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 <stdio.h>
  31 #include <sys/types.h>
  32 #include <sys/prctl.h>
  33 #include <semaphore.h>
  34 #include <math.h>
  35 #include <limits.h>
  36 #include <ctype.h>
  37
  38 #include <linux/list.h>
  39 #include <linux/hash.h>
  40 #include <linux/kernel.h>
  41 #include <linux/zalloc.h>
  42 #include <linux/err.h>
  43 #include <linux/stringify.h>
  44
  45 static struct perf_session *session;
  46 static struct target target;
  47
  48 /* based on kernel/lockdep.c */
  49 #define LOCKHASH_BITS           12
  50 #define LOCKHASH_SIZE           (1UL << LOCKHASH_BITS)
  51
  52 static struct hlist_head *lockhash_table;
  53
  54 #define __lockhashfn(key)       hash_long((unsigned long)key, LOCKHASH_BITS)
  55 #define lockhashentry(key)      (lockhash_table + __lockhashfn((key)))
  56
  57 static struct rb_root           thread_stats;
  58
  59 static bool combine_locks;
  60 static bool show_thread_stats;
  61 static bool show_lock_addrs;
  62 static bool show_lock_owner;
  63 static bool use_bpf;
  64 static unsigned long bpf_map_entries = MAX_ENTRIES;
  65 static int max_stack_depth = CONTENTION_STACK_DEPTH;
  66 static int stack_skip = CONTENTION_STACK_SKIP;
  67 static int print_nr_entries = INT_MAX / 2;
  68 static LIST_HEAD(callstack_filters);
  69 static const char *output_name = NULL;
  70 static FILE *lock_output;
  71
  72 struct callstack_filter {
  73         struct list_head list;
  74         char name[];
  75 };
  76
  77 static struct lock_filter filters;
  78
  79 static enum lock_aggr_mode aggr_mode = LOCK_AGGR_ADDR;
  80
  81 static bool needs_callstack(void)
  82 {
  83         return !list_empty(&callstack_filters);
  84 }
  85
  86 static struct thread_stat *thread_stat_find(u32 tid)
  87 {
  88         struct rb_node *node;
  89         struct thread_stat *st;
  90
  91         node = thread_stats.rb_node;
  92         while (node) {
  93                 st = container_of(node, struct thread_stat, rb);
  94                 if (st->tid == tid)
  95                         return st;
  96                 else if (tid < st->tid)
  97                         node = node->rb_left;
  98                 else
  99                         node = node->rb_right;
 100         }
 101
 102         return NULL;
 103 }
 104
 105 static void thread_stat_insert(struct thread_stat *new)
 106 {
 107         struct rb_node **rb = &thread_stats.rb_node;
 108         struct rb_node *parent = NULL;
 109         struct thread_stat *p;
 110
 111         while (*rb) {
 112                 p = container_of(*rb, struct thread_stat, rb);
 113                 parent = *rb;
 114
 115                 if (new->tid < p->tid)
 116                         rb = &(*rb)->rb_left;
 117                 else if (new->tid > p->tid)
 118                         rb = &(*rb)->rb_right;
 119                 else
 120                         BUG_ON("inserting invalid thread_stat\n");
 121         }
 122
 123         rb_link_node(&new->rb, parent, rb);
 124         rb_insert_color(&new->rb, &thread_stats);
 125 }
 126
 127 static struct thread_stat *thread_stat_findnew_after_first(u32 tid)
 128 {
 129         struct thread_stat *st;
 130
 131         st = thread_stat_find(tid);
 132         if (st)
 133                 return st;
 134
 135         st = zalloc(sizeof(struct thread_stat));
 136         if (!st) {
 137                 pr_err("memory allocation failed\n");
 138                 return NULL;
 139         }
 140
 141         st->tid = tid;
 142         INIT_LIST_HEAD(&st->seq_list);
 143
 144         thread_stat_insert(st);
 145
 146         return st;
 147 }
 148
 149 static struct thread_stat *thread_stat_findnew_first(u32 tid);
 150 static struct thread_stat *(*thread_stat_findnew)(u32 tid) =
 151         thread_stat_findnew_first;
 152
 153 static struct thread_stat *thread_stat_findnew_first(u32 tid)
 154 {
 155         struct thread_stat *st;
 156
 157         st = zalloc(sizeof(struct thread_stat));
 158         if (!st) {
 159                 pr_err("memory allocation failed\n");
 160                 return NULL;
 161         }
 162         st->tid = tid;
 163         INIT_LIST_HEAD(&st->seq_list);
 164
 165         rb_link_node(&st->rb, NULL, &thread_stats.rb_node);
 166         rb_insert_color(&st->rb, &thread_stats);
 167
 168         thread_stat_findnew = thread_stat_findnew_after_first;
 169         return st;
 170 }
 171
 172 /* build simple key function one is bigger than two */
 173 #define SINGLE_KEY(member)                                              \
 174         static int lock_stat_key_ ## member(struct lock_stat *one,      \
 175                                          struct lock_stat *two)         \
 176         {                                                               \
 177                 return one->member > two->member;                       \
 178         }
 179
 180 SINGLE_KEY(nr_acquired)
 181 SINGLE_KEY(nr_contended)
 182 SINGLE_KEY(avg_wait_time)
 183 SINGLE_KEY(wait_time_total)
 184 SINGLE_KEY(wait_time_max)
 185
 186 static int lock_stat_key_wait_time_min(struct lock_stat *one,
 187                                         struct lock_stat *two)
 188 {
 189         u64 s1 = one->wait_time_min;
 190         u64 s2 = two->wait_time_min;
 191         if (s1 == ULLONG_MAX)
 192                 s1 = 0;
 193         if (s2 == ULLONG_MAX)
 194                 s2 = 0;
 195         return s1 > s2;
 196 }
 197
 198 struct lock_key {
 199         /*
 200          * name: the value for specify by user
 201          * this should be simpler than raw name of member
 202          * e.g. nr_acquired -> acquired, wait_time_total -> wait_total
 203          */
 204         const char              *name;
 205         /* header: the string printed on the header line */
 206         const char              *header;
 207         /* len: the printing width of the field */
 208         int                     len;
 209         /* key: a pointer to function to compare two lock stats for sorting */
 210         int                     (*key)(struct lock_stat*, struct lock_stat*);
 211         /* print: a pointer to function to print a given lock stats */
 212         void                    (*print)(struct lock_key*, struct lock_stat*);
 213         /* list: list entry to link this */
 214         struct list_head        list;
 215 };
 216
 217 static void lock_stat_key_print_time(unsigned long long nsec, int len)
 218 {
 219         static const struct {
 220                 float base;
 221                 const char *unit;
 222         } table[] = {
 223                 { 1e9 * 3600, "h " },
 224                 { 1e9 * 60, "m " },
 225                 { 1e9, "s " },
 226                 { 1e6, "ms" },
 227                 { 1e3, "us" },
 228                 { 0, NULL },
 229         };
 230
 231         /* for CSV output */
 232         if (len == 0) {
 233                 fprintf(lock_output, "%llu", nsec);
 234                 return;
 235         }
 236
 237         for (int i = 0; table[i].unit; i++) {
 238                 if (nsec < table[i].base)
 239                         continue;
 240
 241                 fprintf(lock_output, "%*.2f %s", len - 3, nsec / table[i].base, table[i].unit);
 242                 return;
 243         }
 244
 245         fprintf(lock_output, "%*llu %s", len - 3, nsec, "ns");
 246 }
 247
 248 #define PRINT_KEY(member)                                               \
 249 static void lock_stat_key_print_ ## member(struct lock_key *key,        \
 250                                            struct lock_stat *ls)        \
 251 {                                                                       \
 252         fprintf(lock_output, "%*llu", key->len, (unsigned long long)ls->member);\
 253 }
 254
 255 #define PRINT_TIME(member)                                              \
 256 static void lock_stat_key_print_ ## member(struct lock_key *key,        \
 257                                            struct lock_stat *ls)        \
 258 {                                                                       \
 259         lock_stat_key_print_time((unsigned long long)ls->member, key->len);     \
 260 }
 261
 262 PRINT_KEY(nr_acquired)
 263 PRINT_KEY(nr_contended)
 264 PRINT_TIME(avg_wait_time)
 265 PRINT_TIME(wait_time_total)
 266 PRINT_TIME(wait_time_max)
 267
 268 static void lock_stat_key_print_wait_time_min(struct lock_key *key,
 269                                               struct lock_stat *ls)
 270 {
 271         u64 wait_time = ls->wait_time_min;
 272
 273         if (wait_time == ULLONG_MAX)
 274                 wait_time = 0;
 275
 276         lock_stat_key_print_time(wait_time, key->len);
 277 }
 278
 279
 280 static const char               *sort_key = "acquired";
 281
 282 static int                      (*compare)(struct lock_stat *, struct lock_stat *);
 283
 284 static struct rb_root           sorted; /* place to store intermediate data */
 285 static struct rb_root           result; /* place to store sorted data */
 286
 287 static LIST_HEAD(lock_keys);
 288 static const char               *output_fields;
 289
 290 #define DEF_KEY_LOCK(name, header, fn_suffix, len)                      \
 291         { #name, header, len, lock_stat_key_ ## fn_suffix, lock_stat_key_print_ ## fn_suffix, {} }
 292 static struct lock_key report_keys[] = {
 293         DEF_KEY_LOCK(acquired, "acquired", nr_acquired, 10),
 294         DEF_KEY_LOCK(contended, "contended", nr_contended, 10),
 295         DEF_KEY_LOCK(avg_wait, "avg wait", avg_wait_time, 12),
 296         DEF_KEY_LOCK(wait_total, "total wait", wait_time_total, 12),
 297         DEF_KEY_LOCK(wait_max, "max wait", wait_time_max, 12),
 298         DEF_KEY_LOCK(wait_min, "min wait", wait_time_min, 12),
 299
 300         /* extra comparisons much complicated should be here */
 301         { }
 302 };
 303
 304 static struct lock_key contention_keys[] = {
 305         DEF_KEY_LOCK(contended, "contended", nr_contended, 10),
 306         DEF_KEY_LOCK(wait_total, "total wait", wait_time_total, 12),
 307         DEF_KEY_LOCK(wait_max, "max wait", wait_time_max, 12),
 308         DEF_KEY_LOCK(wait_min, "min wait", wait_time_min, 12),
 309         DEF_KEY_LOCK(avg_wait, "avg wait", avg_wait_time, 12),
 310
 311         /* extra comparisons much complicated should be here */
 312         { }
 313 };
 314
 315 static int select_key(bool contention)
 316 {
 317         int i;
 318         struct lock_key *keys = report_keys;
 319
 320         if (contention)
 321                 keys = contention_keys;
 322
 323         for (i = 0; keys[i].name; i++) {
 324                 if (!strcmp(keys[i].name, sort_key)) {
 325                         compare = keys[i].key;
 326
 327                         /* selected key should be in the output fields */
 328                         if (list_empty(&keys[i].list))
 329                                 list_add_tail(&keys[i].list, &lock_keys);
 330
 331                         return 0;
 332                 }
 333         }
 334
 335         pr_err("Unknown compare key: %s\n", sort_key);
 336         return -1;
 337 }
 338
 339 static int add_output_field(bool contention, char *name)
 340 {
 341         int i;
 342         struct lock_key *keys = report_keys;
 343
 344         if (contention)
 345                 keys = contention_keys;
 346
 347         for (i = 0; keys[i].name; i++) {
 348                 if (strcmp(keys[i].name, name))
 349                         continue;
 350
 351                 /* prevent double link */
 352                 if (list_empty(&keys[i].list))
 353                         list_add_tail(&keys[i].list, &lock_keys);
 354
 355                 return 0;
 356         }
 357
 358         pr_err("Unknown output field: %s\n", name);
 359         return -1;
 360 }
 361
 362 static int setup_output_field(bool contention, const char *str)
 363 {
 364         char *tok, *tmp, *orig;
 365         int i, ret = 0;
 366         struct lock_key *keys = report_keys;
 367
 368         if (contention)
 369                 keys = contention_keys;
 370
 371         /* no output field given: use all of them */
 372         if (str == NULL) {
 373                 for (i = 0; keys[i].name; i++)
 374                         list_add_tail(&keys[i].list, &lock_keys);
 375                 return 0;
 376         }
 377
 378         for (i = 0; keys[i].name; i++)
 379                 INIT_LIST_HEAD(&keys[i].list);
 380
 381         orig = tmp = strdup(str);
 382         if (orig == NULL)
 383                 return -ENOMEM;
 384
 385         while ((tok = strsep(&tmp, ",")) != NULL){
 386                 ret = add_output_field(contention, tok);
 387                 if (ret < 0)
 388                         break;
 389         }
 390         free(orig);
 391
 392         return ret;
 393 }
 394
 395 static void combine_lock_stats(struct lock_stat *st)
 396 {
 397         struct rb_node **rb = &sorted.rb_node;
 398         struct rb_node *parent = NULL;
 399         struct lock_stat *p;
 400         int ret;
 401
 402         while (*rb) {
 403                 p = container_of(*rb, struct lock_stat, rb);
 404                 parent = *rb;
 405
 406                 if (st->name && p->name)
 407                         ret = strcmp(st->name, p->name);
 408                 else
 409                         ret = !!st->name - !!p->name;
 410
 411                 if (ret == 0) {
 412                         p->nr_acquired += st->nr_acquired;
 413                         p->nr_contended += st->nr_contended;
 414                         p->wait_time_total += st->wait_time_total;
 415
 416                         if (p->nr_contended)
 417                                 p->avg_wait_time = p->wait_time_total / p->nr_contended;
 418
 419                         if (p->wait_time_min > st->wait_time_min)
 420                                 p->wait_time_min = st->wait_time_min;
 421                         if (p->wait_time_max < st->wait_time_max)
 422                                 p->wait_time_max = st->wait_time_max;
 423
 424                         p->broken |= st->broken;
 425                         st->combined = 1;
 426                         return;
 427                 }
 428
 429                 if (ret < 0)
 430                         rb = &(*rb)->rb_left;
 431                 else
 432                         rb = &(*rb)->rb_right;
 433         }
 434
 435         rb_link_node(&st->rb, parent, rb);
 436         rb_insert_color(&st->rb, &sorted);
 437 }
 438
 439 static void insert_to_result(struct lock_stat *st,
 440                              int (*bigger)(struct lock_stat *, struct lock_stat *))
 441 {
 442         struct rb_node **rb = &result.rb_node;
 443         struct rb_node *parent = NULL;
 444         struct lock_stat *p;
 445
 446         if (combine_locks && st->combined)
 447                 return;
 448
 449         while (*rb) {
 450                 p = container_of(*rb, struct lock_stat, rb);
 451                 parent = *rb;
 452
 453                 if (bigger(st, p))
 454                         rb = &(*rb)->rb_left;
 455                 else
 456                         rb = &(*rb)->rb_right;
 457         }
 458
 459         rb_link_node(&st->rb, parent, rb);
 460         rb_insert_color(&st->rb, &result);
 461 }
 462
 463 /* returns left most element of result, and erase it */
 464 static struct lock_stat *pop_from_result(void)
 465 {
 466         struct rb_node *node = result.rb_node;
 467
 468         if (!node)
 469                 return NULL;
 470
 471         while (node->rb_left)
 472                 node = node->rb_left;
 473
 474         rb_erase(node, &result);
 475         return container_of(node, struct lock_stat, rb);
 476 }
 477
 478 struct lock_stat *lock_stat_find(u64 addr)
 479 {
 480         struct hlist_head *entry = lockhashentry(addr);
 481         struct lock_stat *ret;
 482
 483         hlist_for_each_entry(ret, entry, hash_entry) {
 484                 if (ret->addr == addr)
 485                         return ret;
 486         }
 487         return NULL;
 488 }
 489
 490 struct lock_stat *lock_stat_findnew(u64 addr, const char *name, int flags)
 491 {
 492         struct hlist_head *entry = lockhashentry(addr);
 493         struct lock_stat *ret, *new;
 494
 495         hlist_for_each_entry(ret, entry, hash_entry) {
 496                 if (ret->addr == addr)
 497                         return ret;
 498         }
 499
 500         new = zalloc(sizeof(struct lock_stat));
 501         if (!new)
 502                 goto alloc_failed;
 503
 504         new->addr = addr;
 505         new->name = strdup(name);
 506         if (!new->name) {
 507                 free(new);
 508                 goto alloc_failed;
 509         }
 510
 511         new->flags = flags;
 512         new->wait_time_min = ULLONG_MAX;
 513
 514         hlist_add_head(&new->hash_entry, entry);
 515         return new;
 516
 517 alloc_failed:
 518         pr_err("memory allocation failed\n");
 519         return NULL;
 520 }
 521
 522 bool match_callstack_filter(struct machine *machine, u64 *callstack)
 523 {
 524         struct map *kmap;
 525         struct symbol *sym;
 526         u64 ip;
 527
 528         if (list_empty(&callstack_filters))
 529                 return true;
 530
 531         for (int i = 0; i < max_stack_depth; i++) {
 532                 struct callstack_filter *filter;
 533
 534                 if (!callstack || !callstack[i])
 535                         break;
 536
 537                 ip = callstack[i];
 538                 sym = machine__find_kernel_symbol(machine, ip, &kmap);
 539                 if (sym == NULL)
 540                         continue;
 541
 542                 list_for_each_entry(filter, &callstack_filters, list) {
 543                         if (strstr(sym->name, filter->name))
 544                                 return true;
 545                 }
 546         }
 547         return false;
 548 }
 549
 550 struct trace_lock_handler {
 551         /* it's used on CONFIG_LOCKDEP */
 552         int (*acquire_event)(struct evsel *evsel,
 553                              struct perf_sample *sample);
 554
 555         /* it's used on CONFIG_LOCKDEP && CONFIG_LOCK_STAT */
 556         int (*acquired_event)(struct evsel *evsel,
 557                               struct perf_sample *sample);
 558
 559         /* it's used on CONFIG_LOCKDEP && CONFIG_LOCK_STAT */
 560         int (*contended_event)(struct evsel *evsel,
 561                                struct perf_sample *sample);
 562
 563         /* it's used on CONFIG_LOCKDEP */
 564         int (*release_event)(struct evsel *evsel,
 565                              struct perf_sample *sample);
 566
 567         /* it's used when CONFIG_LOCKDEP is off */
 568         int (*contention_begin_event)(struct evsel *evsel,
 569                                       struct perf_sample *sample);
 570
 571         /* it's used when CONFIG_LOCKDEP is off */
 572         int (*contention_end_event)(struct evsel *evsel,
 573                                     struct perf_sample *sample);
 574 };
 575
 576 static struct lock_seq_stat *get_seq(struct thread_stat *ts, u64 addr)
 577 {
 578         struct lock_seq_stat *seq;
 579
 580         list_for_each_entry(seq, &ts->seq_list, list) {
 581                 if (seq->addr == addr)
 582                         return seq;
 583         }
 584
 585         seq = zalloc(sizeof(struct lock_seq_stat));
 586         if (!seq) {
 587                 pr_err("memory allocation failed\n");
 588                 return NULL;
 589         }
 590         seq->state = SEQ_STATE_UNINITIALIZED;
 591         seq->addr = addr;
 592
 593         list_add(&seq->list, &ts->seq_list);
 594         return seq;
 595 }
 596
 597 enum broken_state {
 598         BROKEN_ACQUIRE,
 599         BROKEN_ACQUIRED,
 600         BROKEN_CONTENDED,
 601         BROKEN_RELEASE,
 602         BROKEN_MAX,
 603 };
 604
 605 static int bad_hist[BROKEN_MAX];
 606
 607 enum acquire_flags {
 608         TRY_LOCK = 1,
 609         READ_LOCK = 2,
 610 };
 611
 612 static int get_key_by_aggr_mode_simple(u64 *key, u64 addr, u32 tid)
 613 {
 614         switch (aggr_mode) {
 615         case LOCK_AGGR_ADDR:
 616                 *key = addr;
 617                 break;
 618         case LOCK_AGGR_TASK:
 619                 *key = tid;
 620                 break;
 621         case LOCK_AGGR_CALLER:
 622         default:
 623                 pr_err("Invalid aggregation mode: %d\n", aggr_mode);
 624                 return -EINVAL;
 625         }
 626         return 0;
 627 }
 628
 629 static u64 callchain_id(struct evsel *evsel, struct perf_sample *sample);
 630
 631 static int get_key_by_aggr_mode(u64 *key, u64 addr, struct evsel *evsel,
 632                                  struct perf_sample *sample)
 633 {
 634         if (aggr_mode == LOCK_AGGR_CALLER) {
 635                 *key = callchain_id(evsel, sample);
 636                 return 0;
 637         }
 638         return get_key_by_aggr_mode_simple(key, addr, sample->tid);
 639 }
 640
 641 static int report_lock_acquire_event(struct evsel *evsel,
 642                                      struct perf_sample *sample)
 643 {
 644         struct lock_stat *ls;
 645         struct thread_stat *ts;
 646         struct lock_seq_stat *seq;
 647         const char *name = evsel__strval(evsel, sample, "name");
 648         u64 addr = evsel__intval(evsel, sample, "lockdep_addr");
 649         int flag = evsel__intval(evsel, sample, "flags");
 650         u64 key;
 651         int ret;
 652
 653         ret = get_key_by_aggr_mode_simple(&key, addr, sample->tid);
 654         if (ret < 0)
 655                 return ret;
 656
 657         ls = lock_stat_findnew(key, name, 0);
 658         if (!ls)
 659                 return -ENOMEM;
 660
 661         ts = thread_stat_findnew(sample->tid);
 662         if (!ts)
 663                 return -ENOMEM;
 664
 665         seq = get_seq(ts, addr);
 666         if (!seq)
 667                 return -ENOMEM;
 668
 669         switch (seq->state) {
 670         case SEQ_STATE_UNINITIALIZED:
 671         case SEQ_STATE_RELEASED:
 672                 if (!flag) {
 673                         seq->state = SEQ_STATE_ACQUIRING;
 674                 } else {
 675                         if (flag & TRY_LOCK)
 676                                 ls->nr_trylock++;
 677                         if (flag & READ_LOCK)
 678                                 ls->nr_readlock++;
 679                         seq->state = SEQ_STATE_READ_ACQUIRED;
 680                         seq->read_count = 1;
 681                         ls->nr_acquired++;
 682                 }
 683                 break;
 684         case SEQ_STATE_READ_ACQUIRED:
 685                 if (flag & READ_LOCK) {
 686                         seq->read_count++;
 687                         ls->nr_acquired++;
 688                         goto end;
 689                 } else {
 690                         goto broken;
 691                 }
 692                 break;
 693         case SEQ_STATE_ACQUIRED:
 694         case SEQ_STATE_ACQUIRING:
 695         case SEQ_STATE_CONTENDED:
 696 broken:
 697                 /* broken lock sequence */
 698                 if (!ls->broken) {
 699                         ls->broken = 1;
 700                         bad_hist[BROKEN_ACQUIRE]++;
 701                 }
 702                 list_del_init(&seq->list);
 703                 free(seq);
 704                 goto end;
 705         default:
 706                 BUG_ON("Unknown state of lock sequence found!\n");
 707                 break;
 708         }
 709
 710         ls->nr_acquire++;
 711         seq->prev_event_time = sample->time;
 712 end:
 713         return 0;
 714 }
 715
 716 static int report_lock_acquired_event(struct evsel *evsel,
 717                                       struct perf_sample *sample)
 718 {
 719         struct lock_stat *ls;
 720         struct thread_stat *ts;
 721         struct lock_seq_stat *seq;
 722         u64 contended_term;
 723         const char *name = evsel__strval(evsel, sample, "name");
 724         u64 addr = evsel__intval(evsel, sample, "lockdep_addr");
 725         u64 key;
 726         int ret;
 727
 728         ret = get_key_by_aggr_mode_simple(&key, addr, sample->tid);
 729         if (ret < 0)
 730                 return ret;
 731
 732         ls = lock_stat_findnew(key, name, 0);
 733         if (!ls)
 734                 return -ENOMEM;
 735
 736         ts = thread_stat_findnew(sample->tid);
 737         if (!ts)
 738                 return -ENOMEM;
 739
 740         seq = get_seq(ts, addr);
 741         if (!seq)
 742                 return -ENOMEM;
 743
 744         switch (seq->state) {
 745         case SEQ_STATE_UNINITIALIZED:
 746                 /* orphan event, do nothing */
 747                 return 0;
 748         case SEQ_STATE_ACQUIRING:
 749                 break;
 750         case SEQ_STATE_CONTENDED:
 751                 contended_term = sample->time - seq->prev_event_time;
 752                 ls->wait_time_total += contended_term;
 753                 if (contended_term < ls->wait_time_min)
 754                         ls->wait_time_min = contended_term;
 755                 if (ls->wait_time_max < contended_term)
 756                         ls->wait_time_max = contended_term;
 757                 break;
 758         case SEQ_STATE_RELEASED:
 759         case SEQ_STATE_ACQUIRED:
 760         case SEQ_STATE_READ_ACQUIRED:
 761                 /* broken lock sequence */
 762                 if (!ls->broken) {
 763                         ls->broken = 1;
 764                         bad_hist[BROKEN_ACQUIRED]++;
 765                 }
 766                 list_del_init(&seq->list);
 767                 free(seq);
 768                 goto end;
 769         default:
 770                 BUG_ON("Unknown state of lock sequence found!\n");
 771                 break;
 772         }
 773
 774         seq->state = SEQ_STATE_ACQUIRED;
 775         ls->nr_acquired++;
 776         ls->avg_wait_time = ls->nr_contended ? ls->wait_time_total/ls->nr_contended : 0;
 777         seq->prev_event_time = sample->time;
 778 end:
 779         return 0;
 780 }
 781
 782 static int report_lock_contended_event(struct evsel *evsel,
 783                                        struct perf_sample *sample)
 784 {
 785         struct lock_stat *ls;
 786         struct thread_stat *ts;
 787         struct lock_seq_stat *seq;
 788         const char *name = evsel__strval(evsel, sample, "name");
 789         u64 addr = evsel__intval(evsel, sample, "lockdep_addr");
 790         u64 key;
 791         int ret;
 792
 793         ret = get_key_by_aggr_mode_simple(&key, addr, sample->tid);
 794         if (ret < 0)
 795                 return ret;
 796
 797         ls = lock_stat_findnew(key, name, 0);
 798         if (!ls)
 799                 return -ENOMEM;
 800
 801         ts = thread_stat_findnew(sample->tid);
 802         if (!ts)
 803                 return -ENOMEM;
 804
 805         seq = get_seq(ts, addr);
 806         if (!seq)
 807                 return -ENOMEM;
 808
 809         switch (seq->state) {
 810         case SEQ_STATE_UNINITIALIZED:
 811                 /* orphan event, do nothing */
 812                 return 0;
 813         case SEQ_STATE_ACQUIRING:
 814                 break;
 815         case SEQ_STATE_RELEASED:
 816         case SEQ_STATE_ACQUIRED:
 817         case SEQ_STATE_READ_ACQUIRED:
 818         case SEQ_STATE_CONTENDED:
 819                 /* broken lock sequence */
 820                 if (!ls->broken) {
 821                         ls->broken = 1;
 822                         bad_hist[BROKEN_CONTENDED]++;
 823                 }
 824                 list_del_init(&seq->list);
 825                 free(seq);
 826                 goto end;
 827         default:
 828                 BUG_ON("Unknown state of lock sequence found!\n");
 829                 break;
 830         }
 831
 832         seq->state = SEQ_STATE_CONTENDED;
 833         ls->nr_contended++;
 834         ls->avg_wait_time = ls->wait_time_total/ls->nr_contended;
 835         seq->prev_event_time = sample->time;
 836 end:
 837         return 0;
 838 }
 839
 840 static int report_lock_release_event(struct evsel *evsel,
 841                                      struct perf_sample *sample)
 842 {
 843         struct lock_stat *ls;
 844         struct thread_stat *ts;
 845         struct lock_seq_stat *seq;
 846         const char *name = evsel__strval(evsel, sample, "name");
 847         u64 addr = evsel__intval(evsel, sample, "lockdep_addr");
 848         u64 key;
 849         int ret;
 850
 851         ret = get_key_by_aggr_mode_simple(&key, addr, sample->tid);
 852         if (ret < 0)
 853                 return ret;
 854
 855         ls = lock_stat_findnew(key, name, 0);
 856         if (!ls)
 857                 return -ENOMEM;
 858
 859         ts = thread_stat_findnew(sample->tid);
 860         if (!ts)
 861                 return -ENOMEM;
 862
 863         seq = get_seq(ts, addr);
 864         if (!seq)
 865                 return -ENOMEM;
 866
 867         switch (seq->state) {
 868         case SEQ_STATE_UNINITIALIZED:
 869                 goto end;
 870         case SEQ_STATE_ACQUIRED:
 871                 break;
 872         case SEQ_STATE_READ_ACQUIRED:
 873                 seq->read_count--;
 874                 BUG_ON(seq->read_count < 0);
 875                 if (seq->read_count) {
 876                         ls->nr_release++;
 877                         goto end;
 878                 }
 879                 break;
 880         case SEQ_STATE_ACQUIRING:
 881         case SEQ_STATE_CONTENDED:
 882         case SEQ_STATE_RELEASED:
 883                 /* broken lock sequence */
 884                 if (!ls->broken) {
 885                         ls->broken = 1;
 886                         bad_hist[BROKEN_RELEASE]++;
 887                 }
 888                 goto free_seq;
 889         default:
 890                 BUG_ON("Unknown state of lock sequence found!\n");
 891                 break;
 892         }
 893
 894         ls->nr_release++;
 895 free_seq:
 896         list_del_init(&seq->list);
 897         free(seq);
 898 end:
 899         return 0;
 900 }
 901
 902 static int get_symbol_name_offset(struct map *map, struct symbol *sym, u64 ip,
 903                                   char *buf, int size)
 904 {
 905         u64 offset;
 906
 907         if (map == NULL || sym == NULL) {
 908                 buf[0] = '\0';
 909                 return 0;
 910         }
 911
 912         offset = map__map_ip(map, ip) - sym->start;
 913
 914         if (offset)
 915                 return scnprintf(buf, size, "%s+%#lx", sym->name, offset);
 916         else
 917                 return strlcpy(buf, sym->name, size);
 918 }
 919 static int lock_contention_caller(struct evsel *evsel, struct perf_sample *sample,
 920                                   char *buf, int size)
 921 {
 922         struct thread *thread;
 923         struct callchain_cursor *cursor;
 924         struct machine *machine = &session->machines.host;
 925         struct symbol *sym;
 926         int skip = 0;
 927         int ret;
 928
 929         /* lock names will be replaced to task name later */
 930         if (show_thread_stats)
 931                 return -1;
 932
 933         thread = machine__findnew_thread(machine, -1, sample->pid);
 934         if (thread == NULL)
 935                 return -1;
 936
 937         cursor = get_tls_callchain_cursor();
 938
 939         /* use caller function name from the callchain */
 940         ret = thread__resolve_callchain(thread, cursor, evsel, sample,
 941                                         NULL, NULL, max_stack_depth);
 942         if (ret != 0) {
 943                 thread__put(thread);
 944                 return -1;
 945         }
 946
 947         callchain_cursor_commit(cursor);
 948         thread__put(thread);
 949
 950         while (true) {
 951                 struct callchain_cursor_node *node;
 952
 953                 node = callchain_cursor_current(cursor);
 954                 if (node == NULL)
 955                         break;
 956
 957                 /* skip first few entries - for lock functions */
 958                 if (++skip <= stack_skip)
 959                         goto next;
 960
 961                 sym = node->ms.sym;
 962                 if (sym && !machine__is_lock_function(machine, node->ip)) {
 963                         get_symbol_name_offset(node->ms.map, sym, node->ip,
 964                                                buf, size);
 965                         return 0;
 966                 }
 967
 968 next:
 969                 callchain_cursor_advance(cursor);
 970         }
 971         return -1;
 972 }
 973
 974 static u64 callchain_id(struct evsel *evsel, struct perf_sample *sample)
 975 {
 976         struct callchain_cursor *cursor;
 977         struct machine *machine = &session->machines.host;
 978         struct thread *thread;
 979         u64 hash = 0;
 980         int skip = 0;
 981         int ret;
 982
 983         thread = machine__findnew_thread(machine, -1, sample->pid);
 984         if (thread == NULL)
 985                 return -1;
 986
 987         cursor = get_tls_callchain_cursor();
 988         /* use caller function name from the callchain */
 989         ret = thread__resolve_callchain(thread, cursor, evsel, sample,
 990                                         NULL, NULL, max_stack_depth);
 991         thread__put(thread);
 992
 993         if (ret != 0)
 994                 return -1;
 995
 996         callchain_cursor_commit(cursor);
 997
 998         while (true) {
 999                 struct callchain_cursor_node *node;
1000
1001                 node = callchain_cursor_current(cursor);
1002                 if (node == NULL)
1003                         break;
1004
1005                 /* skip first few entries - for lock functions */
1006                 if (++skip <= stack_skip)
1007                         goto next;
1008
1009                 if (node->ms.sym && machine__is_lock_function(machine, node->ip))
1010                         goto next;
1011
1012                 hash ^= hash_long((unsigned long)node->ip, 64);
1013
1014 next:
1015                 callchain_cursor_advance(cursor);
1016         }
1017         return hash;
1018 }
1019
1020 static u64 *get_callstack(struct perf_sample *sample, int max_stack)
1021 {
1022         u64 *callstack;
1023         u64 i;
1024         int c;
1025
1026         callstack = calloc(max_stack, sizeof(*callstack));
1027         if (callstack == NULL)
1028                 return NULL;
1029
1030         for (i = 0, c = 0; i < sample->callchain->nr && c < max_stack; i++) {
1031                 u64 ip = sample->callchain->ips[i];
1032
1033                 if (ip >= PERF_CONTEXT_MAX)
1034                         continue;
1035
1036                 callstack[c++] = ip;
1037         }
1038         return callstack;
1039 }
1040
1041 static int report_lock_contention_begin_event(struct evsel *evsel,
1042                                               struct perf_sample *sample)
1043 {
1044         struct lock_stat *ls;
1045         struct thread_stat *ts;
1046         struct lock_seq_stat *seq;
1047         u64 addr = evsel__intval(evsel, sample, "lock_addr");
1048         unsigned int flags = evsel__intval(evsel, sample, "flags");
1049         u64 key;
1050         int i, ret;
1051         static bool kmap_loaded;
1052         struct machine *machine = &session->machines.host;
1053         struct map *kmap;
1054         struct symbol *sym;
1055
1056         ret = get_key_by_aggr_mode(&key, addr, evsel, sample);
1057         if (ret < 0)
1058                 return ret;
1059
1060         if (!kmap_loaded) {
1061                 unsigned long *addrs;
1062
1063                 /* make sure it loads the kernel map to find lock symbols */
1064                 map__load(machine__kernel_map(machine));
1065                 kmap_loaded = true;
1066
1067                 /* convert (kernel) symbols to addresses */
1068                 for (i = 0; i < filters.nr_syms; i++) {
1069                         sym = machine__find_kernel_symbol_by_name(machine,
1070                                                                   filters.syms[i],
1071                                                                   &kmap);
1072                         if (sym == NULL) {
1073                                 pr_warning("ignore unknown symbol: %s\n",
1074                                            filters.syms[i]);
1075                                 continue;
1076                         }
1077
1078                         addrs = realloc(filters.addrs,
1079                                         (filters.nr_addrs + 1) * sizeof(*addrs));
1080                         if (addrs == NULL) {
1081                                 pr_warning("memory allocation failure\n");
1082                                 return -ENOMEM;
1083                         }
1084
1085                         addrs[filters.nr_addrs++] = map__unmap_ip(kmap, sym->start);
1086                         filters.addrs = addrs;
1087                 }
1088         }
1089
1090         ls = lock_stat_find(key);
1091         if (!ls) {
1092                 char buf[128];
1093                 const char *name = "";
1094
1095                 switch (aggr_mode) {
1096                 case LOCK_AGGR_ADDR:
1097                         sym = machine__find_kernel_symbol(machine, key, &kmap);
1098                         if (sym)
1099                                 name = sym->name;
1100                         break;
1101                 case LOCK_AGGR_CALLER:
1102                         name = buf;
1103                         if (lock_contention_caller(evsel, sample, buf, sizeof(buf)) < 0)
1104                                 name = "Unknown";
1105                         break;
1106                 case LOCK_AGGR_TASK:
1107                 default:
1108                         break;
1109                 }
1110
1111                 ls = lock_stat_findnew(key, name, flags);
1112                 if (!ls)
1113                         return -ENOMEM;
1114         }
1115
1116         if (filters.nr_types) {
1117                 bool found = false;
1118
1119                 for (i = 0; i < filters.nr_types; i++) {
1120                         if (flags == filters.types[i]) {
1121                                 found = true;
1122                                 break;
1123                         }
1124                 }
1125
1126                 if (!found)
1127                         return 0;
1128         }
1129
1130         if (filters.nr_addrs) {
1131                 bool found = false;
1132
1133                 for (i = 0; i < filters.nr_addrs; i++) {
1134                         if (addr == filters.addrs[i]) {
1135                                 found = true;
1136                                 break;
1137                         }
1138                 }
1139
1140                 if (!found)
1141                         return 0;
1142         }
1143
1144         if (needs_callstack()) {
1145                 u64 *callstack = get_callstack(sample, max_stack_depth);
1146                 if (callstack == NULL)
1147                         return -ENOMEM;
1148
1149                 if (!match_callstack_filter(machine, callstack)) {
1150                         free(callstack);
1151                         return 0;
1152                 }
1153
1154                 if (ls->callstack == NULL)
1155                         ls->callstack = callstack;
1156                 else
1157                         free(callstack);
1158         }
1159
1160         ts = thread_stat_findnew(sample->tid);
1161         if (!ts)
1162                 return -ENOMEM;
1163
1164         seq = get_seq(ts, addr);
1165         if (!seq)
1166                 return -ENOMEM;
1167
1168         switch (seq->state) {
1169         case SEQ_STATE_UNINITIALIZED:
1170         case SEQ_STATE_ACQUIRED:
1171                 break;
1172         case SEQ_STATE_CONTENDED:
1173                 /*
1174                  * It can have nested contention begin with mutex spinning,
1175                  * then we would use the original contention begin event and
1176                  * ignore the second one.
1177                  */
1178                 goto end;
1179         case SEQ_STATE_ACQUIRING:
1180         case SEQ_STATE_READ_ACQUIRED:
1181         case SEQ_STATE_RELEASED:
1182                 /* broken lock sequence */
1183                 if (!ls->broken) {
1184                         ls->broken = 1;
1185                         bad_hist[BROKEN_CONTENDED]++;
1186                 }
1187                 list_del_init(&seq->list);
1188                 free(seq);
1189                 goto end;
1190         default:
1191                 BUG_ON("Unknown state of lock sequence found!\n");
1192                 break;
1193         }
1194
1195         if (seq->state != SEQ_STATE_CONTENDED) {
1196                 seq->state = SEQ_STATE_CONTENDED;
1197                 seq->prev_event_time = sample->time;
1198                 ls->nr_contended++;
1199         }
1200 end:
1201         return 0;
1202 }
1203
1204 static int report_lock_contention_end_event(struct evsel *evsel,
1205                                             struct perf_sample *sample)
1206 {
1207         struct lock_stat *ls;
1208         struct thread_stat *ts;
1209         struct lock_seq_stat *seq;
1210         u64 contended_term;
1211         u64 addr = evsel__intval(evsel, sample, "lock_addr");
1212         u64 key;
1213         int ret;
1214
1215         ret = get_key_by_aggr_mode(&key, addr, evsel, sample);
1216         if (ret < 0)
1217                 return ret;
1218
1219         ls = lock_stat_find(key);
1220         if (!ls)
1221                 return 0;
1222
1223         ts = thread_stat_find(sample->tid);
1224         if (!ts)
1225                 return 0;
1226
1227         seq = get_seq(ts, addr);
1228         if (!seq)
1229                 return -ENOMEM;
1230
1231         switch (seq->state) {
1232         case SEQ_STATE_UNINITIALIZED:
1233                 goto end;
1234         case SEQ_STATE_CONTENDED:
1235                 contended_term = sample->time - seq->prev_event_time;
1236                 ls->wait_time_total += contended_term;
1237                 if (contended_term < ls->wait_time_min)
1238                         ls->wait_time_min = contended_term;
1239                 if (ls->wait_time_max < contended_term)
1240                         ls->wait_time_max = contended_term;
1241                 break;
1242         case SEQ_STATE_ACQUIRING:
1243         case SEQ_STATE_ACQUIRED:
1244         case SEQ_STATE_READ_ACQUIRED:
1245         case SEQ_STATE_RELEASED:
1246                 /* broken lock sequence */
1247                 if (!ls->broken) {
1248                         ls->broken = 1;
1249                         bad_hist[BROKEN_ACQUIRED]++;
1250                 }
1251                 list_del_init(&seq->list);
1252                 free(seq);
1253                 goto end;
1254         default:
1255                 BUG_ON("Unknown state of lock sequence found!\n");
1256                 break;
1257         }
1258
1259         seq->state = SEQ_STATE_ACQUIRED;
1260         ls->nr_acquired++;
1261         ls->avg_wait_time = ls->wait_time_total/ls->nr_acquired;
1262 end:
1263         return 0;
1264 }
1265
1266 /* lock oriented handlers */
1267 /* TODO: handlers for CPU oriented, thread oriented */
1268 static struct trace_lock_handler report_lock_ops  = {
1269         .acquire_event          = report_lock_acquire_event,
1270         .acquired_event         = report_lock_acquired_event,
1271         .contended_event        = report_lock_contended_event,
1272         .release_event          = report_lock_release_event,
1273         .contention_begin_event = report_lock_contention_begin_event,
1274         .contention_end_event   = report_lock_contention_end_event,
1275 };
1276
1277 static struct trace_lock_handler contention_lock_ops  = {
1278         .contention_begin_event = report_lock_contention_begin_event,
1279         .contention_end_event   = report_lock_contention_end_event,
1280 };
1281
1282
1283 static struct trace_lock_handler *trace_handler;
1284
1285 static int evsel__process_lock_acquire(struct evsel *evsel, struct perf_sample *sample)
1286 {
1287         if (trace_handler->acquire_event)
1288                 return trace_handler->acquire_event(evsel, sample);
1289         return 0;
1290 }
1291
1292 static int evsel__process_lock_acquired(struct evsel *evsel, struct perf_sample *sample)
1293 {
1294         if (trace_handler->acquired_event)
1295                 return trace_handler->acquired_event(evsel, sample);
1296         return 0;
1297 }
1298
1299 static int evsel__process_lock_contended(struct evsel *evsel, struct perf_sample *sample)
1300 {
1301         if (trace_handler->contended_event)
1302                 return trace_handler->contended_event(evsel, sample);
1303         return 0;
1304 }
1305
1306 static int evsel__process_lock_release(struct evsel *evsel, struct perf_sample *sample)
1307 {
1308         if (trace_handler->release_event)
1309                 return trace_handler->release_event(evsel, sample);
1310         return 0;
1311 }
1312
1313 static int evsel__process_contention_begin(struct evsel *evsel, struct perf_sample *sample)
1314 {
1315         if (trace_handler->contention_begin_event)
1316                 return trace_handler->contention_begin_event(evsel, sample);
1317         return 0;
1318 }
1319
1320 static int evsel__process_contention_end(struct evsel *evsel, struct perf_sample *sample)
1321 {
1322         if (trace_handler->contention_end_event)
1323                 return trace_handler->contention_end_event(evsel, sample);
1324         return 0;
1325 }
1326
1327 static void print_bad_events(int bad, int total)
1328 {
1329         /* Output for debug, this have to be removed */
1330         int i;
1331         int broken = 0;
1332         const char *name[4] =
1333                 { "acquire", "acquired", "contended", "release" };
1334
1335         for (i = 0; i < BROKEN_MAX; i++)
1336                 broken += bad_hist[i];
1337
1338         if (quiet || total == 0 || (broken == 0 && verbose <= 0))
1339                 return;
1340
1341         fprintf(lock_output, "\n=== output for debug ===\n\n");
1342         fprintf(lock_output, "bad: %d, total: %d\n", bad, total);
1343         fprintf(lock_output, "bad rate: %.2f %%\n", (double)bad / (double)total * 100);
1344         fprintf(lock_output, "histogram of events caused bad sequence\n");
1345         for (i = 0; i < BROKEN_MAX; i++)
1346                 fprintf(lock_output, " %10s: %d\n", name[i], bad_hist[i]);
1347 }
1348
1349 /* TODO: various way to print, coloring, nano or milli sec */
1350 static void print_result(void)
1351 {
1352         struct lock_stat *st;
1353         struct lock_key *key;
1354         char cut_name[20];
1355         int bad, total, printed;
1356
1357         if (!quiet) {
1358                 fprintf(lock_output, "%20s ", "Name");
1359                 list_for_each_entry(key, &lock_keys, list)
1360                         fprintf(lock_output, "%*s ", key->len, key->header);
1361                 fprintf(lock_output, "\n\n");
1362         }
1363
1364         bad = total = printed = 0;
1365         while ((st = pop_from_result())) {
1366                 total++;
1367                 if (st->broken)
1368                         bad++;
1369                 if (!st->nr_acquired)
1370                         continue;
1371
1372                 bzero(cut_name, 20);
1373
1374                 if (strlen(st->name) < 20) {
1375                         /* output raw name */
1376                         const char *name = st->name;
1377
1378                         if (show_thread_stats) {
1379                                 struct thread *t;
1380
1381                                 /* st->addr contains tid of thread */
1382                                 t = perf_session__findnew(session, st->addr);
1383                                 name = thread__comm_str(t);
1384                         }
1385
1386                         fprintf(lock_output, "%20s ", name);
1387                 } else {
1388                         strncpy(cut_name, st->name, 16);
1389                         cut_name[16] = '.';
1390                         cut_name[17] = '.';
1391                         cut_name[18] = '.';
1392                         cut_name[19] = '\0';
1393                         /* cut off name for saving output style */
1394                         fprintf(lock_output, "%20s ", cut_name);
1395                 }
1396
1397                 list_for_each_entry(key, &lock_keys, list) {
1398                         key->print(key, st);
1399                         fprintf(lock_output, " ");
1400                 }
1401                 fprintf(lock_output, "\n");
1402
1403                 if (++printed >= print_nr_entries)
1404                         break;
1405         }
1406
1407         print_bad_events(bad, total);
1408 }
1409
1410 static bool info_threads, info_map;
1411
1412 static void dump_threads(void)
1413 {
1414         struct thread_stat *st;
1415         struct rb_node *node;
1416         struct thread *t;
1417
1418         fprintf(lock_output, "%10s: comm\n", "Thread ID");
1419
1420         node = rb_first(&thread_stats);
1421         while (node) {
1422                 st = container_of(node, struct thread_stat, rb);
1423                 t = perf_session__findnew(session, st->tid);
1424                 fprintf(lock_output, "%10d: %s\n", st->tid, thread__comm_str(t));
1425                 node = rb_next(node);
1426                 thread__put(t);
1427         }
1428 }
1429
1430 static int compare_maps(struct lock_stat *a, struct lock_stat *b)
1431 {
1432         int ret;
1433
1434         if (a->name && b->name)
1435                 ret = strcmp(a->name, b->name);
1436         else
1437                 ret = !!a->name - !!b->name;
1438
1439         if (!ret)
1440                 return a->addr < b->addr;
1441         else
1442                 return ret < 0;
1443 }
1444
1445 static void dump_map(void)
1446 {
1447         unsigned int i;
1448         struct lock_stat *st;
1449
1450         fprintf(lock_output, "Address of instance: name of class\n");
1451         for (i = 0; i < LOCKHASH_SIZE; i++) {
1452                 hlist_for_each_entry(st, &lockhash_table[i], hash_entry) {
1453                         insert_to_result(st, compare_maps);
1454                 }
1455         }
1456
1457         while ((st = pop_from_result()))
1458                 fprintf(lock_output, " %#llx: %s\n", (unsigned long long)st->addr, st->name);
1459 }
1460
1461 static int dump_info(void)
1462 {
1463         int rc = 0;
1464
1465         if (info_threads)
1466                 dump_threads();
1467         else if (info_map)
1468                 dump_map();
1469         else {
1470                 rc = -1;
1471                 pr_err("Unknown type of information\n");
1472         }
1473
1474         return rc;
1475 }
1476
1477 static const struct evsel_str_handler lock_tracepoints[] = {
1478         { "lock:lock_acquire",   evsel__process_lock_acquire,   }, /* CONFIG_LOCKDEP */
1479         { "lock:lock_acquired",  evsel__process_lock_acquired,  }, /* CONFIG_LOCKDEP, CONFIG_LOCK_STAT */
1480         { "lock:lock_contended", evsel__process_lock_contended, }, /* CONFIG_LOCKDEP, CONFIG_LOCK_STAT */
1481         { "lock:lock_release",   evsel__process_lock_release,   }, /* CONFIG_LOCKDEP */
1482 };
1483
1484 static const struct evsel_str_handler contention_tracepoints[] = {
1485         { "lock:contention_begin", evsel__process_contention_begin, },
1486         { "lock:contention_end",   evsel__process_contention_end,   },
1487 };
1488
1489 static int process_event_update(struct perf_tool *tool,
1490                                 union perf_event *event,
1491                                 struct evlist **pevlist)
1492 {
1493         int ret;
1494
1495         ret = perf_event__process_event_update(tool, event, pevlist);
1496         if (ret < 0)
1497                 return ret;
1498
1499         /* this can return -EEXIST since we call it for each evsel */
1500         perf_session__set_tracepoints_handlers(session, lock_tracepoints);
1501         perf_session__set_tracepoints_handlers(session, contention_tracepoints);
1502         return 0;
1503 }
1504
1505 typedef int (*tracepoint_handler)(struct evsel *evsel,
1506                                   struct perf_sample *sample);
1507
1508 static int process_sample_event(struct perf_tool *tool __maybe_unused,
1509                                 union perf_event *event,
1510                                 struct perf_sample *sample,
1511                                 struct evsel *evsel,
1512                                 struct machine *machine)
1513 {
1514         int err = 0;
1515         struct thread *thread = machine__findnew_thread(machine, sample->pid,
1516                                                         sample->tid);
1517
1518         if (thread == NULL) {
1519                 pr_debug("problem processing %d event, skipping it.\n",
1520                         event->header.type);
1521                 return -1;
1522         }
1523
1524         if (evsel->handler != NULL) {
1525                 tracepoint_handler f = evsel->handler;
1526                 err = f(evsel, sample);
1527         }
1528
1529         thread__put(thread);
1530
1531         return err;
1532 }
1533
1534 static void combine_result(void)
1535 {
1536         unsigned int i;
1537         struct lock_stat *st;
1538
1539         if (!combine_locks)
1540                 return;
1541
1542         for (i = 0; i < LOCKHASH_SIZE; i++) {
1543                 hlist_for_each_entry(st, &lockhash_table[i], hash_entry) {
1544                         combine_lock_stats(st);
1545                 }
1546         }
1547 }
1548
1549 static void sort_result(void)
1550 {
1551         unsigned int i;
1552         struct lock_stat *st;
1553
1554         for (i = 0; i < LOCKHASH_SIZE; i++) {
1555                 hlist_for_each_entry(st, &lockhash_table[i], hash_entry) {
1556                         insert_to_result(st, compare);
1557                 }
1558         }
1559 }
1560
1561 static const struct {
1562         unsigned int flags;
1563         const char *str;
1564         const char *name;
1565 } lock_type_table[] = {
1566         { 0,                            "semaphore",    "semaphore" },
1567         { LCB_F_SPIN,                   "spinlock",     "spinlock" },
1568         { LCB_F_SPIN | LCB_F_READ,      "rwlock:R",     "rwlock" },
1569         { LCB_F_SPIN | LCB_F_WRITE,     "rwlock:W",     "rwlock" },
1570         { LCB_F_READ,                   "rwsem:R",      "rwsem" },
1571         { LCB_F_WRITE,                  "rwsem:W",      "rwsem" },
1572         { LCB_F_RT,                     "rt-mutex",     "rt-mutex" },
1573         { LCB_F_RT | LCB_F_READ,        "rwlock-rt:R",  "rwlock-rt" },
1574         { LCB_F_RT | LCB_F_WRITE,       "rwlock-rt:W",  "rwlock-rt" },
1575         { LCB_F_PERCPU | LCB_F_READ,    "pcpu-sem:R",   "percpu-rwsem" },
1576         { LCB_F_PERCPU | LCB_F_WRITE,   "pcpu-sem:W",   "percpu-rwsem" },
1577         { LCB_F_MUTEX,                  "mutex",        "mutex" },
1578         { LCB_F_MUTEX | LCB_F_SPIN,     "mutex",        "mutex" },
1579         /* alias for get_type_flag() */
1580         { LCB_F_MUTEX | LCB_F_SPIN,     "mutex-spin",   "mutex" },
1581 };
1582
1583 static const char *get_type_str(unsigned int flags)
1584 {
1585         flags &= LCB_F_MAX_FLAGS - 1;
1586
1587         for (unsigned int i = 0; i < ARRAY_SIZE(lock_type_table); i++) {
1588                 if (lock_type_table[i].flags == flags)
1589                         return lock_type_table[i].str;
1590         }
1591         return "unknown";
1592 }
1593
1594 static const char *get_type_name(unsigned int flags)
1595 {
1596         flags &= LCB_F_MAX_FLAGS - 1;
1597
1598         for (unsigned int i = 0; i < ARRAY_SIZE(lock_type_table); i++) {
1599                 if (lock_type_table[i].flags == flags)
1600                         return lock_type_table[i].name;
1601         }
1602         return "unknown";
1603 }
1604
1605 static unsigned int get_type_flag(const char *str)
1606 {
1607         for (unsigned int i = 0; i < ARRAY_SIZE(lock_type_table); i++) {
1608                 if (!strcmp(lock_type_table[i].name, str))
1609                         return lock_type_table[i].flags;
1610         }
1611         for (unsigned int i = 0; i < ARRAY_SIZE(lock_type_table); i++) {
1612                 if (!strcmp(lock_type_table[i].str, str))
1613                         return lock_type_table[i].flags;
1614         }
1615         return UINT_MAX;
1616 }
1617
1618 static void lock_filter_finish(void)
1619 {
1620         zfree(&filters.types);
1621         filters.nr_types = 0;
1622
1623         zfree(&filters.addrs);
1624         filters.nr_addrs = 0;
1625
1626         for (int i = 0; i < filters.nr_syms; i++)
1627                 free(filters.syms[i]);
1628
1629         zfree(&filters.syms);
1630         filters.nr_syms = 0;
1631 }
1632
1633 static void sort_contention_result(void)
1634 {
1635         sort_result();
1636 }
1637
1638 static void print_header_stdio(void)
1639 {
1640         struct lock_key *key;
1641
1642         list_for_each_entry(key, &lock_keys, list)
1643                 fprintf(lock_output, "%*s ", key->len, key->header);
1644
1645         switch (aggr_mode) {
1646         case LOCK_AGGR_TASK:
1647                 fprintf(lock_output, "  %10s   %s\n\n", "pid",
1648                         show_lock_owner ? "owner" : "comm");
1649                 break;
1650         case LOCK_AGGR_CALLER:
1651                 fprintf(lock_output, "  %10s   %s\n\n", "type", "caller");
1652                 break;
1653         case LOCK_AGGR_ADDR:
1654                 fprintf(lock_output, "  %16s   %s\n\n", "address", "symbol");
1655                 break;
1656         default:
1657                 break;
1658         }
1659 }
1660
1661 static void print_header_csv(const char *sep)
1662 {
1663         struct lock_key *key;
1664
1665         fprintf(lock_output, "# output: ");
1666         list_for_each_entry(key, &lock_keys, list)
1667                 fprintf(lock_output, "%s%s ", key->header, sep);
1668
1669         switch (aggr_mode) {
1670         case LOCK_AGGR_TASK:
1671                 fprintf(lock_output, "%s%s %s\n", "pid", sep,
1672                         show_lock_owner ? "owner" : "comm");
1673                 break;
1674         case LOCK_AGGR_CALLER:
1675                 fprintf(lock_output, "%s%s %s", "type", sep, "caller");
1676                 if (verbose > 0)
1677                         fprintf(lock_output, "%s %s", sep, "stacktrace");
1678                 fprintf(lock_output, "\n");
1679                 break;
1680         case LOCK_AGGR_ADDR:
1681                 fprintf(lock_output, "%s%s %s%s %s\n", "address", sep, "symbol", sep, "type");
1682                 break;
1683         default:
1684                 break;
1685         }
1686 }
1687
1688 static void print_header(void)
1689 {
1690         if (!quiet) {
1691                 if (symbol_conf.field_sep)
1692                         print_header_csv(symbol_conf.field_sep);
1693                 else
1694                         print_header_stdio();
1695         }
1696 }
1697
1698 static void print_lock_stat_stdio(struct lock_contention *con, struct lock_stat *st)
1699 {
1700         struct lock_key *key;
1701         struct thread *t;
1702         int pid;
1703
1704         list_for_each_entry(key, &lock_keys, list) {
1705                 key->print(key, st);
1706                 fprintf(lock_output, " ");
1707         }
1708
1709         switch (aggr_mode) {
1710         case LOCK_AGGR_CALLER:
1711                 fprintf(lock_output, "  %10s   %s\n", get_type_str(st->flags), st->name);
1712                 break;
1713         case LOCK_AGGR_TASK:
1714                 pid = st->addr;
1715                 t = perf_session__findnew(session, pid);
1716                 fprintf(lock_output, "  %10d   %s\n",
1717                         pid, pid == -1 ? "Unknown" : thread__comm_str(t));
1718                 break;
1719         case LOCK_AGGR_ADDR:
1720                 fprintf(lock_output, "  %016llx   %s (%s)\n", (unsigned long long)st->addr,
1721                         st->name, get_type_name(st->flags));
1722                 break;
1723         default:
1724                 break;
1725         }
1726
1727         if (aggr_mode == LOCK_AGGR_CALLER && verbose > 0) {
1728                 struct map *kmap;
1729                 struct symbol *sym;
1730                 char buf[128];
1731                 u64 ip;
1732
1733                 for (int i = 0; i < max_stack_depth; i++) {
1734                         if (!st->callstack || !st->callstack[i])
1735                                 break;
1736
1737                         ip = st->callstack[i];
1738                         sym = machine__find_kernel_symbol(con->machine, ip, &kmap);
1739                         get_symbol_name_offset(kmap, sym, ip, buf, sizeof(buf));
1740                         fprintf(lock_output, "\t\t\t%#lx  %s\n", (unsigned long)ip, buf);
1741                 }
1742         }
1743 }
1744
1745 static void print_lock_stat_csv(struct lock_contention *con, struct lock_stat *st,
1746                                 const char *sep)
1747 {
1748         struct lock_key *key;
1749         struct thread *t;
1750         int pid;
1751
1752         list_for_each_entry(key, &lock_keys, list) {
1753                 key->print(key, st);
1754                 fprintf(lock_output, "%s ", sep);
1755         }
1756
1757         switch (aggr_mode) {
1758         case LOCK_AGGR_CALLER:
1759                 fprintf(lock_output, "%s%s %s", get_type_str(st->flags), sep, st->name);
1760                 if (verbose <= 0)
1761                         fprintf(lock_output, "\n");
1762                 break;
1763         case LOCK_AGGR_TASK:
1764                 pid = st->addr;
1765                 t = perf_session__findnew(session, pid);
1766                 fprintf(lock_output, "%d%s %s\n", pid, sep,
1767                         pid == -1 ? "Unknown" : thread__comm_str(t));
1768                 break;
1769         case LOCK_AGGR_ADDR:
1770                 fprintf(lock_output, "%llx%s %s%s %s\n", (unsigned long long)st->addr, sep,
1771                         st->name, sep, get_type_name(st->flags));
1772                 break;
1773         default:
1774                 break;
1775         }
1776
1777         if (aggr_mode == LOCK_AGGR_CALLER && verbose > 0) {
1778                 struct map *kmap;
1779                 struct symbol *sym;
1780                 char buf[128];
1781                 u64 ip;
1782
1783                 for (int i = 0; i < max_stack_depth; i++) {
1784                         if (!st->callstack || !st->callstack[i])
1785                                 break;
1786
1787                         ip = st->callstack[i];
1788                         sym = machine__find_kernel_symbol(con->machine, ip, &kmap);
1789                         get_symbol_name_offset(kmap, sym, ip, buf, sizeof(buf));
1790                         fprintf(lock_output, "%s %#lx %s", i ? ":" : sep, (unsigned long) ip, buf);
1791                 }
1792                 fprintf(lock_output, "\n");
1793         }
1794 }
1795
1796 static void print_lock_stat(struct lock_contention *con, struct lock_stat *st)
1797 {
1798         if (symbol_conf.field_sep)
1799                 print_lock_stat_csv(con, st, symbol_conf.field_sep);
1800         else
1801                 print_lock_stat_stdio(con, st);
1802 }
1803
1804 static void print_footer_stdio(int total, int bad, struct lock_contention_fails *fails)
1805 {
1806         /* Output for debug, this have to be removed */
1807         int broken = fails->task + fails->stack + fails->time + fails->data;
1808
1809         if (!use_bpf)
1810                 print_bad_events(bad, total);
1811
1812         if (quiet || total == 0 || (broken == 0 && verbose <= 0))
1813                 return;
1814
1815         total += broken;
1816         fprintf(lock_output, "\n=== output for debug ===\n\n");
1817         fprintf(lock_output, "bad: %d, total: %d\n", broken, total);
1818         fprintf(lock_output, "bad rate: %.2f %%\n", 100.0 * broken / total);
1819
1820         fprintf(lock_output, "histogram of failure reasons\n");
1821         fprintf(lock_output, " %10s: %d\n", "task", fails->task);
1822         fprintf(lock_output, " %10s: %d\n", "stack", fails->stack);
1823         fprintf(lock_output, " %10s: %d\n", "time", fails->time);
1824         fprintf(lock_output, " %10s: %d\n", "data", fails->data);
1825 }
1826
1827 static void print_footer_csv(int total, int bad, struct lock_contention_fails *fails,
1828                              const char *sep)
1829 {
1830         /* Output for debug, this have to be removed */
1831         if (use_bpf)
1832                 bad = fails->task + fails->stack + fails->time + fails->data;
1833
1834         if (quiet || total == 0 || (bad == 0 && verbose <= 0))
1835                 return;
1836
1837         total += bad;
1838         fprintf(lock_output, "# debug: total=%d%s bad=%d", total, sep, bad);
1839
1840         if (use_bpf) {
1841                 fprintf(lock_output, "%s bad_%s=%d", sep, "task", fails->task);
1842                 fprintf(lock_output, "%s bad_%s=%d", sep, "stack", fails->stack);
1843                 fprintf(lock_output, "%s bad_%s=%d", sep, "time", fails->time);
1844                 fprintf(lock_output, "%s bad_%s=%d", sep, "data", fails->data);
1845         } else {
1846                 int i;
1847                 const char *name[4] = { "acquire", "acquired", "contended", "release" };
1848
1849                 for (i = 0; i < BROKEN_MAX; i++)
1850                         fprintf(lock_output, "%s bad_%s=%d", sep, name[i], bad_hist[i]);
1851         }
1852         fprintf(lock_output, "\n");
1853 }
1854
1855 static void print_footer(int total, int bad, struct lock_contention_fails *fails)
1856 {
1857         if (symbol_conf.field_sep)
1858                 print_footer_csv(total, bad, fails, symbol_conf.field_sep);
1859         else
1860                 print_footer_stdio(total, bad, fails);
1861 }
1862
1863 static void print_contention_result(struct lock_contention *con)
1864 {
1865         struct lock_stat *st;
1866         int bad, total, printed;
1867
1868         if (!quiet)
1869                 print_header();
1870
1871         bad = total = printed = 0;
1872
1873         while ((st = pop_from_result())) {
1874                 total += use_bpf ? st->nr_contended : 1;
1875                 if (st->broken)
1876                         bad++;
1877
1878                 if (!st->wait_time_total)
1879                         continue;
1880
1881                 print_lock_stat(con, st);
1882
1883                 if (++printed >= print_nr_entries)
1884                         break;
1885         }
1886
1887         if (print_nr_entries) {
1888                 /* update the total/bad stats */
1889                 while ((st = pop_from_result())) {
1890                         total += use_bpf ? st->nr_contended : 1;
1891                         if (st->broken)
1892                                 bad++;
1893                 }
1894         }
1895         /* some entries are collected but hidden by the callstack filter */
1896         total += con->nr_filtered;
1897
1898         print_footer(total, bad, &con->fails);
1899 }
1900
1901 static bool force;
1902
1903 static int __cmd_report(bool display_info)
1904 {
1905         int err = -EINVAL;
1906         struct perf_tool eops = {
1907                 .attr            = perf_event__process_attr,
1908                 .event_update    = process_event_update,
1909                 .sample          = process_sample_event,
1910                 .comm            = perf_event__process_comm,
1911                 .mmap            = perf_event__process_mmap,
1912                 .namespaces      = perf_event__process_namespaces,
1913                 .tracing_data    = perf_event__process_tracing_data,
1914                 .ordered_events  = true,
1915         };
1916         struct perf_data data = {
1917                 .path  = input_name,
1918                 .mode  = PERF_DATA_MODE_READ,
1919                 .force = force,
1920         };
1921
1922         session = perf_session__new(&data, &eops);
1923         if (IS_ERR(session)) {
1924                 pr_err("Initializing perf session failed\n");
1925                 return PTR_ERR(session);
1926         }
1927
1928         symbol_conf.allow_aliases = true;
1929         symbol__init(&session->header.env);
1930
1931         if (!data.is_pipe) {
1932                 if (!perf_session__has_traces(session, "lock record"))
1933                         goto out_delete;
1934
1935                 if (perf_session__set_tracepoints_handlers(session, lock_tracepoints)) {
1936                         pr_err("Initializing perf session tracepoint handlers failed\n");
1937                         goto out_delete;
1938                 }
1939
1940                 if (perf_session__set_tracepoints_handlers(session, contention_tracepoints)) {
1941                         pr_err("Initializing perf session tracepoint handlers failed\n");
1942                         goto out_delete;
1943                 }
1944         }
1945
1946         if (setup_output_field(false, output_fields))
1947                 goto out_delete;
1948
1949         if (select_key(false))
1950                 goto out_delete;
1951
1952         if (show_thread_stats)
1953                 aggr_mode = LOCK_AGGR_TASK;
1954
1955         err = perf_session__process_events(session);
1956         if (err)
1957                 goto out_delete;
1958
1959         setup_pager();
1960         if (display_info) /* used for info subcommand */
1961                 err = dump_info();
1962         else {
1963                 combine_result();
1964                 sort_result();
1965                 print_result();
1966         }
1967
1968 out_delete:
1969         perf_session__delete(session);
1970         return err;
1971 }
1972
1973 static void sighandler(int sig __maybe_unused)
1974 {
1975 }
1976
1977 static int check_lock_contention_options(const struct option *options,
1978                                          const char * const *usage)
1979
1980 {
1981         if (show_thread_stats && show_lock_addrs) {
1982                 pr_err("Cannot use thread and addr mode together\n");
1983                 parse_options_usage(usage, options, "threads", 0);
1984                 parse_options_usage(NULL, options, "lock-addr", 0);
1985                 return -1;
1986         }
1987
1988         if (show_lock_owner && !use_bpf) {
1989                 pr_err("Lock owners are available only with BPF\n");
1990                 parse_options_usage(usage, options, "lock-owner", 0);
1991                 parse_options_usage(NULL, options, "use-bpf", 0);
1992                 return -1;
1993         }
1994
1995         if (show_lock_owner && show_lock_addrs) {
1996                 pr_err("Cannot use owner and addr mode together\n");
1997                 parse_options_usage(usage, options, "lock-owner", 0);
1998                 parse_options_usage(NULL, options, "lock-addr", 0);
1999                 return -1;
2000         }
2001
2002         if (symbol_conf.field_sep) {
2003                 if (strstr(symbol_conf.field_sep, ":") || /* part of type flags */
2004                     strstr(symbol_conf.field_sep, "+") || /* part of caller offset */
2005                     strstr(symbol_conf.field_sep, ".")) { /* can be in a symbol name */
2006                         pr_err("Cannot use the separator that is already used\n");
2007                         parse_options_usage(usage, options, "x", 1);
2008                         return -1;
2009                 }
2010         }
2011
2012         if (show_lock_owner)
2013                 show_thread_stats = true;
2014
2015         return 0;
2016 }
2017
2018 static int __cmd_contention(int argc, const char **argv)
2019 {
2020         int err = -EINVAL;
2021         struct perf_tool eops = {
2022                 .attr            = perf_event__process_attr,
2023                 .event_update    = process_event_update,
2024                 .sample          = process_sample_event,
2025                 .comm            = perf_event__process_comm,
2026                 .mmap            = perf_event__process_mmap,
2027                 .tracing_data    = perf_event__process_tracing_data,
2028                 .ordered_events  = true,
2029         };
2030         struct perf_data data = {
2031                 .path  = input_name,
2032                 .mode  = PERF_DATA_MODE_READ,
2033                 .force = force,
2034         };
2035         struct lock_contention con = {
2036                 .target = &target,
2037                 .map_nr_entries = bpf_map_entries,
2038                 .max_stack = max_stack_depth,
2039                 .stack_skip = stack_skip,
2040                 .filters = &filters,
2041                 .save_callstack = needs_callstack(),
2042                 .owner = show_lock_owner,
2043         };
2044
2045         lockhash_table = calloc(LOCKHASH_SIZE, sizeof(*lockhash_table));
2046         if (!lockhash_table)
2047                 return -ENOMEM;
2048
2049         con.result = &lockhash_table[0];
2050
2051         session = perf_session__new(use_bpf ? NULL : &data, &eops);
2052         if (IS_ERR(session)) {
2053                 pr_err("Initializing perf session failed\n");
2054                 err = PTR_ERR(session);
2055                 goto out_delete;
2056         }
2057
2058         con.machine = &session->machines.host;
2059
2060         con.aggr_mode = aggr_mode = show_thread_stats ? LOCK_AGGR_TASK :
2061                 show_lock_addrs ? LOCK_AGGR_ADDR : LOCK_AGGR_CALLER;
2062
2063         if (con.aggr_mode == LOCK_AGGR_CALLER)
2064                 con.save_callstack = true;
2065
2066         symbol_conf.allow_aliases = true;
2067         symbol__init(&session->header.env);
2068
2069         if (use_bpf) {
2070                 err = target__validate(&target);
2071                 if (err) {
2072                         char errbuf[512];
2073
2074                         target__strerror(&target, err, errbuf, 512);
2075                         pr_err("%s\n", errbuf);
2076                         goto out_delete;
2077                 }
2078
2079                 signal(SIGINT, sighandler);
2080                 signal(SIGCHLD, sighandler);
2081                 signal(SIGTERM, sighandler);
2082
2083                 con.evlist = evlist__new();
2084                 if (con.evlist == NULL) {
2085                         err = -ENOMEM;
2086                         goto out_delete;
2087                 }
2088
2089                 err = evlist__create_maps(con.evlist, &target);
2090                 if (err < 0)
2091                         goto out_delete;
2092
2093                 if (argc) {
2094                         err = evlist__prepare_workload(con.evlist, &target,
2095                                                        argv, false, NULL);
2096                         if (err < 0)
2097                                 goto out_delete;
2098                 }
2099
2100                 if (lock_contention_prepare(&con) < 0) {
2101                         pr_err("lock contention BPF setup failed\n");
2102                         goto out_delete;
2103                 }
2104         } else if (!data.is_pipe) {
2105                 if (!perf_session__has_traces(session, "lock record"))
2106                         goto out_delete;
2107
2108                 if (!evlist__find_evsel_by_str(session->evlist,
2109                                                "lock:contention_begin")) {
2110                         pr_err("lock contention evsel not found\n");
2111                         goto out_delete;
2112                 }
2113
2114                 if (perf_session__set_tracepoints_handlers(session,
2115                                                 contention_tracepoints)) {
2116                         pr_err("Initializing perf session tracepoint handlers failed\n");
2117                         goto out_delete;
2118                 }
2119         }
2120
2121         if (setup_output_field(true, output_fields))
2122                 goto out_delete;
2123
2124         if (select_key(true))
2125                 goto out_delete;
2126
2127         if (symbol_conf.field_sep) {
2128                 int i;
2129                 struct lock_key *keys = contention_keys;
2130
2131                 /* do not align output in CSV format */
2132                 for (i = 0; keys[i].name; i++)
2133                         keys[i].len = 0;
2134         }
2135
2136         if (use_bpf) {
2137                 lock_contention_start();
2138                 if (argc)
2139                         evlist__start_workload(con.evlist);
2140
2141                 /* wait for signal */
2142                 pause();
2143
2144                 lock_contention_stop();
2145                 lock_contention_read(&con);
2146         } else {
2147                 err = perf_session__process_events(session);
2148                 if (err)
2149                         goto out_delete;
2150         }
2151
2152         setup_pager();
2153
2154         sort_contention_result();
2155         print_contention_result(&con);
2156
2157 out_delete:
2158         lock_filter_finish();
2159         evlist__delete(con.evlist);
2160         lock_contention_finish();
2161         perf_session__delete(session);
2162         zfree(&lockhash_table);
2163         return err;
2164 }
2165
2166
2167 static int __cmd_record(int argc, const char **argv)
2168 {
2169         const char *record_args[] = {
2170                 "record", "-R", "-m", "1024", "-c", "1", "--synth", "task",
2171         };
2172         const char *callgraph_args[] = {
2173                 "--call-graph", "fp," __stringify(CONTENTION_STACK_DEPTH),
2174         };
2175         unsigned int rec_argc, i, j, ret;
2176         unsigned int nr_tracepoints;
2177         unsigned int nr_callgraph_args = 0;
2178         const char **rec_argv;
2179         bool has_lock_stat = true;
2180
2181         for (i = 0; i < ARRAY_SIZE(lock_tracepoints); i++) {
2182                 if (!is_valid_tracepoint(lock_tracepoints[i].name)) {
2183                         pr_debug("tracepoint %s is not enabled. "
2184                                  "Are CONFIG_LOCKDEP and CONFIG_LOCK_STAT enabled?\n",
2185                                  lock_tracepoints[i].name);
2186                         has_lock_stat = false;
2187                         break;
2188                 }
2189         }
2190
2191         if (has_lock_stat)
2192                 goto setup_args;
2193
2194         for (i = 0; i < ARRAY_SIZE(contention_tracepoints); i++) {
2195                 if (!is_valid_tracepoint(contention_tracepoints[i].name)) {
2196                         pr_err("tracepoint %s is not enabled.\n",
2197                                contention_tracepoints[i].name);
2198                         return 1;
2199                 }
2200         }
2201
2202         nr_callgraph_args = ARRAY_SIZE(callgraph_args);
2203
2204 setup_args:
2205         rec_argc = ARRAY_SIZE(record_args) + nr_callgraph_args + argc - 1;
2206
2207         if (has_lock_stat)
2208                 nr_tracepoints = ARRAY_SIZE(lock_tracepoints);
2209         else
2210                 nr_tracepoints = ARRAY_SIZE(contention_tracepoints);
2211
2212         /* factor of 2 is for -e in front of each tracepoint */
2213         rec_argc += 2 * nr_tracepoints;
2214
2215         rec_argv = calloc(rec_argc + 1, sizeof(char *));
2216         if (!rec_argv)
2217                 return -ENOMEM;
2218
2219         for (i = 0; i < ARRAY_SIZE(record_args); i++)
2220                 rec_argv[i] = strdup(record_args[i]);
2221
2222         for (j = 0; j < nr_tracepoints; j++) {
2223                 const char *ev_name;
2224
2225                 if (has_lock_stat)
2226                         ev_name = strdup(lock_tracepoints[j].name);
2227                 else
2228                         ev_name = strdup(contention_tracepoints[j].name);
2229
2230                 if (!ev_name)
2231                         return -ENOMEM;
2232
2233                 rec_argv[i++] = "-e";
2234                 rec_argv[i++] = ev_name;
2235         }
2236
2237         for (j = 0; j < nr_callgraph_args; j++, i++)
2238                 rec_argv[i] = callgraph_args[j];
2239
2240         for (j = 1; j < (unsigned int)argc; j++, i++)
2241                 rec_argv[i] = argv[j];
2242
2243         BUG_ON(i != rec_argc);
2244
2245         ret = cmd_record(i, rec_argv);
2246         free(rec_argv);
2247         return ret;
2248 }
2249
2250 static int parse_map_entry(const struct option *opt, const char *str,
2251                             int unset __maybe_unused)
2252 {
2253         unsigned long *len = (unsigned long *)opt->value;
2254         unsigned long val;
2255         char *endptr;
2256
2257         errno = 0;
2258         val = strtoul(str, &endptr, 0);
2259         if (*endptr != '\0' || errno != 0) {
2260                 pr_err("invalid BPF map length: %s\n", str);
2261                 return -1;
2262         }
2263
2264         *len = val;
2265         return 0;
2266 }
2267
2268 static int parse_max_stack(const struct option *opt, const char *str,
2269                            int unset __maybe_unused)
2270 {
2271         unsigned long *len = (unsigned long *)opt->value;
2272         long val;
2273         char *endptr;
2274
2275         errno = 0;
2276         val = strtol(str, &endptr, 0);
2277         if (*endptr != '\0' || errno != 0) {
2278                 pr_err("invalid max stack depth: %s\n", str);
2279                 return -1;
2280         }
2281
2282         if (val < 0 || val > sysctl__max_stack()) {
2283                 pr_err("invalid max stack depth: %ld\n", val);
2284                 return -1;
2285         }
2286
2287         *len = val;
2288         return 0;
2289 }
2290
2291 static bool add_lock_type(unsigned int flags)
2292 {
2293         unsigned int *tmp;
2294
2295         tmp = realloc(filters.types, (filters.nr_types + 1) * sizeof(*filters.types));
2296         if (tmp == NULL)
2297                 return false;
2298
2299         tmp[filters.nr_types++] = flags;
2300         filters.types = tmp;
2301         return true;
2302 }
2303
2304 static int parse_lock_type(const struct option *opt __maybe_unused, const char *str,
2305                            int unset __maybe_unused)
2306 {
2307         char *s, *tmp, *tok;
2308         int ret = 0;
2309
2310         s = strdup(str);
2311         if (s == NULL)
2312                 return -1;
2313
2314         for (tok = strtok_r(s, ", ", &tmp); tok; tok = strtok_r(NULL, ", ", &tmp)) {
2315                 unsigned int flags = get_type_flag(tok);
2316
2317                 if (flags == -1U) {
2318                         pr_err("Unknown lock flags: %s\n", tok);
2319                         ret = -1;
2320                         break;
2321                 }
2322
2323                 if (!add_lock_type(flags)) {
2324                         ret = -1;
2325                         break;
2326                 }
2327         }
2328
2329         free(s);
2330         return ret;
2331 }
2332
2333 static bool add_lock_addr(unsigned long addr)
2334 {
2335         unsigned long *tmp;
2336
2337         tmp = realloc(filters.addrs, (filters.nr_addrs + 1) * sizeof(*filters.addrs));
2338         if (tmp == NULL) {
2339                 pr_err("Memory allocation failure\n");
2340                 return false;
2341         }
2342
2343         tmp[filters.nr_addrs++] = addr;
2344         filters.addrs = tmp;
2345         return true;
2346 }
2347
2348 static bool add_lock_sym(char *name)
2349 {
2350         char **tmp;
2351         char *sym = strdup(name);
2352
2353         if (sym == NULL) {
2354                 pr_err("Memory allocation failure\n");
2355                 return false;
2356         }
2357
2358         tmp = realloc(filters.syms, (filters.nr_syms + 1) * sizeof(*filters.syms));
2359         if (tmp == NULL) {
2360                 pr_err("Memory allocation failure\n");
2361                 free(sym);
2362                 return false;
2363         }
2364
2365         tmp[filters.nr_syms++] = sym;
2366         filters.syms = tmp;
2367         return true;
2368 }
2369
2370 static int parse_lock_addr(const struct option *opt __maybe_unused, const char *str,
2371                            int unset __maybe_unused)
2372 {
2373         char *s, *tmp, *tok;
2374         int ret = 0;
2375         u64 addr;
2376
2377         s = strdup(str);
2378         if (s == NULL)
2379                 return -1;
2380
2381         for (tok = strtok_r(s, ", ", &tmp); tok; tok = strtok_r(NULL, ", ", &tmp)) {
2382                 char *end;
2383
2384                 addr = strtoul(tok, &end, 16);
2385                 if (*end == '\0') {
2386                         if (!add_lock_addr(addr)) {
2387                                 ret = -1;
2388                                 break;
2389                         }
2390                         continue;
2391                 }
2392
2393                 /*
2394                  * At this moment, we don't have kernel symbols.  Save the symbols
2395                  * in a separate list and resolve them to addresses later.
2396                  */
2397                 if (!add_lock_sym(tok)) {
2398                         ret = -1;
2399                         break;
2400                 }
2401         }
2402
2403         free(s);
2404         return ret;
2405 }
2406
2407 static int parse_call_stack(const struct option *opt __maybe_unused, const char *str,
2408                            int unset __maybe_unused)
2409 {
2410         char *s, *tmp, *tok;
2411         int ret = 0;
2412
2413         s = strdup(str);
2414         if (s == NULL)
2415                 return -1;
2416
2417         for (tok = strtok_r(s, ", ", &tmp); tok; tok = strtok_r(NULL, ", ", &tmp)) {
2418                 struct callstack_filter *entry;
2419
2420                 entry = malloc(sizeof(*entry) + strlen(tok) + 1);
2421                 if (entry == NULL) {
2422                         pr_err("Memory allocation failure\n");
2423                         return -1;
2424                 }
2425
2426                 strcpy(entry->name, tok);
2427                 list_add_tail(&entry->list, &callstack_filters);
2428         }
2429
2430         free(s);
2431         return ret;
2432 }
2433
2434 static int parse_output(const struct option *opt __maybe_unused, const char *str,
2435                         int unset __maybe_unused)
2436 {
2437         const char **name = (const char **)opt->value;
2438
2439         if (str == NULL)
2440                 return -1;
2441
2442         lock_output = fopen(str, "w");
2443         if (lock_output == NULL) {
2444                 pr_err("Cannot open %s\n", str);
2445                 return -1;
2446         }
2447
2448         *name = str;
2449         return 0;
2450 }
2451
2452 int cmd_lock(int argc, const char **argv)
2453 {
2454         const struct option lock_options[] = {
2455         OPT_STRING('i', "input", &input_name, "file", "input file name"),
2456         OPT_CALLBACK(0, "output", &output_name, "file", "output file name", parse_output),
2457         OPT_INCR('v', "verbose", &verbose, "be more verbose (show symbol address, etc)"),
2458         OPT_BOOLEAN('D', "dump-raw-trace", &dump_trace, "dump raw trace in ASCII"),
2459         OPT_BOOLEAN('f', "force", &force, "don't complain, do it"),
2460         OPT_STRING(0, "vmlinux", &symbol_conf.vmlinux_name,
2461                    "file", "vmlinux pathname"),
2462         OPT_STRING(0, "kallsyms", &symbol_conf.kallsyms_name,
2463                    "file", "kallsyms pathname"),
2464         OPT_BOOLEAN('q', "quiet", &quiet, "Do not show any warnings or messages"),
2465         OPT_END()
2466         };
2467
2468         const struct option info_options[] = {
2469         OPT_BOOLEAN('t', "threads", &info_threads,
2470                     "dump thread list in perf.data"),
2471         OPT_BOOLEAN('m', "map", &info_map,
2472                     "map of lock instances (address:name table)"),
2473         OPT_PARENT(lock_options)
2474         };
2475
2476         const struct option report_options[] = {
2477         OPT_STRING('k', "key", &sort_key, "acquired",
2478                     "key for sorting (acquired / contended / avg_wait / wait_total / wait_max / wait_min)"),
2479         OPT_STRING('F', "field", &output_fields, NULL,
2480                     "output fields (acquired / contended / avg_wait / wait_total / wait_max / wait_min)"),
2481         /* TODO: type */
2482         OPT_BOOLEAN('c', "combine-locks", &combine_locks,
2483                     "combine locks in the same class"),
2484         OPT_BOOLEAN('t', "threads", &show_thread_stats,
2485                     "show per-thread lock stats"),
2486         OPT_INTEGER('E', "entries", &print_nr_entries, "display this many functions"),
2487         OPT_PARENT(lock_options)
2488         };
2489
2490         struct option contention_options[] = {
2491         OPT_STRING('k', "key", &sort_key, "wait_total",
2492                     "key for sorting (contended / wait_total / wait_max / wait_min / avg_wait)"),
2493         OPT_STRING('F', "field", &output_fields, "contended,wait_total,wait_max,avg_wait",
2494                     "output fields (contended / wait_total / wait_max / wait_min / avg_wait)"),
2495         OPT_BOOLEAN('t', "threads", &show_thread_stats,
2496                     "show per-thread lock stats"),
2497         OPT_BOOLEAN('b', "use-bpf", &use_bpf, "use BPF program to collect lock contention stats"),
2498         OPT_BOOLEAN('a', "all-cpus", &target.system_wide,
2499                     "System-wide collection from all CPUs"),
2500         OPT_STRING('C', "cpu", &target.cpu_list, "cpu",
2501                     "List of cpus to monitor"),
2502         OPT_STRING('p', "pid", &target.pid, "pid",
2503                    "Trace on existing process id"),
2504         OPT_STRING(0, "tid", &target.tid, "tid",
2505                    "Trace on existing thread id (exclusive to --pid)"),
2506         OPT_CALLBACK('M', "map-nr-entries", &bpf_map_entries, "num",
2507                      "Max number of BPF map entries", parse_map_entry),
2508         OPT_CALLBACK(0, "max-stack", &max_stack_depth, "num",
2509                      "Set the maximum stack depth when collecting lopck contention, "
2510                      "Default: " __stringify(CONTENTION_STACK_DEPTH), parse_max_stack),
2511         OPT_INTEGER(0, "stack-skip", &stack_skip,
2512                     "Set the number of stack depth to skip when finding a lock caller, "
2513                     "Default: " __stringify(CONTENTION_STACK_SKIP)),
2514         OPT_INTEGER('E', "entries", &print_nr_entries, "display this many functions"),
2515         OPT_BOOLEAN('l', "lock-addr", &show_lock_addrs, "show lock stats by address"),
2516         OPT_CALLBACK('Y', "type-filter", NULL, "FLAGS",
2517                      "Filter specific type of locks", parse_lock_type),
2518         OPT_CALLBACK('L', "lock-filter", NULL, "ADDRS/NAMES",
2519                      "Filter specific address/symbol of locks", parse_lock_addr),
2520         OPT_CALLBACK('S', "callstack-filter", NULL, "NAMES",
2521                      "Filter specific function in the callstack", parse_call_stack),
2522         OPT_BOOLEAN('o', "lock-owner", &show_lock_owner, "show lock owners instead of waiters"),
2523         OPT_STRING_NOEMPTY('x', "field-separator", &symbol_conf.field_sep, "separator",
2524                    "print result in CSV format with custom separator"),
2525         OPT_PARENT(lock_options)
2526         };
2527
2528         const char * const info_usage[] = {
2529                 "perf lock info [<options>]",
2530                 NULL
2531         };
2532         const char *const lock_subcommands[] = { "record", "report", "script",
2533                                                  "info", "contention", NULL };
2534         const char *lock_usage[] = {
2535                 NULL,
2536                 NULL
2537         };
2538         const char * const report_usage[] = {
2539                 "perf lock report [<options>]",
2540                 NULL
2541         };
2542         const char * const contention_usage[] = {
2543                 "perf lock contention [<options>]",
2544                 NULL
2545         };
2546         unsigned int i;
2547         int rc = 0;
2548
2549         lockhash_table = calloc(LOCKHASH_SIZE, sizeof(*lockhash_table));
2550         if (!lockhash_table)
2551                 return -ENOMEM;
2552
2553         for (i = 0; i < LOCKHASH_SIZE; i++)
2554                 INIT_HLIST_HEAD(lockhash_table + i);
2555
2556         lock_output = stderr;
2557         argc = parse_options_subcommand(argc, argv, lock_options, lock_subcommands,
2558                                         lock_usage, PARSE_OPT_STOP_AT_NON_OPTION);
2559         if (!argc)
2560                 usage_with_options(lock_usage, lock_options);
2561
2562         if (strlen(argv[0]) > 2 && strstarts("record", argv[0])) {
2563                 return __cmd_record(argc, argv);
2564         } else if (strlen(argv[0]) > 2 && strstarts("report", argv[0])) {
2565                 trace_handler = &report_lock_ops;
2566                 if (argc) {
2567                         argc = parse_options(argc, argv,
2568                                              report_options, report_usage, 0);
2569                         if (argc)
2570                                 usage_with_options(report_usage, report_options);
2571                 }
2572                 rc = __cmd_report(false);
2573         } else if (!strcmp(argv[0], "script")) {
2574                 /* Aliased to 'perf script' */
2575                 rc = cmd_script(argc, argv);
2576         } else if (!strcmp(argv[0], "info")) {
2577                 if (argc) {
2578                         argc = parse_options(argc, argv,
2579                                              info_options, info_usage, 0);
2580                         if (argc)
2581                                 usage_with_options(info_usage, info_options);
2582                 }
2583                 /* recycling report_lock_ops */
2584                 trace_handler = &report_lock_ops;
2585                 rc = __cmd_report(true);
2586         } else if (strlen(argv[0]) > 2 && strstarts("contention", argv[0])) {
2587                 trace_handler = &contention_lock_ops;
2588                 sort_key = "wait_total";
2589                 output_fields = "contended,wait_total,wait_max,avg_wait";
2590
2591 #ifndef HAVE_BPF_SKEL
2592                 set_option_nobuild(contention_options, 'b', "use-bpf",
2593                                    "no BUILD_BPF_SKEL=1", false);
2594 #endif
2595                 if (argc) {
2596                         argc = parse_options(argc, argv, contention_options,
2597                                              contention_usage, 0);
2598                 }
2599
2600                 if (check_lock_contention_options(contention_options,
2601                                                   contention_usage) < 0)
2602                         return -1;
2603
2604                 rc = __cmd_contention(argc, argv);
2605         } else {
2606                 usage_with_options(lock_usage, lock_options);
2607         }
2608
2609         zfree(&lockhash_table);
2610         return rc;
2611 }