tools/perf/builtin-lock.c

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