tools/perf/builtin-lock.c

   1 // SPDX-License-Identifier: GPL-2.0
   2 #include <errno.h>
   3 #include <inttypes.h>
   4 #include "builtin.h"
   5 #include "perf.h"
   6
   7 #include "util/evlist.h"
   8 #include "util/evsel.h"
   9 #include "util/util.h"
  10 #include "util/cache.h"
  11 #include "util/symbol.h"
  12 #include "util/thread.h"
  13 #include "util/header.h"
  14
  15 #include <subcmd/parse-options.h>
  16 #include "util/trace-event.h"
  17
  18 #include "util/debug.h"
  19 #include "util/session.h"
  20 #include "util/tool.h"
  21 #include "util/data.h"
  22
  23 #include <sys/types.h>
  24 #include <sys/prctl.h>
  25 #include <semaphore.h>
  26 #include <pthread.h>
  27 #include <math.h>
  28 #include <limits.h>
  29
  30 #include <linux/list.h>
  31 #include <linux/hash.h>
  32 #include <linux/kernel.h>
  33
  34 static struct perf_session *session;
  35
  36 /* based on kernel/lockdep.c */
  37 #define LOCKHASH_BITS           12
  38 #define LOCKHASH_SIZE           (1UL << LOCKHASH_BITS)
  39
  40 static struct list_head lockhash_table[LOCKHASH_SIZE];
  41
  42 #define __lockhashfn(key)       hash_long((unsigned long)key, LOCKHASH_BITS)
  43 #define lockhashentry(key)      (lockhash_table + __lockhashfn((key)))
  44
  45 struct lock_stat {
  46         struct list_head        hash_entry;
  47         struct rb_node          rb;             /* used for sorting */
  48
  49         /*
  50          * FIXME: perf_evsel__intval() returns u64,
  51          * so address of lockdep_map should be dealed as 64bit.
  52          * Is there more better solution?
  53          */
  54         void                    *addr;          /* address of lockdep_map, used as ID */
  55         char                    *name;          /* for strcpy(), we cannot use const */
  56
  57         unsigned int            nr_acquire;
  58         unsigned int            nr_acquired;
  59         unsigned int            nr_contended;
  60         unsigned int            nr_release;
  61
  62         unsigned int            nr_readlock;
  63         unsigned int            nr_trylock;
  64
  65         /* these times are in nano sec. */
  66         u64                     avg_wait_time;
  67         u64                     wait_time_total;
  68         u64                     wait_time_min;
  69         u64                     wait_time_max;
  70
  71         int                     discard; /* flag of blacklist */
  72 };
  73
  74 /*
  75  * States of lock_seq_stat
  76  *
  77  * UNINITIALIZED is required for detecting first event of acquire.
  78  * As the nature of lock events, there is no guarantee
  79  * that the first event for the locks are acquire,
  80  * it can be acquired, contended or release.
  81  */
  82 #define SEQ_STATE_UNINITIALIZED      0         /* initial state */
  83 #define SEQ_STATE_RELEASED      1
  84 #define SEQ_STATE_ACQUIRING     2
  85 #define SEQ_STATE_ACQUIRED      3
  86 #define SEQ_STATE_READ_ACQUIRED 4
  87 #define SEQ_STATE_CONTENDED     5
  88
  89 /*
  90  * MAX_LOCK_DEPTH
  91  * Imported from include/linux/sched.h.
  92  * Should this be synchronized?
  93  */
  94 #define MAX_LOCK_DEPTH 48
  95
  96 /*
  97  * struct lock_seq_stat:
  98  * Place to put on state of one lock sequence
  99  * 1) acquire -> acquired -> release
 100  * 2) acquire -> contended -> acquired -> release
 101  * 3) acquire (with read or try) -> release
 102  * 4) Are there other patterns?
 103  */
 104 struct lock_seq_stat {
 105         struct list_head        list;
 106         int                     state;
 107         u64                     prev_event_time;
 108         void                    *addr;
 109
 110         int                     read_count;
 111 };
 112
 113 struct thread_stat {
 114         struct rb_node          rb;
 115
 116         u32                     tid;
 117         struct list_head        seq_list;
 118 };
 119
 120 static struct rb_root           thread_stats;
 121
 122 static struct thread_stat *thread_stat_find(u32 tid)
 123 {
 124         struct rb_node *node;
 125         struct thread_stat *st;
 126
 127         node = thread_stats.rb_node;
 128         while (node) {
 129                 st = container_of(node, struct thread_stat, rb);
 130                 if (st->tid == tid)
 131                         return st;
 132                 else if (tid < st->tid)
 133                         node = node->rb_left;
 134                 else
 135                         node = node->rb_right;
 136         }
 137
 138         return NULL;
 139 }
 140
 141 static void thread_stat_insert(struct thread_stat *new)
 142 {
 143         struct rb_node **rb = &thread_stats.rb_node;
 144         struct rb_node *parent = NULL;
 145         struct thread_stat *p;
 146
 147         while (*rb) {
 148                 p = container_of(*rb, struct thread_stat, rb);
 149                 parent = *rb;
 150
 151                 if (new->tid < p->tid)
 152                         rb = &(*rb)->rb_left;
 153                 else if (new->tid > p->tid)
 154                         rb = &(*rb)->rb_right;
 155                 else
 156                         BUG_ON("inserting invalid thread_stat\n");
 157         }
 158
 159         rb_link_node(&new->rb, parent, rb);
 160         rb_insert_color(&new->rb, &thread_stats);
 161 }
 162
 163 static struct thread_stat *thread_stat_findnew_after_first(u32 tid)
 164 {
 165         struct thread_stat *st;
 166
 167         st = thread_stat_find(tid);
 168         if (st)
 169                 return st;
 170
 171         st = zalloc(sizeof(struct thread_stat));
 172         if (!st) {
 173                 pr_err("memory allocation failed\n");
 174                 return NULL;
 175         }
 176
 177         st->tid = tid;
 178         INIT_LIST_HEAD(&st->seq_list);
 179
 180         thread_stat_insert(st);
 181
 182         return st;
 183 }
 184
 185 static struct thread_stat *thread_stat_findnew_first(u32 tid);
 186 static struct thread_stat *(*thread_stat_findnew)(u32 tid) =
 187         thread_stat_findnew_first;
 188
 189 static struct thread_stat *thread_stat_findnew_first(u32 tid)
 190 {
 191         struct thread_stat *st;
 192
 193         st = zalloc(sizeof(struct thread_stat));
 194         if (!st) {
 195                 pr_err("memory allocation failed\n");
 196                 return NULL;
 197         }
 198         st->tid = tid;
 199         INIT_LIST_HEAD(&st->seq_list);
 200
 201         rb_link_node(&st->rb, NULL, &thread_stats.rb_node);
 202         rb_insert_color(&st->rb, &thread_stats);
 203
 204         thread_stat_findnew = thread_stat_findnew_after_first;
 205         return st;
 206 }
 207
 208 /* build simple key function one is bigger than two */
 209 #define SINGLE_KEY(member)                                              \
 210         static int lock_stat_key_ ## member(struct lock_stat *one,      \
 211                                          struct lock_stat *two)         \
 212         {                                                               \
 213                 return one->member > two->member;                       \
 214         }
 215
 216 SINGLE_KEY(nr_acquired)
 217 SINGLE_KEY(nr_contended)
 218 SINGLE_KEY(avg_wait_time)
 219 SINGLE_KEY(wait_time_total)
 220 SINGLE_KEY(wait_time_max)
 221
 222 static int lock_stat_key_wait_time_min(struct lock_stat *one,
 223                                         struct lock_stat *two)
 224 {
 225         u64 s1 = one->wait_time_min;
 226         u64 s2 = two->wait_time_min;
 227         if (s1 == ULLONG_MAX)
 228                 s1 = 0;
 229         if (s2 == ULLONG_MAX)
 230                 s2 = 0;
 231         return s1 > s2;
 232 }
 233
 234 struct lock_key {
 235         /*
 236          * name: the value for specify by user
 237          * this should be simpler than raw name of member
 238          * e.g. nr_acquired -> acquired, wait_time_total -> wait_total
 239          */
 240         const char              *name;
 241         int                     (*key)(struct lock_stat*, struct lock_stat*);
 242 };
 243
 244 static const char               *sort_key = "acquired";
 245
 246 static int                      (*compare)(struct lock_stat *, struct lock_stat *);
 247
 248 static struct rb_root           result; /* place to store sorted data */
 249
 250 #define DEF_KEY_LOCK(name, fn_suffix)   \
 251         { #name, lock_stat_key_ ## fn_suffix }
 252 struct lock_key keys[] = {
 253         DEF_KEY_LOCK(acquired, nr_acquired),
 254         DEF_KEY_LOCK(contended, nr_contended),
 255         DEF_KEY_LOCK(avg_wait, avg_wait_time),
 256         DEF_KEY_LOCK(wait_total, wait_time_total),
 257         DEF_KEY_LOCK(wait_min, wait_time_min),
 258         DEF_KEY_LOCK(wait_max, wait_time_max),
 259
 260         /* extra comparisons much complicated should be here */
 261
 262         { NULL, NULL }
 263 };
 264
 265 static int select_key(void)
 266 {
 267         int i;
 268
 269         for (i = 0; keys[i].name; i++) {
 270                 if (!strcmp(keys[i].name, sort_key)) {
 271                         compare = keys[i].key;
 272                         return 0;
 273                 }
 274         }
 275
 276         pr_err("Unknown compare key: %s\n", sort_key);
 277
 278         return -1;
 279 }
 280
 281 static void insert_to_result(struct lock_stat *st,
 282                              int (*bigger)(struct lock_stat *, struct lock_stat *))
 283 {
 284         struct rb_node **rb = &result.rb_node;
 285         struct rb_node *parent = NULL;
 286         struct lock_stat *p;
 287
 288         while (*rb) {
 289                 p = container_of(*rb, struct lock_stat, rb);
 290                 parent = *rb;
 291
 292                 if (bigger(st, p))
 293                         rb = &(*rb)->rb_left;
 294                 else
 295                         rb = &(*rb)->rb_right;
 296         }
 297
 298         rb_link_node(&st->rb, parent, rb);
 299         rb_insert_color(&st->rb, &result);
 300 }
 301
 302 /* returns left most element of result, and erase it */
 303 static struct lock_stat *pop_from_result(void)
 304 {
 305         struct rb_node *node = result.rb_node;
 306
 307         if (!node)
 308                 return NULL;
 309
 310         while (node->rb_left)
 311                 node = node->rb_left;
 312
 313         rb_erase(node, &result);
 314         return container_of(node, struct lock_stat, rb);
 315 }
 316
 317 static struct lock_stat *lock_stat_findnew(void *addr, const char *name)
 318 {
 319         struct list_head *entry = lockhashentry(addr);
 320         struct lock_stat *ret, *new;
 321
 322         list_for_each_entry(ret, entry, hash_entry) {
 323                 if (ret->addr == addr)
 324                         return ret;
 325         }
 326
 327         new = zalloc(sizeof(struct lock_stat));
 328         if (!new)
 329                 goto alloc_failed;
 330
 331         new->addr = addr;
 332         new->name = zalloc(sizeof(char) * strlen(name) + 1);
 333         if (!new->name) {
 334                 free(new);
 335                 goto alloc_failed;
 336         }
 337
 338         strcpy(new->name, name);
 339         new->wait_time_min = ULLONG_MAX;
 340
 341         list_add(&new->hash_entry, entry);
 342         return new;
 343
 344 alloc_failed:
 345         pr_err("memory allocation failed\n");
 346         return NULL;
 347 }
 348
 349 struct trace_lock_handler {
 350         int (*acquire_event)(struct perf_evsel *evsel,
 351                              struct perf_sample *sample);
 352
 353         int (*acquired_event)(struct perf_evsel *evsel,
 354                               struct perf_sample *sample);
 355
 356         int (*contended_event)(struct perf_evsel *evsel,
 357                                struct perf_sample *sample);
 358
 359         int (*release_event)(struct perf_evsel *evsel,
 360                              struct perf_sample *sample);
 361 };
 362
 363 static struct lock_seq_stat *get_seq(struct thread_stat *ts, void *addr)
 364 {
 365         struct lock_seq_stat *seq;
 366
 367         list_for_each_entry(seq, &ts->seq_list, list) {
 368                 if (seq->addr == addr)
 369                         return seq;
 370         }
 371
 372         seq = zalloc(sizeof(struct lock_seq_stat));
 373         if (!seq) {
 374                 pr_err("memory allocation failed\n");
 375                 return NULL;
 376         }
 377         seq->state = SEQ_STATE_UNINITIALIZED;
 378         seq->addr = addr;
 379
 380         list_add(&seq->list, &ts->seq_list);
 381         return seq;
 382 }
 383
 384 enum broken_state {
 385         BROKEN_ACQUIRE,
 386         BROKEN_ACQUIRED,
 387         BROKEN_CONTENDED,
 388         BROKEN_RELEASE,
 389         BROKEN_MAX,
 390 };
 391
 392 static int bad_hist[BROKEN_MAX];
 393
 394 enum acquire_flags {
 395         TRY_LOCK = 1,
 396         READ_LOCK = 2,
 397 };
 398
 399 static int report_lock_acquire_event(struct perf_evsel *evsel,
 400                                      struct perf_sample *sample)
 401 {
 402         void *addr;
 403         struct lock_stat *ls;
 404         struct thread_stat *ts;
 405         struct lock_seq_stat *seq;
 406         const char *name = perf_evsel__strval(evsel, sample, "name");
 407         u64 tmp = perf_evsel__intval(evsel, sample, "lockdep_addr");
 408         int flag = perf_evsel__intval(evsel, sample, "flag");
 409
 410         memcpy(&addr, &tmp, sizeof(void *));
 411
 412         ls = lock_stat_findnew(addr, name);
 413         if (!ls)
 414                 return -ENOMEM;
 415         if (ls->discard)
 416                 return 0;
 417
 418         ts = thread_stat_findnew(sample->tid);
 419         if (!ts)
 420                 return -ENOMEM;
 421
 422         seq = get_seq(ts, addr);
 423         if (!seq)
 424                 return -ENOMEM;
 425
 426         switch (seq->state) {
 427         case SEQ_STATE_UNINITIALIZED:
 428         case SEQ_STATE_RELEASED:
 429                 if (!flag) {
 430                         seq->state = SEQ_STATE_ACQUIRING;
 431                 } else {
 432                         if (flag & TRY_LOCK)
 433                                 ls->nr_trylock++;
 434                         if (flag & READ_LOCK)
 435                                 ls->nr_readlock++;
 436                         seq->state = SEQ_STATE_READ_ACQUIRED;
 437                         seq->read_count = 1;
 438                         ls->nr_acquired++;
 439                 }
 440                 break;
 441         case SEQ_STATE_READ_ACQUIRED:
 442                 if (flag & READ_LOCK) {
 443                         seq->read_count++;
 444                         ls->nr_acquired++;
 445                         goto end;
 446                 } else {
 447                         goto broken;
 448                 }
 449                 break;
 450         case SEQ_STATE_ACQUIRED:
 451         case SEQ_STATE_ACQUIRING:
 452         case SEQ_STATE_CONTENDED:
 453 broken:
 454                 /* broken lock sequence, discard it */
 455                 ls->discard = 1;
 456                 bad_hist[BROKEN_ACQUIRE]++;
 457                 list_del(&seq->list);
 458                 free(seq);
 459                 goto end;
 460         default:
 461                 BUG_ON("Unknown state of lock sequence found!\n");
 462                 break;
 463         }
 464
 465         ls->nr_acquire++;
 466         seq->prev_event_time = sample->time;
 467 end:
 468         return 0;
 469 }
 470
 471 static int report_lock_acquired_event(struct perf_evsel *evsel,
 472                                       struct perf_sample *sample)
 473 {
 474         void *addr;
 475         struct lock_stat *ls;
 476         struct thread_stat *ts;
 477         struct lock_seq_stat *seq;
 478         u64 contended_term;
 479         const char *name = perf_evsel__strval(evsel, sample, "name");
 480         u64 tmp = perf_evsel__intval(evsel, sample, "lockdep_addr");
 481
 482         memcpy(&addr, &tmp, sizeof(void *));
 483
 484         ls = lock_stat_findnew(addr, name);
 485         if (!ls)
 486                 return -ENOMEM;
 487         if (ls->discard)
 488                 return 0;
 489
 490         ts = thread_stat_findnew(sample->tid);
 491         if (!ts)
 492                 return -ENOMEM;
 493
 494         seq = get_seq(ts, addr);
 495         if (!seq)
 496                 return -ENOMEM;
 497
 498         switch (seq->state) {
 499         case SEQ_STATE_UNINITIALIZED:
 500                 /* orphan event, do nothing */
 501                 return 0;
 502         case SEQ_STATE_ACQUIRING:
 503                 break;
 504         case SEQ_STATE_CONTENDED:
 505                 contended_term = sample->time - seq->prev_event_time;
 506                 ls->wait_time_total += contended_term;
 507                 if (contended_term < ls->wait_time_min)
 508                         ls->wait_time_min = contended_term;
 509                 if (ls->wait_time_max < contended_term)
 510                         ls->wait_time_max = contended_term;
 511                 break;
 512         case SEQ_STATE_RELEASED:
 513         case SEQ_STATE_ACQUIRED:
 514         case SEQ_STATE_READ_ACQUIRED:
 515                 /* broken lock sequence, discard it */
 516                 ls->discard = 1;
 517                 bad_hist[BROKEN_ACQUIRED]++;
 518                 list_del(&seq->list);
 519                 free(seq);
 520                 goto end;
 521         default:
 522                 BUG_ON("Unknown state of lock sequence found!\n");
 523                 break;
 524         }
 525
 526         seq->state = SEQ_STATE_ACQUIRED;
 527         ls->nr_acquired++;
 528         ls->avg_wait_time = ls->nr_contended ? ls->wait_time_total/ls->nr_contended : 0;
 529         seq->prev_event_time = sample->time;
 530 end:
 531         return 0;
 532 }
 533
 534 static int report_lock_contended_event(struct perf_evsel *evsel,
 535                                        struct perf_sample *sample)
 536 {
 537         void *addr;
 538         struct lock_stat *ls;
 539         struct thread_stat *ts;
 540         struct lock_seq_stat *seq;
 541         const char *name = perf_evsel__strval(evsel, sample, "name");
 542         u64 tmp = perf_evsel__intval(evsel, sample, "lockdep_addr");
 543
 544         memcpy(&addr, &tmp, sizeof(void *));
 545
 546         ls = lock_stat_findnew(addr, name);
 547         if (!ls)
 548                 return -ENOMEM;
 549         if (ls->discard)
 550                 return 0;
 551
 552         ts = thread_stat_findnew(sample->tid);
 553         if (!ts)
 554                 return -ENOMEM;
 555
 556         seq = get_seq(ts, addr);
 557         if (!seq)
 558                 return -ENOMEM;
 559
 560         switch (seq->state) {
 561         case SEQ_STATE_UNINITIALIZED:
 562                 /* orphan event, do nothing */
 563                 return 0;
 564         case SEQ_STATE_ACQUIRING:
 565                 break;
 566         case SEQ_STATE_RELEASED:
 567         case SEQ_STATE_ACQUIRED:
 568         case SEQ_STATE_READ_ACQUIRED:
 569         case SEQ_STATE_CONTENDED:
 570                 /* broken lock sequence, discard it */
 571                 ls->discard = 1;
 572                 bad_hist[BROKEN_CONTENDED]++;
 573                 list_del(&seq->list);
 574                 free(seq);
 575                 goto end;
 576         default:
 577                 BUG_ON("Unknown state of lock sequence found!\n");
 578                 break;
 579         }
 580
 581         seq->state = SEQ_STATE_CONTENDED;
 582         ls->nr_contended++;
 583         ls->avg_wait_time = ls->wait_time_total/ls->nr_contended;
 584         seq->prev_event_time = sample->time;
 585 end:
 586         return 0;
 587 }
 588
 589 static int report_lock_release_event(struct perf_evsel *evsel,
 590                                      struct perf_sample *sample)
 591 {
 592         void *addr;
 593         struct lock_stat *ls;
 594         struct thread_stat *ts;
 595         struct lock_seq_stat *seq;
 596         const char *name = perf_evsel__strval(evsel, sample, "name");
 597         u64 tmp = perf_evsel__intval(evsel, sample, "lockdep_addr");
 598
 599         memcpy(&addr, &tmp, sizeof(void *));
 600
 601         ls = lock_stat_findnew(addr, name);
 602         if (!ls)
 603                 return -ENOMEM;
 604         if (ls->discard)
 605                 return 0;
 606
 607         ts = thread_stat_findnew(sample->tid);
 608         if (!ts)
 609                 return -ENOMEM;
 610
 611         seq = get_seq(ts, addr);
 612         if (!seq)
 613                 return -ENOMEM;
 614
 615         switch (seq->state) {
 616         case SEQ_STATE_UNINITIALIZED:
 617                 goto end;
 618         case SEQ_STATE_ACQUIRED:
 619                 break;
 620         case SEQ_STATE_READ_ACQUIRED:
 621                 seq->read_count--;
 622                 BUG_ON(seq->read_count < 0);
 623                 if (seq->read_count) {
 624                         ls->nr_release++;
 625                         goto end;
 626                 }
 627                 break;
 628         case SEQ_STATE_ACQUIRING:
 629         case SEQ_STATE_CONTENDED:
 630         case SEQ_STATE_RELEASED:
 631                 /* broken lock sequence, discard it */
 632                 ls->discard = 1;
 633                 bad_hist[BROKEN_RELEASE]++;
 634                 goto free_seq;
 635         default:
 636                 BUG_ON("Unknown state of lock sequence found!\n");
 637                 break;
 638         }
 639
 640         ls->nr_release++;
 641 free_seq:
 642         list_del(&seq->list);
 643         free(seq);
 644 end:
 645         return 0;
 646 }
 647
 648 /* lock oriented handlers */
 649 /* TODO: handlers for CPU oriented, thread oriented */
 650 static struct trace_lock_handler report_lock_ops  = {
 651         .acquire_event          = report_lock_acquire_event,
 652         .acquired_event         = report_lock_acquired_event,
 653         .contended_event        = report_lock_contended_event,
 654         .release_event          = report_lock_release_event,
 655 };
 656
 657 static struct trace_lock_handler *trace_handler;
 658
 659 static int perf_evsel__process_lock_acquire(struct perf_evsel *evsel,
 660                                              struct perf_sample *sample)
 661 {
 662         if (trace_handler->acquire_event)
 663                 return trace_handler->acquire_event(evsel, sample);
 664         return 0;
 665 }
 666
 667 static int perf_evsel__process_lock_acquired(struct perf_evsel *evsel,
 668                                               struct perf_sample *sample)
 669 {
 670         if (trace_handler->acquired_event)
 671                 return trace_handler->acquired_event(evsel, sample);
 672         return 0;
 673 }
 674
 675 static int perf_evsel__process_lock_contended(struct perf_evsel *evsel,
 676                                               struct perf_sample *sample)
 677 {
 678         if (trace_handler->contended_event)
 679                 return trace_handler->contended_event(evsel, sample);
 680         return 0;
 681 }
 682
 683 static int perf_evsel__process_lock_release(struct perf_evsel *evsel,
 684                                             struct perf_sample *sample)
 685 {
 686         if (trace_handler->release_event)
 687                 return trace_handler->release_event(evsel, sample);
 688         return 0;
 689 }
 690
 691 static void print_bad_events(int bad, int total)
 692 {
 693         /* Output for debug, this have to be removed */
 694         int i;
 695         const char *name[4] =
 696                 { "acquire", "acquired", "contended", "release" };
 697
 698         pr_info("\n=== output for debug===\n\n");
 699         pr_info("bad: %d, total: %d\n", bad, total);
 700         pr_info("bad rate: %.2f %%\n", (double)bad / (double)total * 100);
 701         pr_info("histogram of events caused bad sequence\n");
 702         for (i = 0; i < BROKEN_MAX; i++)
 703                 pr_info(" %10s: %d\n", name[i], bad_hist[i]);
 704 }
 705
 706 /* TODO: various way to print, coloring, nano or milli sec */
 707 static void print_result(void)
 708 {
 709         struct lock_stat *st;
 710         char cut_name[20];
 711         int bad, total;
 712
 713         pr_info("%20s ", "Name");
 714         pr_info("%10s ", "acquired");
 715         pr_info("%10s ", "contended");
 716
 717         pr_info("%15s ", "avg wait (ns)");
 718         pr_info("%15s ", "total wait (ns)");
 719         pr_info("%15s ", "max wait (ns)");
 720         pr_info("%15s ", "min wait (ns)");
 721
 722         pr_info("\n\n");
 723
 724         bad = total = 0;
 725         while ((st = pop_from_result())) {
 726                 total++;
 727                 if (st->discard) {
 728                         bad++;
 729                         continue;
 730                 }
 731                 bzero(cut_name, 20);
 732
 733                 if (strlen(st->name) < 16) {
 734                         /* output raw name */
 735                         pr_info("%20s ", st->name);
 736                 } else {
 737                         strncpy(cut_name, st->name, 16);
 738                         cut_name[16] = '.';
 739                         cut_name[17] = '.';
 740                         cut_name[18] = '.';
 741                         cut_name[19] = '\0';
 742                         /* cut off name for saving output style */
 743                         pr_info("%20s ", cut_name);
 744                 }
 745
 746                 pr_info("%10u ", st->nr_acquired);
 747                 pr_info("%10u ", st->nr_contended);
 748
 749                 pr_info("%15" PRIu64 " ", st->avg_wait_time);
 750                 pr_info("%15" PRIu64 " ", st->wait_time_total);
 751                 pr_info("%15" PRIu64 " ", st->wait_time_max);
 752                 pr_info("%15" PRIu64 " ", st->wait_time_min == ULLONG_MAX ?
 753                        0 : st->wait_time_min);
 754                 pr_info("\n");
 755         }
 756
 757         print_bad_events(bad, total);
 758 }
 759
 760 static bool info_threads, info_map;
 761
 762 static void dump_threads(void)
 763 {
 764         struct thread_stat *st;
 765         struct rb_node *node;
 766         struct thread *t;
 767
 768         pr_info("%10s: comm\n", "Thread ID");
 769
 770         node = rb_first(&thread_stats);
 771         while (node) {
 772                 st = container_of(node, struct thread_stat, rb);
 773                 t = perf_session__findnew(session, st->tid);
 774                 pr_info("%10d: %s\n", st->tid, thread__comm_str(t));
 775                 node = rb_next(node);
 776                 thread__put(t);
 777         };
 778 }
 779
 780 static void dump_map(void)
 781 {
 782         unsigned int i;
 783         struct lock_stat *st;
 784
 785         pr_info("Address of instance: name of class\n");
 786         for (i = 0; i < LOCKHASH_SIZE; i++) {
 787                 list_for_each_entry(st, &lockhash_table[i], hash_entry) {
 788                         pr_info(" %p: %s\n", st->addr, st->name);
 789                 }
 790         }
 791 }
 792
 793 static int dump_info(void)
 794 {
 795         int rc = 0;
 796
 797         if (info_threads)
 798                 dump_threads();
 799         else if (info_map)
 800                 dump_map();
 801         else {
 802                 rc = -1;
 803                 pr_err("Unknown type of information\n");
 804         }
 805
 806         return rc;
 807 }
 808
 809 typedef int (*tracepoint_handler)(struct perf_evsel *evsel,
 810                                   struct perf_sample *sample);
 811
 812 static int process_sample_event(struct perf_tool *tool __maybe_unused,
 813                                 union perf_event *event,
 814                                 struct perf_sample *sample,
 815                                 struct perf_evsel *evsel,
 816                                 struct machine *machine)
 817 {
 818         int err = 0;
 819         struct thread *thread = machine__findnew_thread(machine, sample->pid,
 820                                                         sample->tid);
 821
 822         if (thread == NULL) {
 823                 pr_debug("problem processing %d event, skipping it.\n",
 824                         event->header.type);
 825                 return -1;
 826         }
 827
 828         if (evsel->handler != NULL) {
 829                 tracepoint_handler f = evsel->handler;
 830                 err = f(evsel, sample);
 831         }
 832
 833         thread__put(thread);
 834
 835         return err;
 836 }
 837
 838 static void sort_result(void)
 839 {
 840         unsigned int i;
 841         struct lock_stat *st;
 842
 843         for (i = 0; i < LOCKHASH_SIZE; i++) {
 844                 list_for_each_entry(st, &lockhash_table[i], hash_entry) {
 845                         insert_to_result(st, compare);
 846                 }
 847         }
 848 }
 849
 850 static const struct perf_evsel_str_handler lock_tracepoints[] = {
 851         { "lock:lock_acquire",   perf_evsel__process_lock_acquire,   }, /* CONFIG_LOCKDEP */
 852         { "lock:lock_acquired",  perf_evsel__process_lock_acquired,  }, /* CONFIG_LOCKDEP, CONFIG_LOCK_STAT */
 853         { "lock:lock_contended", perf_evsel__process_lock_contended, }, /* CONFIG_LOCKDEP, CONFIG_LOCK_STAT */
 854         { "lock:lock_release",   perf_evsel__process_lock_release,   }, /* CONFIG_LOCKDEP */
 855 };
 856
 857 static bool force;
 858
 859 static int __cmd_report(bool display_info)
 860 {
 861         int err = -EINVAL;
 862         struct perf_tool eops = {
 863                 .sample          = process_sample_event,
 864                 .comm            = perf_event__process_comm,
 865                 .namespaces      = perf_event__process_namespaces,
 866                 .ordered_events  = true,
 867         };
 868         struct perf_data data = {
 869                 .file      = {
 870                         .path = input_name,
 871                 },
 872                 .mode      = PERF_DATA_MODE_READ,
 873                 .force     = force,
 874         };
 875
 876         session = perf_session__new(&data, false, &eops);
 877         if (!session) {
 878                 pr_err("Initializing perf session failed\n");
 879                 return -1;
 880         }
 881
 882         symbol__init(&session->header.env);
 883
 884         if (!perf_session__has_traces(session, "lock record"))
 885                 goto out_delete;
 886
 887         if (perf_session__set_tracepoints_handlers(session, lock_tracepoints)) {
 888                 pr_err("Initializing perf session tracepoint handlers failed\n");
 889                 goto out_delete;
 890         }
 891
 892         if (select_key())
 893                 goto out_delete;
 894
 895         err = perf_session__process_events(session);
 896         if (err)
 897                 goto out_delete;
 898
 899         setup_pager();
 900         if (display_info) /* used for info subcommand */
 901                 err = dump_info();
 902         else {
 903                 sort_result();
 904                 print_result();
 905         }
 906
 907 out_delete:
 908         perf_session__delete(session);
 909         return err;
 910 }
 911
 912 static int __cmd_record(int argc, const char **argv)
 913 {
 914         const char *record_args[] = {
 915                 "record", "-R", "-m", "1024", "-c", "1",
 916         };
 917         unsigned int rec_argc, i, j, ret;
 918         const char **rec_argv;
 919
 920         for (i = 0; i < ARRAY_SIZE(lock_tracepoints); i++) {
 921                 if (!is_valid_tracepoint(lock_tracepoints[i].name)) {
 922                                 pr_err("tracepoint %s is not enabled. "
 923                                        "Are CONFIG_LOCKDEP and CONFIG_LOCK_STAT enabled?\n",
 924                                        lock_tracepoints[i].name);
 925                                 return 1;
 926                 }
 927         }
 928
 929         rec_argc = ARRAY_SIZE(record_args) + argc - 1;
 930         /* factor of 2 is for -e in front of each tracepoint */
 931         rec_argc += 2 * ARRAY_SIZE(lock_tracepoints);
 932
 933         rec_argv = calloc(rec_argc + 1, sizeof(char *));
 934         if (!rec_argv)
 935                 return -ENOMEM;
 936
 937         for (i = 0; i < ARRAY_SIZE(record_args); i++)
 938                 rec_argv[i] = strdup(record_args[i]);
 939
 940         for (j = 0; j < ARRAY_SIZE(lock_tracepoints); j++) {
 941                 rec_argv[i++] = "-e";
 942                 rec_argv[i++] = strdup(lock_tracepoints[j].name);
 943         }
 944
 945         for (j = 1; j < (unsigned int)argc; j++, i++)
 946                 rec_argv[i] = argv[j];
 947
 948         BUG_ON(i != rec_argc);
 949
 950         ret = cmd_record(i, rec_argv);
 951         free(rec_argv);
 952         return ret;
 953 }
 954
 955 int cmd_lock(int argc, const char **argv)
 956 {
 957         const struct option lock_options[] = {
 958         OPT_STRING('i', "input", &input_name, "file", "input file name"),
 959         OPT_INCR('v', "verbose", &verbose, "be more verbose (show symbol address, etc)"),
 960         OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace, "dump raw trace in ASCII"),
 961         OPT_BOOLEAN('f', "force", &force, "don't complain, do it"),
 962         OPT_END()
 963         };
 964
 965         const struct option info_options[] = {
 966         OPT_BOOLEAN('t', "threads", &info_threads,
 967                     "dump thread list in perf.data"),
 968         OPT_BOOLEAN('m', "map", &info_map,
 969                     "map of lock instances (address:name table)"),
 970         OPT_PARENT(lock_options)
 971         };
 972
 973         const struct option report_options[] = {
 974         OPT_STRING('k', "key", &sort_key, "acquired",
 975                     "key for sorting (acquired / contended / avg_wait / wait_total / wait_max / wait_min)"),
 976         /* TODO: type */
 977         OPT_PARENT(lock_options)
 978         };
 979
 980         const char * const info_usage[] = {
 981                 "perf lock info [<options>]",
 982                 NULL
 983         };
 984         const char *const lock_subcommands[] = { "record", "report", "script",
 985                                                  "info", NULL };
 986         const char *lock_usage[] = {
 987                 NULL,
 988                 NULL
 989         };
 990         const char * const report_usage[] = {
 991                 "perf lock report [<options>]",
 992                 NULL
 993         };
 994         unsigned int i;
 995         int rc = 0;
 996
 997         for (i = 0; i < LOCKHASH_SIZE; i++)
 998                 INIT_LIST_HEAD(lockhash_table + i);
 999
1000         argc = parse_options_subcommand(argc, argv, lock_options, lock_subcommands,
1001                                         lock_usage, PARSE_OPT_STOP_AT_NON_OPTION);
1002         if (!argc)
1003                 usage_with_options(lock_usage, lock_options);
1004
1005         if (!strncmp(argv[0], "rec", 3)) {
1006                 return __cmd_record(argc, argv);
1007         } else if (!strncmp(argv[0], "report", 6)) {
1008                 trace_handler = &report_lock_ops;
1009                 if (argc) {
1010                         argc = parse_options(argc, argv,
1011                                              report_options, report_usage, 0);
1012                         if (argc)
1013                                 usage_with_options(report_usage, report_options);
1014                 }
1015                 rc = __cmd_report(false);
1016         } else if (!strcmp(argv[0], "script")) {
1017                 /* Aliased to 'perf script' */
1018                 return cmd_script(argc, argv);
1019         } else if (!strcmp(argv[0], "info")) {
1020                 if (argc) {
1021                         argc = parse_options(argc, argv,
1022                                              info_options, info_usage, 0);
1023                         if (argc)
1024                                 usage_with_options(info_usage, info_options);
1025                 }
1026                 /* recycling report_lock_ops */
1027                 trace_handler = &report_lock_ops;
1028                 rc = __cmd_report(true);
1029         } else {
1030                 usage_with_options(lock_usage, lock_options);
1031         }
1032
1033         return rc;
1034 }