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" // for struct evsel_str_handler
   8 #include "util/evsel.h"
   9 #include "util/symbol.h"
  10 #include "util/thread.h"
  11 #include "util/header.h"
  12
  13 #include <subcmd/pager.h>
  14 #include <subcmd/parse-options.h>
  15 #include "util/trace-event.h"
  16
  17 #include "util/debug.h"
  18 #include "util/session.h"
  19 #include "util/tool.h"
  20 #include "util/data.h"
  21 #include "util/string2.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 #include <linux/zalloc.h>
  34 #include <linux/err.h>
  35
  36 static struct perf_session *session;
  37
  38 /* based on kernel/lockdep.c */
  39 #define LOCKHASH_BITS           12
  40 #define LOCKHASH_SIZE           (1UL << LOCKHASH_BITS)
  41
  42 static struct hlist_head lockhash_table[LOCKHASH_SIZE];
  43
  44 #define __lockhashfn(key)       hash_long((unsigned long)key, LOCKHASH_BITS)
  45 #define lockhashentry(key)      (lockhash_table + __lockhashfn((key)))
  46
  47 struct lock_stat {
  48         struct hlist_node       hash_entry;
  49         struct rb_node          rb;             /* used for sorting */
  50
  51         u64                     addr;           /* address of lockdep_map, used as ID */
  52         char                    *name;          /* for strcpy(), we cannot use const */
  53
  54         unsigned int            nr_acquire;
  55         unsigned int            nr_acquired;
  56         unsigned int            nr_contended;
  57         unsigned int            nr_release;
  58
  59         unsigned int            nr_readlock;
  60         unsigned int            nr_trylock;
  61
  62         /* these times are in nano sec. */
  63         u64                     avg_wait_time;
  64         u64                     wait_time_total;
  65         u64                     wait_time_min;
  66         u64                     wait_time_max;
  67
  68         int                     broken; /* flag of blacklist */
  69         int                     combined;
  70 };
  71
  72 /*
  73  * States of lock_seq_stat
  74  *
  75  * UNINITIALIZED is required for detecting first event of acquire.
  76  * As the nature of lock events, there is no guarantee
  77  * that the first event for the locks are acquire,
  78  * it can be acquired, contended or release.
  79  */
  80 #define SEQ_STATE_UNINITIALIZED      0         /* initial state */
  81 #define SEQ_STATE_RELEASED      1
  82 #define SEQ_STATE_ACQUIRING     2
  83 #define SEQ_STATE_ACQUIRED      3
  84 #define SEQ_STATE_READ_ACQUIRED 4
  85 #define SEQ_STATE_CONTENDED     5
  86
  87 /*
  88  * MAX_LOCK_DEPTH
  89  * Imported from include/linux/sched.h.
  90  * Should this be synchronized?
  91  */
  92 #define MAX_LOCK_DEPTH 48
  93
  94 /*
  95  * struct lock_seq_stat:
  96  * Place to put on state of one lock sequence
  97  * 1) acquire -> acquired -> release
  98  * 2) acquire -> contended -> acquired -> release
  99  * 3) acquire (with read or try) -> release
 100  * 4) Are there other patterns?
 101  */
 102 struct lock_seq_stat {
 103         struct list_head        list;
 104         int                     state;
 105         u64                     prev_event_time;
 106         u64                     addr;
 107
 108         int                     read_count;
 109 };
 110
 111 struct thread_stat {
 112         struct rb_node          rb;
 113
 114         u32                     tid;
 115         struct list_head        seq_list;
 116 };
 117
 118 static struct rb_root           thread_stats;
 119
 120 static bool combine_locks;
 121 static bool show_thread_stats;
 122
 123 static struct thread_stat *thread_stat_find(u32 tid)
 124 {
 125         struct rb_node *node;
 126         struct thread_stat *st;
 127
 128         node = thread_stats.rb_node;
 129         while (node) {
 130                 st = container_of(node, struct thread_stat, rb);
 131                 if (st->tid == tid)
 132                         return st;
 133                 else if (tid < st->tid)
 134                         node = node->rb_left;
 135                 else
 136                         node = node->rb_right;
 137         }
 138
 139         return NULL;
 140 }
 141
 142 static void thread_stat_insert(struct thread_stat *new)
 143 {
 144         struct rb_node **rb = &thread_stats.rb_node;
 145         struct rb_node *parent = NULL;
 146         struct thread_stat *p;
 147
 148         while (*rb) {
 149                 p = container_of(*rb, struct thread_stat, rb);
 150                 parent = *rb;
 151
 152                 if (new->tid < p->tid)
 153                         rb = &(*rb)->rb_left;
 154                 else if (new->tid > p->tid)
 155                         rb = &(*rb)->rb_right;
 156                 else
 157                         BUG_ON("inserting invalid thread_stat\n");
 158         }
 159
 160         rb_link_node(&new->rb, parent, rb);
 161         rb_insert_color(&new->rb, &thread_stats);
 162 }
 163
 164 static struct thread_stat *thread_stat_findnew_after_first(u32 tid)
 165 {
 166         struct thread_stat *st;
 167
 168         st = thread_stat_find(tid);
 169         if (st)
 170                 return st;
 171
 172         st = zalloc(sizeof(struct thread_stat));
 173         if (!st) {
 174                 pr_err("memory allocation failed\n");
 175                 return NULL;
 176         }
 177
 178         st->tid = tid;
 179         INIT_LIST_HEAD(&st->seq_list);
 180
 181         thread_stat_insert(st);
 182
 183         return st;
 184 }
 185
 186 static struct thread_stat *thread_stat_findnew_first(u32 tid);
 187 static struct thread_stat *(*thread_stat_findnew)(u32 tid) =
 188         thread_stat_findnew_first;
 189
 190 static struct thread_stat *thread_stat_findnew_first(u32 tid)
 191 {
 192         struct thread_stat *st;
 193
 194         st = zalloc(sizeof(struct thread_stat));
 195         if (!st) {
 196                 pr_err("memory allocation failed\n");
 197                 return NULL;
 198         }
 199         st->tid = tid;
 200         INIT_LIST_HEAD(&st->seq_list);
 201
 202         rb_link_node(&st->rb, NULL, &thread_stats.rb_node);
 203         rb_insert_color(&st->rb, &thread_stats);
 204
 205         thread_stat_findnew = thread_stat_findnew_after_first;
 206         return st;
 207 }
 208
 209 /* build simple key function one is bigger than two */
 210 #define SINGLE_KEY(member)                                              \
 211         static int lock_stat_key_ ## member(struct lock_stat *one,      \
 212                                          struct lock_stat *two)         \
 213         {                                                               \
 214                 return one->member > two->member;                       \
 215         }
 216
 217 SINGLE_KEY(nr_acquired)
 218 SINGLE_KEY(nr_contended)
 219 SINGLE_KEY(avg_wait_time)
 220 SINGLE_KEY(wait_time_total)
 221 SINGLE_KEY(wait_time_max)
 222
 223 static int lock_stat_key_wait_time_min(struct lock_stat *one,
 224                                         struct lock_stat *two)
 225 {
 226         u64 s1 = one->wait_time_min;
 227         u64 s2 = two->wait_time_min;
 228         if (s1 == ULLONG_MAX)
 229                 s1 = 0;
 230         if (s2 == ULLONG_MAX)
 231                 s2 = 0;
 232         return s1 > s2;
 233 }
 234
 235 struct lock_key {
 236         /*
 237          * name: the value for specify by user
 238          * this should be simpler than raw name of member
 239          * e.g. nr_acquired -> acquired, wait_time_total -> wait_total
 240          */
 241         const char              *name;
 242         /* header: the string printed on the header line */
 243         const char              *header;
 244         /* len: the printing width of the field */
 245         int                     len;
 246         /* key: a pointer to function to compare two lock stats for sorting */
 247         int                     (*key)(struct lock_stat*, struct lock_stat*);
 248         /* print: a pointer to function to print a given lock stats */
 249         void                    (*print)(struct lock_key*, struct lock_stat*);
 250         /* list: list entry to link this */
 251         struct list_head        list;
 252 };
 253
 254 #define PRINT_KEY(member)                                               \
 255 static void lock_stat_key_print_ ## member(struct lock_key *key,        \
 256                                            struct lock_stat *ls)        \
 257 {                                                                       \
 258         pr_info("%*llu", key->len, (unsigned long long)ls->member);     \
 259 }
 260
 261 PRINT_KEY(nr_acquired)
 262 PRINT_KEY(nr_contended)
 263 PRINT_KEY(avg_wait_time)
 264 PRINT_KEY(wait_time_total)
 265 PRINT_KEY(wait_time_max)
 266
 267 static void lock_stat_key_print_wait_time_min(struct lock_key *key,
 268                                               struct lock_stat *ls)
 269 {
 270         u64 wait_time = ls->wait_time_min;
 271
 272         if (wait_time == ULLONG_MAX)
 273                 wait_time = 0;
 274
 275         pr_info("%*"PRIu64, key->len, wait_time);
 276 }
 277
 278
 279 static const char               *sort_key = "acquired";
 280
 281 static int                      (*compare)(struct lock_stat *, struct lock_stat *);
 282
 283 static struct rb_root           sorted; /* place to store intermediate data */
 284 static struct rb_root           result; /* place to store sorted data */
 285
 286 static LIST_HEAD(lock_keys);
 287 static const char               *output_fields;
 288
 289 #define DEF_KEY_LOCK(name, header, fn_suffix, len)                      \
 290         { #name, header, len, lock_stat_key_ ## fn_suffix, lock_stat_key_print_ ## fn_suffix, {} }
 291 struct lock_key keys[] = {
 292         DEF_KEY_LOCK(acquired, "acquired", nr_acquired, 10),
 293         DEF_KEY_LOCK(contended, "contended", nr_contended, 10),
 294         DEF_KEY_LOCK(avg_wait, "avg wait (ns)", avg_wait_time, 15),
 295         DEF_KEY_LOCK(wait_total, "total wait (ns)", wait_time_total, 15),
 296         DEF_KEY_LOCK(wait_max, "max wait (ns)", wait_time_max, 15),
 297         DEF_KEY_LOCK(wait_min, "min wait (ns)", wait_time_min, 15),
 298
 299         /* extra comparisons much complicated should be here */
 300         { }
 301 };
 302
 303 static int select_key(void)
 304 {
 305         int i;
 306
 307         for (i = 0; keys[i].name; i++) {
 308                 if (!strcmp(keys[i].name, sort_key)) {
 309                         compare = keys[i].key;
 310
 311                         /* selected key should be in the output fields */
 312                         if (list_empty(&keys[i].list))
 313                                 list_add_tail(&keys[i].list, &lock_keys);
 314
 315                         return 0;
 316                 }
 317         }
 318
 319         pr_err("Unknown compare key: %s\n", sort_key);
 320         return -1;
 321 }
 322
 323 static int add_output_field(struct list_head *head, char *name)
 324 {
 325         int i;
 326
 327         for (i = 0; keys[i].name; i++) {
 328                 if (strcmp(keys[i].name, name))
 329                         continue;
 330
 331                 /* prevent double link */
 332                 if (list_empty(&keys[i].list))
 333                         list_add_tail(&keys[i].list, head);
 334
 335                 return 0;
 336         }
 337
 338         pr_err("Unknown output field: %s\n", name);
 339         return -1;
 340 }
 341
 342 static int setup_output_field(const char *str)
 343 {
 344         char *tok, *tmp, *orig;
 345         int i, ret = 0;
 346
 347         /* no output field given: use all of them */
 348         if (str == NULL) {
 349                 for (i = 0; keys[i].name; i++)
 350                         list_add_tail(&keys[i].list, &lock_keys);
 351                 return 0;
 352         }
 353
 354         for (i = 0; keys[i].name; i++)
 355                 INIT_LIST_HEAD(&keys[i].list);
 356
 357         orig = tmp = strdup(str);
 358         if (orig == NULL)
 359                 return -ENOMEM;
 360
 361         while ((tok = strsep(&tmp, ",")) != NULL){
 362                 ret = add_output_field(&lock_keys, tok);
 363                 if (ret < 0)
 364                         break;
 365         }
 366         free(orig);
 367
 368         return ret;
 369 }
 370
 371 static void combine_lock_stats(struct lock_stat *st)
 372 {
 373         struct rb_node **rb = &sorted.rb_node;
 374         struct rb_node *parent = NULL;
 375         struct lock_stat *p;
 376         int ret;
 377
 378         while (*rb) {
 379                 p = container_of(*rb, struct lock_stat, rb);
 380                 parent = *rb;
 381
 382                 if (st->name && p->name)
 383                         ret = strcmp(st->name, p->name);
 384                 else
 385                         ret = !!st->name - !!p->name;
 386
 387                 if (ret == 0) {
 388                         p->nr_acquired += st->nr_acquired;
 389                         p->nr_contended += st->nr_contended;
 390                         p->wait_time_total += st->wait_time_total;
 391
 392                         if (p->nr_contended)
 393                                 p->avg_wait_time = p->wait_time_total / p->nr_contended;
 394
 395                         if (p->wait_time_min > st->wait_time_min)
 396                                 p->wait_time_min = st->wait_time_min;
 397                         if (p->wait_time_max < st->wait_time_max)
 398                                 p->wait_time_max = st->wait_time_max;
 399
 400                         p->broken |= st->broken;
 401                         st->combined = 1;
 402                         return;
 403                 }
 404
 405                 if (ret < 0)
 406                         rb = &(*rb)->rb_left;
 407                 else
 408                         rb = &(*rb)->rb_right;
 409         }
 410
 411         rb_link_node(&st->rb, parent, rb);
 412         rb_insert_color(&st->rb, &sorted);
 413 }
 414
 415 static void insert_to_result(struct lock_stat *st,
 416                              int (*bigger)(struct lock_stat *, struct lock_stat *))
 417 {
 418         struct rb_node **rb = &result.rb_node;
 419         struct rb_node *parent = NULL;
 420         struct lock_stat *p;
 421
 422         if (combine_locks && st->combined)
 423                 return;
 424
 425         while (*rb) {
 426                 p = container_of(*rb, struct lock_stat, rb);
 427                 parent = *rb;
 428
 429                 if (bigger(st, p))
 430                         rb = &(*rb)->rb_left;
 431                 else
 432                         rb = &(*rb)->rb_right;
 433         }
 434
 435         rb_link_node(&st->rb, parent, rb);
 436         rb_insert_color(&st->rb, &result);
 437 }
 438
 439 /* returns left most element of result, and erase it */
 440 static struct lock_stat *pop_from_result(void)
 441 {
 442         struct rb_node *node = result.rb_node;
 443
 444         if (!node)
 445                 return NULL;
 446
 447         while (node->rb_left)
 448                 node = node->rb_left;
 449
 450         rb_erase(node, &result);
 451         return container_of(node, struct lock_stat, rb);
 452 }
 453
 454 static struct lock_stat *lock_stat_findnew(u64 addr, const char *name)
 455 {
 456         struct hlist_head *entry = lockhashentry(addr);
 457         struct lock_stat *ret, *new;
 458
 459         hlist_for_each_entry(ret, entry, hash_entry) {
 460                 if (ret->addr == addr)
 461                         return ret;
 462         }
 463
 464         new = zalloc(sizeof(struct lock_stat));
 465         if (!new)
 466                 goto alloc_failed;
 467
 468         new->addr = addr;
 469         new->name = zalloc(sizeof(char) * strlen(name) + 1);
 470         if (!new->name) {
 471                 free(new);
 472                 goto alloc_failed;
 473         }
 474
 475         strcpy(new->name, name);
 476         new->wait_time_min = ULLONG_MAX;
 477
 478         hlist_add_head(&new->hash_entry, entry);
 479         return new;
 480
 481 alloc_failed:
 482         pr_err("memory allocation failed\n");
 483         return NULL;
 484 }
 485
 486 struct trace_lock_handler {
 487         int (*acquire_event)(struct evsel *evsel,
 488                              struct perf_sample *sample);
 489
 490         int (*acquired_event)(struct evsel *evsel,
 491                               struct perf_sample *sample);
 492
 493         int (*contended_event)(struct evsel *evsel,
 494                                struct perf_sample *sample);
 495
 496         int (*release_event)(struct evsel *evsel,
 497                              struct perf_sample *sample);
 498 };
 499
 500 static struct lock_seq_stat *get_seq(struct thread_stat *ts, u64 addr)
 501 {
 502         struct lock_seq_stat *seq;
 503
 504         list_for_each_entry(seq, &ts->seq_list, list) {
 505                 if (seq->addr == addr)
 506                         return seq;
 507         }
 508
 509         seq = zalloc(sizeof(struct lock_seq_stat));
 510         if (!seq) {
 511                 pr_err("memory allocation failed\n");
 512                 return NULL;
 513         }
 514         seq->state = SEQ_STATE_UNINITIALIZED;
 515         seq->addr = addr;
 516
 517         list_add(&seq->list, &ts->seq_list);
 518         return seq;
 519 }
 520
 521 enum broken_state {
 522         BROKEN_ACQUIRE,
 523         BROKEN_ACQUIRED,
 524         BROKEN_CONTENDED,
 525         BROKEN_RELEASE,
 526         BROKEN_MAX,
 527 };
 528
 529 static int bad_hist[BROKEN_MAX];
 530
 531 enum acquire_flags {
 532         TRY_LOCK = 1,
 533         READ_LOCK = 2,
 534 };
 535
 536 static int report_lock_acquire_event(struct evsel *evsel,
 537                                      struct perf_sample *sample)
 538 {
 539         struct lock_stat *ls;
 540         struct thread_stat *ts;
 541         struct lock_seq_stat *seq;
 542         const char *name = evsel__strval(evsel, sample, "name");
 543         u64 addr = evsel__intval(evsel, sample, "lockdep_addr");
 544         int flag = evsel__intval(evsel, sample, "flags");
 545
 546         /* abuse ls->addr for tid */
 547         if (show_thread_stats)
 548                 addr = sample->tid;
 549
 550         ls = lock_stat_findnew(addr, name);
 551         if (!ls)
 552                 return -ENOMEM;
 553
 554         ts = thread_stat_findnew(sample->tid);
 555         if (!ts)
 556                 return -ENOMEM;
 557
 558         seq = get_seq(ts, addr);
 559         if (!seq)
 560                 return -ENOMEM;
 561
 562         switch (seq->state) {
 563         case SEQ_STATE_UNINITIALIZED:
 564         case SEQ_STATE_RELEASED:
 565                 if (!flag) {
 566                         seq->state = SEQ_STATE_ACQUIRING;
 567                 } else {
 568                         if (flag & TRY_LOCK)
 569                                 ls->nr_trylock++;
 570                         if (flag & READ_LOCK)
 571                                 ls->nr_readlock++;
 572                         seq->state = SEQ_STATE_READ_ACQUIRED;
 573                         seq->read_count = 1;
 574                         ls->nr_acquired++;
 575                 }
 576                 break;
 577         case SEQ_STATE_READ_ACQUIRED:
 578                 if (flag & READ_LOCK) {
 579                         seq->read_count++;
 580                         ls->nr_acquired++;
 581                         goto end;
 582                 } else {
 583                         goto broken;
 584                 }
 585                 break;
 586         case SEQ_STATE_ACQUIRED:
 587         case SEQ_STATE_ACQUIRING:
 588         case SEQ_STATE_CONTENDED:
 589 broken:
 590                 /* broken lock sequence */
 591                 if (!ls->broken) {
 592                         ls->broken = 1;
 593                         bad_hist[BROKEN_ACQUIRE]++;
 594                 }
 595                 list_del_init(&seq->list);
 596                 free(seq);
 597                 goto end;
 598         default:
 599                 BUG_ON("Unknown state of lock sequence found!\n");
 600                 break;
 601         }
 602
 603         ls->nr_acquire++;
 604         seq->prev_event_time = sample->time;
 605 end:
 606         return 0;
 607 }
 608
 609 static int report_lock_acquired_event(struct evsel *evsel,
 610                                       struct perf_sample *sample)
 611 {
 612         struct lock_stat *ls;
 613         struct thread_stat *ts;
 614         struct lock_seq_stat *seq;
 615         u64 contended_term;
 616         const char *name = evsel__strval(evsel, sample, "name");
 617         u64 addr = evsel__intval(evsel, sample, "lockdep_addr");
 618
 619         if (show_thread_stats)
 620                 addr = sample->tid;
 621
 622         ls = lock_stat_findnew(addr, name);
 623         if (!ls)
 624                 return -ENOMEM;
 625
 626         ts = thread_stat_findnew(sample->tid);
 627         if (!ts)
 628                 return -ENOMEM;
 629
 630         seq = get_seq(ts, addr);
 631         if (!seq)
 632                 return -ENOMEM;
 633
 634         switch (seq->state) {
 635         case SEQ_STATE_UNINITIALIZED:
 636                 /* orphan event, do nothing */
 637                 return 0;
 638         case SEQ_STATE_ACQUIRING:
 639                 break;
 640         case SEQ_STATE_CONTENDED:
 641                 contended_term = sample->time - seq->prev_event_time;
 642                 ls->wait_time_total += contended_term;
 643                 if (contended_term < ls->wait_time_min)
 644                         ls->wait_time_min = contended_term;
 645                 if (ls->wait_time_max < contended_term)
 646                         ls->wait_time_max = contended_term;
 647                 break;
 648         case SEQ_STATE_RELEASED:
 649         case SEQ_STATE_ACQUIRED:
 650         case SEQ_STATE_READ_ACQUIRED:
 651                 /* broken lock sequence */
 652                 if (!ls->broken) {
 653                         ls->broken = 1;
 654                         bad_hist[BROKEN_ACQUIRED]++;
 655                 }
 656                 list_del_init(&seq->list);
 657                 free(seq);
 658                 goto end;
 659         default:
 660                 BUG_ON("Unknown state of lock sequence found!\n");
 661                 break;
 662         }
 663
 664         seq->state = SEQ_STATE_ACQUIRED;
 665         ls->nr_acquired++;
 666         ls->avg_wait_time = ls->nr_contended ? ls->wait_time_total/ls->nr_contended : 0;
 667         seq->prev_event_time = sample->time;
 668 end:
 669         return 0;
 670 }
 671
 672 static int report_lock_contended_event(struct evsel *evsel,
 673                                        struct perf_sample *sample)
 674 {
 675         struct lock_stat *ls;
 676         struct thread_stat *ts;
 677         struct lock_seq_stat *seq;
 678         const char *name = evsel__strval(evsel, sample, "name");
 679         u64 addr = evsel__intval(evsel, sample, "lockdep_addr");
 680
 681         if (show_thread_stats)
 682                 addr = sample->tid;
 683
 684         ls = lock_stat_findnew(addr, name);
 685         if (!ls)
 686                 return -ENOMEM;
 687
 688         ts = thread_stat_findnew(sample->tid);
 689         if (!ts)
 690                 return -ENOMEM;
 691
 692         seq = get_seq(ts, addr);
 693         if (!seq)
 694                 return -ENOMEM;
 695
 696         switch (seq->state) {
 697         case SEQ_STATE_UNINITIALIZED:
 698                 /* orphan event, do nothing */
 699                 return 0;
 700         case SEQ_STATE_ACQUIRING:
 701                 break;
 702         case SEQ_STATE_RELEASED:
 703         case SEQ_STATE_ACQUIRED:
 704         case SEQ_STATE_READ_ACQUIRED:
 705         case SEQ_STATE_CONTENDED:
 706                 /* broken lock sequence */
 707                 if (!ls->broken) {
 708                         ls->broken = 1;
 709                         bad_hist[BROKEN_CONTENDED]++;
 710                 }
 711                 list_del_init(&seq->list);
 712                 free(seq);
 713                 goto end;
 714         default:
 715                 BUG_ON("Unknown state of lock sequence found!\n");
 716                 break;
 717         }
 718
 719         seq->state = SEQ_STATE_CONTENDED;
 720         ls->nr_contended++;
 721         ls->avg_wait_time = ls->wait_time_total/ls->nr_contended;
 722         seq->prev_event_time = sample->time;
 723 end:
 724         return 0;
 725 }
 726
 727 static int report_lock_release_event(struct evsel *evsel,
 728                                      struct perf_sample *sample)
 729 {
 730         struct lock_stat *ls;
 731         struct thread_stat *ts;
 732         struct lock_seq_stat *seq;
 733         const char *name = evsel__strval(evsel, sample, "name");
 734         u64 addr = evsel__intval(evsel, sample, "lockdep_addr");
 735
 736         if (show_thread_stats)
 737                 addr = sample->tid;
 738
 739         ls = lock_stat_findnew(addr, name);
 740         if (!ls)
 741                 return -ENOMEM;
 742
 743         ts = thread_stat_findnew(sample->tid);
 744         if (!ts)
 745                 return -ENOMEM;
 746
 747         seq = get_seq(ts, addr);
 748         if (!seq)
 749                 return -ENOMEM;
 750
 751         switch (seq->state) {
 752         case SEQ_STATE_UNINITIALIZED:
 753                 goto end;
 754         case SEQ_STATE_ACQUIRED:
 755                 break;
 756         case SEQ_STATE_READ_ACQUIRED:
 757                 seq->read_count--;
 758                 BUG_ON(seq->read_count < 0);
 759                 if (seq->read_count) {
 760                         ls->nr_release++;
 761                         goto end;
 762                 }
 763                 break;
 764         case SEQ_STATE_ACQUIRING:
 765         case SEQ_STATE_CONTENDED:
 766         case SEQ_STATE_RELEASED:
 767                 /* broken lock sequence */
 768                 if (!ls->broken) {
 769                         ls->broken = 1;
 770                         bad_hist[BROKEN_RELEASE]++;
 771                 }
 772                 goto free_seq;
 773         default:
 774                 BUG_ON("Unknown state of lock sequence found!\n");
 775                 break;
 776         }
 777
 778         ls->nr_release++;
 779 free_seq:
 780         list_del_init(&seq->list);
 781         free(seq);
 782 end:
 783         return 0;
 784 }
 785
 786 /* lock oriented handlers */
 787 /* TODO: handlers for CPU oriented, thread oriented */
 788 static struct trace_lock_handler report_lock_ops  = {
 789         .acquire_event          = report_lock_acquire_event,
 790         .acquired_event         = report_lock_acquired_event,
 791         .contended_event        = report_lock_contended_event,
 792         .release_event          = report_lock_release_event,
 793 };
 794
 795 static struct trace_lock_handler *trace_handler;
 796
 797 static int evsel__process_lock_acquire(struct evsel *evsel, struct perf_sample *sample)
 798 {
 799         if (trace_handler->acquire_event)
 800                 return trace_handler->acquire_event(evsel, sample);
 801         return 0;
 802 }
 803
 804 static int evsel__process_lock_acquired(struct evsel *evsel, struct perf_sample *sample)
 805 {
 806         if (trace_handler->acquired_event)
 807                 return trace_handler->acquired_event(evsel, sample);
 808         return 0;
 809 }
 810
 811 static int evsel__process_lock_contended(struct evsel *evsel, struct perf_sample *sample)
 812 {
 813         if (trace_handler->contended_event)
 814                 return trace_handler->contended_event(evsel, sample);
 815         return 0;
 816 }
 817
 818 static int evsel__process_lock_release(struct evsel *evsel, struct perf_sample *sample)
 819 {
 820         if (trace_handler->release_event)
 821                 return trace_handler->release_event(evsel, sample);
 822         return 0;
 823 }
 824
 825 static void print_bad_events(int bad, int total)
 826 {
 827         /* Output for debug, this have to be removed */
 828         int i;
 829         const char *name[4] =
 830                 { "acquire", "acquired", "contended", "release" };
 831
 832         pr_info("\n=== output for debug===\n\n");
 833         pr_info("bad: %d, total: %d\n", bad, total);
 834         pr_info("bad rate: %.2f %%\n", (double)bad / (double)total * 100);
 835         pr_info("histogram of events caused bad sequence\n");
 836         for (i = 0; i < BROKEN_MAX; i++)
 837                 pr_info(" %10s: %d\n", name[i], bad_hist[i]);
 838 }
 839
 840 /* TODO: various way to print, coloring, nano or milli sec */
 841 static void print_result(void)
 842 {
 843         struct lock_stat *st;
 844         struct lock_key *key;
 845         char cut_name[20];
 846         int bad, total;
 847
 848         pr_info("%20s ", "Name");
 849         list_for_each_entry(key, &lock_keys, list)
 850                 pr_info("%*s ", key->len, key->header);
 851         pr_info("\n\n");
 852
 853         bad = total = 0;
 854         while ((st = pop_from_result())) {
 855                 total++;
 856                 if (st->broken)
 857                         bad++;
 858                 if (!st->nr_acquired)
 859                         continue;
 860
 861                 bzero(cut_name, 20);
 862
 863                 if (strlen(st->name) < 20) {
 864                         /* output raw name */
 865                         const char *name = st->name;
 866
 867                         if (show_thread_stats) {
 868                                 struct thread *t;
 869
 870                                 /* st->addr contains tid of thread */
 871                                 t = perf_session__findnew(session, st->addr);
 872                                 name = thread__comm_str(t);
 873                         }
 874
 875                         pr_info("%20s ", name);
 876                 } else {
 877                         strncpy(cut_name, st->name, 16);
 878                         cut_name[16] = '.';
 879                         cut_name[17] = '.';
 880                         cut_name[18] = '.';
 881                         cut_name[19] = '\0';
 882                         /* cut off name for saving output style */
 883                         pr_info("%20s ", cut_name);
 884                 }
 885
 886                 list_for_each_entry(key, &lock_keys, list) {
 887                         key->print(key, st);
 888                         pr_info(" ");
 889                 }
 890                 pr_info("\n");
 891         }
 892
 893         print_bad_events(bad, total);
 894 }
 895
 896 static bool info_threads, info_map;
 897
 898 static void dump_threads(void)
 899 {
 900         struct thread_stat *st;
 901         struct rb_node *node;
 902         struct thread *t;
 903
 904         pr_info("%10s: comm\n", "Thread ID");
 905
 906         node = rb_first(&thread_stats);
 907         while (node) {
 908                 st = container_of(node, struct thread_stat, rb);
 909                 t = perf_session__findnew(session, st->tid);
 910                 pr_info("%10d: %s\n", st->tid, thread__comm_str(t));
 911                 node = rb_next(node);
 912                 thread__put(t);
 913         }
 914 }
 915
 916 static int compare_maps(struct lock_stat *a, struct lock_stat *b)
 917 {
 918         int ret;
 919
 920         if (a->name && b->name)
 921                 ret = strcmp(a->name, b->name);
 922         else
 923                 ret = !!a->name - !!b->name;
 924
 925         if (!ret)
 926                 return a->addr < b->addr;
 927         else
 928                 return ret < 0;
 929 }
 930
 931 static void dump_map(void)
 932 {
 933         unsigned int i;
 934         struct lock_stat *st;
 935
 936         pr_info("Address of instance: name of class\n");
 937         for (i = 0; i < LOCKHASH_SIZE; i++) {
 938                 hlist_for_each_entry(st, &lockhash_table[i], hash_entry) {
 939                         insert_to_result(st, compare_maps);
 940                 }
 941         }
 942
 943         while ((st = pop_from_result()))
 944                 pr_info(" %#llx: %s\n", (unsigned long long)st->addr, st->name);
 945 }
 946
 947 static int dump_info(void)
 948 {
 949         int rc = 0;
 950
 951         if (info_threads)
 952                 dump_threads();
 953         else if (info_map)
 954                 dump_map();
 955         else {
 956                 rc = -1;
 957                 pr_err("Unknown type of information\n");
 958         }
 959
 960         return rc;
 961 }
 962
 963 typedef int (*tracepoint_handler)(struct evsel *evsel,
 964                                   struct perf_sample *sample);
 965
 966 static int process_sample_event(struct perf_tool *tool __maybe_unused,
 967                                 union perf_event *event,
 968                                 struct perf_sample *sample,
 969                                 struct evsel *evsel,
 970                                 struct machine *machine)
 971 {
 972         int err = 0;
 973         struct thread *thread = machine__findnew_thread(machine, sample->pid,
 974                                                         sample->tid);
 975
 976         if (thread == NULL) {
 977                 pr_debug("problem processing %d event, skipping it.\n",
 978                         event->header.type);
 979                 return -1;
 980         }
 981
 982         if (evsel->handler != NULL) {
 983                 tracepoint_handler f = evsel->handler;
 984                 err = f(evsel, sample);
 985         }
 986
 987         thread__put(thread);
 988
 989         return err;
 990 }
 991
 992 static void combine_result(void)
 993 {
 994         unsigned int i;
 995         struct lock_stat *st;
 996
 997         if (!combine_locks)
 998                 return;
 999
1000         for (i = 0; i < LOCKHASH_SIZE; i++) {
1001                 hlist_for_each_entry(st, &lockhash_table[i], hash_entry) {
1002                         combine_lock_stats(st);
1003                 }
1004         }
1005 }
1006
1007 static void sort_result(void)
1008 {
1009         unsigned int i;
1010         struct lock_stat *st;
1011
1012         for (i = 0; i < LOCKHASH_SIZE; i++) {
1013                 hlist_for_each_entry(st, &lockhash_table[i], hash_entry) {
1014                         insert_to_result(st, compare);
1015                 }
1016         }
1017 }
1018
1019 static const struct evsel_str_handler lock_tracepoints[] = {
1020         { "lock:lock_acquire",   evsel__process_lock_acquire,   }, /* CONFIG_LOCKDEP */
1021         { "lock:lock_acquired",  evsel__process_lock_acquired,  }, /* CONFIG_LOCKDEP, CONFIG_LOCK_STAT */
1022         { "lock:lock_contended", evsel__process_lock_contended, }, /* CONFIG_LOCKDEP, CONFIG_LOCK_STAT */
1023         { "lock:lock_release",   evsel__process_lock_release,   }, /* CONFIG_LOCKDEP */
1024 };
1025
1026 static bool force;
1027
1028 static int __cmd_report(bool display_info)
1029 {
1030         int err = -EINVAL;
1031         struct perf_tool eops = {
1032                 .sample          = process_sample_event,
1033                 .comm            = perf_event__process_comm,
1034                 .namespaces      = perf_event__process_namespaces,
1035                 .ordered_events  = true,
1036         };
1037         struct perf_data data = {
1038                 .path  = input_name,
1039                 .mode  = PERF_DATA_MODE_READ,
1040                 .force = force,
1041         };
1042
1043         session = perf_session__new(&data, &eops);
1044         if (IS_ERR(session)) {
1045                 pr_err("Initializing perf session failed\n");
1046                 return PTR_ERR(session);
1047         }
1048
1049         symbol__init(&session->header.env);
1050
1051         if (!perf_session__has_traces(session, "lock record"))
1052                 goto out_delete;
1053
1054         if (perf_session__set_tracepoints_handlers(session, lock_tracepoints)) {
1055                 pr_err("Initializing perf session tracepoint handlers failed\n");
1056                 goto out_delete;
1057         }
1058
1059         if (setup_output_field(output_fields))
1060                 goto out_delete;
1061
1062         if (select_key())
1063                 goto out_delete;
1064
1065         err = perf_session__process_events(session);
1066         if (err)
1067                 goto out_delete;
1068
1069         setup_pager();
1070         if (display_info) /* used for info subcommand */
1071                 err = dump_info();
1072         else {
1073                 combine_result();
1074                 sort_result();
1075                 print_result();
1076         }
1077
1078 out_delete:
1079         perf_session__delete(session);
1080         return err;
1081 }
1082
1083 static int __cmd_record(int argc, const char **argv)
1084 {
1085         const char *record_args[] = {
1086                 "record", "-R", "-m", "1024", "-c", "1", "--synth", "task",
1087         };
1088         unsigned int rec_argc, i, j, ret;
1089         const char **rec_argv;
1090
1091         for (i = 0; i < ARRAY_SIZE(lock_tracepoints); i++) {
1092                 if (!is_valid_tracepoint(lock_tracepoints[i].name)) {
1093                                 pr_err("tracepoint %s is not enabled. "
1094                                        "Are CONFIG_LOCKDEP and CONFIG_LOCK_STAT enabled?\n",
1095                                        lock_tracepoints[i].name);
1096                                 return 1;
1097                 }
1098         }
1099
1100         rec_argc = ARRAY_SIZE(record_args) + argc - 1;
1101         /* factor of 2 is for -e in front of each tracepoint */
1102         rec_argc += 2 * ARRAY_SIZE(lock_tracepoints);
1103
1104         rec_argv = calloc(rec_argc + 1, sizeof(char *));
1105         if (!rec_argv)
1106                 return -ENOMEM;
1107
1108         for (i = 0; i < ARRAY_SIZE(record_args); i++)
1109                 rec_argv[i] = strdup(record_args[i]);
1110
1111         for (j = 0; j < ARRAY_SIZE(lock_tracepoints); j++) {
1112                 rec_argv[i++] = "-e";
1113                 rec_argv[i++] = strdup(lock_tracepoints[j].name);
1114         }
1115
1116         for (j = 1; j < (unsigned int)argc; j++, i++)
1117                 rec_argv[i] = argv[j];
1118
1119         BUG_ON(i != rec_argc);
1120
1121         ret = cmd_record(i, rec_argv);
1122         free(rec_argv);
1123         return ret;
1124 }
1125
1126 int cmd_lock(int argc, const char **argv)
1127 {
1128         const struct option lock_options[] = {
1129         OPT_STRING('i', "input", &input_name, "file", "input file name"),
1130         OPT_INCR('v', "verbose", &verbose, "be more verbose (show symbol address, etc)"),
1131         OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace, "dump raw trace in ASCII"),
1132         OPT_BOOLEAN('f', "force", &force, "don't complain, do it"),
1133         OPT_END()
1134         };
1135
1136         const struct option info_options[] = {
1137         OPT_BOOLEAN('t', "threads", &info_threads,
1138                     "dump thread list in perf.data"),
1139         OPT_BOOLEAN('m', "map", &info_map,
1140                     "map of lock instances (address:name table)"),
1141         OPT_PARENT(lock_options)
1142         };
1143
1144         const struct option report_options[] = {
1145         OPT_STRING('k', "key", &sort_key, "acquired",
1146                     "key for sorting (acquired / contended / avg_wait / wait_total / wait_max / wait_min)"),
1147         OPT_STRING('F', "field", &output_fields, NULL,
1148                     "output fields (acquired / contended / avg_wait / wait_total / wait_max / wait_min)"),
1149         /* TODO: type */
1150         OPT_BOOLEAN('c', "combine-locks", &combine_locks,
1151                     "combine locks in the same class"),
1152         OPT_BOOLEAN('t', "threads", &show_thread_stats,
1153                     "show per-thread lock stats"),
1154         OPT_PARENT(lock_options)
1155         };
1156
1157         const char * const info_usage[] = {
1158                 "perf lock info [<options>]",
1159                 NULL
1160         };
1161         const char *const lock_subcommands[] = { "record", "report", "script",
1162                                                  "info", NULL };
1163         const char *lock_usage[] = {
1164                 NULL,
1165                 NULL
1166         };
1167         const char * const report_usage[] = {
1168                 "perf lock report [<options>]",
1169                 NULL
1170         };
1171         unsigned int i;
1172         int rc = 0;
1173
1174         for (i = 0; i < LOCKHASH_SIZE; i++)
1175                 INIT_HLIST_HEAD(lockhash_table + i);
1176
1177         argc = parse_options_subcommand(argc, argv, lock_options, lock_subcommands,
1178                                         lock_usage, PARSE_OPT_STOP_AT_NON_OPTION);
1179         if (!argc)
1180                 usage_with_options(lock_usage, lock_options);
1181
1182         if (strlen(argv[0]) > 2 && strstarts("record", argv[0])) {
1183                 return __cmd_record(argc, argv);
1184         } else if (strlen(argv[0]) > 2 && strstarts("report", argv[0])) {
1185                 trace_handler = &report_lock_ops;
1186                 if (argc) {
1187                         argc = parse_options(argc, argv,
1188                                              report_options, report_usage, 0);
1189                         if (argc)
1190                                 usage_with_options(report_usage, report_options);
1191                 }
1192                 rc = __cmd_report(false);
1193         } else if (!strcmp(argv[0], "script")) {
1194                 /* Aliased to 'perf script' */
1195                 return cmd_script(argc, argv);
1196         } else if (!strcmp(argv[0], "info")) {
1197                 if (argc) {
1198                         argc = parse_options(argc, argv,
1199                                              info_options, info_usage, 0);
1200                         if (argc)
1201                                 usage_with_options(info_usage, info_options);
1202                 }
1203                 /* recycling report_lock_ops */
1204                 trace_handler = &report_lock_ops;
1205                 rc = __cmd_report(true);
1206         } else {
1207                 usage_with_options(lock_usage, lock_options);
1208         }
1209
1210         return rc;
1211 }