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 #include "util/target.h"
  13 #include "util/callchain.h"
  14 #include "util/lock-contention.h"
  15 #include "util/bpf_skel/lock_data.h"
  16
  17 #include <subcmd/pager.h>
  18 #include <subcmd/parse-options.h>
  19 #include "util/trace-event.h"
  20 #include "util/tracepoint.h"
  21
  22 #include "util/debug.h"
  23 #include "util/session.h"
  24 #include "util/tool.h"
  25 #include "util/data.h"
  26 #include "util/string2.h"
  27 #include "util/map.h"
  28 #include "util/util.h"
  29
  30 #include <sys/types.h>
  31 #include <sys/prctl.h>
  32 #include <semaphore.h>
  33 #include <math.h>
  34 #include <limits.h>
  35 #include <ctype.h>
  36
  37 #include <linux/list.h>
  38 #include <linux/hash.h>
  39 #include <linux/kernel.h>
  40 #include <linux/zalloc.h>
  41 #include <linux/err.h>
  42 #include <linux/stringify.h>
  43
  44 static struct perf_session *session;
  45 static struct target target;
  46
  47 /* based on kernel/lockdep.c */
  48 #define LOCKHASH_BITS           12
  49 #define LOCKHASH_SIZE           (1UL << LOCKHASH_BITS)
  50
  51 static struct hlist_head lockhash_table[LOCKHASH_SIZE];
  52
  53 #define __lockhashfn(key)       hash_long((unsigned long)key, LOCKHASH_BITS)
  54 #define lockhashentry(key)      (lockhash_table + __lockhashfn((key)))
  55
  56 static struct rb_root           thread_stats;
  57
  58 static bool combine_locks;
  59 static bool show_thread_stats;
  60 static bool show_lock_addrs;
  61 static bool show_lock_owner;
  62 static bool use_bpf;
  63 static unsigned long bpf_map_entries = MAX_ENTRIES;
  64 static int max_stack_depth = CONTENTION_STACK_DEPTH;
  65 static int stack_skip = CONTENTION_STACK_SKIP;
  66 static int print_nr_entries = INT_MAX / 2;
  67 static LIST_HEAD(callstack_filters);
  68
  69 struct callstack_filter {
  70         struct list_head list;
  71         char name[];
  72 };
  73
  74 static struct lock_filter filters;
  75
  76 static enum lock_aggr_mode aggr_mode = LOCK_AGGR_ADDR;
  77
  78 static bool needs_callstack(void)
  79 {
  80         return !list_empty(&callstack_filters);
  81 }
  82
  83 static struct thread_stat *thread_stat_find(u32 tid)
  84 {
  85         struct rb_node *node;
  86         struct thread_stat *st;
  87
  88         node = thread_stats.rb_node;
  89         while (node) {
  90                 st = container_of(node, struct thread_stat, rb);
  91                 if (st->tid == tid)
  92                         return st;
  93                 else if (tid < st->tid)
  94                         node = node->rb_left;
  95                 else
  96                         node = node->rb_right;
  97         }
  98
  99         return NULL;
 100 }
 101
 102 static void thread_stat_insert(struct thread_stat *new)
 103 {
 104         struct rb_node **rb = &thread_stats.rb_node;
 105         struct rb_node *parent = NULL;
 106         struct thread_stat *p;
 107
 108         while (*rb) {
 109                 p = container_of(*rb, struct thread_stat, rb);
 110                 parent = *rb;
 111
 112                 if (new->tid < p->tid)
 113                         rb = &(*rb)->rb_left;
 114                 else if (new->tid > p->tid)
 115                         rb = &(*rb)->rb_right;
 116                 else
 117                         BUG_ON("inserting invalid thread_stat\n");
 118         }
 119
 120         rb_link_node(&new->rb, parent, rb);
 121         rb_insert_color(&new->rb, &thread_stats);
 122 }
 123
 124 static struct thread_stat *thread_stat_findnew_after_first(u32 tid)
 125 {
 126         struct thread_stat *st;
 127
 128         st = thread_stat_find(tid);
 129         if (st)
 130                 return st;
 131
 132         st = zalloc(sizeof(struct thread_stat));
 133         if (!st) {
 134                 pr_err("memory allocation failed\n");
 135                 return NULL;
 136         }
 137
 138         st->tid = tid;
 139         INIT_LIST_HEAD(&st->seq_list);
 140
 141         thread_stat_insert(st);
 142
 143         return st;
 144 }
 145
 146 static struct thread_stat *thread_stat_findnew_first(u32 tid);
 147 static struct thread_stat *(*thread_stat_findnew)(u32 tid) =
 148         thread_stat_findnew_first;
 149
 150 static struct thread_stat *thread_stat_findnew_first(u32 tid)
 151 {
 152         struct thread_stat *st;
 153
 154         st = zalloc(sizeof(struct thread_stat));
 155         if (!st) {
 156                 pr_err("memory allocation failed\n");
 157                 return NULL;
 158         }
 159         st->tid = tid;
 160         INIT_LIST_HEAD(&st->seq_list);
 161
 162         rb_link_node(&st->rb, NULL, &thread_stats.rb_node);
 163         rb_insert_color(&st->rb, &thread_stats);
 164
 165         thread_stat_findnew = thread_stat_findnew_after_first;
 166         return st;
 167 }
 168
 169 /* build simple key function one is bigger than two */
 170 #define SINGLE_KEY(member)                                              \
 171         static int lock_stat_key_ ## member(struct lock_stat *one,      \
 172                                          struct lock_stat *two)         \
 173         {                                                               \
 174                 return one->member > two->member;                       \
 175         }
 176
 177 SINGLE_KEY(nr_acquired)
 178 SINGLE_KEY(nr_contended)
 179 SINGLE_KEY(avg_wait_time)
 180 SINGLE_KEY(wait_time_total)
 181 SINGLE_KEY(wait_time_max)
 182
 183 static int lock_stat_key_wait_time_min(struct lock_stat *one,
 184                                         struct lock_stat *two)
 185 {
 186         u64 s1 = one->wait_time_min;
 187         u64 s2 = two->wait_time_min;
 188         if (s1 == ULLONG_MAX)
 189                 s1 = 0;
 190         if (s2 == ULLONG_MAX)
 191                 s2 = 0;
 192         return s1 > s2;
 193 }
 194
 195 struct lock_key {
 196         /*
 197          * name: the value for specify by user
 198          * this should be simpler than raw name of member
 199          * e.g. nr_acquired -> acquired, wait_time_total -> wait_total
 200          */
 201         const char              *name;
 202         /* header: the string printed on the header line */
 203         const char              *header;
 204         /* len: the printing width of the field */
 205         int                     len;
 206         /* key: a pointer to function to compare two lock stats for sorting */
 207         int                     (*key)(struct lock_stat*, struct lock_stat*);
 208         /* print: a pointer to function to print a given lock stats */
 209         void                    (*print)(struct lock_key*, struct lock_stat*);
 210         /* list: list entry to link this */
 211         struct list_head        list;
 212 };
 213
 214 static void lock_stat_key_print_time(unsigned long long nsec, int len)
 215 {
 216         static const struct {
 217                 float base;
 218                 const char *unit;
 219         } table[] = {
 220                 { 1e9 * 3600, "h " },
 221                 { 1e9 * 60, "m " },
 222                 { 1e9, "s " },
 223                 { 1e6, "ms" },
 224                 { 1e3, "us" },
 225                 { 0, NULL },
 226         };
 227
 228         for (int i = 0; table[i].unit; i++) {
 229                 if (nsec < table[i].base)
 230                         continue;
 231
 232                 pr_info("%*.2f %s", len - 3, nsec / table[i].base, table[i].unit);
 233                 return;
 234         }
 235
 236         pr_info("%*llu %s", len - 3, nsec, "ns");
 237 }
 238
 239 #define PRINT_KEY(member)                                               \
 240 static void lock_stat_key_print_ ## member(struct lock_key *key,        \
 241                                            struct lock_stat *ls)        \
 242 {                                                                       \
 243         pr_info("%*llu", key->len, (unsigned long long)ls->member);     \
 244 }
 245
 246 #define PRINT_TIME(member)                                              \
 247 static void lock_stat_key_print_ ## member(struct lock_key *key,        \
 248                                            struct lock_stat *ls)        \
 249 {                                                                       \
 250         lock_stat_key_print_time((unsigned long long)ls->member, key->len);     \
 251 }
 252
 253 PRINT_KEY(nr_acquired)
 254 PRINT_KEY(nr_contended)
 255 PRINT_TIME(avg_wait_time)
 256 PRINT_TIME(wait_time_total)
 257 PRINT_TIME(wait_time_max)
 258
 259 static void lock_stat_key_print_wait_time_min(struct lock_key *key,
 260                                               struct lock_stat *ls)
 261 {
 262         u64 wait_time = ls->wait_time_min;
 263
 264         if (wait_time == ULLONG_MAX)
 265                 wait_time = 0;
 266
 267         lock_stat_key_print_time(wait_time, key->len);
 268 }
 269
 270
 271 static const char               *sort_key = "acquired";
 272
 273 static int                      (*compare)(struct lock_stat *, struct lock_stat *);
 274
 275 static struct rb_root           sorted; /* place to store intermediate data */
 276 static struct rb_root           result; /* place to store sorted data */
 277
 278 static LIST_HEAD(lock_keys);
 279 static const char               *output_fields;
 280
 281 #define DEF_KEY_LOCK(name, header, fn_suffix, len)                      \
 282         { #name, header, len, lock_stat_key_ ## fn_suffix, lock_stat_key_print_ ## fn_suffix, {} }
 283 static struct lock_key report_keys[] = {
 284         DEF_KEY_LOCK(acquired, "acquired", nr_acquired, 10),
 285         DEF_KEY_LOCK(contended, "contended", nr_contended, 10),
 286         DEF_KEY_LOCK(avg_wait, "avg wait", avg_wait_time, 12),
 287         DEF_KEY_LOCK(wait_total, "total wait", wait_time_total, 12),
 288         DEF_KEY_LOCK(wait_max, "max wait", wait_time_max, 12),
 289         DEF_KEY_LOCK(wait_min, "min wait", wait_time_min, 12),
 290
 291         /* extra comparisons much complicated should be here */
 292         { }
 293 };
 294
 295 static struct lock_key contention_keys[] = {
 296         DEF_KEY_LOCK(contended, "contended", nr_contended, 10),
 297         DEF_KEY_LOCK(wait_total, "total wait", wait_time_total, 12),
 298         DEF_KEY_LOCK(wait_max, "max wait", wait_time_max, 12),
 299         DEF_KEY_LOCK(wait_min, "min wait", wait_time_min, 12),
 300         DEF_KEY_LOCK(avg_wait, "avg wait", avg_wait_time, 12),
 301
 302         /* extra comparisons much complicated should be here */
 303         { }
 304 };
 305
 306 static int select_key(bool contention)
 307 {
 308         int i;
 309         struct lock_key *keys = report_keys;
 310
 311         if (contention)
 312                 keys = contention_keys;
 313
 314         for (i = 0; keys[i].name; i++) {
 315                 if (!strcmp(keys[i].name, sort_key)) {
 316                         compare = keys[i].key;
 317
 318                         /* selected key should be in the output fields */
 319                         if (list_empty(&keys[i].list))
 320                                 list_add_tail(&keys[i].list, &lock_keys);
 321
 322                         return 0;
 323                 }
 324         }
 325
 326         pr_err("Unknown compare key: %s\n", sort_key);
 327         return -1;
 328 }
 329
 330 static int add_output_field(bool contention, char *name)
 331 {
 332         int i;
 333         struct lock_key *keys = report_keys;
 334
 335         if (contention)
 336                 keys = contention_keys;
 337
 338         for (i = 0; keys[i].name; i++) {
 339                 if (strcmp(keys[i].name, name))
 340                         continue;
 341
 342                 /* prevent double link */
 343                 if (list_empty(&keys[i].list))
 344                         list_add_tail(&keys[i].list, &lock_keys);
 345
 346                 return 0;
 347         }
 348
 349         pr_err("Unknown output field: %s\n", name);
 350         return -1;
 351 }
 352
 353 static int setup_output_field(bool contention, const char *str)
 354 {
 355         char *tok, *tmp, *orig;
 356         int i, ret = 0;
 357         struct lock_key *keys = report_keys;
 358
 359         if (contention)
 360                 keys = contention_keys;
 361
 362         /* no output field given: use all of them */
 363         if (str == NULL) {
 364                 for (i = 0; keys[i].name; i++)
 365                         list_add_tail(&keys[i].list, &lock_keys);
 366                 return 0;
 367         }
 368
 369         for (i = 0; keys[i].name; i++)
 370                 INIT_LIST_HEAD(&keys[i].list);
 371
 372         orig = tmp = strdup(str);
 373         if (orig == NULL)
 374                 return -ENOMEM;
 375
 376         while ((tok = strsep(&tmp, ",")) != NULL){
 377                 ret = add_output_field(contention, tok);
 378                 if (ret < 0)
 379                         break;
 380         }
 381         free(orig);
 382
 383         return ret;
 384 }
 385
 386 static void combine_lock_stats(struct lock_stat *st)
 387 {
 388         struct rb_node **rb = &sorted.rb_node;
 389         struct rb_node *parent = NULL;
 390         struct lock_stat *p;
 391         int ret;
 392
 393         while (*rb) {
 394                 p = container_of(*rb, struct lock_stat, rb);
 395                 parent = *rb;
 396
 397                 if (st->name && p->name)
 398                         ret = strcmp(st->name, p->name);
 399                 else
 400                         ret = !!st->name - !!p->name;
 401
 402                 if (ret == 0) {
 403                         p->nr_acquired += st->nr_acquired;
 404                         p->nr_contended += st->nr_contended;
 405                         p->wait_time_total += st->wait_time_total;
 406
 407                         if (p->nr_contended)
 408                                 p->avg_wait_time = p->wait_time_total / p->nr_contended;
 409
 410                         if (p->wait_time_min > st->wait_time_min)
 411                                 p->wait_time_min = st->wait_time_min;
 412                         if (p->wait_time_max < st->wait_time_max)
 413                                 p->wait_time_max = st->wait_time_max;
 414
 415                         p->broken |= st->broken;
 416                         st->combined = 1;
 417                         return;
 418                 }
 419
 420                 if (ret < 0)
 421                         rb = &(*rb)->rb_left;
 422                 else
 423                         rb = &(*rb)->rb_right;
 424         }
 425
 426         rb_link_node(&st->rb, parent, rb);
 427         rb_insert_color(&st->rb, &sorted);
 428 }
 429
 430 static void insert_to_result(struct lock_stat *st,
 431                              int (*bigger)(struct lock_stat *, struct lock_stat *))
 432 {
 433         struct rb_node **rb = &result.rb_node;
 434         struct rb_node *parent = NULL;
 435         struct lock_stat *p;
 436
 437         if (combine_locks && st->combined)
 438                 return;
 439
 440         while (*rb) {
 441                 p = container_of(*rb, struct lock_stat, rb);
 442                 parent = *rb;
 443
 444                 if (bigger(st, p))
 445                         rb = &(*rb)->rb_left;
 446                 else
 447                         rb = &(*rb)->rb_right;
 448         }
 449
 450         rb_link_node(&st->rb, parent, rb);
 451         rb_insert_color(&st->rb, &result);
 452 }
 453
 454 /* returns left most element of result, and erase it */
 455 static struct lock_stat *pop_from_result(void)
 456 {
 457         struct rb_node *node = result.rb_node;
 458
 459         if (!node)
 460                 return NULL;
 461
 462         while (node->rb_left)
 463                 node = node->rb_left;
 464
 465         rb_erase(node, &result);
 466         return container_of(node, struct lock_stat, rb);
 467 }
 468
 469 struct lock_stat *lock_stat_find(u64 addr)
 470 {
 471         struct hlist_head *entry = lockhashentry(addr);
 472         struct lock_stat *ret;
 473
 474         hlist_for_each_entry(ret, entry, hash_entry) {
 475                 if (ret->addr == addr)
 476                         return ret;
 477         }
 478         return NULL;
 479 }
 480
 481 struct lock_stat *lock_stat_findnew(u64 addr, const char *name, int flags)
 482 {
 483         struct hlist_head *entry = lockhashentry(addr);
 484         struct lock_stat *ret, *new;
 485
 486         hlist_for_each_entry(ret, entry, hash_entry) {
 487                 if (ret->addr == addr)
 488                         return ret;
 489         }
 490
 491         new = zalloc(sizeof(struct lock_stat));
 492         if (!new)
 493                 goto alloc_failed;
 494
 495         new->addr = addr;
 496         new->name = strdup(name);
 497         if (!new->name) {
 498                 free(new);
 499                 goto alloc_failed;
 500         }
 501
 502         new->flags = flags;
 503         new->wait_time_min = ULLONG_MAX;
 504
 505         hlist_add_head(&new->hash_entry, entry);
 506         return new;
 507
 508 alloc_failed:
 509         pr_err("memory allocation failed\n");
 510         return NULL;
 511 }
 512
 513 bool match_callstack_filter(struct machine *machine, u64 *callstack)
 514 {
 515         struct map *kmap;
 516         struct symbol *sym;
 517         u64 ip;
 518
 519         if (list_empty(&callstack_filters))
 520                 return true;
 521
 522         for (int i = 0; i < max_stack_depth; i++) {
 523                 struct callstack_filter *filter;
 524
 525                 if (!callstack || !callstack[i])
 526                         break;
 527
 528                 ip = callstack[i];
 529                 sym = machine__find_kernel_symbol(machine, ip, &kmap);
 530                 if (sym == NULL)
 531                         continue;
 532
 533                 list_for_each_entry(filter, &callstack_filters, list) {
 534                         if (strstr(sym->name, filter->name))
 535                                 return true;
 536                 }
 537         }
 538         return false;
 539 }
 540
 541 struct trace_lock_handler {
 542         /* it's used on CONFIG_LOCKDEP */
 543         int (*acquire_event)(struct evsel *evsel,
 544                              struct perf_sample *sample);
 545
 546         /* it's used on CONFIG_LOCKDEP && CONFIG_LOCK_STAT */
 547         int (*acquired_event)(struct evsel *evsel,
 548                               struct perf_sample *sample);
 549
 550         /* it's used on CONFIG_LOCKDEP && CONFIG_LOCK_STAT */
 551         int (*contended_event)(struct evsel *evsel,
 552                                struct perf_sample *sample);
 553
 554         /* it's used on CONFIG_LOCKDEP */
 555         int (*release_event)(struct evsel *evsel,
 556                              struct perf_sample *sample);
 557
 558         /* it's used when CONFIG_LOCKDEP is off */
 559         int (*contention_begin_event)(struct evsel *evsel,
 560                                       struct perf_sample *sample);
 561
 562         /* it's used when CONFIG_LOCKDEP is off */
 563         int (*contention_end_event)(struct evsel *evsel,
 564                                     struct perf_sample *sample);
 565 };
 566
 567 static struct lock_seq_stat *get_seq(struct thread_stat *ts, u64 addr)
 568 {
 569         struct lock_seq_stat *seq;
 570
 571         list_for_each_entry(seq, &ts->seq_list, list) {
 572                 if (seq->addr == addr)
 573                         return seq;
 574         }
 575
 576         seq = zalloc(sizeof(struct lock_seq_stat));
 577         if (!seq) {
 578                 pr_err("memory allocation failed\n");
 579                 return NULL;
 580         }
 581         seq->state = SEQ_STATE_UNINITIALIZED;
 582         seq->addr = addr;
 583
 584         list_add(&seq->list, &ts->seq_list);
 585         return seq;
 586 }
 587
 588 enum broken_state {
 589         BROKEN_ACQUIRE,
 590         BROKEN_ACQUIRED,
 591         BROKEN_CONTENDED,
 592         BROKEN_RELEASE,
 593         BROKEN_MAX,
 594 };
 595
 596 static int bad_hist[BROKEN_MAX];
 597
 598 enum acquire_flags {
 599         TRY_LOCK = 1,
 600         READ_LOCK = 2,
 601 };
 602
 603 static int get_key_by_aggr_mode_simple(u64 *key, u64 addr, u32 tid)
 604 {
 605         switch (aggr_mode) {
 606         case LOCK_AGGR_ADDR:
 607                 *key = addr;
 608                 break;
 609         case LOCK_AGGR_TASK:
 610                 *key = tid;
 611                 break;
 612         case LOCK_AGGR_CALLER:
 613         default:
 614                 pr_err("Invalid aggregation mode: %d\n", aggr_mode);
 615                 return -EINVAL;
 616         }
 617         return 0;
 618 }
 619
 620 static u64 callchain_id(struct evsel *evsel, struct perf_sample *sample);
 621
 622 static int get_key_by_aggr_mode(u64 *key, u64 addr, struct evsel *evsel,
 623                                  struct perf_sample *sample)
 624 {
 625         if (aggr_mode == LOCK_AGGR_CALLER) {
 626                 *key = callchain_id(evsel, sample);
 627                 return 0;
 628         }
 629         return get_key_by_aggr_mode_simple(key, addr, sample->tid);
 630 }
 631
 632 static int report_lock_acquire_event(struct evsel *evsel,
 633                                      struct perf_sample *sample)
 634 {
 635         struct lock_stat *ls;
 636         struct thread_stat *ts;
 637         struct lock_seq_stat *seq;
 638         const char *name = evsel__strval(evsel, sample, "name");
 639         u64 addr = evsel__intval(evsel, sample, "lockdep_addr");
 640         int flag = evsel__intval(evsel, sample, "flags");
 641         u64 key;
 642         int ret;
 643
 644         ret = get_key_by_aggr_mode_simple(&key, addr, sample->tid);
 645         if (ret < 0)
 646                 return ret;
 647
 648         ls = lock_stat_findnew(key, name, 0);
 649         if (!ls)
 650                 return -ENOMEM;
 651
 652         ts = thread_stat_findnew(sample->tid);
 653         if (!ts)
 654                 return -ENOMEM;
 655
 656         seq = get_seq(ts, addr);
 657         if (!seq)
 658                 return -ENOMEM;
 659
 660         switch (seq->state) {
 661         case SEQ_STATE_UNINITIALIZED:
 662         case SEQ_STATE_RELEASED:
 663                 if (!flag) {
 664                         seq->state = SEQ_STATE_ACQUIRING;
 665                 } else {
 666                         if (flag & TRY_LOCK)
 667                                 ls->nr_trylock++;
 668                         if (flag & READ_LOCK)
 669                                 ls->nr_readlock++;
 670                         seq->state = SEQ_STATE_READ_ACQUIRED;
 671                         seq->read_count = 1;
 672                         ls->nr_acquired++;
 673                 }
 674                 break;
 675         case SEQ_STATE_READ_ACQUIRED:
 676                 if (flag & READ_LOCK) {
 677                         seq->read_count++;
 678                         ls->nr_acquired++;
 679                         goto end;
 680                 } else {
 681                         goto broken;
 682                 }
 683                 break;
 684         case SEQ_STATE_ACQUIRED:
 685         case SEQ_STATE_ACQUIRING:
 686         case SEQ_STATE_CONTENDED:
 687 broken:
 688                 /* broken lock sequence */
 689                 if (!ls->broken) {
 690                         ls->broken = 1;
 691                         bad_hist[BROKEN_ACQUIRE]++;
 692                 }
 693                 list_del_init(&seq->list);
 694                 free(seq);
 695                 goto end;
 696         default:
 697                 BUG_ON("Unknown state of lock sequence found!\n");
 698                 break;
 699         }
 700
 701         ls->nr_acquire++;
 702         seq->prev_event_time = sample->time;
 703 end:
 704         return 0;
 705 }
 706
 707 static int report_lock_acquired_event(struct evsel *evsel,
 708                                       struct perf_sample *sample)
 709 {
 710         struct lock_stat *ls;
 711         struct thread_stat *ts;
 712         struct lock_seq_stat *seq;
 713         u64 contended_term;
 714         const char *name = evsel__strval(evsel, sample, "name");
 715         u64 addr = evsel__intval(evsel, sample, "lockdep_addr");
 716         u64 key;
 717         int ret;
 718
 719         ret = get_key_by_aggr_mode_simple(&key, addr, sample->tid);
 720         if (ret < 0)
 721                 return ret;
 722
 723         ls = lock_stat_findnew(key, name, 0);
 724         if (!ls)
 725                 return -ENOMEM;
 726
 727         ts = thread_stat_findnew(sample->tid);
 728         if (!ts)
 729                 return -ENOMEM;
 730
 731         seq = get_seq(ts, addr);
 732         if (!seq)
 733                 return -ENOMEM;
 734
 735         switch (seq->state) {
 736         case SEQ_STATE_UNINITIALIZED:
 737                 /* orphan event, do nothing */
 738                 return 0;
 739         case SEQ_STATE_ACQUIRING:
 740                 break;
 741         case SEQ_STATE_CONTENDED:
 742                 contended_term = sample->time - seq->prev_event_time;
 743                 ls->wait_time_total += contended_term;
 744                 if (contended_term < ls->wait_time_min)
 745                         ls->wait_time_min = contended_term;
 746                 if (ls->wait_time_max < contended_term)
 747                         ls->wait_time_max = contended_term;
 748                 break;
 749         case SEQ_STATE_RELEASED:
 750         case SEQ_STATE_ACQUIRED:
 751         case SEQ_STATE_READ_ACQUIRED:
 752                 /* broken lock sequence */
 753                 if (!ls->broken) {
 754                         ls->broken = 1;
 755                         bad_hist[BROKEN_ACQUIRED]++;
 756                 }
 757                 list_del_init(&seq->list);
 758                 free(seq);
 759                 goto end;
 760         default:
 761                 BUG_ON("Unknown state of lock sequence found!\n");
 762                 break;
 763         }
 764
 765         seq->state = SEQ_STATE_ACQUIRED;
 766         ls->nr_acquired++;
 767         ls->avg_wait_time = ls->nr_contended ? ls->wait_time_total/ls->nr_contended : 0;
 768         seq->prev_event_time = sample->time;
 769 end:
 770         return 0;
 771 }
 772
 773 static int report_lock_contended_event(struct evsel *evsel,
 774                                        struct perf_sample *sample)
 775 {
 776         struct lock_stat *ls;
 777         struct thread_stat *ts;
 778         struct lock_seq_stat *seq;
 779         const char *name = evsel__strval(evsel, sample, "name");
 780         u64 addr = evsel__intval(evsel, sample, "lockdep_addr");
 781         u64 key;
 782         int ret;
 783
 784         ret = get_key_by_aggr_mode_simple(&key, addr, sample->tid);
 785         if (ret < 0)
 786                 return ret;
 787
 788         ls = lock_stat_findnew(key, name, 0);
 789         if (!ls)
 790                 return -ENOMEM;
 791
 792         ts = thread_stat_findnew(sample->tid);
 793         if (!ts)
 794                 return -ENOMEM;
 795
 796         seq = get_seq(ts, addr);
 797         if (!seq)
 798                 return -ENOMEM;
 799
 800         switch (seq->state) {
 801         case SEQ_STATE_UNINITIALIZED:
 802                 /* orphan event, do nothing */
 803                 return 0;
 804         case SEQ_STATE_ACQUIRING:
 805                 break;
 806         case SEQ_STATE_RELEASED:
 807         case SEQ_STATE_ACQUIRED:
 808         case SEQ_STATE_READ_ACQUIRED:
 809         case SEQ_STATE_CONTENDED:
 810                 /* broken lock sequence */
 811                 if (!ls->broken) {
 812                         ls->broken = 1;
 813                         bad_hist[BROKEN_CONTENDED]++;
 814                 }
 815                 list_del_init(&seq->list);
 816                 free(seq);
 817                 goto end;
 818         default:
 819                 BUG_ON("Unknown state of lock sequence found!\n");
 820                 break;
 821         }
 822
 823         seq->state = SEQ_STATE_CONTENDED;
 824         ls->nr_contended++;
 825         ls->avg_wait_time = ls->wait_time_total/ls->nr_contended;
 826         seq->prev_event_time = sample->time;
 827 end:
 828         return 0;
 829 }
 830
 831 static int report_lock_release_event(struct evsel *evsel,
 832                                      struct perf_sample *sample)
 833 {
 834         struct lock_stat *ls;
 835         struct thread_stat *ts;
 836         struct lock_seq_stat *seq;
 837         const char *name = evsel__strval(evsel, sample, "name");
 838         u64 addr = evsel__intval(evsel, sample, "lockdep_addr");
 839         u64 key;
 840         int ret;
 841
 842         ret = get_key_by_aggr_mode_simple(&key, addr, sample->tid);
 843         if (ret < 0)
 844                 return ret;
 845
 846         ls = lock_stat_findnew(key, name, 0);
 847         if (!ls)
 848                 return -ENOMEM;
 849
 850         ts = thread_stat_findnew(sample->tid);
 851         if (!ts)
 852                 return -ENOMEM;
 853
 854         seq = get_seq(ts, addr);
 855         if (!seq)
 856                 return -ENOMEM;
 857
 858         switch (seq->state) {
 859         case SEQ_STATE_UNINITIALIZED:
 860                 goto end;
 861         case SEQ_STATE_ACQUIRED:
 862                 break;
 863         case SEQ_STATE_READ_ACQUIRED:
 864                 seq->read_count--;
 865                 BUG_ON(seq->read_count < 0);
 866                 if (seq->read_count) {
 867                         ls->nr_release++;
 868                         goto end;
 869                 }
 870                 break;
 871         case SEQ_STATE_ACQUIRING:
 872         case SEQ_STATE_CONTENDED:
 873         case SEQ_STATE_RELEASED:
 874                 /* broken lock sequence */
 875                 if (!ls->broken) {
 876                         ls->broken = 1;
 877                         bad_hist[BROKEN_RELEASE]++;
 878                 }
 879                 goto free_seq;
 880         default:
 881                 BUG_ON("Unknown state of lock sequence found!\n");
 882                 break;
 883         }
 884
 885         ls->nr_release++;
 886 free_seq:
 887         list_del_init(&seq->list);
 888         free(seq);
 889 end:
 890         return 0;
 891 }
 892
 893 static int get_symbol_name_offset(struct map *map, struct symbol *sym, u64 ip,
 894                                   char *buf, int size)
 895 {
 896         u64 offset;
 897
 898         if (map == NULL || sym == NULL) {
 899                 buf[0] = '\0';
 900                 return 0;
 901         }
 902
 903         offset = map__map_ip(map, ip) - sym->start;
 904
 905         if (offset)
 906                 return scnprintf(buf, size, "%s+%#lx", sym->name, offset);
 907         else
 908                 return strlcpy(buf, sym->name, size);
 909 }
 910 static int lock_contention_caller(struct evsel *evsel, struct perf_sample *sample,
 911                                   char *buf, int size)
 912 {
 913         struct thread *thread;
 914         struct callchain_cursor *cursor = &callchain_cursor;
 915         struct machine *machine = &session->machines.host;
 916         struct symbol *sym;
 917         int skip = 0;
 918         int ret;
 919
 920         /* lock names will be replaced to task name later */
 921         if (show_thread_stats)
 922                 return -1;
 923
 924         thread = machine__findnew_thread(machine, -1, sample->pid);
 925         if (thread == NULL)
 926                 return -1;
 927
 928         /* use caller function name from the callchain */
 929         ret = thread__resolve_callchain(thread, cursor, evsel, sample,
 930                                         NULL, NULL, max_stack_depth);
 931         if (ret != 0) {
 932                 thread__put(thread);
 933                 return -1;
 934         }
 935
 936         callchain_cursor_commit(cursor);
 937         thread__put(thread);
 938
 939         while (true) {
 940                 struct callchain_cursor_node *node;
 941
 942                 node = callchain_cursor_current(cursor);
 943                 if (node == NULL)
 944                         break;
 945
 946                 /* skip first few entries - for lock functions */
 947                 if (++skip <= stack_skip)
 948                         goto next;
 949
 950                 sym = node->ms.sym;
 951                 if (sym && !machine__is_lock_function(machine, node->ip)) {
 952                         get_symbol_name_offset(node->ms.map, sym, node->ip,
 953                                                buf, size);
 954                         return 0;
 955                 }
 956
 957 next:
 958                 callchain_cursor_advance(cursor);
 959         }
 960         return -1;
 961 }
 962
 963 static u64 callchain_id(struct evsel *evsel, struct perf_sample *sample)
 964 {
 965         struct callchain_cursor *cursor = &callchain_cursor;
 966         struct machine *machine = &session->machines.host;
 967         struct thread *thread;
 968         u64 hash = 0;
 969         int skip = 0;
 970         int ret;
 971
 972         thread = machine__findnew_thread(machine, -1, sample->pid);
 973         if (thread == NULL)
 974                 return -1;
 975
 976         /* use caller function name from the callchain */
 977         ret = thread__resolve_callchain(thread, cursor, evsel, sample,
 978                                         NULL, NULL, max_stack_depth);
 979         thread__put(thread);
 980
 981         if (ret != 0)
 982                 return -1;
 983
 984         callchain_cursor_commit(cursor);
 985
 986         while (true) {
 987                 struct callchain_cursor_node *node;
 988
 989                 node = callchain_cursor_current(cursor);
 990                 if (node == NULL)
 991                         break;
 992
 993                 /* skip first few entries - for lock functions */
 994                 if (++skip <= stack_skip)
 995                         goto next;
 996
 997                 if (node->ms.sym && machine__is_lock_function(machine, node->ip))
 998                         goto next;
 999
1000                 hash ^= hash_long((unsigned long)node->ip, 64);
1001
1002 next:
1003                 callchain_cursor_advance(cursor);
1004         }
1005         return hash;
1006 }
1007
1008 static u64 *get_callstack(struct perf_sample *sample, int max_stack)
1009 {
1010         u64 *callstack;
1011         u64 i;
1012         int c;
1013
1014         callstack = calloc(max_stack, sizeof(*callstack));
1015         if (callstack == NULL)
1016                 return NULL;
1017
1018         for (i = 0, c = 0; i < sample->callchain->nr && c < max_stack; i++) {
1019                 u64 ip = sample->callchain->ips[i];
1020
1021                 if (ip >= PERF_CONTEXT_MAX)
1022                         continue;
1023
1024                 callstack[c++] = ip;
1025         }
1026         return callstack;
1027 }
1028
1029 static int report_lock_contention_begin_event(struct evsel *evsel,
1030                                               struct perf_sample *sample)
1031 {
1032         struct lock_stat *ls;
1033         struct thread_stat *ts;
1034         struct lock_seq_stat *seq;
1035         u64 addr = evsel__intval(evsel, sample, "lock_addr");
1036         unsigned int flags = evsel__intval(evsel, sample, "flags");
1037         u64 key;
1038         int i, ret;
1039         static bool kmap_loaded;
1040         struct machine *machine = &session->machines.host;
1041         struct map *kmap;
1042         struct symbol *sym;
1043
1044         ret = get_key_by_aggr_mode(&key, addr, evsel, sample);
1045         if (ret < 0)
1046                 return ret;
1047
1048         if (!kmap_loaded) {
1049                 unsigned long *addrs;
1050
1051                 /* make sure it loads the kernel map to find lock symbols */
1052                 map__load(machine__kernel_map(machine));
1053                 kmap_loaded = true;
1054
1055                 /* convert (kernel) symbols to addresses */
1056                 for (i = 0; i < filters.nr_syms; i++) {
1057                         sym = machine__find_kernel_symbol_by_name(machine,
1058                                                                   filters.syms[i],
1059                                                                   &kmap);
1060                         if (sym == NULL) {
1061                                 pr_warning("ignore unknown symbol: %s\n",
1062                                            filters.syms[i]);
1063                                 continue;
1064                         }
1065
1066                         addrs = realloc(filters.addrs,
1067                                         (filters.nr_addrs + 1) * sizeof(*addrs));
1068                         if (addrs == NULL) {
1069                                 pr_warning("memory allocation failure\n");
1070                                 return -ENOMEM;
1071                         }
1072
1073                         addrs[filters.nr_addrs++] = map__unmap_ip(kmap, sym->start);
1074                         filters.addrs = addrs;
1075                 }
1076         }
1077
1078         ls = lock_stat_find(key);
1079         if (!ls) {
1080                 char buf[128];
1081                 const char *name = "";
1082
1083                 switch (aggr_mode) {
1084                 case LOCK_AGGR_ADDR:
1085                         sym = machine__find_kernel_symbol(machine, key, &kmap);
1086                         if (sym)
1087                                 name = sym->name;
1088                         break;
1089                 case LOCK_AGGR_CALLER:
1090                         name = buf;
1091                         if (lock_contention_caller(evsel, sample, buf, sizeof(buf)) < 0)
1092                                 name = "Unknown";
1093                         break;
1094                 case LOCK_AGGR_TASK:
1095                 default:
1096                         break;
1097                 }
1098
1099                 ls = lock_stat_findnew(key, name, flags);
1100                 if (!ls)
1101                         return -ENOMEM;
1102         }
1103
1104         if (filters.nr_types) {
1105                 bool found = false;
1106
1107                 for (i = 0; i < filters.nr_types; i++) {
1108                         if (flags == filters.types[i]) {
1109                                 found = true;
1110                                 break;
1111                         }
1112                 }
1113
1114                 if (!found)
1115                         return 0;
1116         }
1117
1118         if (filters.nr_addrs) {
1119                 bool found = false;
1120
1121                 for (i = 0; i < filters.nr_addrs; i++) {
1122                         if (addr == filters.addrs[i]) {
1123                                 found = true;
1124                                 break;
1125                         }
1126                 }
1127
1128                 if (!found)
1129                         return 0;
1130         }
1131
1132         if (needs_callstack()) {
1133                 u64 *callstack = get_callstack(sample, max_stack_depth);
1134                 if (callstack == NULL)
1135                         return -ENOMEM;
1136
1137                 if (!match_callstack_filter(machine, callstack)) {
1138                         free(callstack);
1139                         return 0;
1140                 }
1141
1142                 if (ls->callstack == NULL)
1143                         ls->callstack = callstack;
1144                 else
1145                         free(callstack);
1146         }
1147
1148         ts = thread_stat_findnew(sample->tid);
1149         if (!ts)
1150                 return -ENOMEM;
1151
1152         seq = get_seq(ts, addr);
1153         if (!seq)
1154                 return -ENOMEM;
1155
1156         switch (seq->state) {
1157         case SEQ_STATE_UNINITIALIZED:
1158         case SEQ_STATE_ACQUIRED:
1159                 break;
1160         case SEQ_STATE_CONTENDED:
1161                 /*
1162                  * It can have nested contention begin with mutex spinning,
1163                  * then we would use the original contention begin event and
1164                  * ignore the second one.
1165                  */
1166                 goto end;
1167         case SEQ_STATE_ACQUIRING:
1168         case SEQ_STATE_READ_ACQUIRED:
1169         case SEQ_STATE_RELEASED:
1170                 /* broken lock sequence */
1171                 if (!ls->broken) {
1172                         ls->broken = 1;
1173                         bad_hist[BROKEN_CONTENDED]++;
1174                 }
1175                 list_del_init(&seq->list);
1176                 free(seq);
1177                 goto end;
1178         default:
1179                 BUG_ON("Unknown state of lock sequence found!\n");
1180                 break;
1181         }
1182
1183         if (seq->state != SEQ_STATE_CONTENDED) {
1184                 seq->state = SEQ_STATE_CONTENDED;
1185                 seq->prev_event_time = sample->time;
1186                 ls->nr_contended++;
1187         }
1188 end:
1189         return 0;
1190 }
1191
1192 static int report_lock_contention_end_event(struct evsel *evsel,
1193                                             struct perf_sample *sample)
1194 {
1195         struct lock_stat *ls;
1196         struct thread_stat *ts;
1197         struct lock_seq_stat *seq;
1198         u64 contended_term;
1199         u64 addr = evsel__intval(evsel, sample, "lock_addr");
1200         u64 key;
1201         int ret;
1202
1203         ret = get_key_by_aggr_mode(&key, addr, evsel, sample);
1204         if (ret < 0)
1205                 return ret;
1206
1207         ls = lock_stat_find(key);
1208         if (!ls)
1209                 return 0;
1210
1211         ts = thread_stat_find(sample->tid);
1212         if (!ts)
1213                 return 0;
1214
1215         seq = get_seq(ts, addr);
1216         if (!seq)
1217                 return -ENOMEM;
1218
1219         switch (seq->state) {
1220         case SEQ_STATE_UNINITIALIZED:
1221                 goto end;
1222         case SEQ_STATE_CONTENDED:
1223                 contended_term = sample->time - seq->prev_event_time;
1224                 ls->wait_time_total += contended_term;
1225                 if (contended_term < ls->wait_time_min)
1226                         ls->wait_time_min = contended_term;
1227                 if (ls->wait_time_max < contended_term)
1228                         ls->wait_time_max = contended_term;
1229                 break;
1230         case SEQ_STATE_ACQUIRING:
1231         case SEQ_STATE_ACQUIRED:
1232         case SEQ_STATE_READ_ACQUIRED:
1233         case SEQ_STATE_RELEASED:
1234                 /* broken lock sequence */
1235                 if (!ls->broken) {
1236                         ls->broken = 1;
1237                         bad_hist[BROKEN_ACQUIRED]++;
1238                 }
1239                 list_del_init(&seq->list);
1240                 free(seq);
1241                 goto end;
1242         default:
1243                 BUG_ON("Unknown state of lock sequence found!\n");
1244                 break;
1245         }
1246
1247         seq->state = SEQ_STATE_ACQUIRED;
1248         ls->nr_acquired++;
1249         ls->avg_wait_time = ls->wait_time_total/ls->nr_acquired;
1250 end:
1251         return 0;
1252 }
1253
1254 /* lock oriented handlers */
1255 /* TODO: handlers for CPU oriented, thread oriented */
1256 static struct trace_lock_handler report_lock_ops  = {
1257         .acquire_event          = report_lock_acquire_event,
1258         .acquired_event         = report_lock_acquired_event,
1259         .contended_event        = report_lock_contended_event,
1260         .release_event          = report_lock_release_event,
1261         .contention_begin_event = report_lock_contention_begin_event,
1262         .contention_end_event   = report_lock_contention_end_event,
1263 };
1264
1265 static struct trace_lock_handler contention_lock_ops  = {
1266         .contention_begin_event = report_lock_contention_begin_event,
1267         .contention_end_event   = report_lock_contention_end_event,
1268 };
1269
1270
1271 static struct trace_lock_handler *trace_handler;
1272
1273 static int evsel__process_lock_acquire(struct evsel *evsel, struct perf_sample *sample)
1274 {
1275         if (trace_handler->acquire_event)
1276                 return trace_handler->acquire_event(evsel, sample);
1277         return 0;
1278 }
1279
1280 static int evsel__process_lock_acquired(struct evsel *evsel, struct perf_sample *sample)
1281 {
1282         if (trace_handler->acquired_event)
1283                 return trace_handler->acquired_event(evsel, sample);
1284         return 0;
1285 }
1286
1287 static int evsel__process_lock_contended(struct evsel *evsel, struct perf_sample *sample)
1288 {
1289         if (trace_handler->contended_event)
1290                 return trace_handler->contended_event(evsel, sample);
1291         return 0;
1292 }
1293
1294 static int evsel__process_lock_release(struct evsel *evsel, struct perf_sample *sample)
1295 {
1296         if (trace_handler->release_event)
1297                 return trace_handler->release_event(evsel, sample);
1298         return 0;
1299 }
1300
1301 static int evsel__process_contention_begin(struct evsel *evsel, struct perf_sample *sample)
1302 {
1303         if (trace_handler->contention_begin_event)
1304                 return trace_handler->contention_begin_event(evsel, sample);
1305         return 0;
1306 }
1307
1308 static int evsel__process_contention_end(struct evsel *evsel, struct perf_sample *sample)
1309 {
1310         if (trace_handler->contention_end_event)
1311                 return trace_handler->contention_end_event(evsel, sample);
1312         return 0;
1313 }
1314
1315 static void print_bad_events(int bad, int total)
1316 {
1317         /* Output for debug, this have to be removed */
1318         int i;
1319         int broken = 0;
1320         const char *name[4] =
1321                 { "acquire", "acquired", "contended", "release" };
1322
1323         for (i = 0; i < BROKEN_MAX; i++)
1324                 broken += bad_hist[i];
1325
1326         if (quiet || total == 0 || (broken == 0 && verbose <= 0))
1327                 return;
1328
1329         pr_info("\n=== output for debug ===\n\n");
1330         pr_info("bad: %d, total: %d\n", bad, total);
1331         pr_info("bad rate: %.2f %%\n", (double)bad / (double)total * 100);
1332         pr_info("histogram of events caused bad sequence\n");
1333         for (i = 0; i < BROKEN_MAX; i++)
1334                 pr_info(" %10s: %d\n", name[i], bad_hist[i]);
1335 }
1336
1337 /* TODO: various way to print, coloring, nano or milli sec */
1338 static void print_result(void)
1339 {
1340         struct lock_stat *st;
1341         struct lock_key *key;
1342         char cut_name[20];
1343         int bad, total, printed;
1344
1345         if (!quiet) {
1346                 pr_info("%20s ", "Name");
1347                 list_for_each_entry(key, &lock_keys, list)
1348                         pr_info("%*s ", key->len, key->header);
1349                 pr_info("\n\n");
1350         }
1351
1352         bad = total = printed = 0;
1353         while ((st = pop_from_result())) {
1354                 total++;
1355                 if (st->broken)
1356                         bad++;
1357                 if (!st->nr_acquired)
1358                         continue;
1359
1360                 bzero(cut_name, 20);
1361
1362                 if (strlen(st->name) < 20) {
1363                         /* output raw name */
1364                         const char *name = st->name;
1365
1366                         if (show_thread_stats) {
1367                                 struct thread *t;
1368
1369                                 /* st->addr contains tid of thread */
1370                                 t = perf_session__findnew(session, st->addr);
1371                                 name = thread__comm_str(t);
1372                         }
1373
1374                         pr_info("%20s ", name);
1375                 } else {
1376                         strncpy(cut_name, st->name, 16);
1377                         cut_name[16] = '.';
1378                         cut_name[17] = '.';
1379                         cut_name[18] = '.';
1380                         cut_name[19] = '\0';
1381                         /* cut off name for saving output style */
1382                         pr_info("%20s ", cut_name);
1383                 }
1384
1385                 list_for_each_entry(key, &lock_keys, list) {
1386                         key->print(key, st);
1387                         pr_info(" ");
1388                 }
1389                 pr_info("\n");
1390
1391                 if (++printed >= print_nr_entries)
1392                         break;
1393         }
1394
1395         print_bad_events(bad, total);
1396 }
1397
1398 static bool info_threads, info_map;
1399
1400 static void dump_threads(void)
1401 {
1402         struct thread_stat *st;
1403         struct rb_node *node;
1404         struct thread *t;
1405
1406         pr_info("%10s: comm\n", "Thread ID");
1407
1408         node = rb_first(&thread_stats);
1409         while (node) {
1410                 st = container_of(node, struct thread_stat, rb);
1411                 t = perf_session__findnew(session, st->tid);
1412                 pr_info("%10d: %s\n", st->tid, thread__comm_str(t));
1413                 node = rb_next(node);
1414                 thread__put(t);
1415         }
1416 }
1417
1418 static int compare_maps(struct lock_stat *a, struct lock_stat *b)
1419 {
1420         int ret;
1421
1422         if (a->name && b->name)
1423                 ret = strcmp(a->name, b->name);
1424         else
1425                 ret = !!a->name - !!b->name;
1426
1427         if (!ret)
1428                 return a->addr < b->addr;
1429         else
1430                 return ret < 0;
1431 }
1432
1433 static void dump_map(void)
1434 {
1435         unsigned int i;
1436         struct lock_stat *st;
1437
1438         pr_info("Address of instance: name of class\n");
1439         for (i = 0; i < LOCKHASH_SIZE; i++) {
1440                 hlist_for_each_entry(st, &lockhash_table[i], hash_entry) {
1441                         insert_to_result(st, compare_maps);
1442                 }
1443         }
1444
1445         while ((st = pop_from_result()))
1446                 pr_info(" %#llx: %s\n", (unsigned long long)st->addr, st->name);
1447 }
1448
1449 static int dump_info(void)
1450 {
1451         int rc = 0;
1452
1453         if (info_threads)
1454                 dump_threads();
1455         else if (info_map)
1456                 dump_map();
1457         else {
1458                 rc = -1;
1459                 pr_err("Unknown type of information\n");
1460         }
1461
1462         return rc;
1463 }
1464
1465 static const struct evsel_str_handler lock_tracepoints[] = {
1466         { "lock:lock_acquire",   evsel__process_lock_acquire,   }, /* CONFIG_LOCKDEP */
1467         { "lock:lock_acquired",  evsel__process_lock_acquired,  }, /* CONFIG_LOCKDEP, CONFIG_LOCK_STAT */
1468         { "lock:lock_contended", evsel__process_lock_contended, }, /* CONFIG_LOCKDEP, CONFIG_LOCK_STAT */
1469         { "lock:lock_release",   evsel__process_lock_release,   }, /* CONFIG_LOCKDEP */
1470 };
1471
1472 static const struct evsel_str_handler contention_tracepoints[] = {
1473         { "lock:contention_begin", evsel__process_contention_begin, },
1474         { "lock:contention_end",   evsel__process_contention_end,   },
1475 };
1476
1477 static int process_event_update(struct perf_tool *tool,
1478                                 union perf_event *event,
1479                                 struct evlist **pevlist)
1480 {
1481         int ret;
1482
1483         ret = perf_event__process_event_update(tool, event, pevlist);
1484         if (ret < 0)
1485                 return ret;
1486
1487         /* this can return -EEXIST since we call it for each evsel */
1488         perf_session__set_tracepoints_handlers(session, lock_tracepoints);
1489         perf_session__set_tracepoints_handlers(session, contention_tracepoints);
1490         return 0;
1491 }
1492
1493 typedef int (*tracepoint_handler)(struct evsel *evsel,
1494                                   struct perf_sample *sample);
1495
1496 static int process_sample_event(struct perf_tool *tool __maybe_unused,
1497                                 union perf_event *event,
1498                                 struct perf_sample *sample,
1499                                 struct evsel *evsel,
1500                                 struct machine *machine)
1501 {
1502         int err = 0;
1503         struct thread *thread = machine__findnew_thread(machine, sample->pid,
1504                                                         sample->tid);
1505
1506         if (thread == NULL) {
1507                 pr_debug("problem processing %d event, skipping it.\n",
1508                         event->header.type);
1509                 return -1;
1510         }
1511
1512         if (evsel->handler != NULL) {
1513                 tracepoint_handler f = evsel->handler;
1514                 err = f(evsel, sample);
1515         }
1516
1517         thread__put(thread);
1518
1519         return err;
1520 }
1521
1522 static void combine_result(void)
1523 {
1524         unsigned int i;
1525         struct lock_stat *st;
1526
1527         if (!combine_locks)
1528                 return;
1529
1530         for (i = 0; i < LOCKHASH_SIZE; i++) {
1531                 hlist_for_each_entry(st, &lockhash_table[i], hash_entry) {
1532                         combine_lock_stats(st);
1533                 }
1534         }
1535 }
1536
1537 static void sort_result(void)
1538 {
1539         unsigned int i;
1540         struct lock_stat *st;
1541
1542         for (i = 0; i < LOCKHASH_SIZE; i++) {
1543                 hlist_for_each_entry(st, &lockhash_table[i], hash_entry) {
1544                         insert_to_result(st, compare);
1545                 }
1546         }
1547 }
1548
1549 static const struct {
1550         unsigned int flags;
1551         const char *str;
1552         const char *name;
1553 } lock_type_table[] = {
1554         { 0,                            "semaphore",    "semaphore" },
1555         { LCB_F_SPIN,                   "spinlock",     "spinlock" },
1556         { LCB_F_SPIN | LCB_F_READ,      "rwlock:R",     "rwlock" },
1557         { LCB_F_SPIN | LCB_F_WRITE,     "rwlock:W",     "rwlock" },
1558         { LCB_F_READ,                   "rwsem:R",      "rwsem" },
1559         { LCB_F_WRITE,                  "rwsem:W",      "rwsem" },
1560         { LCB_F_RT,                     "rt-mutex",     "rt-mutex" },
1561         { LCB_F_RT | LCB_F_READ,        "rwlock-rt:R",  "rwlock-rt" },
1562         { LCB_F_RT | LCB_F_WRITE,       "rwlock-rt:W",  "rwlock-rt" },
1563         { LCB_F_PERCPU | LCB_F_READ,    "pcpu-sem:R",   "percpu-rwsem" },
1564         { LCB_F_PERCPU | LCB_F_WRITE,   "pcpu-sem:W",   "percpu-rwsem" },
1565         { LCB_F_MUTEX,                  "mutex",        "mutex" },
1566         { LCB_F_MUTEX | LCB_F_SPIN,     "mutex",        "mutex" },
1567         /* alias for get_type_flag() */
1568         { LCB_F_MUTEX | LCB_F_SPIN,     "mutex-spin",   "mutex" },
1569 };
1570
1571 static const char *get_type_str(unsigned int flags)
1572 {
1573         flags &= LCB_F_MAX_FLAGS - 1;
1574
1575         for (unsigned int i = 0; i < ARRAY_SIZE(lock_type_table); i++) {
1576                 if (lock_type_table[i].flags == flags)
1577                         return lock_type_table[i].str;
1578         }
1579         return "unknown";
1580 }
1581
1582 static const char *get_type_name(unsigned int flags)
1583 {
1584         flags &= LCB_F_MAX_FLAGS - 1;
1585
1586         for (unsigned int i = 0; i < ARRAY_SIZE(lock_type_table); i++) {
1587                 if (lock_type_table[i].flags == flags)
1588                         return lock_type_table[i].name;
1589         }
1590         return "unknown";
1591 }
1592
1593 static unsigned int get_type_flag(const char *str)
1594 {
1595         for (unsigned int i = 0; i < ARRAY_SIZE(lock_type_table); i++) {
1596                 if (!strcmp(lock_type_table[i].name, str))
1597                         return lock_type_table[i].flags;
1598         }
1599         for (unsigned int i = 0; i < ARRAY_SIZE(lock_type_table); i++) {
1600                 if (!strcmp(lock_type_table[i].str, str))
1601                         return lock_type_table[i].flags;
1602         }
1603         return UINT_MAX;
1604 }
1605
1606 static void lock_filter_finish(void)
1607 {
1608         zfree(&filters.types);
1609         filters.nr_types = 0;
1610
1611         zfree(&filters.addrs);
1612         filters.nr_addrs = 0;
1613
1614         for (int i = 0; i < filters.nr_syms; i++)
1615                 free(filters.syms[i]);
1616
1617         zfree(&filters.syms);
1618         filters.nr_syms = 0;
1619 }
1620
1621 static void sort_contention_result(void)
1622 {
1623         sort_result();
1624 }
1625
1626 static void print_bpf_events(int total, struct lock_contention_fails *fails)
1627 {
1628         /* Output for debug, this have to be removed */
1629         int broken = fails->task + fails->stack + fails->time + fails->data;
1630
1631         if (quiet || total == 0 || (broken == 0 && verbose <= 0))
1632                 return;
1633
1634         total += broken;
1635         pr_info("\n=== output for debug ===\n\n");
1636         pr_info("bad: %d, total: %d\n", broken, total);
1637         pr_info("bad rate: %.2f %%\n", (double)broken / (double)total * 100);
1638
1639         pr_info("histogram of failure reasons\n");
1640         pr_info(" %10s: %d\n", "task", fails->task);
1641         pr_info(" %10s: %d\n", "stack", fails->stack);
1642         pr_info(" %10s: %d\n", "time", fails->time);
1643         pr_info(" %10s: %d\n", "data", fails->data);
1644 }
1645
1646 static void print_contention_result(struct lock_contention *con)
1647 {
1648         struct lock_stat *st;
1649         struct lock_key *key;
1650         int bad, total, printed;
1651
1652         if (!quiet) {
1653                 list_for_each_entry(key, &lock_keys, list)
1654                         pr_info("%*s ", key->len, key->header);
1655
1656                 switch (aggr_mode) {
1657                 case LOCK_AGGR_TASK:
1658                         pr_info("  %10s   %s\n\n", "pid",
1659                                 show_lock_owner ? "owner" : "comm");
1660                         break;
1661                 case LOCK_AGGR_CALLER:
1662                         pr_info("  %10s   %s\n\n", "type", "caller");
1663                         break;
1664                 case LOCK_AGGR_ADDR:
1665                         pr_info("  %16s   %s\n\n", "address", "symbol");
1666                         break;
1667                 default:
1668                         break;
1669                 }
1670         }
1671
1672         bad = total = printed = 0;
1673
1674         while ((st = pop_from_result())) {
1675                 struct thread *t;
1676                 int pid;
1677
1678                 total += use_bpf ? st->nr_contended : 1;
1679                 if (st->broken)
1680                         bad++;
1681
1682                 if (!st->wait_time_total)
1683                         continue;
1684
1685                 list_for_each_entry(key, &lock_keys, list) {
1686                         key->print(key, st);
1687                         pr_info(" ");
1688                 }
1689
1690                 switch (aggr_mode) {
1691                 case LOCK_AGGR_CALLER:
1692                         pr_info("  %10s   %s\n", get_type_str(st->flags), st->name);
1693                         break;
1694                 case LOCK_AGGR_TASK:
1695                         pid = st->addr;
1696                         t = perf_session__findnew(session, pid);
1697                         pr_info("  %10d   %s\n",
1698                                 pid, pid == -1 ? "Unknown" : thread__comm_str(t));
1699                         break;
1700                 case LOCK_AGGR_ADDR:
1701                         pr_info("  %016llx   %s (%s)\n", (unsigned long long)st->addr,
1702                                 st->name, get_type_name(st->flags));
1703                         break;
1704                 default:
1705                         break;
1706                 }
1707
1708                 if (aggr_mode == LOCK_AGGR_CALLER && verbose > 0) {
1709                         struct map *kmap;
1710                         struct symbol *sym;
1711                         char buf[128];
1712                         u64 ip;
1713
1714                         for (int i = 0; i < max_stack_depth; i++) {
1715                                 if (!st->callstack || !st->callstack[i])
1716                                         break;
1717
1718                                 ip = st->callstack[i];
1719                                 sym = machine__find_kernel_symbol(con->machine, ip, &kmap);
1720                                 get_symbol_name_offset(kmap, sym, ip, buf, sizeof(buf));
1721                                 pr_info("\t\t\t%#lx  %s\n", (unsigned long)ip, buf);
1722                         }
1723                 }
1724
1725                 if (++printed >= print_nr_entries)
1726                         break;
1727         }
1728
1729         if (print_nr_entries) {
1730                 /* update the total/bad stats */
1731                 while ((st = pop_from_result())) {
1732                         total += use_bpf ? st->nr_contended : 1;
1733                         if (st->broken)
1734                                 bad++;
1735                 }
1736         }
1737         /* some entries are collected but hidden by the callstack filter */
1738         total += con->nr_filtered;
1739
1740         if (use_bpf)
1741                 print_bpf_events(total, &con->fails);
1742         else
1743                 print_bad_events(bad, total);
1744 }
1745
1746 static bool force;
1747
1748 static int __cmd_report(bool display_info)
1749 {
1750         int err = -EINVAL;
1751         struct perf_tool eops = {
1752                 .attr            = perf_event__process_attr,
1753                 .event_update    = process_event_update,
1754                 .sample          = process_sample_event,
1755                 .comm            = perf_event__process_comm,
1756                 .mmap            = perf_event__process_mmap,
1757                 .namespaces      = perf_event__process_namespaces,
1758                 .tracing_data    = perf_event__process_tracing_data,
1759                 .ordered_events  = true,
1760         };
1761         struct perf_data data = {
1762                 .path  = input_name,
1763                 .mode  = PERF_DATA_MODE_READ,
1764                 .force = force,
1765         };
1766
1767         session = perf_session__new(&data, &eops);
1768         if (IS_ERR(session)) {
1769                 pr_err("Initializing perf session failed\n");
1770                 return PTR_ERR(session);
1771         }
1772
1773         /* for lock function check */
1774         symbol_conf.sort_by_name = true;
1775         symbol_conf.allow_aliases = true;
1776         symbol__init(&session->header.env);
1777
1778         if (!data.is_pipe) {
1779                 if (!perf_session__has_traces(session, "lock record"))
1780                         goto out_delete;
1781
1782                 if (perf_session__set_tracepoints_handlers(session, lock_tracepoints)) {
1783                         pr_err("Initializing perf session tracepoint handlers failed\n");
1784                         goto out_delete;
1785                 }
1786
1787                 if (perf_session__set_tracepoints_handlers(session, contention_tracepoints)) {
1788                         pr_err("Initializing perf session tracepoint handlers failed\n");
1789                         goto out_delete;
1790                 }
1791         }
1792
1793         if (setup_output_field(false, output_fields))
1794                 goto out_delete;
1795
1796         if (select_key(false))
1797                 goto out_delete;
1798
1799         if (show_thread_stats)
1800                 aggr_mode = LOCK_AGGR_TASK;
1801
1802         err = perf_session__process_events(session);
1803         if (err)
1804                 goto out_delete;
1805
1806         setup_pager();
1807         if (display_info) /* used for info subcommand */
1808                 err = dump_info();
1809         else {
1810                 combine_result();
1811                 sort_result();
1812                 print_result();
1813         }
1814
1815 out_delete:
1816         perf_session__delete(session);
1817         return err;
1818 }
1819
1820 static void sighandler(int sig __maybe_unused)
1821 {
1822 }
1823
1824 static int check_lock_contention_options(const struct option *options,
1825                                          const char * const *usage)
1826
1827 {
1828         if (show_thread_stats && show_lock_addrs) {
1829                 pr_err("Cannot use thread and addr mode together\n");
1830                 parse_options_usage(usage, options, "threads", 0);
1831                 parse_options_usage(NULL, options, "lock-addr", 0);
1832                 return -1;
1833         }
1834
1835         if (show_lock_owner && !use_bpf) {
1836                 pr_err("Lock owners are available only with BPF\n");
1837                 parse_options_usage(usage, options, "lock-owner", 0);
1838                 parse_options_usage(NULL, options, "use-bpf", 0);
1839                 return -1;
1840         }
1841
1842         if (show_lock_owner && show_lock_addrs) {
1843                 pr_err("Cannot use owner and addr mode together\n");
1844                 parse_options_usage(usage, options, "lock-owner", 0);
1845                 parse_options_usage(NULL, options, "lock-addr", 0);
1846                 return -1;
1847         }
1848
1849         if (show_lock_owner)
1850                 show_thread_stats = true;
1851
1852         return 0;
1853 }
1854
1855 static int __cmd_contention(int argc, const char **argv)
1856 {
1857         int err = -EINVAL;
1858         struct perf_tool eops = {
1859                 .attr            = perf_event__process_attr,
1860                 .event_update    = process_event_update,
1861                 .sample          = process_sample_event,
1862                 .comm            = perf_event__process_comm,
1863                 .mmap            = perf_event__process_mmap,
1864                 .tracing_data    = perf_event__process_tracing_data,
1865                 .ordered_events  = true,
1866         };
1867         struct perf_data data = {
1868                 .path  = input_name,
1869                 .mode  = PERF_DATA_MODE_READ,
1870                 .force = force,
1871         };
1872         struct lock_contention con = {
1873                 .target = &target,
1874                 .result = &lockhash_table[0],
1875                 .map_nr_entries = bpf_map_entries,
1876                 .max_stack = max_stack_depth,
1877                 .stack_skip = stack_skip,
1878                 .filters = &filters,
1879                 .save_callstack = needs_callstack(),
1880                 .owner = show_lock_owner,
1881         };
1882
1883         session = perf_session__new(use_bpf ? NULL : &data, &eops);
1884         if (IS_ERR(session)) {
1885                 pr_err("Initializing perf session failed\n");
1886                 return PTR_ERR(session);
1887         }
1888
1889         con.machine = &session->machines.host;
1890
1891         con.aggr_mode = aggr_mode = show_thread_stats ? LOCK_AGGR_TASK :
1892                 show_lock_addrs ? LOCK_AGGR_ADDR : LOCK_AGGR_CALLER;
1893
1894         if (con.aggr_mode == LOCK_AGGR_CALLER)
1895                 con.save_callstack = true;
1896
1897         /* for lock function check */
1898         symbol_conf.sort_by_name = true;
1899         symbol_conf.allow_aliases = true;
1900         symbol__init(&session->header.env);
1901
1902         if (use_bpf) {
1903                 err = target__validate(&target);
1904                 if (err) {
1905                         char errbuf[512];
1906
1907                         target__strerror(&target, err, errbuf, 512);
1908                         pr_err("%s\n", errbuf);
1909                         goto out_delete;
1910                 }
1911
1912                 signal(SIGINT, sighandler);
1913                 signal(SIGCHLD, sighandler);
1914                 signal(SIGTERM, sighandler);
1915
1916                 con.evlist = evlist__new();
1917                 if (con.evlist == NULL) {
1918                         err = -ENOMEM;
1919                         goto out_delete;
1920                 }
1921
1922                 err = evlist__create_maps(con.evlist, &target);
1923                 if (err < 0)
1924                         goto out_delete;
1925
1926                 if (argc) {
1927                         err = evlist__prepare_workload(con.evlist, &target,
1928                                                        argv, false, NULL);
1929                         if (err < 0)
1930                                 goto out_delete;
1931                 }
1932
1933                 if (lock_contention_prepare(&con) < 0) {
1934                         pr_err("lock contention BPF setup failed\n");
1935                         goto out_delete;
1936                 }
1937         } else if (!data.is_pipe) {
1938                 if (!perf_session__has_traces(session, "lock record"))
1939                         goto out_delete;
1940
1941                 if (!evlist__find_evsel_by_str(session->evlist,
1942                                                "lock:contention_begin")) {
1943                         pr_err("lock contention evsel not found\n");
1944                         goto out_delete;
1945                 }
1946
1947                 if (perf_session__set_tracepoints_handlers(session,
1948                                                 contention_tracepoints)) {
1949                         pr_err("Initializing perf session tracepoint handlers failed\n");
1950                         goto out_delete;
1951                 }
1952         }
1953
1954         if (setup_output_field(true, output_fields))
1955                 goto out_delete;
1956
1957         if (select_key(true))
1958                 goto out_delete;
1959
1960         if (use_bpf) {
1961                 lock_contention_start();
1962                 if (argc)
1963                         evlist__start_workload(con.evlist);
1964
1965                 /* wait for signal */
1966                 pause();
1967
1968                 lock_contention_stop();
1969                 lock_contention_read(&con);
1970         } else {
1971                 err = perf_session__process_events(session);
1972                 if (err)
1973                         goto out_delete;
1974         }
1975
1976         setup_pager();
1977
1978         sort_contention_result();
1979         print_contention_result(&con);
1980
1981 out_delete:
1982         lock_filter_finish();
1983         evlist__delete(con.evlist);
1984         lock_contention_finish();
1985         perf_session__delete(session);
1986         return err;
1987 }
1988
1989
1990 static int __cmd_record(int argc, const char **argv)
1991 {
1992         const char *record_args[] = {
1993                 "record", "-R", "-m", "1024", "-c", "1", "--synth", "task",
1994         };
1995         const char *callgraph_args[] = {
1996                 "--call-graph", "fp," __stringify(CONTENTION_STACK_DEPTH),
1997         };
1998         unsigned int rec_argc, i, j, ret;
1999         unsigned int nr_tracepoints;
2000         unsigned int nr_callgraph_args = 0;
2001         const char **rec_argv;
2002         bool has_lock_stat = true;
2003
2004         for (i = 0; i < ARRAY_SIZE(lock_tracepoints); i++) {
2005                 if (!is_valid_tracepoint(lock_tracepoints[i].name)) {
2006                         pr_debug("tracepoint %s is not enabled. "
2007                                  "Are CONFIG_LOCKDEP and CONFIG_LOCK_STAT enabled?\n",
2008                                  lock_tracepoints[i].name);
2009                         has_lock_stat = false;
2010                         break;
2011                 }
2012         }
2013
2014         if (has_lock_stat)
2015                 goto setup_args;
2016
2017         for (i = 0; i < ARRAY_SIZE(contention_tracepoints); i++) {
2018                 if (!is_valid_tracepoint(contention_tracepoints[i].name)) {
2019                         pr_err("tracepoint %s is not enabled.\n",
2020                                contention_tracepoints[i].name);
2021                         return 1;
2022                 }
2023         }
2024
2025         nr_callgraph_args = ARRAY_SIZE(callgraph_args);
2026
2027 setup_args:
2028         rec_argc = ARRAY_SIZE(record_args) + nr_callgraph_args + argc - 1;
2029
2030         if (has_lock_stat)
2031                 nr_tracepoints = ARRAY_SIZE(lock_tracepoints);
2032         else
2033                 nr_tracepoints = ARRAY_SIZE(contention_tracepoints);
2034
2035         /* factor of 2 is for -e in front of each tracepoint */
2036         rec_argc += 2 * nr_tracepoints;
2037
2038         rec_argv = calloc(rec_argc + 1, sizeof(char *));
2039         if (!rec_argv)
2040                 return -ENOMEM;
2041
2042         for (i = 0; i < ARRAY_SIZE(record_args); i++)
2043                 rec_argv[i] = strdup(record_args[i]);
2044
2045         for (j = 0; j < nr_tracepoints; j++) {
2046                 const char *ev_name;
2047
2048                 if (has_lock_stat)
2049                         ev_name = strdup(lock_tracepoints[j].name);
2050                 else
2051                         ev_name = strdup(contention_tracepoints[j].name);
2052
2053                 if (!ev_name)
2054                         return -ENOMEM;
2055
2056                 rec_argv[i++] = "-e";
2057                 rec_argv[i++] = ev_name;
2058         }
2059
2060         for (j = 0; j < nr_callgraph_args; j++, i++)
2061                 rec_argv[i] = callgraph_args[j];
2062
2063         for (j = 1; j < (unsigned int)argc; j++, i++)
2064                 rec_argv[i] = argv[j];
2065
2066         BUG_ON(i != rec_argc);
2067
2068         ret = cmd_record(i, rec_argv);
2069         free(rec_argv);
2070         return ret;
2071 }
2072
2073 static int parse_map_entry(const struct option *opt, const char *str,
2074                             int unset __maybe_unused)
2075 {
2076         unsigned long *len = (unsigned long *)opt->value;
2077         unsigned long val;
2078         char *endptr;
2079
2080         errno = 0;
2081         val = strtoul(str, &endptr, 0);
2082         if (*endptr != '\0' || errno != 0) {
2083                 pr_err("invalid BPF map length: %s\n", str);
2084                 return -1;
2085         }
2086
2087         *len = val;
2088         return 0;
2089 }
2090
2091 static int parse_max_stack(const struct option *opt, const char *str,
2092                            int unset __maybe_unused)
2093 {
2094         unsigned long *len = (unsigned long *)opt->value;
2095         long val;
2096         char *endptr;
2097
2098         errno = 0;
2099         val = strtol(str, &endptr, 0);
2100         if (*endptr != '\0' || errno != 0) {
2101                 pr_err("invalid max stack depth: %s\n", str);
2102                 return -1;
2103         }
2104
2105         if (val < 0 || val > sysctl__max_stack()) {
2106                 pr_err("invalid max stack depth: %ld\n", val);
2107                 return -1;
2108         }
2109
2110         *len = val;
2111         return 0;
2112 }
2113
2114 static bool add_lock_type(unsigned int flags)
2115 {
2116         unsigned int *tmp;
2117
2118         tmp = realloc(filters.types, (filters.nr_types + 1) * sizeof(*filters.types));
2119         if (tmp == NULL)
2120                 return false;
2121
2122         tmp[filters.nr_types++] = flags;
2123         filters.types = tmp;
2124         return true;
2125 }
2126
2127 static int parse_lock_type(const struct option *opt __maybe_unused, const char *str,
2128                            int unset __maybe_unused)
2129 {
2130         char *s, *tmp, *tok;
2131         int ret = 0;
2132
2133         s = strdup(str);
2134         if (s == NULL)
2135                 return -1;
2136
2137         for (tok = strtok_r(s, ", ", &tmp); tok; tok = strtok_r(NULL, ", ", &tmp)) {
2138                 unsigned int flags = get_type_flag(tok);
2139
2140                 if (flags == -1U) {
2141                         pr_err("Unknown lock flags: %s\n", tok);
2142                         ret = -1;
2143                         break;
2144                 }
2145
2146                 if (!add_lock_type(flags)) {
2147                         ret = -1;
2148                         break;
2149                 }
2150         }
2151
2152         free(s);
2153         return ret;
2154 }
2155
2156 static bool add_lock_addr(unsigned long addr)
2157 {
2158         unsigned long *tmp;
2159
2160         tmp = realloc(filters.addrs, (filters.nr_addrs + 1) * sizeof(*filters.addrs));
2161         if (tmp == NULL) {
2162                 pr_err("Memory allocation failure\n");
2163                 return false;
2164         }
2165
2166         tmp[filters.nr_addrs++] = addr;
2167         filters.addrs = tmp;
2168         return true;
2169 }
2170
2171 static bool add_lock_sym(char *name)
2172 {
2173         char **tmp;
2174         char *sym = strdup(name);
2175
2176         if (sym == NULL) {
2177                 pr_err("Memory allocation failure\n");
2178                 return false;
2179         }
2180
2181         tmp = realloc(filters.syms, (filters.nr_syms + 1) * sizeof(*filters.syms));
2182         if (tmp == NULL) {
2183                 pr_err("Memory allocation failure\n");
2184                 free(sym);
2185                 return false;
2186         }
2187
2188         tmp[filters.nr_syms++] = sym;
2189         filters.syms = tmp;
2190         return true;
2191 }
2192
2193 static int parse_lock_addr(const struct option *opt __maybe_unused, const char *str,
2194                            int unset __maybe_unused)
2195 {
2196         char *s, *tmp, *tok;
2197         int ret = 0;
2198         u64 addr;
2199
2200         s = strdup(str);
2201         if (s == NULL)
2202                 return -1;
2203
2204         for (tok = strtok_r(s, ", ", &tmp); tok; tok = strtok_r(NULL, ", ", &tmp)) {
2205                 char *end;
2206
2207                 addr = strtoul(tok, &end, 16);
2208                 if (*end == '\0') {
2209                         if (!add_lock_addr(addr)) {
2210                                 ret = -1;
2211                                 break;
2212                         }
2213                         continue;
2214                 }
2215
2216                 /*
2217                  * At this moment, we don't have kernel symbols.  Save the symbols
2218                  * in a separate list and resolve them to addresses later.
2219                  */
2220                 if (!add_lock_sym(tok)) {
2221                         ret = -1;
2222                         break;
2223                 }
2224         }
2225
2226         free(s);
2227         return ret;
2228 }
2229
2230 static int parse_call_stack(const struct option *opt __maybe_unused, const char *str,
2231                            int unset __maybe_unused)
2232 {
2233         char *s, *tmp, *tok;
2234         int ret = 0;
2235
2236         s = strdup(str);
2237         if (s == NULL)
2238                 return -1;
2239
2240         for (tok = strtok_r(s, ", ", &tmp); tok; tok = strtok_r(NULL, ", ", &tmp)) {
2241                 struct callstack_filter *entry;
2242
2243                 entry = malloc(sizeof(*entry) + strlen(tok) + 1);
2244                 if (entry == NULL) {
2245                         pr_err("Memory allocation failure\n");
2246                         return -1;
2247                 }
2248
2249                 strcpy(entry->name, tok);
2250                 list_add_tail(&entry->list, &callstack_filters);
2251         }
2252
2253         free(s);
2254         return ret;
2255 }
2256
2257 int cmd_lock(int argc, const char **argv)
2258 {
2259         const struct option lock_options[] = {
2260         OPT_STRING('i', "input", &input_name, "file", "input file name"),
2261         OPT_INCR('v', "verbose", &verbose, "be more verbose (show symbol address, etc)"),
2262         OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace, "dump raw trace in ASCII"),
2263         OPT_BOOLEAN('f', "force", &force, "don't complain, do it"),
2264         OPT_STRING(0, "vmlinux", &symbol_conf.vmlinux_name,
2265                    "file", "vmlinux pathname"),
2266         OPT_STRING(0, "kallsyms", &symbol_conf.kallsyms_name,
2267                    "file", "kallsyms pathname"),
2268         OPT_BOOLEAN('q', "quiet", &quiet, "Do not show any warnings or messages"),
2269         OPT_END()
2270         };
2271
2272         const struct option info_options[] = {
2273         OPT_BOOLEAN('t', "threads", &info_threads,
2274                     "dump thread list in perf.data"),
2275         OPT_BOOLEAN('m', "map", &info_map,
2276                     "map of lock instances (address:name table)"),
2277         OPT_PARENT(lock_options)
2278         };
2279
2280         const struct option report_options[] = {
2281         OPT_STRING('k', "key", &sort_key, "acquired",
2282                     "key for sorting (acquired / contended / avg_wait / wait_total / wait_max / wait_min)"),
2283         OPT_STRING('F', "field", &output_fields, NULL,
2284                     "output fields (acquired / contended / avg_wait / wait_total / wait_max / wait_min)"),
2285         /* TODO: type */
2286         OPT_BOOLEAN('c', "combine-locks", &combine_locks,
2287                     "combine locks in the same class"),
2288         OPT_BOOLEAN('t', "threads", &show_thread_stats,
2289                     "show per-thread lock stats"),
2290         OPT_INTEGER('E', "entries", &print_nr_entries, "display this many functions"),
2291         OPT_PARENT(lock_options)
2292         };
2293
2294         struct option contention_options[] = {
2295         OPT_STRING('k', "key", &sort_key, "wait_total",
2296                     "key for sorting (contended / wait_total / wait_max / wait_min / avg_wait)"),
2297         OPT_STRING('F', "field", &output_fields, "contended,wait_total,wait_max,avg_wait",
2298                     "output fields (contended / wait_total / wait_max / wait_min / avg_wait)"),
2299         OPT_BOOLEAN('t', "threads", &show_thread_stats,
2300                     "show per-thread lock stats"),
2301         OPT_BOOLEAN('b', "use-bpf", &use_bpf, "use BPF program to collect lock contention stats"),
2302         OPT_BOOLEAN('a', "all-cpus", &target.system_wide,
2303                     "System-wide collection from all CPUs"),
2304         OPT_STRING('C', "cpu", &target.cpu_list, "cpu",
2305                     "List of cpus to monitor"),
2306         OPT_STRING('p', "pid", &target.pid, "pid",
2307                    "Trace on existing process id"),
2308         OPT_STRING(0, "tid", &target.tid, "tid",
2309                    "Trace on existing thread id (exclusive to --pid)"),
2310         OPT_CALLBACK('M', "map-nr-entries", &bpf_map_entries, "num",
2311                      "Max number of BPF map entries", parse_map_entry),
2312         OPT_CALLBACK(0, "max-stack", &max_stack_depth, "num",
2313                      "Set the maximum stack depth when collecting lopck contention, "
2314                      "Default: " __stringify(CONTENTION_STACK_DEPTH), parse_max_stack),
2315         OPT_INTEGER(0, "stack-skip", &stack_skip,
2316                     "Set the number of stack depth to skip when finding a lock caller, "
2317                     "Default: " __stringify(CONTENTION_STACK_SKIP)),
2318         OPT_INTEGER('E', "entries", &print_nr_entries, "display this many functions"),
2319         OPT_BOOLEAN('l', "lock-addr", &show_lock_addrs, "show lock stats by address"),
2320         OPT_CALLBACK('Y', "type-filter", NULL, "FLAGS",
2321                      "Filter specific type of locks", parse_lock_type),
2322         OPT_CALLBACK('L', "lock-filter", NULL, "ADDRS/NAMES",
2323                      "Filter specific address/symbol of locks", parse_lock_addr),
2324         OPT_CALLBACK('S', "callstack-filter", NULL, "NAMES",
2325                      "Filter specific function in the callstack", parse_call_stack),
2326         OPT_BOOLEAN('o', "lock-owner", &show_lock_owner, "show lock owners instead of waiters"),
2327         OPT_PARENT(lock_options)
2328         };
2329
2330         const char * const info_usage[] = {
2331                 "perf lock info [<options>]",
2332                 NULL
2333         };
2334         const char *const lock_subcommands[] = { "record", "report", "script",
2335                                                  "info", "contention", NULL };
2336         const char *lock_usage[] = {
2337                 NULL,
2338                 NULL
2339         };
2340         const char * const report_usage[] = {
2341                 "perf lock report [<options>]",
2342                 NULL
2343         };
2344         const char * const contention_usage[] = {
2345                 "perf lock contention [<options>]",
2346                 NULL
2347         };
2348         unsigned int i;
2349         int rc = 0;
2350
2351         for (i = 0; i < LOCKHASH_SIZE; i++)
2352                 INIT_HLIST_HEAD(lockhash_table + i);
2353
2354         argc = parse_options_subcommand(argc, argv, lock_options, lock_subcommands,
2355                                         lock_usage, PARSE_OPT_STOP_AT_NON_OPTION);
2356         if (!argc)
2357                 usage_with_options(lock_usage, lock_options);
2358
2359         if (strlen(argv[0]) > 2 && strstarts("record", argv[0])) {
2360                 return __cmd_record(argc, argv);
2361         } else if (strlen(argv[0]) > 2 && strstarts("report", argv[0])) {
2362                 trace_handler = &report_lock_ops;
2363                 if (argc) {
2364                         argc = parse_options(argc, argv,
2365                                              report_options, report_usage, 0);
2366                         if (argc)
2367                                 usage_with_options(report_usage, report_options);
2368                 }
2369                 rc = __cmd_report(false);
2370         } else if (!strcmp(argv[0], "script")) {
2371                 /* Aliased to 'perf script' */
2372                 return cmd_script(argc, argv);
2373         } else if (!strcmp(argv[0], "info")) {
2374                 if (argc) {
2375                         argc = parse_options(argc, argv,
2376                                              info_options, info_usage, 0);
2377                         if (argc)
2378                                 usage_with_options(info_usage, info_options);
2379                 }
2380                 /* recycling report_lock_ops */
2381                 trace_handler = &report_lock_ops;
2382                 rc = __cmd_report(true);
2383         } else if (strlen(argv[0]) > 2 && strstarts("contention", argv[0])) {
2384                 trace_handler = &contention_lock_ops;
2385                 sort_key = "wait_total";
2386                 output_fields = "contended,wait_total,wait_max,avg_wait";
2387
2388 #ifndef HAVE_BPF_SKEL
2389                 set_option_nobuild(contention_options, 'b', "use-bpf",
2390                                    "no BUILD_BPF_SKEL=1", false);
2391 #endif
2392                 if (argc) {
2393                         argc = parse_options(argc, argv, contention_options,
2394                                              contention_usage, 0);
2395                 }
2396
2397                 if (check_lock_contention_options(contention_options,
2398                                                   contention_usage) < 0)
2399                         return -1;
2400
2401                 rc = __cmd_contention(argc, argv);
2402         } else {
2403                 usage_with_options(lock_usage, lock_options);
2404         }
2405
2406         return rc;
2407 }