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