
| 2963 | if (!mark_lock(curr, hlock,
|
| 2964 | LOCK_ENABLED_SOFTIRQ))
|
| 2965 | return 0;
|
| 2966 | }
|
| 2967 | }
|
| 2968 |
|
| 2969 | |
| 2970 | * We reuse the irq context infrastructure more broadly as a general
|
| 2971 | * context checking code. This tests GFP_FS recursion (a lock taken
|
| 2972 | * during reclaim for a GFP_FS allocation is held over a GFP_FS
|
| 2973 | * allocation).
|
| 2974 |
|
| 2975 | if (!hlock->trylock && (curr->lockdep_reclaim_gfp & __GFP_FS)) {
|
| 2976 | if (hlock->read) {
|
| 2977 | if (!mark_lock(curr, hlock, LOCK_USED_IN_RECLAIM_FS_READ))
|
| 2978 | return 0;
|
| 2979 | } else {
|
| 2980 | if (!mark_lock(curr, hlock, LOCK_USED_IN_RECLAIM_FS))
|
| 2981 | return 0;
|
| 2982 | }
|
| 2983 | }
|
| 2984 |
|
| 2985 | return 1;
|
| 2986 | }
|
| 2987 |
|
| 2988 | static inline unsigned int task_irq_context(struct task_struct *task)
|
| 2989 | {
|
| 2990 | return 2 * !!task->hardirq_context + !!task->softirq_context;
|
| 2991 | }
|
| 2992 |
|
| 2993 | static int separate_irq_context(struct task_struct *curr,
|
| 2994 | struct held_lock *hlock)
|
| 2995 | {
|
| 2996 | unsigned int depth = curr->lockdep_depth;
|
| 2997 |
|
| 2998 | |
| 2999 | * Keep track of points where we cross into an interrupt context:
|
| 3000 |
|
| 3001 | if (depth) {
|
| 3002 | struct held_lock *prev_hlock;
|
| 3003 |
|
| 3004 | prev_hlock = curr->held_locks + depth-1;
|
| 3005 | |
| 3006 | * If we cross into another context, reset the
|
| 3007 | * hash key (this also prevents the checking and the
|
| 3008 | * adding of the dependency to 'prev'):
|
| 3009 |
|
| 3010 | if (prev_hlock->irq_context != hlock->irq_context)
|
| 3011 | return 1;
|
| 3012 | }
|
| 3013 | return 0;
|
| 3014 | }
|
| 3015 |
|
| 3016 | #else
|
| 3017 |
|
| 3018 | static inline
|
| 3019 | int mark_lock_irq(struct task_struct *curr, struct held_lock *this,
|
| 3020 | enum lock_usage_bit new_bit)
|
| 3021 | {
|
| 3022 | WARN_ON(1);
|
| 3023 | return 1;
|
| 3024 | }
|
| 3025 |
|
| 3026 | static inline int mark_irqflags(struct task_struct *curr,
|
| 3027 | struct held_lock *hlock)
|
| 3028 | {
|
| 3029 | return 1;
|
| 3030 | }
|
| 3031 |
|
| 3032 | static inline unsigned int task_irq_context(struct task_struct *task)
|
| 3033 | {
|
| 3034 | return 0;
|
| 3035 | }
|
| 3036 |
|
| 3037 | static inline int separate_irq_context(struct task_struct *curr,
|
| 3038 | struct held_lock *hlock)
|
| 3039 | {
|
| 3040 | return 0;
|
| 3041 | }
|
| 3042 |
|
| 3043 | void lockdep_trace_alloc(gfp_t gfp_mask)
|
| 3044 | {
|
| 3045 | }
|
| 3046 |
|
| 3047 | #endif
|
| 3048 |
|
| 3049 | |
| 3050 | * Mark a lock with a usage bit, and validate the state transition:
|
| 3051 |
|
| 3052 | static int mark_lock(struct task_struct *curr, struct held_lock *this,
|
| 3053 | enum lock_usage_bit new_bit)
|
| 3054 | {
|
| 3055 | unsigned int new_mask = 1 << new_bit, ret = 1;
|
| 3056 |
|
| 3057 | |
| 3058 | * If already set then do not dirty the cacheline,
|
| 3059 | * nor do any checks:
|
| 3060 |
|
| 3061 | if (likely(hlock_class(this)->usage_mask & new_mask))
|
| 3062 | return 1;
|
| 3063 |
|
| 3064 | if (!graph_lock())
|
| 3065 | return 0;
|
| 3066 | |
| 3067 | * Make sure we didn't race:
|
| 3068 |
|
| 3069 | if (unlikely(hlock_class(this)->usage_mask & new_mask)) {
|
| 3070 | graph_unlock();
|
| 3071 | return 1;
|
| 3072 | }
|
| 3073 |
|
| 3074 | hlock_class(this)->usage_mask |= new_mask;
|
| 3075 |
|
| 3076 | if (!save_trace(hlock_class(this)->usage_traces + new_bit))
|
| 3077 | return 0;
|
| 3078 |
|
| 3079 | switch (new_bit) {
|
| 3080 | #define LOCKDEP_STATE(__STATE) \
|
| 3081 | case LOCK_USED_IN_##__STATE: \
|
| 3082 | case LOCK_USED_IN_##__STATE##_READ: \
|
| 3083 | case LOCK_ENABLED_##__STATE: \
|
| 3084 | case LOCK_ENABLED_##__STATE##_READ:
|
| 3085 | #include "lockdep_states.h"
|
| 3086 | #undef LOCKDEP_STATE
|
| 3087 | ret = mark_lock_irq(curr, this, new_bit);
|
| 3088 | if (!ret)
|
| 3089 | return 0;
|
| 3090 | break;
|
| 3091 | case LOCK_USED:
|
| 3092 | debug_atomic_dec(nr_unused_locks);
|
| 3093 | break;
|
| 3094 | default:
|
| 3095 | if (!debug_locks_off_graph_unlock())
|
| 3096 | return 0;
|
| 3097 | WARN_ON(1);
|
| 3098 | return 0;
|
| 3099 | }
|
| 3100 |
|
| 3101 | graph_unlock();
|
| 3102 |
|
| 3103 | |
| 3104 | * We must printk outside of the graph_lock:
|
| 3105 |
|
| 3106 | if (ret == 2) {
|
| 3107 | printk("\nmarked lock as {%s}:\n", usage_str[new_bit]);
|
| 3108 | print_lock(this);
|
| 3109 | print_irqtrace_events(curr);
|
| 3110 | dump_stack();
|
| 3111 | }
|
| 3112 |
|
| 3113 | return ret;
|
| 3114 | }
|
| 3115 |
|
| 3116 | |
| 3117 | * Initialize a lock instance's lock-class mapping info:
|
| 3118 |
|
| 3119 | void lockdep_init_map(struct lockdep_map *lock, const char *name,
|
| 3120 | struct lock_class_key *key, int subclass)
|
| 3121 | {
|
| 3122 | int i;
|
| 3123 |
|
| 3124 | kmemcheck_mark_initialized(lock, sizeof(*lock));
|
| 3125 |
|
| 3126 | for (i = 0; i < NR_LOCKDEP_CACHING_CLASSES; i++)
|
| 3127 | lock->class_cache[i] = NULL;
|
| 3128 |
|
| 3129 | #ifdef CONFIG_LOCK_STAT
|
| 3130 | lock->cpu = raw_smp_processor_id();
|
| 3131 | #endif
|
| 3132 |
|
| 3133 | |
| 3134 | * Can't be having no nameless bastards around this place!
|
| 3135 |
|
| 3136 | if (DEBUG_LOCKS_WARN_ON(!name)) {
|
| 3137 | lock->name = "NULL";
|
| 3138 | return;
|
| 3139 | }
|
| 3140 |
|
| 3141 | lock->name = name;
|
| 3142 |
|
| 3143 | |
| 3144 | * No key, no joy, we need to hash something.
|
| 3145 |
|
| 3146 | if (DEBUG_LOCKS_WARN_ON(!key))
|
| 3147 | return;
|
| 3148 | |
| 3149 | * Sanity check, the lock-class key must be persistent:
|
| 3150 |
|
| 3151 | if (!static_obj(key)) {
|
| 3152 | printk("BUG: key %p not in .data!\n", key);
|
| 3153 | |
| 3154 | * What it says above ^^^^^, I suggest you read it.
|
| 3155 |
|
| 3156 | DEBUG_LOCKS_WARN_ON(1);
|
| 3157 | return;
|
| 3158 | }
|
| 3159 | lock->key = key;
|
| 3160 |
|
| 3161 | if (unlikely(!debug_locks))
|
| 3162 | return;
|
| 3163 |
|
| 3164 | if (subclass) {
|
| 3165 | unsigned long flags;
|
| 3166 |
|
| 3167 | if (DEBUG_LOCKS_WARN_ON(current->lockdep_recursion))
|
| 3168 | return;
|
| 3169 |
|
| 3170 | raw_local_irq_save(flags);
|
| 3171 | current->lockdep_recursion = 1;
|
| 3172 | register_lock_class(lock, subclass, 1);
|
| 3173 | current->lockdep_recursion = 0;
|
| 3174 | raw_local_irq_restore(flags);
|
| 3175 | }
|
| 3176 | }
|
| 3177 | EXPORT_SYMBOL_GPL(lockdep_init_map);
|
| 3178 |
|
| 3179 | struct lock_class_key __lockdep_no_validate__;
|
| 3180 | EXPORT_SYMBOL_GPL(__lockdep_no_validate__);
|
| 3181 |
|
| 3182 | static int
|
| 3183 | print_lock_nested_lock_not_held(struct task_struct *curr,
|
| 3184 | struct held_lock *hlock,
|
| 3185 | unsigned long ip)
|
| 3186 | {
|
| 3187 | if (!debug_locks_off())
|
| 3188 | return 0;
|
| 3189 | if (debug_locks_silent)
|
| 3190 | return 0;
|
| 3191 |
|
| 3192 | pr_warn("\n");
|
| 3193 | pr_warn("==================================\n");
|
| 3194 | pr_warn("WARNING: Nested lock was not taken\n");
|
| 3195 | print_kernel_ident();
|
| 3196 | pr_warn("----------------------------------\n");
|
| 3197 |
|
| 3198 | pr_warn("%s/%d is trying to lock:\n", curr->comm, task_pid_nr(curr));
|
| 3199 | print_lock(hlock);
|
| 3200 |
|
| 3201 | pr_warn("\nbut this task is not holding:\n");
|
| 3202 | pr_warn("%s\n", hlock->nest_lock->name);
|
| 3203 |
|
| 3204 | pr_warn("\nstack backtrace:\n");
|
| 3205 | dump_stack();
|
| 3206 |
|
| 3207 | pr_warn("\nother info that might help us debug this:\n");
|
| 3208 | lockdep_print_held_locks(curr);
|
| 3209 |
|
| 3210 | pr_warn("\nstack backtrace:\n");
|
| 3211 | dump_stack();
|
| 3212 |
|
| 3213 | return 0;
|
| 3214 | }
|
| 3215 |
|
| 3216 | static int __lock_is_held(struct lockdep_map *lock, int read);
|
| 3217 |
|
| 3218 | |
| 3219 | * This gets called for every mutex_lock*()/spin_lock*() operation.
|
| 3220 | * We maintain the dependency maps and validate the locking attempt:
|
| 3221 |
|
| 3222 | static int __lock_acquire(struct lockdep_map *lock, unsigned int subclass,
|
| 3223 | int trylock, int read, int check, int hardirqs_off,
|
| 3224 | struct lockdep_map *nest_lock, unsigned long ip,
|
| 3225 | int references, int pin_count)
|
| 3226 | {
|
| 3227 | struct task_struct *curr = current;
|
| 3228 | struct lock_class *class = NULL;
|
| 3229 | struct held_lock *hlock;
|
| 3230 | unsigned int depth;
|
| 3231 | int chain_head = 0;
|
| 3232 | int class_idx;
|
| 3233 | u64 chain_key;
|
| 3234 |
|
| 3235 | if (unlikely(!debug_locks))
|
| 3236 | return 0;
|
| 3237 |
|
| 3238 | |
| 3239 | * Lockdep should run with IRQs disabled, otherwise we could
|
| 3240 | * get an interrupt which would want to take locks, which would
|
| 3241 | * end up in lockdep and have you got a head-ache already?
|
| 3242 |
|
| 3243 | if (DEBUG_LOCKS_WARN_ON(!irqs_disabled()))
|
| 3244 | return 0;
|
| 3245 |
|
| 3246 | if (!prove_locking || lock->key == &__lockdep_no_validate__)
|
| 3247 | check = 0;
|
| 3248 |
|
| 3249 | if (subclass < NR_LOCKDEP_CACHING_CLASSES)
|
| 3250 | class = lock->class_cache[subclass];
|
| 3251 | |
| 3252 | * Not cached?
|
| 3253 |
|
| 3254 | if (unlikely(!class)) {
|
| 3255 | class = register_lock_class(lock, subclass, 0);
|
| 3256 | if (!class)
|
| 3257 | return 0;
|
| 3258 | }
|
| 3259 | atomic_inc((atomic_t *)&class->ops);
|
| 3260 | if (very_verbose(class)) {
|
| 3261 | printk("\nacquire class [%p] %s", class->key, class->name);
|
| 3262 | if (class->name_version > 1)
|
| 3263 | printk(KERN_CONT "#%d", class->name_version);
|
| 3264 | printk(KERN_CONT "\n");
|
| 3265 | dump_stack();
|
| 3266 | }
|
| 3267 |
|
| 3268 | |
| 3269 | * Add the lock to the list of currently held locks.
|
| 3270 | * (we dont increase the depth just yet, up until the
|
| 3271 | * dependency checks are done)
|
| 3272 |
|
| 3273 | depth = curr->lockdep_depth;
|
| 3274 | |
| 3275 | * Ran out of static storage for our per-task lock stack again have we?
|
| 3276 |
|
| 3277 | if (DEBUG_LOCKS_WARN_ON(depth >= MAX_LOCK_DEPTH))
|
| 3278 | return 0;
|
| 3279 |
|
| 3280 | class_idx = class - lock_classes + 1;
|
| 3281 |
|
| 3282 | if (depth) {
|
| 3283 | hlock = curr->held_locks + depth - 1;
|
| 3284 | if (hlock->class_idx == class_idx && nest_lock) {
|
| 3285 | if (hlock->references) {
|
| 3286 | |
| 3287 | * Check: unsigned int references:12, overflow.
|
| 3288 |
|
| 3289 | if (DEBUG_LOCKS_WARN_ON(hlock->references == (1 << 12)-1))
|
| 3290 | return 0;
|
| 3291 |
|
| 3292 | hlock->references++;
|
| 3293 | } else {
|
| 3294 | hlock->references = 2;
|
| 3295 | }
|
| 3296 |
|
| 3297 | return 1;
|
| 3298 | }
|
| 3299 | }
|
| 3300 |
|
| 3301 | hlock = curr->held_locks + depth;
|
| 3302 | |
| 3303 | * Plain impossible, we just registered it and checked it weren't no
|
| 3304 | * NULL like.. I bet this mushroom I ate was good!
|
| 3305 |
|
| 3306 | if (DEBUG_LOCKS_WARN_ON(!class))
|
| 3307 | return 0;
|
| 3308 | hlock->class_idx = class_idx;
|
| 3309 | hlock->acquire_ip = ip;
|
| 3310 | hlock->instance = lock;
|
| 3311 | hlock->nest_lock = nest_lock;
|
| 3312 | hlock->irq_context = task_irq_context(curr);
|
| 3313 | hlock->trylock = trylock;
|
| 3314 | hlock->read = read;
|
| 3315 | hlock->check = check;
|
| 3316 | hlock->hardirqs_off = !!hardirqs_off;
|
| 3317 | hlock->references = references;
|
| 3318 | #ifdef CONFIG_LOCK_STAT
|
| 3319 | hlock->waittime_stamp = 0;
|
| 3320 | hlock->holdtime_stamp = lockstat_clock();
|
| 3321 | #endif
|
| 3322 | hlock->pin_count = pin_count;
|
| 3323 |
|
| 3324 | if (check && !mark_irqflags(curr, hlock))
|
| 3325 | return 0;
|
| 3326 |
|
| 3327 |
|
| 3328 | if (!mark_lock(curr, hlock, LOCK_USED))
|
| 3329 | return 0;
|
| 3330 |
|
| 3331 | |
| 3332 | * Calculate the chain hash: it's the combined hash of all the
|
| 3333 | * lock keys along the dependency chain. We save the hash value
|
| 3334 | * at every step so that we can get the current hash easily
|
| 3335 | * after unlock. The chain hash is then used to cache dependency
|
| 3336 | * results.
|
| 3337 | *
|
| 3338 | * The 'key ID' is what is the most compact key value to drive
|
| 3339 | * the hash, not class->key.
|
| 3340 |
|
| 3341 | |
| 3342 | * Whoops, we did it again.. ran straight out of our static allocation.
|
| 3343 |
|
| 3344 | if (DEBUG_LOCKS_WARN_ON(class_idx > MAX_LOCKDEP_KEYS))
|
| 3345 | return 0;
|
| 3346 |
|
| 3347 | chain_key = curr->curr_chain_key;
|
| 3348 | if (!depth) {
|
| 3349 | |
| 3350 | * How can we have a chain hash when we ain't got no keys?!
|
| 3351 |
|
| 3352 | if (DEBUG_LOCKS_WARN_ON(chain_key != 0))
|
| 3353 | return 0;
|
| 3354 | chain_head = 1;
|
| 3355 | }
|
| 3356 |
|
| 3357 | hlock->prev_chain_key = chain_key;
|
| 3358 | if (separate_irq_context(curr, hlock)) {
|
| 3359 | chain_key = 0;
|
| 3360 | chain_head = 1;
|
| 3361 | }
|
| 3362 | chain_key = iterate_chain_key(chain_key, class_idx);
|
| 3363 |
|
| 3364 | if (nest_lock && !__lock_is_held(nest_lock, -1))
|
| 3365 | return print_lock_nested_lock_not_held(curr, hlock, ip);
|
| 3366 |
|
| 3367 | if (!validate_chain(curr, lock, hlock, chain_head, chain_key))
|
| 3368 | return 0;
|
| 3369 |
|
| 3370 | curr->curr_chain_key = chain_key;
|
| 3371 | curr->lockdep_depth++;
|
| 3372 | check_chain_key(curr);
|
| 3373 | #ifdef CONFIG_DEBUG_LOCKDEP
|
| 3374 | if (unlikely(!debug_locks))
|
| 3375 | return 0;
|
| 3376 | #endif
|
| 3377 | if (unlikely(curr->lockdep_depth >= MAX_LOCK_DEPTH)) {
|
| 3378 | debug_locks_off();
|
| 3379 | print_lockdep_off("BUG: MAX_LOCK_DEPTH too low!");
|
| 3380 | printk(KERN_DEBUG "depth: %i max: %lu!\n",
|
| 3381 | curr->lockdep_depth, MAX_LOCK_DEPTH);
|
| 3382 |
|
| 3383 | lockdep_print_held_locks(current);
|
| 3384 | debug_show_all_locks();
|
| 3385 | dump_stack();
|
| 3386 |
|
| 3387 | return 0;
|
| 3388 | }
|
| 3389 |
|
| 3390 | if (unlikely(curr->lockdep_depth > max_lockdep_depth))
|
| 3391 | max_lockdep_depth = curr->lockdep_depth;
|
| 3392 |
|
| 3393 | return 1;
|
| 3394 | }
|
| 3395 |
|
| 3396 | static int
|
| 3397 | print_unlock_imbalance_bug(struct task_struct *curr, struct lockdep_map *lock,
|
| 3398 | unsigned long ip)
|
| 3399 | {
|
| 3400 | if (!debug_locks_off())
|
| 3401 | return 0;
|
| 3402 | if (debug_locks_silent)
|
| 3403 | return 0;
|
| 3404 |
|
| 3405 | pr_warn("\n");
|
| 3406 | pr_warn("=====================================\n");
|
| 3407 | pr_warn("WARNING: bad unlock balance detected!\n");
|
| 3408 | print_kernel_ident();
|
| 3409 | pr_warn("-------------------------------------\n");
|
| 3410 | pr_warn("%s/%d is trying to release lock (",
|
| 3411 | curr->comm, task_pid_nr(curr));
|
| 3412 | print_lockdep_cache(lock);
|
| 3413 | pr_cont(") at:\n");
|
| 3414 | print_ip_sym(ip);
|
| 3415 | pr_warn("but there are no more locks to release!\n");
|
| 3416 | pr_warn("\nother info that might help us debug this:\n");
|
| 3417 | lockdep_print_held_locks(curr);
|
| 3418 |
|
| 3419 | pr_warn("\nstack backtrace:\n");
|
| 3420 | dump_stack();
|
| 3421 |
|
| 3422 | return 0;
|
| 3423 | }
|
| 3424 |
|
| 3425 | static int match_held_lock(struct held_lock *hlock, struct lockdep_map *lock)
|
| 3426 | {
|
| 3427 | if (hlock->instance == lock)
|
| 3428 | return 1;
|
| 3429 |
|
| 3430 | if (hlock->references) {
|
| 3431 | struct lock_class *class = lock->class_cache[0];
|
| 3432 |
|
| 3433 | if (!class)
|
| 3434 | class = look_up_lock_class(lock, 0);
|
| 3435 |
|
| 3436 | |
| 3437 | * If look_up_lock_class() failed to find a class, we're trying
|
| 3438 | * to test if we hold a lock that has never yet been acquired.
|
| 3439 | * Clearly if the lock hasn't been acquired _ever_, we're not
|
| 3440 | * holding it either, so report failure.
|
| 3441 |
|
| 3442 | if (IS_ERR_OR_NULL(class))
|
| 3443 | return 0;
|
| 3444 |
|
| 3445 | |
| 3446 | * References, but not a lock we're actually ref-counting?
|
| 3447 | * State got messed up, follow the sites that change ->references
|
| 3448 | * and try to make sense of it.
|
| 3449 |
|
| 3450 | if (DEBUG_LOCKS_WARN_ON(!hlock->nest_lock))
|
| 3451 | return 0;
|
| 3452 |
|
| 3453 | if (hlock->class_idx == class - lock_classes + 1)
|
| 3454 | return 1;
|
| 3455 | }
|
| 3456 |
|
| 3457 | return 0;
|
| 3458 | }
|
| 3459 |
|
| 3460 |
|
| 3461 | static struct held_lock *find_held_lock(struct task_struct *curr,
|
| 3462 | struct lockdep_map *lock,
|
| 3463 | unsigned int depth, int *idx)
|
| 3464 | {
|
| 3465 | struct held_lock *ret, *hlock, *prev_hlock;
|
| 3466 | int i;
|
| 3467 |
|
| 3468 | i = depth - 1;
|
| 3469 | hlock = curr->held_locks + i;
|
| 3470 | ret = hlock;
|
| 3471 | if (match_held_lock(hlock, lock))
|
| 3472 | goto out;
|
| 3473 |
|
| 3474 | ret = NULL;
|
| 3475 | for (i--, prev_hlock = hlock--;
|
| 3476 | i >= 0;
|
| 3477 | i--, prev_hlock = hlock--) {
|
| 3478 | |
| 3479 | * We must not cross into another context:
|
| 3480 |
|
| 3481 | if (prev_hlock->irq_context != hlock->irq_context) {
|
| 3482 | ret = NULL;
|
| 3483 | break;
|
| 3484 | }
|
| 3485 | if (match_held_lock(hlock, lock)) {
|
| 3486 | ret = hlock;
|
| 3487 | break;
|
| 3488 | }
|
| 3489 | }
|
| 3490 |
|
| 3491 | out:
|
| 3492 | *idx = i;
|
| 3493 | return ret;
|
| 3494 | }
|
| 3495 |
|
| 3496 | static int reacquire_held_locks(struct task_struct *curr, unsigned int depth,
|
| 3497 | int idx)
|
| 3498 | {
|
| 3499 | struct held_lock *hlock;
|
| 3500 |
|
| 3501 | for (hlock = curr->held_locks + idx; idx < depth; idx++, hlock++) {
|
| 3502 | if (!__lock_acquire(hlock->instance,
|
| 3503 | hlock_class(hlock)->subclass,
|
| 3504 | hlock->trylock,
|
| 3505 | hlock->read, hlock->check,
|
| 3506 | hlock->hardirqs_off,
|
| 3507 | hlock->nest_lock, hlock->acquire_ip,
|
| 3508 | hlock->references, hlock->pin_count))
|
| 3509 | return 1;
|
| 3510 | }
|
| 3511 | return 0;
|
| 3512 | }
|
| 3513 |
|
| 3514 | static int
|
| 3515 | __lock_set_class(struct lockdep_map *lock, const char *name,
|
| 3516 | struct lock_class_key *key, unsigned int subclass,
|
| 3517 | unsigned long ip)
|
| 3518 | {
|
| 3519 | struct task_struct *curr = current;
|
| 3520 | struct held_lock *hlock;
|
| 3521 | struct lock_class *class;
|
| 3522 | unsigned int depth;
|
| 3523 | int i;
|
| 3524 |
|
| 3525 | depth = curr->lockdep_depth;
|
| 3526 | |
| 3527 | * This function is about (re)setting the class of a held lock,
|
| 3528 | * yet we're not actually holding any locks. Naughty user!
|
| 3529 |
|
| 3530 | if (DEBUG_LOCKS_WARN_ON(!depth))
|
| 3531 | return 0;
|
| 3532 |
|
| 3533 | hlock = find_held_lock(curr, lock, depth, &i);
|
| 3534 | if (!hlock)
|
| 3535 | return print_unlock_imbalance_bug(curr, lock, ip);
|
| 3536 |
|
| 3537 | lockdep_init_map(lock, name, key, 0);
|
| 3538 | class = register_lock_class(lock, subclass, 0);
|
| 3539 | hlock->class_idx = class - lock_classes + 1;
|
| 3540 |
|
| 3541 | curr->lockdep_depth = i;
|
| 3542 | curr->curr_chain_key = hlock->prev_chain_key;
|
| 3543 |
|
| 3544 | if (reacquire_held_locks(curr, depth, i))
|
| 3545 | return 0;
|
| 3546 |
|
| 3547 | |
| 3548 | * I took it apart and put it back together again, except now I have
|
| 3549 | * these 'spare' parts.. where shall I put them.
|
| 3550 |
|
| 3551 | if (DEBUG_LOCKS_WARN_ON(curr->lockdep_depth != depth))
|
| 3552 | return 0;
|
| 3553 | return 1;
|
| 3554 | }
|
| 3555 |
|
| 3556 | static int __lock_downgrade(struct lockdep_map *lock, unsigned long ip)
|
| 3557 | {
|
| 3558 | struct task_struct *curr = current;
|
| 3559 | struct held_lock *hlock;
|
| 3560 | unsigned int depth;
|
| 3561 | int i;
|
| 3562 |
|
| 3563 | depth = curr->lockdep_depth;
|
| 3564 | |
| 3565 | * This function is about (re)setting the class of a held lock,
|
| 3566 | * yet we're not actually holding any locks. Naughty user!
|
| 3567 |
|
| 3568 | if (DEBUG_LOCKS_WARN_ON(!depth))
|
| 3569 | return 0;
|
| 3570 |
|
| 3571 | hlock = find_held_lock(curr, lock, depth, &i);
|
| 3572 | if (!hlock)
|
| 3573 | return print_unlock_imbalance_bug(curr, lock, ip);
|
| 3574 |
|
| 3575 | curr->lockdep_depth = i;
|
| 3576 | curr->curr_chain_key = hlock->prev_chain_key;
|
| 3577 |
|
| 3578 | WARN(hlock->read, "downgrading a read lock");
|
| 3579 | hlock->read = 1;
|
| 3580 | hlock->acquire_ip = ip;
|
| 3581 |
|
| 3582 | if (reacquire_held_locks(curr, depth, i))
|
| 3583 | return 0;
|
| 3584 |
|
| 3585 | |
| 3586 | * I took it apart and put it back together again, except now I have
|
| 3587 | * these 'spare' parts.. where shall I put them.
|
| 3588 |
|
| 3589 | if (DEBUG_LOCKS_WARN_ON(curr->lockdep_depth != depth))
|
| 3590 | return 0;
|
| 3591 | return 1;
|
| 3592 | }
|
| 3593 |
|
| 3594 | |
| 3595 | * Remove the lock to the list of currently held locks - this gets
|
| 3596 | * called on mutex_unlock()/spin_unlock*() (or on a failed
|
| 3597 | * mutex_lock_interruptible()).
|
| 3598 | *
|
| 3599 | * @nested is an hysterical artifact, needs a tree wide cleanup.
|
| 3600 |
|
| 3601 | static int
|
| 3602 | __lock_release(struct lockdep_map *lock, int nested, unsigned long ip)
|
| 3603 | {
|
| 3604 | struct task_struct *curr = current;
|
| 3605 | struct held_lock *hlock;
|
| 3606 | unsigned int depth;
|
| 3607 | int i;
|
| 3608 |
|
| 3609 | if (unlikely(!debug_locks))
|
| 3610 | return 0;
|
| 3611 |
|
| 3612 | depth = curr->lockdep_depth;
|
| 3613 | |
| 3614 | * So we're all set to release this lock.. wait what lock? We don't
|
| 3615 | * own any locks, you've been drinking again?
|
| 3616 |
|
| 3617 | if (DEBUG_LOCKS_WARN_ON(depth <= 0))
|
| 3618 | return print_unlock_imbalance_bug(curr, lock, ip);
|
| 3619 |
|
| 3620 | |
| 3621 | * Check whether the lock exists in the current stack
|
| 3622 | * of held locks:
|
| 3623 |
|
| 3624 | hlock = find_held_lock(curr, lock, depth, &i);
|
| 3625 | if (!hlock)
|
| 3626 | return print_unlock_imbalance_bug(curr, lock, ip);
|
| 3627 |
|
| 3628 | if (hlock->instance == lock)
|
| 3629 | lock_release_holdtime(hlock);
|
| 3630 |
|
| 3631 | WARN(hlock->pin_count, "releasing a pinned lock\n");
|
| 3632 |
|
| 3633 | if (hlock->references) {
|
| 3634 | hlock->references--;
|
| 3635 | if (hlock->references) {
|
| 3636 | |
| 3637 | * We had, and after removing one, still have
|
| 3638 | * references, the current lock stack is still
|
| 3639 | * valid. We're done!
|
| 3640 |
|
| 3641 | return 1;
|
| 3642 | }
|
| 3643 | }
|
| 3644 |
|
| 3645 | |
| 3646 | * We have the right lock to unlock, 'hlock' points to it.
|
| 3647 | * Now we remove it from the stack, and add back the other
|
| 3648 | * entries (if any), recalculating the hash along the way:
|
| 3649 |
|
| 3650 |
|
| 3651 | curr->lockdep_depth = i;
|
| 3652 | curr->curr_chain_key = hlock->prev_chain_key;
|
| 3653 |
|
| 3654 | if (reacquire_held_locks(curr, depth, i + 1))
|
| 3655 | return 0;
|
| 3656 |
|
| 3657 | |
| 3658 | * We had N bottles of beer on the wall, we drank one, but now
|
| 3659 | * there's not N-1 bottles of beer left on the wall...
|
| 3660 |
|
| 3661 | if (DEBUG_LOCKS_WARN_ON(curr->lockdep_depth != depth - 1))
|
| 3662 | return 0;
|
| 3663 |
|
| 3664 | return 1;
|
| 3665 | }
|
| 3666 |
|
| 3667 | static int __lock_is_held(struct lockdep_map *lock, int read)
|
| 3668 | {
|
| 3669 | struct task_struct *curr = current;
|
| 3670 | int i;
|
| 3671 |
|
| 3672 | for (i = 0; i < curr->lockdep_depth; i++) {
|
| 3673 | struct held_lock *hlock = curr->held_locks + i;
|
| 3674 |
|
| 3675 | if (match_held_lock(hlock, lock)) {
|
| 3676 | if (read == -1 || hlock->read == read)
|
| 3677 | return 1;
|
| 3678 |
|
| 3679 | return 0;
|
| 3680 | }
|
| 3681 | }
|
| 3682 |
|
| 3683 | return 0;
|
| 3684 | }
|
| 3685 |
|
| 3686 | static struct pin_cookie __lock_pin_lock(struct lockdep_map *lock)
|
| 3687 | {
|
| 3688 | struct pin_cookie cookie = NIL_COOKIE;
|
| 3689 | struct task_struct *curr = current;
|
| 3690 | int i;
|
| 3691 |
|
| 3692 | if (unlikely(!debug_locks))
|
| 3693 | return cookie;
|
| 3694 |
|
| 3695 | for (i = 0; i < curr->lockdep_depth; i++) {
|
| 3696 | struct held_lock *hlock = curr->held_locks + i;
|
| 3697 |
|
| 3698 | if (match_held_lock(hlock, lock)) {
|
| 3699 | |
| 3700 | * Grab 16bits of randomness; this is sufficient to not
|
| 3701 | * be guessable and still allows some pin nesting in
|
| 3702 | * our u32 pin_count.
|
| 3703 |
|
| 3704 | cookie.val = 1 + (prandom_u32() >> 16);
|
| 3705 | hlock->pin_count += cookie.val;
|
| 3706 | return cookie;
|
| 3707 | }
|
| 3708 | }
|
| 3709 |
|
| 3710 | WARN(1, "pinning an unheld lock\n");
|
| 3711 | return cookie;
|
| 3712 | }
|
| 3713 |
|
| 3714 | static void __lock_repin_lock(struct lockdep_map *lock, struct pin_cookie cookie)
|
| 3715 | {
|
| 3716 | struct task_struct *curr = current;
|
| 3717 | int i;
|
| 3718 |
|
| 3719 | if (unlikely(!debug_locks))
|
| 3720 | return;
|
| 3721 |
|
| 3722 | for (i = 0; i < curr->lockdep_depth; i++) {
|
| 3723 | struct held_lock *hlock = curr->held_locks + i;
|
| 3724 |
|
| 3725 | if (match_held_lock(hlock, lock)) {
|
| 3726 | hlock->pin_count += cookie.val;
|
| 3727 | return;
|
| 3728 | }
|
| 3729 | }
|
| 3730 |
|
| 3731 | WARN(1, "pinning an unheld lock\n");
|
| 3732 | }
|
| 3733 |
|
| 3734 | static void __lock_unpin_lock(struct lockdep_map *lock, struct pin_cookie cookie)
|
| 3735 | {
|
| 3736 | struct task_struct *curr = current;
|
| 3737 | int i;
|
| 3738 |
|
| 3739 | if (unlikely(!debug_locks))
|
| 3740 | return;
|
| 3741 |
|
| 3742 | for (i = 0; i < curr->lockdep_depth; i++) {
|
| 3743 | struct held_lock *hlock = curr->held_locks + i;
|
| 3744 |
|
| 3745 | if (match_held_lock(hlock, lock)) {
|
| 3746 | if (WARN(!hlock->pin_count, "unpinning an unpinned lock\n"))
|
| 3747 | return;
|
| 3748 |
|
| 3749 | hlock->pin_count -= cookie.val;
|
| 3750 |
|
| 3751 | if (WARN((int)hlock->pin_count < 0, "pin count corrupted\n"))
|
| 3752 | hlock->pin_count = 0;
|
| 3753 |
|
| 3754 | return;
|
| 3755 | }
|
| 3756 | }
|
| 3757 |
|
| 3758 | WARN(1, "unpinning an unheld lock\n");
|
| 3759 | }
|
| 3760 |
|
| 3761 | |
| 3762 | * Check whether we follow the irq-flags state precisely:
|
| 3763 |
|
| 3764 | static void check_flags(unsigned long flags)
|
| 3765 | {
|
| 3766 | #if defined(CONFIG_PROVE_LOCKING) && defined(CONFIG_DEBUG_LOCKDEP) && \
|
| 3767 | defined(CONFIG_TRACE_IRQFLAGS)
|
| 3768 | if (!debug_locks)
|
| 3769 | return;
|
| 3770 |
|
| 3771 | if (irqs_disabled_flags(flags)) {
|
| 3772 | if (DEBUG_LOCKS_WARN_ON(current->hardirqs_enabled)) {
|
| 3773 | printk("possible reason: unannotated irqs-off.\n");
|
| 3774 | }
|
| 3775 | } else {
|
| 3776 | if (DEBUG_LOCKS_WARN_ON(!current->hardirqs_enabled)) {
|
| 3777 | printk("possible reason: unannotated irqs-on.\n");
|
| 3778 | }
|
| 3779 | }
|
| 3780 |
|
| 3781 | |
| 3782 | * We dont accurately track softirq state in e.g.
|
| 3783 | * hardirq contexts (such as on 4KSTACKS), so only
|
| 3784 | * check if not in hardirq contexts:
|
| 3785 |
|
| 3786 | if (!hardirq_count()) {
|
| 3787 | if (softirq_count()) {
|
| 3788 |
|
| 3789 | DEBUG_LOCKS_WARN_ON(current->softirqs_enabled);
|
| 3790 | } else {
|
| 3791 |
|
| 3792 | DEBUG_LOCKS_WARN_ON(!current->softirqs_enabled);
|
| 3793 | }
|
| 3794 | }
|
| 3795 |
|
| 3796 | if (!debug_locks)
|
| 3797 | print_irqtrace_events(current);
|
| 3798 | #endif
|
| 3799 | }
|
| 3800 |
|
| 3801 | void lock_set_class(struct lockdep_map *lock, const char *name,
|
| 3802 | struct lock_class_key *key, unsigned int subclass,
|
| 3803 | unsigned long ip)
|
| 3804 | {
|
| 3805 | unsigned long flags;
|
| 3806 |
|
| 3807 | if (unlikely(current->lockdep_recursion))
|
| 3808 | return;
|
| 3809 |
|
| 3810 | raw_local_irq_save(flags);
|
| 3811 | current->lockdep_recursion = 1;
|
| 3812 | check_flags(flags);
|
| 3813 | if (__lock_set_class(lock, name, key, subclass, ip))
|
| 3814 | check_chain_key(current);
|
| 3815 | current->lockdep_recursion = 0;
|
| 3816 | raw_local_irq_restore(flags);
|
| 3817 | }
|
| 3818 | EXPORT_SYMBOL_GPL(lock_set_class);
|
| 3819 |
|
| 3820 | void lock_downgrade(struct lockdep_map *lock, unsigned long ip)
|
| 3821 | {
|
| 3822 | unsigned long flags;
|
| 3823 |
|
| 3824 | if (unlikely(current->lockdep_recursion))
|
| 3825 | return;
|
| 3826 |
|
| 3827 | raw_local_irq_save(flags);
|
| 3828 | current->lockdep_recursion = 1;
|
| 3829 | check_flags(flags);
|
| 3830 | if (__lock_downgrade(lock, ip))
|
| 3831 | check_chain_key(current);
|
| 3832 | current->lockdep_recursion = 0;
|
| 3833 | raw_local_irq_restore(flags);
|
| 3834 | }
|
| 3835 | EXPORT_SYMBOL_GPL(lock_downgrade);
|
| 3836 |
|
| 3837 | |
| 3838 | * We are not always called with irqs disabled - do that here,
|
| 3839 | * and also avoid lockdep recursion:
|
| 3840 |
|
| 3841 | void lock_acquire(struct lockdep_map *lock, unsigned int subclass,
|
| 3842 | int trylock, int read, int check,
|
| 3843 | struct lockdep_map *nest_lock, unsigned long ip)
|
| 3844 | {
|
| 3845 | unsigned long flags;
|
| 3846 |
|
| 3847 | if (unlikely(current->lockdep_recursion))
|
| 3848 | return;
|
| 3849 |
|
| 3850 | raw_local_irq_save(flags);
|
| 3851 | check_flags(flags);
|
| 3852 |
|
| 3853 | current->lockdep_recursion = 1;
|
| 3854 | trace_lock_acquire(lock, subclass, trylock, read, check, nest_lock, ip);
|
| 3855 | __lock_acquire(lock, subclass, trylock, read, check,
|
| 3856 | irqs_disabled_flags(flags), nest_lock, ip, 0, 0);
|
| 3857 | current->lockdep_recursion = 0;
|
| 3858 | raw_local_irq_restore(flags);
|
| 3859 | }
|
| 3860 | EXPORT_SYMBOL_GPL(lock_acquire);
|
| 3861 |
|
| 3862 | void lock_release(struct lockdep_map *lock, int nested,
|
| 3863 | unsigned long ip)
|
| 3864 | {
|
| 3865 | unsigned long flags;
|
| 3866 |
|
| 3867 | if (unlikely(current->lockdep_recursion))
|
| 3868 | return;
|
| 3869 |
|
| 3870 | raw_local_irq_save(flags);
|
| 3871 | check_flags(flags);
|
| 3872 | current->lockdep_recursion = 1;
|
| 3873 | trace_lock_release(lock, ip);
|
| 3874 | if (__lock_release(lock, nested, ip))
|
| 3875 | check_chain_key(current);
|
| 3876 | current->lockdep_recursion = 0;
|
| 3877 | raw_local_irq_restore(flags);
|
| 3878 | }
|
| 3879 | EXPORT_SYMBOL_GPL(lock_release);
|
| 3880 |
|
| 3881 | int lock_is_held_type(struct lockdep_map *lock, int read)
|
| 3882 | {
|
| 3883 | unsigned long flags;
|
| 3884 | int ret = 0;
|
| 3885 |
|
| 3886 | if (unlikely(current->lockdep_recursion))
|
| 3887 | return 1;
|
| 3888 |
|
| 3889 | raw_local_irq_save(flags);
|
| 3890 | check_flags(flags);
|
| 3891 |
|
| 3892 | current->lockdep_recursion = 1;
|
| 3893 | ret = __lock_is_held(lock, read);
|
| 3894 | current->lockdep_recursion = 0;
|
| 3895 | raw_local_irq_restore(flags);
|
| 3896 |
|
| 3897 | return ret;
|
| 3898 | }
|
| 3899 | EXPORT_SYMBOL_GPL(lock_is_held_type);
|
| 3900 |
|
| 3901 | struct pin_cookie lock_pin_lock(struct lockdep_map *lock)
|
| 3902 | {
|
| 3903 | struct pin_cookie cookie = NIL_COOKIE;
|
| 3904 | unsigned long flags;
|
| 3905 |
|
| 3906 | if (unlikely(current->lockdep_recursion))
|
| 3907 | return cookie;
|
| 3908 |
|
| 3909 | raw_local_irq_save(flags);
|
| 3910 | check_flags(flags);
|
| 3911 |
|
| 3912 | current->lockdep_recursion = 1;
|
| 3913 | cookie = __lock_pin_lock(lock);
|
| 3914 | current->lockdep_recursion = 0;
|
| 3915 | raw_local_irq_restore(flags);
|
| 3916 |
|
| 3917 | return cookie;
|
| 3918 | }
|
| 3919 | EXPORT_SYMBOL_GPL(lock_pin_lock);
|
| 3920 |
|
| 3921 | void lock_repin_lock(struct lockdep_map *lock, struct pin_cookie cookie)
|
| 3922 | {
|
| 3923 | unsigned long flags;
|
| 3924 |
|
| 3925 | if (unlikely(current->lockdep_recursion))
|
| 3926 | return;
|
| 3927 |
|
| 3928 | raw_local_irq_save(flags);
|
| 3929 | check_flags(flags);
|
| 3930 |
|
| 3931 | current->lockdep_recursion = 1;
|
| 3932 | __lock_repin_lock(lock, cookie);
|
| 3933 | current->lockdep_recursion = 0;
|
| 3934 | raw_local_irq_restore(flags);
|
| 3935 | }
|
| 3936 | EXPORT_SYMBOL_GPL(lock_repin_lock);
|
| 3937 |
|
| 3938 | void lock_unpin_lock(struct lockdep_map *lock, struct pin_cookie cookie)
|
| 3939 | {
|
| 3940 | unsigned long flags;
|
| 3941 |
|
| 3942 | if (unlikely(current->lockdep_recursion))
|
| 3943 | return;
|
| 3944 |
|
| 3945 | raw_local_irq_save(flags);
|
| 3946 | check_flags(flags);
|
| 3947 |
|
| 3948 | current->lockdep_recursion = 1;
|
| 3949 | __lock_unpin_lock(lock, cookie);
|
| 3950 | current->lockdep_recursion = 0;
|
| 3951 | raw_local_irq_restore(flags);
|
| 3952 | }
|
| 3953 | EXPORT_SYMBOL_GPL(lock_unpin_lock);
|
| 3954 |
|
| 3955 | void lockdep_set_current_reclaim_state(gfp_t gfp_mask)
|
| 3956 | {
|
| 3957 | current->lockdep_reclaim_gfp = current_gfp_context(gfp_mask);
|
| 3958 | }
|
| 3959 | EXPORT_SYMBOL_GPL(lockdep_set_current_reclaim_state);
|
| 3960 |
|
| 3961 | void lockdep_clear_current_reclaim_state(void)
|
| 3962 | {
|
| 3963 | current->lockdep_reclaim_gfp = 0;
|
| 3964 | }
|
| 3965 | EXPORT_SYMBOL_GPL(lockdep_clear_current_reclaim_state);
|
| 3966 |
|
| 3967 | #ifdef CONFIG_LOCK_STAT
|
| 3968 | static int
|
| 3969 | print_lock_contention_bug(struct task_struct *curr, struct lockdep_map *lock,
|
| 3970 | unsigned long ip)
|
| 3971 | {
|
| 3972 | if (!debug_locks_off())
|
| 3973 | return 0;
|
| 3974 | if (debug_locks_silent)
|
| 3975 | return 0;
|
| 3976 |
|
| 3977 | pr_warn("\n");
|
| 3978 | pr_warn("=================================\n");
|
| 3979 | pr_warn("WARNING: bad contention detected!\n");
|
| 3980 | print_kernel_ident();
|
| 3981 | pr_warn("---------------------------------\n");
|
| 3982 | pr_warn("%s/%d is trying to contend lock (",
|
| 3983 | curr->comm, task_pid_nr(curr));
|
| 3984 | print_lockdep_cache(lock);
|
| 3985 | pr_cont(") at:\n");
|
| 3986 | print_ip_sym(ip);
|
| 3987 | pr_warn("but there are no locks held!\n");
|
| 3988 | pr_warn("\nother info that might help us debug this:\n");
|
| 3989 | lockdep_print_held_locks(curr);
|
| 3990 |
|
| 3991 | pr_warn("\nstack backtrace:\n");
|
| 3992 | dump_stack();
|
| 3993 |
|
| 3994 | return 0;
|
| 3995 | }
|
| 3996 |
|
| 3997 | static void
|
| 3998 | __lock_contended(struct lockdep_map *lock, unsigned long ip)
|
| 3999 | {
|
| 4000 | struct task_struct *curr = current;
|
| 4001 | struct held_lock *hlock;
|
| 4002 | struct lock_class_stats *stats;
|
| 4003 | unsigned int depth;
|
| 4004 | int i, contention_point, contending_point;
|
| 4005 |
|
| 4006 | depth = curr->lockdep_depth;
|
| 4007 | |
| 4008 | * Whee, we contended on this lock, except it seems we're not
|
| 4009 | * actually trying to acquire anything much at all..
|
| 4010 |
|
| 4011 | if (DEBUG_LOCKS_WARN_ON(!depth))
|
| 4012 | return;
|
| 4013 |
|
| 4014 | hlock = find_held_lock(curr, lock, depth, &i);
|
| 4015 | if (!hlock) {
|
| 4016 | print_lock_contention_bug(curr, lock, ip);
|
| 4017 | return;
|
| 4018 | }
|
| 4019 |
|
| 4020 | if (hlock->instance != lock)
|
| 4021 | return;
|
| 4022 |
|
| 4023 | hlock->waittime_stamp = lockstat_clock();
|
| 4024 |
|
| 4025 | contention_point = lock_point(hlock_class(hlock)->contention_point, ip);
|
| 4026 | contending_point = lock_point(hlock_class(hlock)->contending_point,
|
| 4027 | lock->ip);
|
| 4028 |
|
| 4029 | stats = get_lock_stats(hlock_class(hlock));
|
| 4030 | if (contention_point < LOCKSTAT_POINTS)
|
| 4031 | stats->contention_point[contention_point]++;
|
| 4032 | if (contending_point < LOCKSTAT_POINTS)
|
| 4033 | stats->contending_point[contending_point]++;
|
| 4034 | if (lock->cpu != smp_processor_id())
|
| 4035 | stats->bounces[bounce_contended + !!hlock->read]++;
|
| 4036 | put_lock_stats(stats);
|
| 4037 | }
|
| 4038 |
|
| 4039 | static void
|
| 4040 | __lock_acquired(struct lockdep_map *lock, unsigned long ip)
|
| 4041 | {
|
| 4042 | struct task_struct *curr = current;
|
| 4043 | struct held_lock *hlock;
|
| 4044 | struct lock_class_stats *stats;
|
| 4045 | unsigned int depth;
|
| 4046 | u64 now, waittime = 0;
|
| 4047 | int i, cpu;
|
| 4048 |
|
| 4049 | depth = curr->lockdep_depth;
|
| 4050 | |
| 4051 | * Yay, we acquired ownership of this lock we didn't try to
|
| 4052 | * acquire, how the heck did that happen?
|
| 4053 |
|
| 4054 | if (DEBUG_LOCKS_WARN_ON(!depth))
|
| 4055 | return;
|
| 4056 |
|
| 4057 | hlock = find_held_lock(curr, lock, depth, &i);
|
| 4058 | if (!hlock) {
|
| 4059 | print_lock_contention_bug(curr, lock, _RET_IP_);
|
| 4060 | return;
|
| 4061 | }
|
| 4062 |
|
| 4063 | if (hlock->instance != lock)
|
| 4064 | return;
|
| 4065 |
|
| 4066 | cpu = smp_processor_id();
|
| 4067 | if (hlock->waittime_stamp) {
|
| 4068 | now = lockstat_clock();
|
| 4069 | waittime = now - hlock->waittime_stamp;
|
| 4070 | hlock->holdtime_stamp = now;
|
| 4071 | }
|
| 4072 |
|
| 4073 | trace_lock_acquired(lock, ip);
|
| 4074 |
|
| 4075 | stats = get_lock_stats(hlock_class(hlock));
|
| 4076 | if (waittime) {
|
| 4077 | if (hlock->read)
|
| 4078 | lock_time_inc(&stats->read_waittime, waittime);
|
| 4079 | else
|
| 4080 | lock_time_inc(&stats->write_waittime, waittime);
|
| 4081 | }
|
| 4082 | if (lock->cpu != cpu)
|
| 4083 | stats->bounces[bounce_acquired + !!hlock->read]++;
|
| 4084 | put_lock_stats(stats);
|
| 4085 |
|
| 4086 | lock->cpu = cpu;
|
| 4087 | lock->ip = ip;
|
| 4088 | }
|
| 4089 |
|
| 4090 | void lock_contended(struct lockdep_map *lock, unsigned long ip)
|
| 4091 | {
|
| 4092 | unsigned long flags;
|
| 4093 |
|
| 4094 | if (unlikely(!lock_stat))
|
| 4095 | return;
|
| 4096 |
|
| 4097 | if (unlikely(current->lockdep_recursion))
|
| 4098 | return;
|
| 4099 |
|
| 4100 | raw_local_irq_save(flags);
|
| 4101 | check_flags(flags);
|
| 4102 | current->lockdep_recursion = 1;
|
| 4103 | trace_lock_contended(lock, ip);
|
| 4104 | __lock_contended(lock, ip);
|
| 4105 | current->lockdep_recursion = 0;
|
| 4106 | raw_local_irq_restore(flags);
|
| 4107 | }
|
| 4108 | EXPORT_SYMBOL_GPL(lock_contended);
|
| 4109 |
|
| 4110 | void lock_acquired(struct lockdep_map *lock, unsigned long ip)
|
| 4111 | {
|
| 4112 | unsigned long flags;
|
| 4113 |
|
| 4114 | if (unlikely(!lock_stat))
|
| 4115 | return;
|
| 4116 |
|
| 4117 | if (unlikely(current->lockdep_recursion))
|
| 4118 | return;
|
| 4119 |
|
| 4120 | raw_local_irq_save(flags);
|
| 4121 | check_flags(flags);
|
| 4122 | current->lockdep_recursion = 1;
|
| 4123 | __lock_acquired(lock, ip);
|
| 4124 | current->lockdep_recursion = 0;
|
| 4125 | raw_local_irq_restore(flags);
|
| 4126 | }
|
| 4127 | EXPORT_SYMBOL_GPL(lock_acquired);
|
| 4128 | #endif
|
| 4129 |
|
| 4130 | |
| 4131 | * Used by the testsuite, sanitize the validator state
|
| 4132 | * after a simulated failure:
|
| 4133 |
|
| 4134 |
|
| 4135 | void lockdep_reset(void)
|
| 4136 | {
|
| 4137 | unsigned long flags;
|
| 4138 | int i;
|
| 4139 |
|
| 4140 | raw_local_irq_save(flags);
|
| 4141 | current->curr_chain_key = 0;
|
| 4142 | current->lockdep_depth = 0;
|
| 4143 | current->lockdep_recursion = 0;
|
| 4144 | memset(current->held_locks, 0, MAX_LOCK_DEPTH*sizeof(struct held_lock));
|
| 4145 | nr_hardirq_chains = 0;
|
| 4146 | nr_softirq_chains = 0;
|
| 4147 | nr_process_chains = 0;
|
| 4148 | debug_locks = 1;
|
| 4149 | for (i = 0; i < CHAINHASH_SIZE; i++)
|
| 4150 | INIT_HLIST_HEAD(chainhash_table + i);
|
| 4151 | raw_local_irq_restore(flags);
|
| 4152 | }
|
| 4153 |
|
| 4154 | static void zap_class(struct lock_class *class)
|
| 4155 | {
|
| 4156 | int i;
|
| 4157 |
|
| 4158 | |
| 4159 | * Remove all dependencies this lock is
|
| 4160 | * involved in:
|
| 4161 |
|
| 4162 | for (i = 0; i < nr_list_entries; i++) {
|
| 4163 | if (list_entries[i].class == class)
|
| 4164 | list_del_rcu(&list_entries[i].entry);
|
| 4165 | }
|
| 4166 | |
| 4167 | * Unhash the class and remove it from the all_lock_classes list:
|
| 4168 |
|
| 4169 | hlist_del_rcu(&class->hash_entry);
|
| 4170 | list_del_rcu(&class->lock_entry);
|
| 4171 |
|
| 4172 | RCU_INIT_POINTER(class->key, NULL);
|
| 4173 | RCU_INIT_POINTER(class->name, NULL);
|
| 4174 | }
|
| 4175 |
|
| 4176 | static inline int within(const void *addr, void *start, unsigned long size)
|
| 4177 | {
|
| 4178 | return addr >= start && addr < start + size;
|
| 4179 | }
|
| 4180 |
|
| 4181 | |
| 4182 | * Used in module.c to remove lock classes from memory that is going to be
|
| 4183 | * freed; and possibly re-used by other modules.
|
| 4184 | *
|
| 4185 | * We will have had one sync_sched() before getting here, so we're guaranteed
|
| 4186 | * nobody will look up these exact classes -- they're properly dead but still
|
| 4187 | * allocated.
|
| 4188 |
|
| 4189 | void lockdep_free_key_range(void *start, unsigned long size)
|
| 4190 | {
|
| 4191 | struct lock_class *class;
|
| 4192 | struct hlist_head *head;
|
| 4193 | unsigned long flags;
|
| 4194 | int i;
|
| 4195 | int locked;
|
| 4196 |
|
| 4197 | raw_local_irq_save(flags);
|
| 4198 | locked = graph_lock();
|
| 4199 |
|
| 4200 | |
| 4201 | * Unhash all classes that were created by this module:
|
| 4202 |
|
| 4203 | for (i = 0; i < CLASSHASH_SIZE; i++) {
|
| 4204 | head = classhash_table + i;
|
| 4205 | hlist_for_each_entry_rcu(class, head, hash_entry) {
|
| 4206 | if (within(class->key, start, size))
|
| 4207 | zap_class(class);
|
| 4208 | else if (within(class->name, start, size))
|
| 4209 | zap_class(class);
|
| 4210 | }
|
| 4211 | }
|
| 4212 |
|
| 4213 | if (locked)
|
| 4214 | graph_unlock();
|
| 4215 | raw_local_irq_restore(flags);
|
| 4216 |
|
| 4217 | |
| 4218 | * Wait for any possible iterators from look_up_lock_class() to pass
|
| 4219 | * before continuing to free the memory they refer to.
|
| 4220 | *
|
| 4221 | * sync_sched() is sufficient because the read-side is IRQ disable.
|
| 4222 |
|
| 4223 | synchronize_sched();
|
| 4224 |
|
| 4225 | |
| 4226 | * XXX at this point we could return the resources to the pool;
|
| 4227 | * instead we leak them. We would need to change to bitmap allocators
|
| 4228 | * instead of the linear allocators we have now.
|
| 4229 |
|
| 4230 | }
|
| 4231 |
|
| 4232 | void lockdep_reset_lock(struct lockdep_map *lock)
|
| 4233 | {
|
| 4234 | struct lock_class *class;
|
| 4235 | struct hlist_head *head;
|
| 4236 | unsigned long flags;
|
| 4237 | int i, j;
|
| 4238 | int locked;
|
| 4239 |
|
| 4240 | raw_local_irq_save(flags);
|
| 4241 |
|
| 4242 | |
| 4243 | * Remove all classes this lock might have:
|
| 4244 |
|
| 4245 | for (j = 0; j < MAX_LOCKDEP_SUBCLASSES; j++) {
|
| 4246 | |
| 4247 | * If the class exists we look it up and zap it:
|
| 4248 |
|
| 4249 | class = look_up_lock_class(lock, j);
|
| 4250 | if (!IS_ERR_OR_NULL(class))
|
| 4251 | zap_class(class);
|
| 4252 | }
|
| 4253 | |
| 4254 | * Debug check: in the end all mapped classes should
|
| 4255 | * be gone.
|
| 4256 |
|
| 4257 | locked = graph_lock();
|
| 4258 | for (i = 0; i < CLASSHASH_SIZE; i++) {
|
| 4259 | head = classhash_table + i;
|
| 4260 | hlist_for_each_entry_rcu(class, head, hash_entry) {
|
| 4261 | int match = 0;
|
| 4262 |
|
| 4263 | for (j = 0; j < NR_LOCKDEP_CACHING_CLASSES; j++)
|
| 4264 | match |= class == lock->class_cache[j];
|
| 4265 |
|
| 4266 | if (unlikely(match)) {
|
| 4267 | if (debug_locks_off_graph_unlock()) {
|
| 4268 | |
| 4269 | * We all just reset everything, how did it match?
|
| 4270 |
|
| 4271 | WARN_ON(1);
|
| 4272 | }
|
| 4273 | goto out_restore;
|
| 4274 | }
|
| 4275 | }
|
| 4276 | }
|
| 4277 | if (locked)
|
| 4278 | graph_unlock();
|
| 4279 |
|
| 4280 | out_restore:
|
| 4281 | raw_local_irq_restore(flags);
|
| 4282 | }
|
| 4283 |
|
| 4284 | void __init lockdep_info(void)
|
| 4285 | {
|
| 4286 | printk("Lock dependency validator: Copyright (c) 2006 Red Hat, Inc., Ingo Molnar\n");
|
| 4287 |
|
| 4288 | printk("... MAX_LOCKDEP_SUBCLASSES: %lu\n", MAX_LOCKDEP_SUBCLASSES);
|
| 4289 | printk("... MAX_LOCK_DEPTH: %lu\n", MAX_LOCK_DEPTH);
|
| 4290 | printk("... MAX_LOCKDEP_KEYS: %lu\n", MAX_LOCKDEP_KEYS);
|
| 4291 | printk("... CLASSHASH_SIZE: %lu\n", CLASSHASH_SIZE);
|
| 4292 | printk("... MAX_LOCKDEP_ENTRIES: %lu\n", MAX_LOCKDEP_ENTRIES);
|
| 4293 | printk("... MAX_LOCKDEP_CHAINS: %lu\n", MAX_LOCKDEP_CHAINS);
|
| 4294 | printk("... CHAINHASH_SIZE: %lu\n", CHAINHASH_SIZE);
|
| 4295 |
|
| 4296 | printk(" memory used by lock dependency info: %lu kB\n",
|
| 4297 | (sizeof(struct lock_class) * MAX_LOCKDEP_KEYS +
|
| 4298 | sizeof(struct list_head) * CLASSHASH_SIZE +
|
| 4299 | sizeof(struct lock_list) * MAX_LOCKDEP_ENTRIES +
|
| 4300 | sizeof(struct lock_chain) * MAX_LOCKDEP_CHAINS +
|
| 4301 | sizeof(struct list_head) * CHAINHASH_SIZE
|
| 4302 | #ifdef CONFIG_PROVE_LOCKING
|
| 4303 | + sizeof(struct circular_queue)
|
| 4304 | #endif
|
| 4305 | ) / 1024
|
| 4306 | );
|
| 4307 |
|
| 4308 | printk(" per task-struct memory footprint: %lu bytes\n",
|
| 4309 | sizeof(struct held_lock) * MAX_LOCK_DEPTH);
|
| 4310 | }
|
| 4311 |
|
| 4312 | static void
|
| 4313 | print_freed_lock_bug(struct task_struct *curr, const void *mem_from,
|
| 4314 | const void *mem_to, struct held_lock *hlock)
|
| 4315 | {
|
| 4316 | if (!debug_locks_off())
|
| 4317 | return;
|
| 4318 | if (debug_locks_silent)
|
| 4319 | return;
|
| 4320 |
|
| 4321 | pr_warn("\n");
|
| 4322 | pr_warn("=========================\n");
|
| 4323 | pr_warn("WARNING: held lock freed!\n");
|
| 4324 | print_kernel_ident();
|
| 4325 | pr_warn("-------------------------\n");
|
| 4326 | pr_warn("%s/%d is freeing memory %p-%p, with a lock still held there!\n",
|
| 4327 | curr->comm, task_pid_nr(curr), mem_from, mem_to-1);
|
| 4328 | print_lock(hlock);
|
| 4329 | lockdep_print_held_locks(curr);
|
| 4330 |
|
| 4331 | pr_warn("\nstack backtrace:\n");
|
| 4332 | dump_stack();
|
| 4333 | }
|
| 4334 |
|
| 4335 | static inline int not_in_range(const void* mem_from, unsigned long mem_len,
|
| 4336 | const void* lock_from, unsigned long lock_len)
|
| 4337 | {
|
| 4338 | return lock_from + lock_len <= mem_from ||
|
| 4339 | mem_from + mem_len <= lock_from;
|
| 4340 | }
|
| 4341 |
|
| 4342 | |
| 4343 | * Called when kernel memory is freed (or unmapped), or if a lock
|
| 4344 | * is destroyed or reinitialized - this code checks whether there is
|
| 4345 | * any held lock in the memory range of <from> to <to>:
|
| 4346 |
|
| 4347 | void debug_check_no_locks_freed(const void *mem_from, unsigned long mem_len)
|
| 4348 | {
|
| 4349 | struct task_struct *curr = current;
|
| 4350 | struct held_lock *hlock;
|
| 4351 | unsigned long flags;
|
| 4352 | int i;
|
| 4353 |
|
| 4354 | if (unlikely(!debug_locks))
|
| 4355 | return;
|
| 4356 |
|
| 4357 | local_irq_save(flags);
|
| 4358 | for (i = 0; i < curr->lockdep_depth; i++) {
|
| 4359 | hlock = curr->held_locks + i;
|
| 4360 |
|
| 4361 | if (not_in_range(mem_from, mem_len, hlock->instance,
|
| 4362 | sizeof(*hlock->instance)))
|
| 4363 | continue;
|
| 4364 |
|
| 4365 | print_freed_lock_bug(curr, mem_from, mem_from + mem_len, hlock);
|
| 4366 | break;
|
| 4367 | }
|
| 4368 | local_irq_restore(flags);
|
| 4369 | }
|
| 4370 | EXPORT_SYMBOL_GPL(debug_check_no_locks_freed);
|
| 4371 |
|
| 4372 | static void print_held_locks_bug(void)
|
| 4373 | {
|
| 4374 | if (!debug_locks_off())
|
| 4375 | return;
|
| 4376 | if (debug_locks_silent)
|
| 4377 | return;
|
| 4378 |
|
| 4379 | pr_warn("\n");
|
| 4380 | pr_warn("====================================\n");
|
| 4381 | pr_warn("WARNING: %s/%d still has locks held!\n",
|
| 4382 | current->comm, task_pid_nr(current));
|
| 4383 | print_kernel_ident();
|
| 4384 | pr_warn("------------------------------------\n");
|
| 4385 | lockdep_print_held_locks(current);
|
| 4386 | pr_warn("\nstack backtrace:\n");
|
| 4387 | dump_stack();
|
| 4388 | }
|
| 4389 |
|
| 4390 | void debug_check_no_locks_held(void)
|
| 4391 | {
|
| 4392 | if (unlikely(current->lockdep_depth > 0))
|
| 4393 | print_held_locks_bug();
|
| 4394 | }
|
| 4395 | EXPORT_SYMBOL_GPL(debug_check_no_locks_held);
|
| 4396 |
|
| 4397 | #ifdef __KERNEL__
|
| 4398 | void debug_show_all_locks(void)
|
| 4399 | {
|
| 4400 | struct task_struct *g, *p;
|
| 4401 | int count = 10;
|
| 4402 | int unlock = 1;
|
| 4403 |
|
| 4404 | if (unlikely(!debug_locks)) {
|
| 4405 | pr_warn("INFO: lockdep is turned off.\n");
|
| 4406 | return;
|
| 4407 | }
|
| 4408 | pr_warn("\nShowing all locks held in the system:\n");
|
| 4409 |
|
| 4410 | |
| 4411 | * Here we try to get the tasklist_lock as hard as possible,
|
| 4412 | * if not successful after 2 seconds we ignore it (but keep
|
| 4413 | * trying). This is to enable a debug printout even if a
|
| 4414 | * tasklist_lock-holding task deadlocks or crashes.
|
| 4415 |
|
| 4416 | retry:
|
| 4417 | if (!read_trylock(&tasklist_lock)) {
|
| 4418 | if (count == 10)
|
| 4419 | pr_warn("hm, tasklist_lock locked, retrying... ");
|
| 4420 | if (count) {
|
| 4421 | count--;
|
| 4422 | pr_cont(" #%d", 10-count);
|
| 4423 | mdelay(200);
|
| 4424 | goto retry;
|
| 4425 | }
|
| 4426 | pr_cont(" ignoring it.\n");
|
| 4427 | unlock = 0;
|
| 4428 | } else {
|
| 4429 | if (count != 10)
|
| 4430 | pr_cont(" locked it.\n");
|
| 4431 | }
|
| 4432 |
|
| 4433 | do_each_thread(g, p) {
|
| 4434 | |
| 4435 | * It's not reliable to print a task's held locks
|
| 4436 | * if it's not sleeping (or if it's not the current
|
| 4437 | * task):
|
| 4438 |
|
| 4439 | if (p->state == TASK_RUNNING && p != current)
|
| 4440 | continue;
|
| 4441 | if (p->lockdep_depth)
|
| 4442 | lockdep_print_held_locks(p);
|
| 4443 | if (!unlock)
|
| 4444 | if (read_trylock(&tasklist_lock))
|
| 4445 | unlock = 1;
|
| 4446 | } while_each_thread(g, p);
|
| 4447 |
|
| 4448 | pr_warn("\n");
|
| 4449 | pr_warn("=============================================\n\n");
|
| 4450 |
|
| 4451 | if (unlock)
|
| 4452 | read_unlock(&tasklist_lock);
|
| 4453 | }
|
| 4454 | EXPORT_SYMBOL_GPL(debug_show_all_locks);
|
| 4455 | #endif
|
| 4456 |
|
| 4457 | |
| 4458 | * Careful: only use this function if you are sure that
|
| 4459 | * the task cannot run in parallel!
|
| 4460 |
|
| 4461 | void debug_show_held_locks(struct task_struct *task)
|
| 4462 | {
|
| 4463 | if (unlikely(!debug_locks)) {
|
| 4464 | printk("INFO: lockdep is turned off.\n");
|
| 4465 | return;
|
| 4466 | }
|
| 4467 | lockdep_print_held_locks(task);
|
| 4468 | }
|
| 4469 | EXPORT_SYMBOL_GPL(debug_show_held_locks);
|
| 4470 |
|
| 4471 | asmlinkage __visible void lockdep_sys_exit(void)
|
| 4472 | {
|
| 4473 | struct task_struct *curr = current;
|
| 4474 |
|
| 4475 | if (unlikely(curr->lockdep_depth)) {
|
| 4476 | if (!debug_locks_off())
|
| 4477 | return;
|
| 4478 | pr_warn("\n");
|
| 4479 | pr_warn("================================================\n");
|
| 4480 | pr_warn("WARNING: lock held when returning to user space!\n");
|
| 4481 | print_kernel_ident();
|
| 4482 | pr_warn("------------------------------------------------\n");
|
| 4483 | pr_warn("%s/%d is leaving the kernel with locks still held!\n",
|
| 4484 | curr->comm, curr->pid);
|
| 4485 | lockdep_print_held_locks(curr);
|
| 4486 | }
|
| 4487 | }
|
| 4488 |
|
| 4489 | void lockdep_rcu_suspicious(const char *file, const int line, const char *s)
|
| 4490 | {
|
| 4491 | struct task_struct *curr = current;
|
| 4492 |
|
| 4493 |
|
| 4494 | pr_warn("\n");
|
| 4495 | pr_warn("=============================\n");
|
| 4496 | pr_warn("WARNING: suspicious RCU usage\n");
|
| 4497 | print_kernel_ident();
|
| 4498 | pr_warn("-----------------------------\n");
|
| 4499 | pr_warn("%s:%d %s!\n", file, line, s);
|
| 4500 | pr_warn("\nother info that might help us debug this:\n\n");
|
| 4501 | pr_warn("\n%srcu_scheduler_active = %d, debug_locks = %d\n",
|
| 4502 | !rcu_lockdep_current_cpu_online()
|
| 4503 | ? "RCU used illegally from offline CPU!\n"
|
| 4504 | : !rcu_is_watching()
|
| 4505 | ? "RCU used illegally from idle CPU!\n"
|
| 4506 | : "",
|
| 4507 | rcu_scheduler_active, debug_locks);
|
| 4508 |
|
| 4509 | |
| 4510 | * If a CPU is in the RCU-free window in idle (ie: in the section
|
| 4511 | * between rcu_idle_enter() and rcu_idle_exit(), then RCU
|
| 4512 | * considers that CPU to be in an "extended quiescent state",
|
| 4513 | * which means that RCU will be completely ignoring that CPU.
|
| 4514 | * Therefore, rcu_read_lock() and friends have absolutely no
|
| 4515 | * effect on a CPU running in that state. In other words, even if
|
| 4516 | * such an RCU-idle CPU has called rcu_read_lock(), RCU might well
|
| 4517 | * delete data structures out from under it. RCU really has no
|
| 4518 | * choice here: we need to keep an RCU-free window in idle where
|
| 4519 | * the CPU may possibly enter into low power mode. This way we can
|
| 4520 | * notice an extended quiescent state to other CPUs that started a grace
|
| 4521 | * period. Otherwise we would delay any grace period as long as we run
|
| 4522 | * in the idle task.
|
| 4523 | *
|
| 4524 | * So complain bitterly if someone does call rcu_read_lock(),
|
| 4525 | * rcu_read_lock_bh() and so on from extended quiescent states.
|
| 4526 |
|
| 4527 | if (!rcu_is_watching())
|
| 4528 | pr_warn("RCU used illegally from extended quiescent state!\n");
|
| 4529 |
|
| 4530 | lockdep_print_held_locks(curr);
|
| 4531 | pr_warn("\nstack backtrace:\n");
|
| 4532 | dump_stack();
|
| 4533 | }
|
| 4534 | EXPORT_SYMBOL_GPL(lockdep_rcu_suspicious);
|