tools/perf/builtin-lock.c

   1 #include "builtin.h"
   2 #include "perf.h"
   3
   4 #include "util/evlist.h"
   5 #include "util/evsel.h"
   6 #include "util/util.h"
   7 #include "util/cache.h"
   8 #include "util/symbol.h"
   9 #include "util/thread.h"
  10 #include "util/header.h"
  11
  12 #include <subcmd/parse-options.h>
  13 #include "util/trace-event.h"
  14
  15 #include "util/debug.h"
  16 #include "util/session.h"
  17 #include "util/tool.h"
  18 #include "util/data.h"
  19
  20 #include <sys/types.h>
  21 #include <sys/prctl.h>
  22 #include <semaphore.h>
  23 #include <pthread.h>
  24 #include <math.h>
  25 #include <limits.h>
  26
  27 #include <linux/list.h>
  28 #include <linux/hash.h>
  29
  30 static struct perf_session *session;
  31
  32 /* based on kernel/lockdep.c */
  33 #define LOCKHASH_BITS           12
  34 #define LOCKHASH_SIZE           (1UL << LOCKHASH_BITS)
  35
  36 static struct list_head lockhash_table[LOCKHASH_SIZE];
  37
  38 #define __lockhashfn(key)       hash_long((unsigned long)key, LOCKHASH_BITS)
  39 #define lockhashentry(key)      (lockhash_table + __lockhashfn((key)))
  40
  41 struct lock_stat {
  42         struct list_head        hash_entry;
  43         struct rb_node          rb;             /* used for sorting */
  44
  45         /*
  46          * FIXME: perf_evsel__intval() returns u64,
  47          * so address of lockdep_map should be dealed as 64bit.
  48          * Is there more better solution?
  49          */
  50         void                    *addr;          /* address of lockdep_map, used as ID */
  51         char                    *name;          /* for strcpy(), we cannot use const */
  52
  53         unsigned int            nr_acquire;
  54         unsigned int            nr_acquired;
  55         unsigned int            nr_contended;
  56         unsigned int            nr_release;
  57
  58         unsigned int            nr_readlock;
  59         unsigned int            nr_trylock;
  60
  61         /* these times are in nano sec. */
  62         u64                     avg_wait_time;
  63         u64                     wait_time_total;
  64         u64                     wait_time_min;
  65         u64                     wait_time_max;
  66
  67         int                     discard; /* flag of blacklist */
  68 };
  69
  70 /*
  71  * States of lock_seq_stat
  72  *
  73  * UNINITIALIZED is required for detecting first event of acquire.
  74  * As the nature of lock events, there is no guarantee
  75  * that the first event for the locks are acquire,
  76  * it can be acquired, contended or release.
  77  */
  78 #define SEQ_STATE_UNINITIALIZED      0         /* initial state */
  79 #define SEQ_STATE_RELEASED      1
  80 #define SEQ_STATE_ACQUIRING     2
  81 #define SEQ_STATE_ACQUIRED      3
  82 #define SEQ_STATE_READ_ACQUIRED 4
  83 #define SEQ_STATE_CONTENDED     5
  84
  85 /*
  86  * MAX_LOCK_DEPTH
  87  * Imported from include/linux/sched.h.
  88  * Should this be synchronized?
  89  */
  90 #define MAX_LOCK_DEPTH 48
  91
  92 /*
  93  * struct lock_seq_stat:
  94  * Place to put on state of one lock sequence
  95  * 1) acquire -> acquired -> release
  96  * 2) acquire -> contended -> acquired -> release
  97  * 3) acquire (with read or try) -> release
  98  * 4) Are there other patterns?
  99  */
 100 struct lock_seq_stat {
 101         struct list_head        list;
 102         int                     state;
 103         u64                     prev_event_time;
 104         void                    *addr;
 105
 106         int                     read_count;
 107 };
 108
 109 struct thread_stat {
 110         struct rb_node          rb;
 111
 112         u32                     tid;
 113         struct list_head        seq_list;
 114 };
 115
 116 static struct rb_root           thread_stats;
 117
 118 static struct thread_stat *thread_stat_find(u32 tid)
 119 {
 120         struct rb_node *node;
 121         struct thread_stat *st;
 122
 123         node = thread_stats.rb_node;
 124         while (node) {
 125                 st = container_of(node, struct thread_stat, rb);
 126                 if (st->tid == tid)
 127                         return st;
 128                 else if (tid < st->tid)
 129                         node = node->rb_left;
 130                 else
 131                         node = node->rb_right;
 132         }
 133
 134         return NULL;
 135 }
 136
 137 static void thread_stat_insert(struct thread_stat *new)
 138 {
 139         struct rb_node **rb = &thread_stats.rb_node;
 140         struct rb_node *parent = NULL;
 141         struct thread_stat *p;
 142
 143         while (*rb) {
 144                 p = container_of(*rb, struct thread_stat, rb);
 145                 parent = *rb;
 146
 147                 if (new->tid < p->tid)
 148                         rb = &(*rb)->rb_left;
 149                 else if (new->tid > p->tid)
 150                         rb = &(*rb)->rb_right;
 151                 else
 152                         BUG_ON("inserting invalid thread_stat\n");
 153         }
 154
 155         rb_link_node(&new->rb, parent, rb);
 156         rb_insert_color(&new->rb, &thread_stats);
 157 }
 158
 159 static struct thread_stat *thread_stat_findnew_after_first(u32 tid)
 160 {
 161         struct thread_stat *st;
 162
 163         st = thread_stat_find(tid);
 164         if (st)
 165                 return st;
 166
 167         st = zalloc(sizeof(struct thread_stat));
 168         if (!st) {
 169                 pr_err("memory allocation failed\n");
 170                 return NULL;
 171         }
 172
 173         st->tid = tid;
 174         INIT_LIST_HEAD(&st->seq_list);
 175
 176         thread_stat_insert(st);
 177
 178         return st;
 179 }
 180
 181 static struct thread_stat *thread_stat_findnew_first(u32 tid);
 182 static struct thread_stat *(*thread_stat_findnew)(u32 tid) =
 183         thread_stat_findnew_first;
 184
 185 static struct thread_stat *thread_stat_findnew_first(u32 tid)
 186 {
 187         struct thread_stat *st;
 188
 189         st = zalloc(sizeof(struct thread_stat));
 190         if (!st) {
 191                 pr_err("memory allocation failed\n");
 192                 return NULL;
 193         }
 194         st->tid = tid;
 195         INIT_LIST_HEAD(&st->seq_list);
 196
 197         rb_link_node(&st->rb, NULL, &thread_stats.rb_node);
 198         rb_insert_color(&st->rb, &thread_stats);
 199
 200         thread_stat_findnew = thread_stat_findnew_after_first;
 201         return st;
 202 }
 203
 204 /* build simple key function one is bigger than two */
 205 #define SINGLE_KEY(member)                                              \
 206         static int lock_stat_key_ ## member(struct lock_stat *one,      \
 207                                          struct lock_stat *two)         \
 208         {                                                               \
 209                 return one->member > two->member;                       \
 210         }
 211
 212 SINGLE_KEY(nr_acquired)
 213 SINGLE_KEY(nr_contended)
 214 SINGLE_KEY(avg_wait_time)
 215 SINGLE_KEY(wait_time_total)
 216 SINGLE_KEY(wait_time_max)
 217
 218 static int lock_stat_key_wait_time_min(struct lock_stat *one,
 219                                         struct lock_stat *two)
 220 {
 221         u64 s1 = one->wait_time_min;
 222         u64 s2 = two->wait_time_min;
 223         if (s1 == ULLONG_MAX)
 224                 s1 = 0;
 225         if (s2 == ULLONG_MAX)
 226                 s2 = 0;
 227         return s1 > s2;
 228 }
 229
 230 struct lock_key {
 231         /*
 232          * name: the value for specify by user
 233          * this should be simpler than raw name of member
 234          * e.g. nr_acquired -> acquired, wait_time_total -> wait_total
 235          */
 236         const char              *name;
 237         int                     (*key)(struct lock_stat*, struct lock_stat*);
 238 };
 239
 240 static const char               *sort_key = "acquired";
 241
 242 static int                      (*compare)(struct lock_stat *, struct lock_stat *);
 243
 244 static struct rb_root           result; /* place to store sorted data */
 245
 246 #define DEF_KEY_LOCK(name, fn_suffix)   \
 247         { #name, lock_stat_key_ ## fn_suffix }
 248 struct lock_key keys[] = {
 249         DEF_KEY_LOCK(acquired, nr_acquired),
 250         DEF_KEY_LOCK(contended, nr_contended),
 251         DEF_KEY_LOCK(avg_wait, avg_wait_time),
 252         DEF_KEY_LOCK(wait_total, wait_time_total),
 253         DEF_KEY_LOCK(wait_min, wait_time_min),
 254         DEF_KEY_LOCK(wait_max, wait_time_max),
 255
 256         /* extra comparisons much complicated should be here */
 257
 258         { NULL, NULL }
 259 };
 260
 261 static int select_key(void)
 262 {
 263         int i;
 264
 265         for (i = 0; keys[i].name; i++) {
 266                 if (!strcmp(keys[i].name, sort_key)) {
 267                         compare = keys[i].key;
 268                         return 0;
 269                 }
 270         }
 271
 272         pr_err("Unknown compare key: %s\n", sort_key);
 273
 274         return -1;
 275 }
 276
 277 static void insert_to_result(struct lock_stat *st,
 278                              int (*bigger)(struct lock_stat *, struct lock_stat *))
 279 {
 280         struct rb_node **rb = &result.rb_node;
 281         struct rb_node *parent = NULL;
 282         struct lock_stat *p;
 283
 284         while (*rb) {
 285                 p = container_of(*rb, struct lock_stat, rb);
 286                 parent = *rb;
 287
 288                 if (bigger(st, p))
 289                         rb = &(*rb)->rb_left;
 290                 else
 291                         rb = &(*rb)->rb_right;
 292         }
 293
 294         rb_link_node(&st->rb, parent, rb);
 295         rb_insert_color(&st->rb, &result);
 296 }
 297
 298 /* returns left most element of result, and erase it */
 299 static struct lock_stat *pop_from_result(void)
 300 {
 301         struct rb_node *node = result.rb_node;
 302
 303         if (!node)
 304                 return NULL;
 305
 306         while (node->rb_left)
 307                 node = node->rb_left;
 308
 309         rb_erase(node, &result);
 310         return container_of(node, struct lock_stat, rb);
 311 }
 312
 313 static struct lock_stat *lock_stat_findnew(void *addr, const char *name)
 314 {
 315         struct list_head *entry = lockhashentry(addr);
 316         struct lock_stat *ret, *new;
 317
 318         list_for_each_entry(ret, entry, hash_entry) {
 319                 if (ret->addr == addr)
 320                         return ret;
 321         }
 322
 323         new = zalloc(sizeof(struct lock_stat));
 324         if (!new)
 325                 goto alloc_failed;
 326
 327         new->addr = addr;
 328         new->name = zalloc(sizeof(char) * strlen(name) + 1);
 329         if (!new->name) {
 330                 free(new);
 331                 goto alloc_failed;
 332         }
 333
 334         strcpy(new->name, name);
 335         new->wait_time_min = ULLONG_MAX;
 336
 337         list_add(&new->hash_entry, entry);
 338         return new;
 339
 340 alloc_failed:
 341         pr_err("memory allocation failed\n");
 342         return NULL;
 343 }
 344
 345 struct trace_lock_handler {
 346         int (*acquire_event)(struct perf_evsel *evsel,
 347                              struct perf_sample *sample);
 348
 349         int (*acquired_event)(struct perf_evsel *evsel,
 350                               struct perf_sample *sample);
 351
 352         int (*contended_event)(struct perf_evsel *evsel,
 353                                struct perf_sample *sample);
 354
 355         int (*release_event)(struct perf_evsel *evsel,
 356                              struct perf_sample *sample);
 357 };
 358
 359 static struct lock_seq_stat *get_seq(struct thread_stat *ts, void *addr)
 360 {
 361         struct lock_seq_stat *seq;
 362
 363         list_for_each_entry(seq, &ts->seq_list, list) {
 364                 if (seq->addr == addr)
 365                         return seq;
 366         }
 367
 368         seq = zalloc(sizeof(struct lock_seq_stat));
 369         if (!seq) {
 370                 pr_err("memory allocation failed\n");
 371                 return NULL;
 372         }
 373         seq->state = SEQ_STATE_UNINITIALIZED;
 374         seq->addr = addr;
 375
 376         list_add(&seq->list, &ts->seq_list);
 377         return seq;
 378 }
 379
 380 enum broken_state {
 381         BROKEN_ACQUIRE,
 382         BROKEN_ACQUIRED,
 383         BROKEN_CONTENDED,
 384         BROKEN_RELEASE,
 385         BROKEN_MAX,
 386 };
 387
 388 static int bad_hist[BROKEN_MAX];
 389
 390 enum acquire_flags {
 391         TRY_LOCK = 1,
 392         READ_LOCK = 2,
 393 };
 394
 395 static int report_lock_acquire_event(struct perf_evsel *evsel,
 396                                      struct perf_sample *sample)
 397 {
 398         void *addr;
 399         struct lock_stat *ls;
 400         struct thread_stat *ts;
 401         struct lock_seq_stat *seq;
 402         const char *name = perf_evsel__strval(evsel, sample, "name");
 403         u64 tmp = perf_evsel__intval(evsel, sample, "lockdep_addr");
 404         int flag = perf_evsel__intval(evsel, sample, "flag");
 405
 406         memcpy(&addr, &tmp, sizeof(void *));
 407
 408         ls = lock_stat_findnew(addr, name);
 409         if (!ls)
 410                 return -ENOMEM;
 411         if (ls->discard)
 412                 return 0;
 413
 414         ts = thread_stat_findnew(sample->tid);
 415         if (!ts)
 416                 return -ENOMEM;
 417
 418         seq = get_seq(ts, addr);
 419         if (!seq)
 420                 return -ENOMEM;
 421
 422         switch (seq->state) {
 423         case SEQ_STATE_UNINITIALIZED:
 424         case SEQ_STATE_RELEASED:
 425                 if (!flag) {
 426                         seq->state = SEQ_STATE_ACQUIRING;
 427                 } else {
 428                         if (flag & TRY_LOCK)
 429                                 ls->nr_trylock++;
 430                         if (flag & READ_LOCK)
 431                                 ls->nr_readlock++;
 432                         seq->state = SEQ_STATE_READ_ACQUIRED;
 433                         seq->read_count = 1;
 434                         ls->nr_acquired++;
 435                 }
 436                 break;
 437         case SEQ_STATE_READ_ACQUIRED:
 438                 if (flag & READ_LOCK) {
 439                         seq->read_count++;
 440                         ls->nr_acquired++;
 441                         goto end;
 442                 } else {
 443                         goto broken;
 444                 }
 445                 break;
 446         case SEQ_STATE_ACQUIRED:
 447         case SEQ_STATE_ACQUIRING:
 448         case SEQ_STATE_CONTENDED:
 449 broken:
 450                 /* broken lock sequence, discard it */
 451                 ls->discard = 1;
 452                 bad_hist[BROKEN_ACQUIRE]++;
 453                 list_del(&seq->list);
 454                 free(seq);
 455                 goto end;
 456         default:
 457                 BUG_ON("Unknown state of lock sequence found!\n");
 458                 break;
 459         }
 460
 461         ls->nr_acquire++;
 462         seq->prev_event_time = sample->time;
 463 end:
 464         return 0;
 465 }
 466
 467 static int report_lock_acquired_event(struct perf_evsel *evsel,
 468                                       struct perf_sample *sample)
 469 {
 470         void *addr;
 471         struct lock_stat *ls;
 472         struct thread_stat *ts;
 473         struct lock_seq_stat *seq;
 474         u64 contended_term;
 475         const char *name = perf_evsel__strval(evsel, sample, "name");
 476         u64 tmp = perf_evsel__intval(evsel, sample, "lockdep_addr");
 477
 478         memcpy(&addr, &tmp, sizeof(void *));
 479
 480         ls = lock_stat_findnew(addr, name);
 481         if (!ls)
 482                 return -ENOMEM;
 483         if (ls->discard)
 484                 return 0;
 485
 486         ts = thread_stat_findnew(sample->tid);
 487         if (!ts)
 488                 return -ENOMEM;
 489
 490         seq = get_seq(ts, addr);
 491         if (!seq)
 492                 return -ENOMEM;
 493
 494         switch (seq->state) {
 495         case SEQ_STATE_UNINITIALIZED:
 496                 /* orphan event, do nothing */
 497                 return 0;
 498         case SEQ_STATE_ACQUIRING:
 499                 break;
 500         case SEQ_STATE_CONTENDED:
 501                 contended_term = sample->time - seq->prev_event_time;
 502                 ls->wait_time_total += contended_term;
 503                 if (contended_term < ls->wait_time_min)
 504                         ls->wait_time_min = contended_term;
 505                 if (ls->wait_time_max < contended_term)
 506                         ls->wait_time_max = contended_term;
 507                 break;
 508         case SEQ_STATE_RELEASED:
 509         case SEQ_STATE_ACQUIRED:
 510         case SEQ_STATE_READ_ACQUIRED:
 511                 /* broken lock sequence, discard it */
 512                 ls->discard = 1;
 513                 bad_hist[BROKEN_ACQUIRED]++;
 514                 list_del(&seq->list);
 515                 free(seq);
 516                 goto end;
 517         default:
 518                 BUG_ON("Unknown state of lock sequence found!\n");
 519                 break;
 520         }
 521
 522         seq->state = SEQ_STATE_ACQUIRED;
 523         ls->nr_acquired++;
 524         ls->avg_wait_time = ls->nr_contended ? ls->wait_time_total/ls->nr_contended : 0;
 525         seq->prev_event_time = sample->time;
 526 end:
 527         return 0;
 528 }
 529
 530 static int report_lock_contended_event(struct perf_evsel *evsel,
 531                                        struct perf_sample *sample)
 532 {
 533         void *addr;
 534         struct lock_stat *ls;
 535         struct thread_stat *ts;
 536         struct lock_seq_stat *seq;
 537         const char *name = perf_evsel__strval(evsel, sample, "name");
 538         u64 tmp = perf_evsel__intval(evsel, sample, "lockdep_addr");
 539
 540         memcpy(&addr, &tmp, sizeof(void *));
 541
 542         ls = lock_stat_findnew(addr, name);
 543         if (!ls)
 544                 return -ENOMEM;
 545         if (ls->discard)
 546                 return 0;
 547
 548         ts = thread_stat_findnew(sample->tid);
 549         if (!ts)
 550                 return -ENOMEM;
 551
 552         seq = get_seq(ts, addr);
 553         if (!seq)
 554                 return -ENOMEM;
 555
 556         switch (seq->state) {
 557         case SEQ_STATE_UNINITIALIZED:
 558                 /* orphan event, do nothing */
 559                 return 0;
 560         case SEQ_STATE_ACQUIRING:
 561                 break;
 562         case SEQ_STATE_RELEASED:
 563         case SEQ_STATE_ACQUIRED:
 564         case SEQ_STATE_READ_ACQUIRED:
 565         case SEQ_STATE_CONTENDED:
 566                 /* broken lock sequence, discard it */
 567                 ls->discard = 1;
 568                 bad_hist[BROKEN_CONTENDED]++;
 569                 list_del(&seq->list);
 570                 free(seq);
 571                 goto end;
 572         default:
 573                 BUG_ON("Unknown state of lock sequence found!\n");
 574                 break;
 575         }
 576
 577         seq->state = SEQ_STATE_CONTENDED;
 578         ls->nr_contended++;
 579         ls->avg_wait_time = ls->wait_time_total/ls->nr_contended;
 580         seq->prev_event_time = sample->time;
 581 end:
 582         return 0;
 583 }
 584
 585 static int report_lock_release_event(struct perf_evsel *evsel,
 586                                      struct perf_sample *sample)
 587 {
 588         void *addr;
 589         struct lock_stat *ls;
 590         struct thread_stat *ts;
 591         struct lock_seq_stat *seq;
 592         const char *name = perf_evsel__strval(evsel, sample, "name");
 593         u64 tmp = perf_evsel__intval(evsel, sample, "lockdep_addr");
 594
 595         memcpy(&addr, &tmp, sizeof(void *));
 596
 597         ls = lock_stat_findnew(addr, name);
 598         if (!ls)
 599                 return -ENOMEM;
 600         if (ls->discard)
 601                 return 0;
 602
 603         ts = thread_stat_findnew(sample->tid);
 604         if (!ts)
 605                 return -ENOMEM;
 606
 607         seq = get_seq(ts, addr);
 608         if (!seq)
 609                 return -ENOMEM;
 610
 611         switch (seq->state) {
 612         case SEQ_STATE_UNINITIALIZED:
 613                 goto end;
 614         case SEQ_STATE_ACQUIRED:
 615                 break;
 616         case SEQ_STATE_READ_ACQUIRED:
 617                 seq->read_count--;
 618                 BUG_ON(seq->read_count < 0);
 619                 if (!seq->read_count) {
 620                         ls->nr_release++;
 621                         goto end;
 622                 }
 623                 break;
 624         case SEQ_STATE_ACQUIRING:
 625         case SEQ_STATE_CONTENDED:
 626         case SEQ_STATE_RELEASED:
 627                 /* broken lock sequence, discard it */
 628                 ls->discard = 1;
 629                 bad_hist[BROKEN_RELEASE]++;
 630                 goto free_seq;
 631         default:
 632                 BUG_ON("Unknown state of lock sequence found!\n");
 633                 break;
 634         }
 635
 636         ls->nr_release++;
 637 free_seq:
 638         list_del(&seq->list);
 639         free(seq);
 640 end:
 641         return 0;
 642 }
 643
 644 /* lock oriented handlers */
 645 /* TODO: handlers for CPU oriented, thread oriented */
 646 static struct trace_lock_handler report_lock_ops  = {
 647         .acquire_event          = report_lock_acquire_event,
 648         .acquired_event         = report_lock_acquired_event,
 649         .contended_event        = report_lock_contended_event,
 650         .release_event          = report_lock_release_event,
 651 };
 652
 653 static struct trace_lock_handler *trace_handler;
 654
 655 static int perf_evsel__process_lock_acquire(struct perf_evsel *evsel,
 656                                              struct perf_sample *sample)
 657 {
 658         if (trace_handler->acquire_event)
 659                 return trace_handler->acquire_event(evsel, sample);
 660         return 0;
 661 }
 662
 663 static int perf_evsel__process_lock_acquired(struct perf_evsel *evsel,
 664                                               struct perf_sample *sample)
 665 {
 666         if (trace_handler->acquired_event)
 667                 return trace_handler->acquired_event(evsel, sample);
 668         return 0;
 669 }
 670
 671 static int perf_evsel__process_lock_contended(struct perf_evsel *evsel,
 672                                               struct perf_sample *sample)
 673 {
 674         if (trace_handler->contended_event)
 675                 return trace_handler->contended_event(evsel, sample);
 676         return 0;
 677 }
 678
 679 static int perf_evsel__process_lock_release(struct perf_evsel *evsel,
 680                                             struct perf_sample *sample)
 681 {
 682         if (trace_handler->release_event)
 683                 return trace_handler->release_event(evsel, sample);
 684         return 0;
 685 }
 686
 687 static void print_bad_events(int bad, int total)
 688 {
 689         /* Output for debug, this have to be removed */
 690         int i;
 691         const char *name[4] =
 692                 { "acquire", "acquired", "contended", "release" };
 693
 694         pr_info("\n=== output for debug===\n\n");
 695         pr_info("bad: %d, total: %d\n", bad, total);
 696         pr_info("bad rate: %.2f %%\n", (double)bad / (double)total * 100);
 697         pr_info("histogram of events caused bad sequence\n");
 698         for (i = 0; i < BROKEN_MAX; i++)
 699                 pr_info(" %10s: %d\n", name[i], bad_hist[i]);
 700 }
 701
 702 /* TODO: various way to print, coloring, nano or milli sec */
 703 static void print_result(void)
 704 {
 705         struct lock_stat *st;
 706         char cut_name[20];
 707         int bad, total;
 708
 709         pr_info("%20s ", "Name");
 710         pr_info("%10s ", "acquired");
 711         pr_info("%10s ", "contended");
 712
 713         pr_info("%15s ", "avg wait (ns)");
 714         pr_info("%15s ", "total wait (ns)");
 715         pr_info("%15s ", "max wait (ns)");
 716         pr_info("%15s ", "min wait (ns)");
 717
 718         pr_info("\n\n");
 719
 720         bad = total = 0;
 721         while ((st = pop_from_result())) {
 722                 total++;
 723                 if (st->discard) {
 724                         bad++;
 725                         continue;
 726                 }
 727                 bzero(cut_name, 20);
 728
 729                 if (strlen(st->name) < 16) {
 730                         /* output raw name */
 731                         pr_info("%20s ", st->name);
 732                 } else {
 733                         strncpy(cut_name, st->name, 16);
 734                         cut_name[16] = '.';
 735                         cut_name[17] = '.';
 736                         cut_name[18] = '.';
 737                         cut_name[19] = '\0';
 738                         /* cut off name for saving output style */
 739                         pr_info("%20s ", cut_name);
 740                 }
 741
 742                 pr_info("%10u ", st->nr_acquired);
 743                 pr_info("%10u ", st->nr_contended);
 744
 745                 pr_info("%15" PRIu64 " ", st->avg_wait_time);
 746                 pr_info("%15" PRIu64 " ", st->wait_time_total);
 747                 pr_info("%15" PRIu64 " ", st->wait_time_max);
 748                 pr_info("%15" PRIu64 " ", st->wait_time_min == ULLONG_MAX ?
 749                        0 : st->wait_time_min);
 750                 pr_info("\n");
 751         }
 752
 753         print_bad_events(bad, total);
 754 }
 755
 756 static bool info_threads, info_map;
 757
 758 static void dump_threads(void)
 759 {
 760         struct thread_stat *st;
 761         struct rb_node *node;
 762         struct thread *t;
 763
 764         pr_info("%10s: comm\n", "Thread ID");
 765
 766         node = rb_first(&thread_stats);
 767         while (node) {
 768                 st = container_of(node, struct thread_stat, rb);
 769                 t = perf_session__findnew(session, st->tid);
 770                 pr_info("%10d: %s\n", st->tid, thread__comm_str(t));
 771                 node = rb_next(node);
 772                 thread__put(t);
 773         };
 774 }
 775
 776 static void dump_map(void)
 777 {
 778         unsigned int i;
 779         struct lock_stat *st;
 780
 781         pr_info("Address of instance: name of class\n");
 782         for (i = 0; i < LOCKHASH_SIZE; i++) {
 783                 list_for_each_entry(st, &lockhash_table[i], hash_entry) {
 784                         pr_info(" %p: %s\n", st->addr, st->name);
 785                 }
 786         }
 787 }
 788
 789 static int dump_info(void)
 790 {
 791         int rc = 0;
 792
 793         if (info_threads)
 794                 dump_threads();
 795         else if (info_map)
 796                 dump_map();
 797         else {
 798                 rc = -1;
 799                 pr_err("Unknown type of information\n");
 800         }
 801
 802         return rc;
 803 }
 804
 805 typedef int (*tracepoint_handler)(struct perf_evsel *evsel,
 806                                   struct perf_sample *sample);
 807
 808 static int process_sample_event(struct perf_tool *tool __maybe_unused,
 809                                 union perf_event *event,
 810                                 struct perf_sample *sample,
 811                                 struct perf_evsel *evsel,
 812                                 struct machine *machine)
 813 {
 814         int err = 0;
 815         struct thread *thread = machine__findnew_thread(machine, sample->pid,
 816                                                         sample->tid);
 817
 818         if (thread == NULL) {
 819                 pr_debug("problem processing %d event, skipping it.\n",
 820                         event->header.type);
 821                 return -1;
 822         }
 823
 824         if (evsel->handler != NULL) {
 825                 tracepoint_handler f = evsel->handler;
 826                 err = f(evsel, sample);
 827         }
 828
 829         thread__put(thread);
 830
 831         return err;
 832 }
 833
 834 static void sort_result(void)
 835 {
 836         unsigned int i;
 837         struct lock_stat *st;
 838
 839         for (i = 0; i < LOCKHASH_SIZE; i++) {
 840                 list_for_each_entry(st, &lockhash_table[i], hash_entry) {
 841                         insert_to_result(st, compare);
 842                 }
 843         }
 844 }
 845
 846 static const struct perf_evsel_str_handler lock_tracepoints[] = {
 847         { "lock:lock_acquire",   perf_evsel__process_lock_acquire,   }, /* CONFIG_LOCKDEP */
 848         { "lock:lock_acquired",  perf_evsel__process_lock_acquired,  }, /* CONFIG_LOCKDEP, CONFIG_LOCK_STAT */
 849         { "lock:lock_contended", perf_evsel__process_lock_contended, }, /* CONFIG_LOCKDEP, CONFIG_LOCK_STAT */
 850         { "lock:lock_release",   perf_evsel__process_lock_release,   }, /* CONFIG_LOCKDEP */
 851 };
 852
 853 static bool force;
 854
 855 static int __cmd_report(bool display_info)
 856 {
 857         int err = -EINVAL;
 858         struct perf_tool eops = {
 859                 .sample          = process_sample_event,
 860                 .comm            = perf_event__process_comm,
 861                 .ordered_events  = true,
 862         };
 863         struct perf_data_file file = {
 864                 .path = input_name,
 865                 .mode = PERF_DATA_MODE_READ,
 866                 .force = force,
 867         };
 868
 869         session = perf_session__new(&file, false, &eops);
 870         if (!session) {
 871                 pr_err("Initializing perf session failed\n");
 872                 return -1;
 873         }
 874
 875         symbol__init(&session->header.env);
 876
 877         if (!perf_session__has_traces(session, "lock record"))
 878                 goto out_delete;
 879
 880         if (perf_session__set_tracepoints_handlers(session, lock_tracepoints)) {
 881                 pr_err("Initializing perf session tracepoint handlers failed\n");
 882                 goto out_delete;
 883         }
 884
 885         if (select_key())
 886                 goto out_delete;
 887
 888         err = perf_session__process_events(session);
 889         if (err)
 890                 goto out_delete;
 891
 892         setup_pager();
 893         if (display_info) /* used for info subcommand */
 894                 err = dump_info();
 895         else {
 896                 sort_result();
 897                 print_result();
 898         }
 899
 900 out_delete:
 901         perf_session__delete(session);
 902         return err;
 903 }
 904
 905 static int __cmd_record(int argc, const char **argv)
 906 {
 907         const char *record_args[] = {
 908                 "record", "-R", "-m", "1024", "-c", "1",
 909         };
 910         unsigned int rec_argc, i, j, ret;
 911         const char **rec_argv;
 912
 913         for (i = 0; i < ARRAY_SIZE(lock_tracepoints); i++) {
 914                 if (!is_valid_tracepoint(lock_tracepoints[i].name)) {
 915                                 pr_err("tracepoint %s is not enabled. "
 916                                        "Are CONFIG_LOCKDEP and CONFIG_LOCK_STAT enabled?\n",
 917                                        lock_tracepoints[i].name);
 918                                 return 1;
 919                 }
 920         }
 921
 922         rec_argc = ARRAY_SIZE(record_args) + argc - 1;
 923         /* factor of 2 is for -e in front of each tracepoint */
 924         rec_argc += 2 * ARRAY_SIZE(lock_tracepoints);
 925
 926         rec_argv = calloc(rec_argc + 1, sizeof(char *));
 927         if (!rec_argv)
 928                 return -ENOMEM;
 929
 930         for (i = 0; i < ARRAY_SIZE(record_args); i++)
 931                 rec_argv[i] = strdup(record_args[i]);
 932
 933         for (j = 0; j < ARRAY_SIZE(lock_tracepoints); j++) {
 934                 rec_argv[i++] = "-e";
 935                 rec_argv[i++] = strdup(lock_tracepoints[j].name);
 936         }
 937
 938         for (j = 1; j < (unsigned int)argc; j++, i++)
 939                 rec_argv[i] = argv[j];
 940
 941         BUG_ON(i != rec_argc);
 942
 943         ret = cmd_record(i, rec_argv, NULL);
 944         free(rec_argv);
 945         return ret;
 946 }
 947
 948 int cmd_lock(int argc, const char **argv, const char *prefix __maybe_unused)
 949 {
 950         const struct option info_options[] = {
 951         OPT_BOOLEAN('t', "threads", &info_threads,
 952                     "dump thread list in perf.data"),
 953         OPT_BOOLEAN('m', "map", &info_map,
 954                     "map of lock instances (address:name table)"),
 955         OPT_BOOLEAN('f', "force", &force, "don't complain, do it"),
 956         OPT_END()
 957         };
 958         const struct option lock_options[] = {
 959         OPT_STRING('i', "input", &input_name, "file", "input file name"),
 960         OPT_INCR('v', "verbose", &verbose, "be more verbose (show symbol address, etc)"),
 961         OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace, "dump raw trace in ASCII"),
 962         OPT_END()
 963         };
 964         const struct option report_options[] = {
 965         OPT_STRING('k', "key", &sort_key, "acquired",
 966                     "key for sorting (acquired / contended / avg_wait / wait_total / wait_max / wait_min)"),
 967         OPT_BOOLEAN('f', "force", &force, "don't complain, do it"),
 968         /* TODO: type */
 969         OPT_END()
 970         };
 971         const char * const info_usage[] = {
 972                 "perf lock info [<options>]",
 973                 NULL
 974         };
 975         const char *const lock_subcommands[] = { "record", "report", "script",
 976                                                  "info", NULL };
 977         const char *lock_usage[] = {
 978                 NULL,
 979                 NULL
 980         };
 981         const char * const report_usage[] = {
 982                 "perf lock report [<options>]",
 983                 NULL
 984         };
 985         unsigned int i;
 986         int rc = 0;
 987
 988         for (i = 0; i < LOCKHASH_SIZE; i++)
 989                 INIT_LIST_HEAD(lockhash_table + i);
 990
 991         argc = parse_options_subcommand(argc, argv, lock_options, lock_subcommands,
 992                                         lock_usage, PARSE_OPT_STOP_AT_NON_OPTION);
 993         if (!argc)
 994                 usage_with_options(lock_usage, lock_options);
 995
 996         if (!strncmp(argv[0], "rec", 3)) {
 997                 return __cmd_record(argc, argv);
 998         } else if (!strncmp(argv[0], "report", 6)) {
 999                 trace_handler = &report_lock_ops;
1000                 if (argc) {
1001                         argc = parse_options(argc, argv,
1002                                              report_options, report_usage, 0);
1003                         if (argc)
1004                                 usage_with_options(report_usage, report_options);
1005                 }
1006                 rc = __cmd_report(false);
1007         } else if (!strcmp(argv[0], "script")) {
1008                 /* Aliased to 'perf script' */
1009                 return cmd_script(argc, argv, prefix);
1010         } else if (!strcmp(argv[0], "info")) {
1011                 if (argc) {
1012                         argc = parse_options(argc, argv,
1013                                              info_options, info_usage, 0);
1014                         if (argc)
1015                                 usage_with_options(info_usage, info_options);
1016                 }
1017                 /* recycling report_lock_ops */
1018                 trace_handler = &report_lock_ops;
1019                 rc = __cmd_report(true);
1020         } else {
1021                 usage_with_options(lock_usage, lock_options);
1022         }
1023
1024         return rc;
1025 }